common-pile/pubmed · Datasets at Hugging Face

id string	text string	source string	added string	created string	metadata dict
PMC1166566	Background ========== Multidrug resistance (MDR) phenotype of malignant cells is the major problem in the chemotherapy of neoplasia. P-glycoprotein (Pgp) activity is recognised to be one of the major mechanisms responsible for MDR. Pgp transports many structurally diverse compounds across the cell membrane and confers the MDR phenotype in tumor cells \[[@B1]\]. A number of signaling pathways participate in the regulation of MDR1gene expression and the activity of its product, Pgp \[[@B2]\]. Some of these signaling pathways could participate in coordinated regulation of MDR1/Pgp activity, cell proliferation and cell differentiation. It was shown that retinoic acid (RA) can modulate MDR1gene expression \[[@B3]-[@B5]\]. Retinoids are known to be involved into the regulation of the cell growth, differentiation and apoptosis. In the last decade retinoids became implicated into the treatment of leukemia and some solid tumors \[[@B6]\]. This approach changed the focus of the haematological diseases treatment from the cytotoxicity of the anti-cancer drugs to the reversal of arrested maturation of leukemic cells. Retinoids act via two families of receptors (RARs -- RARα, RARβ, RARγ) and RXRs (RXRα, RXRβ, RXRγ). There is the evidence that RARα is the crucial receptor mediating the biological effects during retinoid signaling in some cells \[[@B7]\]. Cell differentiation caused by the stable overexpression of receptor RARα was shown to result in constitutive over expression of MDR1gene in some cultured cells of solid tumors \[[@B4]\]. However the interconnections between RA/RARα regulated differentiation of leukemic cells and regulation of MDR1/Pgp activity remains unclear. In some leukemic cells RA did not influence MDR1and/or Pgp activity, while in the others it either augmented or reduced MDR1/Pgp expression \[[@B5],[@B8]\]. The aim of this study is to investigate if effects of all-trans-retinoic acid (ATRA) on MDR1/Pgp activity in leukemic cells are connected with RARα expression and with the leukemic cell differentiation. We isolated sublines of cultured leukemic cells characterized by the stable RARα overexpression and investigated the constitutive and ATRA induced MDR1/Pgp activity in these cells. Our data show that various RARα transformed leukemic cell lines acquired more differentiated phenotype. Constitutive level of MDR1gene expression increased in one of RARα overexpressing cell sublines. RARα overexpression did not influence Pgp functional activity while Pgp activity induced by ATRA was elevated in all infectants studied. This shows that the main effect of RARα in the cells studied is its influence on the induced functional activity of Pgp. Methods ======= Cell lines and culture ---------------------- Lines of cultured leukemic cells used in the study: H9 cells (acute human T-cell leukemia) \[[@B9]\], KG-1 cell line (cells of acute myelogenous leukemia) \[[@B10]\], K562 cell line (cells obtained from the patient in blast crisis of chronic myeloid leukemia) \[[@B11]\], NB4 (acute promyelocytic leukemia) \[[@B12]\]. Cells were grown in RPMI-1640 medium supplemented with 10% fetal calf serum (Gibco, USA), 2 mM L-glutamine, 50 μg/ml gentamycin at 37°C in a fully humidified atmosphere of 95% air and 5% CO~2~. All the derived cell lines described in this paper were obtained by retroviral infection and selection with the appropriate antibiotic. ATRA (all-trans-retinoic acid, Sigma, USA) was added to the culture medium at seeding or 24 hours after seeding (see Legends to Figures). Expression vector and retrovilal infection ------------------------------------------ The PA317/LRARSN retroviral vector-producing cell line was used. All the procedure was described earlier \[[@B4]\]. In brief, the vector used contains a cDNA fragment harbouring the complete coding sequence of the RARα gene driven by the Moloney murine leukemia virus long-terminal repeat as well as the SV40 early promoter-driving neomycin phosphotransferase gene (neo) as a selectable marker \[[@B13]\]. The cells (4 × 10^5^per 25-cm^2^flask) were seeded 24 h before infection. Conditioned medium from a retrovirus-producing cell line was filtered through a 0.45-μm membrane (Millipore, USA), diluted 1:1 with medium, containing 1% serum and 8 μg /ml Polybrene and added to the cells for 24 h at 37°C, 5% CO~2~. Further selection were carried out by culturing cells in medium supplemented with 400 μg/ml G418 (Gibco, USA) for at least 21 days. The medium was changed twice a week. The pool of G418-resistant cells was resuspended in culture medium and progressively expanded. Assay of cell growth, apoptosis and differentiation --------------------------------------------------- Cells were seeded into 24-well plates (1 × 10^4^cell per well) and the cell number was counted at days 1, 3, 5 and 8 after seeding. The apoptosis in the populations of the parental and RARαinfected cell lines was performed using the standard procedure \[[@B14]\]. Cells were collected 24 h after seeding, washed with PBS, and fixed in 70% ethanol overnight at 4°C. Fixed cells were suspended in citric buffer and stained with propidium iodide (5 mcg/ml) in PBS for 1 hour at 4°C. DNA content was subsequently measured by FACScan (Becton Dickinson, USA). The immunophenotype of the cells was evaluated as previously described \[[@B15]\]. Surface expression of the following antigens was determined: CD3, CD5, CD7, CD8, CD11b, CD13, CD15, CD33, CD34, HAE3 and HAE9. In brief, cells were incubated with phycoerythrin-labelled monoclonal mouse antibodies for 20 min at 4°C (Becton, Dickinson), washed with RPMI 1640 medium and analyzed with a flow cytometer (Becton Dickinson). Analysis of rhodamine 123 (Rh123) efflux by the cells ----------------------------------------------------- The technique used in the study was described in \[[@B16]\]. Cells were loaded with 5 μg/ml Rh123 (Sigma) for 10 min at 37°C, washed twice with cold PBS, pH 7.2, and incubated for 30 min in dye-free medium at 37°C. After the completion of incubation, cell were washed twice with cold PBS. Cell fluorescence was measured on a flow cytometer FACScan (Becton Dickinson, USA). Each measurement counted 5000 events. Non-viable cells were gated out of the analysis on the basis of side scatter. RNA isolation and reverse transcriptase polymerase chain reaction (RT-PCR) analysis of RARα and MDR1 genes expression --------------------------------------------------------------------------------------------------------------------- The cells were dissolved in TRI reagent (Sigma, USA). Total RNA was isolated as described in the manufacturer\'s manual. For analysis, aliquots of isolated RNA were denatured with formamide and subjected to electrophoresis in 1.8% agarose gels. The samples with clearly visualized 18S and 28S RNA bands were used for further procedures. First-strand cDNA was synthesized using reverse transcriptase M-MuLV (MBI Fermentas, Russia) with 4 μg RNA as a template, 2.5 ng random hexamers, 0.25 mM of each deoxynucleotide triphosphate (SibEnzyme, Russia), dithiothreitol, 4 Units of RNAase inhibitor (MBI Fermentas, Russia) and 100 Units of M-MuLV RT. The reaction was performed at 42°C for 50 min, and 1/60 volume of reaction mixture was used for amplification. PCR was done in a total volume of 25 μl using the thermocycler \"Tercyc\" (DNA-technology, Russia). The PCR mixture consisted of (NH~4~)~2~SO~4~-containing PCR buffer (\"MBI Fermentas\"), 0.160 mM dNTPs mix (\"MBI Fermentas\"), 2 mM MgCl2, 20 pmoles of each specific primer and 0.8 Unit of Taq-polymerase (\"MBI Fermentas\"). PCR was done as follows: 94°C for 2 seconds, Tm (different for each gene) for 10 seconds, 72°C for 5 seconds. Semi-quantitative PCR analysis of RARaand MDR1genes expression were performed using oligomers amplifying a 333 bp and 167 bp products, respectively. Specific gene primers used for RT-PCR are given in Table [1](#T1){ref-type="table"}. The amounts of template cDNAs were normalized by PCR amplification of β2-microglobulin cDNA (internal control). The optimal numbers of PCR cycles were 24 for the b2-microglobulin, 26 for RARα-specific product, 33 for MDR1(for all cells lines except KG1 and KG1/RAR, for these cells the numbers of PCR cycles MDR1-specific product were 26). These numbers of cycles yielded clearly detectable PCR products within an exponential range. PCR products were amplified in separate tubes, resolved by electrophoresis in 2% agarose gel, stained with ethidium bromide and visualized in UV light. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Specific gene primers used for RT-PCR ::: Gene Product size Primer Sequence -------- -------------- -------- ------------------------------- RARα 333 bp F 5\'-GTCTTTGCCTTCGCCAACCAG-3\' R 5\'-GCCCTCTGAGTTCTCCAACA-3\' MDR1 127 bp F 5\'-CCCATCATTGCAATAGCAGG-3\' R 5\'-GTTCAAACTTCTGCTCCTGA-3\' β2m 114 bp F 5\'-ACCCCCACTGAAAAAGATGA-3\' R 5\'-ATCTTCAAACCTCCATGATG-3\' ::: Results ======= Influence of RARα gene overexpression on cell differentiation, proliferation and spontaneous apoptosis ------------------------------------------------------------------------------------------------------ RARαgene was introduced into the cultured leukemic cells of diverse types of differentiation as described in Methods. The sublines of H9, KG-1, K562 and NB4 cells characterized by the capability to grow in the medium supplemented with G418 were isolated (H9/RAR, KG-1/RAR, K562/RAR and NB4/RAR). Semi-quantitative RT-PCR revealed more pronounced expression of RARαmRNA in all transfected cell lines in comparison with the wild type cells (Fig. [1](#F1){ref-type="fig"}). ATRA (5 μM applied for 48 h) increased RARαmRNA in some RARαtransformed cells (H9/RAR, KG-1/RAR, K562/RAR) to a greater extent than in parental cells (Fig. [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *Expression of RARαmRNA in parental and RARαtransfected cell lines. RT PCRM. k- -- water. ATRA (5 μM) was added to cell cultures 24 h after seeding for 48 h. Then the cells were collected and processed as specified in \"Methods\" (RNA isolation and reverse transcriptase polymerase chain reaction (RT-PCR) analysis of RARαgene expression). This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-1) ::: The investigation of the differentiation status of these cells shows that all RARαtransfected sublines differ from the parental cell populations (Fig. [2](#F2){ref-type="fig"}). RARαtransfected H9 culture contains more cell variants expressing CD5 and CD8 antigens than parental cell line (Fig. [2A](#F2){ref-type="fig"}). Thus the number of cells with antigens of later lymphoid differentiation markers increased in RARαoverexpressing H9 cells. There is phenotypic evidence of granulocytic differentiation in KG-1/RAR cell subline as indicated by a reduction in CD13 expression and the increase in the expression of CD11b antigen in comparison with parental cells (Fig. [2B](#F2){ref-type="fig"}). In KG-1/RAR cell population the portion of CD34 cells decreased and the portion of CD33 cells increased (Fig. [2B](#F2){ref-type="fig"}). This also testifies to increased differentiation of these RARαoverexpressing cells. In K562/RAR population the number of the cells of erythroid differentiation (expressing HAE9 and HAE3 antigens) is larger than in K562 population (Fig. [2C](#F2){ref-type="fig"}). Hemoglobin synthesis is increased in K562/RAR culture more than 5-fold in comparison with parental cells (not shown). In NB4/RAR population the number of the cells of myelogenous differentiation (expressing CD11b and CD15 antigens) is larger than in NB4 population (Fig. [2D](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Comparison of antigen expression by the parental and RARαinfected cell lines. Cells were incubated for 30 min at 4°C in the presence of an appropriate monoclonal antibody. After three washes with PBS, cells were incubated for 30 min at 4°C with goat antimouse IgG labeled with phycoerythrin and then analyzed in flow cytometer (Becton Dickinson). ::: ![](1475-2867-5-15-2) ::: The percentage of cells undergoing spontaneous apoptosis increased 2-3-fold in all RARαtransfected cell populations (Fig. [3](#F3){ref-type="fig"}). This could be connected with more differentiated phenotype of RARαtransformed cells. It seems that in the population of H9/RAR cells the increased number of apoptotic cells could be at least in part connected with increased expression of CD95 (Fas/APO1): in this RARαtransformed subline CD95 increased almost 10-fold in comparison with parental cell population (from 2,6% in H9 to 21,4% in H9/RAR culture). However in KG-1 and K562 cell populations the number of CD95 expressing cell did not increase after RARαtransformation. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Spontaneous apoptosis in the populations of the parental and RARαinfected cell lines. Propidium iodide flow cytometry detection of dead cells was performed using the standard procedure. Cells were collected 24 h after seeding, washed with PBS, and fixed in 70% ethanol overnight at 4°C. Fixed cells were suspended in citric buffer and stained with 5 mcg/ml propidium iodide in PBS for 1 hour at 4°C. DNA content was subsequently measured by FACScan (Becton Dickinson, USA). All cells with sub-G~0~DNA content were regarded as dead cells. This figure is representative of 3 separate experiments. ::: ![](1475-2867-5-15-3) ::: As Fig. [4](#F4){ref-type="fig"} shows, RARαtransfected KG-1, K562 and NB4 cells proliferated more slowly than parental cells. However H9/RAR cells did not demonstrate slower proliferation rate. Thus, more differentiated status of the RARαtransformed cell populations was not necessary connected with the decrease in the proliferation rates. All RARαtransformed cells seem to be more sensitive than wild type cells to inhibitory action of ATRA on cell proliferation (Fig. [5](#F5){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Proliferation rates of parental and RARαinfected cells. Cells were seeded into 24-well plates (1 × 10^4^cell per well) and the cell number was counted at days 1,3,5,8 after seeding. This figure is representative of 3 separate experiments. ::: ![](1475-2867-5-15-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Influence of retinoic acid (ATRA, 5 μM) on the proliferation of parental and RARαtransfected cells. Cells were seeded into 24-well plates (1 × 10^4^cell per well), ATRA was added at seeding and the cell number was counted at days 1, 3, 5 and 8 after seeding. This figure is representative of 3 separate experiments. ::: ![](1475-2867-5-15-5) ::: Influence of RARα overexpression on MDR1 gene activity ------------------------------------------------------ We studied intrinsic and ATRA induced expression of MDR1gene in all cell lines by semi-quantitative RT-PCR technique. The basal levels of MDR1mRNA varied in different wild type cells: in H9 and NB4 cells constitutive MDR1gene expression was not revealed, in K562 wild type cells some MDR1mRNA was found, in KG-1 cells the quantity of MDR1mRNA was large (Fig. [6](#F6){ref-type="fig"}). It is noteworthy that the optimal number of PCR cycles were 33 for MDR1-specific product in all cells while for studies of KG1 and KG1/RAR cells we used 26 PCR cycles. In RARαtransfected H9 cells the constitutive expression of MDR1gene slightly increased, while in KG-1/RAR and NB4/RAR cells the constitutive MDR1mRNA quantity was not elevated in comparison with the wild type cells, it seems to be even slightly decreased in K562/RAR (Fig. [6](#F6){ref-type="fig"}). Thus the alterations of the basal level of MDR1expression in RARαtransformed cells seem to vary in different cell lines. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Intrinsic and retinoic acid induced expression of MDR1gene in parental and RARαtransfected cells. k- -- water. ATRA (5 μM) was added to cell cultures 24 h after seeding for 48 h. Then the cells were collected and processed as specified in \"Methods\" (RNA isolation and reverse transcriptase polymerase chain reaction (RT-PCR) analysis of MDR1gene expression). The optimal numbers of PCR cycles were 33 for MDR1-specific product for all cells lines except KG1 and KG1/RAR, for these cells the numbers of PCR cycles for MDR1-specific product were 26. This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-6) ::: ATRA (5 μM applied for 48 h) increased MDR1gene expression in all examined cell lines either in parental or RARαtransfected cells (Fig. [6](#F6){ref-type="fig"}). In H9/RAR cells effect of ATRA on MDR1expression was significantly greater in comparison with parental cells. In other ATRA treated RARαtransformed cell sublines MDR1expression was undistinguishable from ATRA treated parental cells (Fig [6](#F6){ref-type="fig"}). Influence of RARα gene transformation on Rh123efflux by the cells ----------------------------------------------------------------- The retention of Rh123 by the cells is considered as a test for Pgp functional activity \[[@B16],[@B17]\]. Rh123 efflux from the cells was increased in K562/RAR cells in comparison with the parental cell population (Fig. [7B](#F7){ref-type="fig"}, Table [2](#T2){ref-type="table"}). In H9 and KG-1 RARαtransformed cells there were alterations in the Rh123 retention (Fig. [7A](#F7){ref-type="fig"} and [7C](#F7){ref-type="fig"}): in the populations of H9/RAR and KG-1/RAR cultures the fraction of more dull cells decreased in comparison with parental cultures (mean fluorescence intensity of the cell sublines studied are given in the table [2](#T2){ref-type="table"}). This shows that Pgp activity was not elevated in these RARαtransformed cell populations and suggests that there is some decrease in Pgp functional activity. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Evaluation of Rh123 efflux from the parental and cells RARαtransfected cells. Cells were loaded with 5 μg/ml Rh123 for 10 min at 37°C, than washed twice with cold PBS, and incubated for 30 min in dye-free medium at 37°C. After the completion of incubation, cell were washed twice with cold PBS. Cell fluorescence was measured on a flow cytometer FACScan (Becton Dikinson, USA). Each measurement counted 5000 events. Non-viable cells were gated out of the analysis on the basis of side scatter. This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-7) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Influence of RARαtransformation on the intrinsic and induced expression of MDR1gene ::: MDR1expression Rh123 efflux (mean fluorescence intensity) ------------ ------------------ -------------------------------------------- ---------- ---------- H9* \- \+ 450 457 H9/RAR \+ ++ 506 416 K562 \+ ++ 446 370 K562/RAR \+ ++ 403 333 KG-1 ++ +++ 65,7 38,3 KG-1/RAR ++ +++ 90,7 20,2 NB4 \- ++ n.d. n.d. NB4/RAR \- ++ n.d. n.d. ::: There was increase in the portion of Rh123 dull cells after ATRA treatment both in K562 and K562/RAR cell populations (mean fluorescence intensity of both populations decreased approximately on 17%) (Fig. [8C, D](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). ATRA induced Rh123 efflux from H9/RAR cells, while in H9 parental population this drug had no effect (Fig. [8A, B](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). In KG-1/RAR cells ATRA induced very prominent increase in the number of Rh123 dull cells (more than 70% decrease of mean fluorescence intensity), while in the parental cell population ATRA decreased mean fluorescence intensity to a lesser extent (Fig. [8E, F](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). ::: {#F8 .fig} Figure 8 ::: {.caption} ###### *Influence of retinoic acid (ATRA) on Rh123 efflux from the parental and cells RARαtransfected cells. ATRA (5 μM) was added to cell cultures 24 h after seeding for 48 h. Than cells were loaded with 5 μg/ml Rh123 for 10 min at 37°C, than washed twice with cold PBS, and incubated for 30 min in dye-free medium at 37°C. After the completion of incubation, cell were washed twice with cold PBS. Cell fluorescence was measured on a flow cytometer FACScan (Becton Dikinson, USA). Each measurement counted 5000 events. Non-viable cells were gated out of the analysis on the basis of side scatter. This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-8) ::: Discussion ========== The treatment of leukemia with retinoids is aimed on the induction of leukemic cells differentiation. The question is: are there interconnections between RA/RARα regulated differentiation of leukemic cells and MDR1/Pgp activity? In this study we have isolated more differentiated variants of the cultured leukemic cells by the introduction into the cells of RARαgene encoding one of RA receptors. All RARαtransformed leukemic cell populations were characterized by the higher RARαgene expression in comparison with the parental cells. All RARαtransformed leukemic cell populations became more differentiated. This was demonstrated by the studies of the differentiation markers, by the increase in the number of cells dying by spontaneous apoptosis and by the decrease of the proliferation rates of most RARαtransfected cell sublines. Thus, RARαoverexpression could result in the increase of the differentiation of various leukemic cell populations. We compared MDR1gene expression and Pgp functional activity tested by Rh123 retention in parental and RARαtransformed cells. The results are summarized in the Table [2](#T2){ref-type="table"}. Increased constitutive (uninduced) expression of MDR1gene was found in one of four cell lines after RARαtransformation (H9/RAR, Table [2](#T2){ref-type="table"}, Fig. [6](#F6){ref-type="fig"}). In the previous experiments with melanoma and hepatoblastoma human cells we have shown that constitutive expression of MDR1gene was increased after RARαtransfection in both RARαtransformed cell sublines \[[@B4]\]. Thus interconnections between regulation of the basal MDR1and RARαactivities could exist both in the cells of solid tumors and in the leukemic cells. Our data suggest, that in the cell populations of solid tumors RARαoverexpression could be accompanied by constitutive MDR1over-expression more often than in the cells of hematopoietic malignancies. Our study did not reveal the occurrence of the functional Pgp in leukemic cells studied after RARαtransformation. In H9/RAR cells elevation of the constitutive MDR1expression did not lead to the increase in Rh123 efflux (Fig. [7A](#F7){ref-type="fig"}, Table [2](#T2){ref-type="table"}). Some studies also have described discrepancies between Pgp (protein) or MDR1mRNA expression and Pgp function in leukemic cells \[[@B18],[@B19]\]. These discrepancies could occur for a variety of reasons. Anyway, our data show that increase in the differentiation of leukemic cell populations induced by RARαoverexpression did not result in the elevation of constitutive Pgp functional activity. In our previous study we found that RARαoverexpression did not change Pgp functional activity in two RARαtransformed sublines of human cells (melanoma and hepatoblastoma) but did change it in the rat cells \[[@B4]\]. It seems that exogeneous RARαin the cells of human malignancies does not influence basal Pgp functional activity. In KG-1/RAR characterized by the increased differentiaion (Table [1](#T1){ref-type="table"}) we had not found increase in the constitutive MDR1expression and Pgp functional activity decreased (Fig. [7C](#F7){ref-type="fig"}, Table [2](#T2){ref-type="table"}). It is known that blood stem cells and early progenitors expressing CD34 antigen also express high levels of functionally active Pgp \[[@B20]\]. Maturation of these cells is accompanied by the decrease in Pgp expression and even more rapid decrease in Pgp functional activity \[[@B21]\]. It may be suggested that alterations of Pgp function in KG-1/RAR are connected with the differentiation of these cells. The situation with Pgp functional activity induced by ATRA in the cells studied differs from the situation with constitutive activity of this protein. In all three RARαtransfected cells ATRA had induced Pgp fuctional activity (Fig. [8](#F8){ref-type="fig"}. Table [2](#T2){ref-type="table"}). Moreover, in two RARαtransformed sublines (H9/RAR and KG-1/RAR) ATRA activated Pgp, while in the parental cells it had either no effect (H9) or activated Pgp to a lesser extent (KG-1) (Table [2](#T2){ref-type="table"}). These data suggest that RARαparticipate in the control of induced, but not in constitutive Pgp functional activity in leukemic cells. The regulation of MDR1gene transcription and Pgp functional activities are the complex processes \[[@B1],[@B2]\]. The studies of these processes are underway. Our data show that RARα*gene overexpression could influence the induced Pgp functional activity in leukemic cells, i.e. could participate in the occurrence of multidrug resistance in the populations of these malignant cells. It seems that this influence could depend on the cell context. Acknowledgements ================ This work was supported by grants 04-04-48613a and 02-04-48200 from the Russian Foundation for Basic Research.	PubMed Central	2024-06-05T03:55:58.152910	2005-5-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166566/", "journal": "Cancer Cell Int. 2005 May 24; 5:15", "authors": [ { "first": "Tatjana P", "last": "Stromskaya" }, { "first": "Ekaterina Y", "last": "Rybalkina" }, { "first": "Tatjana N", "last": "Zabotina" }, { "first": "Alexander A", "last": "Shishkin" }, { "first": "Alla A", "last": "Stavrovskaya" } ] }
PMC1166567	Introduction ============ The prevalence of type 2 diabetes in adults was estimated at 2.8% worldwide in 2000, and predicted to increase to 4.4% by 2030 \[[@B1]\]. Patients with type 2 diabetes have a risk of cardiovascular disease approximately two- to four-times greater than that in the non-diabetic population \[[@B2]\]. Furthermore, their prognosis is worse; in a Swedish study the 5-year mortality rate after myocardial infarction was 55% for patients with diabetes compared with 30% in patients without diabetes (p \< 0.001), and the re-infarction rates were 42% and 25%, respectively (p \< 0.001) \[[@B3]\]. More recent data reflecting the outcome of new evidence-based interventions in acute myocardial infarction demonstrate that the difference between diabetic and non-diabetic subjects is still present, showing a 1-year mortality in males of 22.3% versus 13.0% in males and 26.1% versus 14.4% in females \[[@B4]\]. In the US National Health and Nutrition Examination Survey (NHANES I), the age-adjusted mortality rate in diabetic patients over 9 years of follow-up was double that in non-diabetic patients, and cardiovascular disease accounted for 75% of the excess mortality in men and 57% in women \[[@B5]\]. The elevated cardiovascular risk in patients with type 2 diabetes is primarily attributed to the clustering of atherogenic risk factors, including dyslipidaemia, hypertension, abdominal obesity, left ventricular hypertrophy, and impaired fibrinolysis \[[@B6]\]. For patients with diabetes, European Diabetes Policy Group guidelines published in 1999 and European guidelines for coronary heart disease prevention published in 1998, both recommend that low-density lipoprotein cholesterol (LDL-C) levels should be \<3.0 mmol/L \[[@B7],[@B8]\]. More recent (2003) European guidelines on cardiovascular disease prevention also recognise type 2 diabetes as a risk factor, and recommend more stringent LDL-C reductions to\<2.5 mmol/L \[[@B9]\]. In addition, this goal is recommended by both the American Diabetes Association and the US National Cholesterol Education Program Adult Treatment Panel III \[[@B10],[@B11]\]. Statins are recognized as first-line therapy for cholesterol lowering \[[@B7],[@B11]\], and have been proven to reduce cardiovascular morbidity and mortality in large outcome trials in various populations \[[@B12]-[@B17]\]. The benefits of statin therapy extend to patients with diabetes, as shown by subgroup analyses of patients with diabetes in several of the major statin outcome studies, including the Cholesterol And Recurrent Events (CARE) study, the Scandinavian Simvastatin Survival Study (4S), the Long-Term Intervention with Pravastatin in Ischemic Disease (LIPID) study and the Heart Protection Study \[[@B18]-[@B21]\]. The Collaborative Atorvastatin Diabetes Study (CARDS) recently investigated the effects of lipid lowering with statin therapy specifically in patients with type 2 diabetes \[[@B22]\]. The primary endpoint, time to the first occurrence of acute coronary events, coronary revascularisation or stroke, was significantly reduced by 37% in patients treated with atorvastatin 10 mg compared with placebo (p = 0.001). In addition, LDL-C levels were significantly reduced by 40% in the atorvastatin 10 mg group compared with the placebo group (p \< 0.001) \[[@B22]\]. Despite the proven benefits of statin therapy, studies suggest that many patients with diabetes fail to achieve lipid goals in clinical practice \[[@B23],[@B24]\]. Statins differ in their lipid-modifying efficacy and their ability to enable patients to achieve lipid goals \[[@B25],[@B26]\]. Trials in patients with hypercholesterolaemia have shown that rosuvastatin is more effective than atorvastatin at reducing LDL-C and achieving US and/or European LDL-C goals over treatment periods ranging from 6 to 52 weeks \[[@B25]-[@B30]\]. The URANUS (Use of Rosuvastatin versus Atorvastatin iN type 2 diabetes mellitUS) study is a direct comparison of the effects of rosuvastatin with atorvastatin on LDL-C, other plasma lipids and LDL-C goal achievement in patients with diabetes. This study was designed to reflect the optimal treatment of diabetic dyslipidaemia in the clinical setting, in that statin dose was titrated upwards from the recommended start dose to that required to enable patients to achieve LDL-C goal. Methods ======= Patients -------- Patients (male or female) aged 18 years or more were eligible for the study if they had a history of type 2 diabetes for at least 3 months; were being treated with diet, oral antidiabetic medication, insulin or a combination of these treatments; and had fasting LDL-C of ≥ 3.3 mmol/L and triglycerides (TG) of\<6.0 mmol/L at enrolment. Exclusion criteria included: type 1 diabetes; uncontrolled type 2 diabetes; uncontrolled hypothyroidism or hypertension; nephrotic syndrome or severe renal failure; active liver disease or hepatic dysfunction; active arterial disease (e.g., unstable angina, myocardial infarction, transient ischaemic attack, cerebrovascular accident, coronary artery bypass grafting or percutaneous transluminal coronary angioplasty within 3 months before beginning the study); serum creatinine kinase (CK) levels \>3 × the upper limit of normal (ULN); body mass index \>35 kg/m^2^; and known hypersensitivity to statins. All patients gave written informed consent, and the study was conducted in accordance with the Declaration of Helsinki. Study design ------------ The trial was a randomised, double-blind, parallel-group study (4522SE/0001) conducted in 51 centres in Sweden. The study design is summarised in figure [1](#F1){ref-type="fig"}. Patients meeting the inclusion and exclusion criteria at enrolment entered a 6-week dietary run-in period and all lipid-lowering therapy was withdrawn at least 14 days before the end of this period. Patients with fasting LDL-C ≥ 3.3 mmol/L were then randomised to a starting dose of either rosuvastatin 10 mg or atorvastatin 10 mg for 4 weeks. This was followed by a 12-week period of dose titration, making a total of 16 weeks of treatment. Patients who had not reached the 1998 European guideline goal of LDL-C\<3.0 mmol/L \[[@B7]\] after 4 weeks were titrated up by doubling the statin dose (rosuvastatin 20 mg or atorvastatin 20 mg). Further dose titrations (to rosuvastatin 40 mg, or atorvastatin 40 mg or 80 mg) were performed at 8 weeks and 12 weeks for patients who were still not at their LDL-C goal. Patients whose LDL-C level was below the goal at 4 weeks continued on the initial dose of study medication; if their LDL-C level exceeded the goal at subsequent visits, study medication was up-titrated. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Study design. ^a^In patients who had not reached the European goal of LDL-C\<3.0 mmol/L after 4 weeks, the statin dose was doubled at each visit, up to a maximum of RSV 40 mg and ATV 80 mg RSV: Rosuvastatin, ATV: Atorvastatin, LDL-C: Low-density lipoprotein cholesterol ::: ![](1475-2840-4-7-1) ::: Concomitant treatment with erythromycin, azole antimycotic agents, vitamin K antagonists, immunosuppressive agents, glitazones or systemic steroids was not permitted during the study. If insulin treatment became necessary, or if the patient took lipid-lowering medication (other than study medication), the patient was discontinued from the trial. Assessments ----------- ### Efficacy The primary endpoint was the percentage change in LDL-C from baseline (randomisation) to 16 weeks. Secondary endpoints included: percentage change in LDL-C from baseline to 4 weeks; percentage of patients achieving the 1998 European LDL-C goal at 4 and 16 weeks; percentage change in total cholesterol (TC), TG, high-density lipoprotein cholesterol (HDL-C), the LDL-C/HDL-C ratio, the non-HDL-C/HDL-C ratio, the TC/HDL-C ratio, apolipoprotein (apo) B, apo A-I and the apo B/apo A-I ratio from baseline to 4 and 16 weeks; and the number of titration steps at 16 weeks. A tertiary endpoint was the difference in overnight urinary albumin excretion (UAE) from baseline to 16 weeks. All patients were instructed to fast for 8 hours prior to giving blood samples. LDL-C levels were measured using a direct method with enzymatic colorimetry (Genzyme Diagnostics, Genzyme Corporation, Cambridge, MA, USA). All analyses were conducted at a central laboratory. ### Safety Adverse events spontaneously reported by the patients, elicited in response to an open question or revealed by observation, were recorded at each visit. Laboratory safety variables included: blood haemoglobin, platelet count, leucocyte count, serum aspartate aminotransferase (ASAT), serum alanine aminotransferase (ALAT), serum alkaline phosphatase, serum bilirubin, CK, serum creatinine and glycated haemoglobin (HbA~1c~). All analyses were performed at a central laboratory. Statistical methods ------------------- In order to have a 90% chance of detecting a difference between the two treatment arms of 6% in the primary endpoint (percentage change in LDL-C from baseline to 16 weeks), 212 patients per arm were required to complete the study. The primary efficacy endpoint was determined using analysis of covariance with change in LDL-C as response variable, treatment and centre as factors, and baseline LDL-C as covariate. Percentage change in other lipid variables from baseline to 4 weeks and 16 weeks, and percentage change in UAE, were analysed in the same way as the primary endpoint. The proportion of patients reaching LDL-C goal was analysed using a Mantel-Haenszel test stratified by centre. All tests were two-sided with a significance level of 5%. In addition, 95% confidence intervals were calculated for the treatment differences in all efficacy variables except the percentage of patients reaching LDL-C goal. All efficacy variables were analysed in the intention-to-treat population (observed data). Analysis of the primary endpoint was also carried out using the last observation carried forward approach. All enrolled patients were evaluated for safety. Results ======= Demographics ------------ A total of 469 patients were randomised, and efficacy data were obtained from 465 patients, 232 in the rosuvastatin group and 233 in the atorvastatin group (figure [2](#F2){ref-type="fig"}). The two groups were well matched at baseline, and demographic details are shown in table [1](#T1){ref-type="table"}. Previous statin treatment was received by 31 patients (13%) in the rosuvastatin group and 39 patients (17%) in the atorvastatin group. Eleven patients in the rosuvastatin group and 12 in the atorvastatin group discontinued during the randomised treatment period (figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Study populations ::: ![](1475-2840-4-7-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient demographics of ITT population ::: Rosuvastatin (n = 232) Atorvastatin (n = 233) ---------------------------------- ---------------------------- ---------------------------- Gender, male/female (%) 128/104 (55.2/44.8) 136/97 (58.4/41.6) Race, white (%) 229 (98.7) 229 (98.3) Mean age, years (SD) 63.5 (8.8) 65.0 (8.6) Mean weight, kg (SD) 84.8 (14.3) 82.5 (13.5) Mean BMI, kg/m^2^(SD) 29.0 (3.6) 28.4 (3.6) Mean baseline LDL-C, mmol/L (SD) 4.6 (0.85) 4.6 (0.82)^a^ Mean baseline HDL-C, mmol/L (SD) 1.2 (0.27) 1.2 (0.27)^a^ Mean baseline TG, mmol/L (SD) 2.0 (1.0) 2.0 (0.93)^a^ ITT:Intention to treat, SD:Standard deviation, BMI:Body mass index, LDL-C:Low-density lipoprotein cholesterol, HDL-C:High-density lipoprotein cholesterol, TG:Triglycerides ^a^n = 232 ::: Efficacy -------- At the end of the titration-to-goal period, rosuvastatin was significantly more effective than atorvastatin on the primary efficacy measure, reducing LDL-C by 52% compared with 46% in the atorvastatin group (p \< 0.0001) (table [2](#T2){ref-type="table"}). In line with its greater efficacy for LDL-C reduction, significantly more rosuvastatin-treated patients reached the 1998 European LDL-C goal after 16 weeks than atorvastatin-treated patients (94% vs 88%, p \< 0.05). Furthermore, more patients achieved the goal on the starting dose of rosuvastatin than atorvastatin (75% vs 54%) (figure [3](#F3){ref-type="fig"}). The greater ability of rosuvastatin to lower LDL-C was also reflected by the number of dose titrations required by each treatment group; a total of 75 titration steps were required by rosuvastatin-treated patients compared with 155 titrations in the atorvastatin group. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Percentage change from baseline in lipid variables at 16 weeks (ITT population). Doses were titrated from week 4 to week 16 in patients who had not reached the 1998 European LDL-C goal (\< 3.0 mmol/L) ::: Variable Least-squares mean percentage change from baseline to 16 weeks Difference (95% CI) p-value ----------------------- -------------------------------------------------------------------- ------------------------- ------------------- ----------- LDL-C -52.3 -45.5 -6.7 (-8.8, -4.7) \< 0.0001 TC -35.4 -31.3 -4.1 (-5.8, -2.4) \< 0.0001 HDL-C 5.3 4.0 1.3 (-1.3, 3.8) NS TG -21.2 -21.1 -0.1 (-5.6, 5.3) NS Non-HDL-C -45.0 -39.6 -5.5 (-7.4, -3.5) \< 0.0001 LDL-C/HDL-C ratio -54.1 -47.0 -7.1 (-9.3, -4.9) \< 0.0001 Non-HDL-C/HDL-C ratio -47.1 -40.9 -6.2 (-8.6, -3.9) \< 0.0001 TC/HDL-C ratio -38.0 -33.1 -5.0 (-6.9, -3.0) \< 0.0001 Apo B -45.2 -40.1 -5.1 (-7.2, -3.1) \< 0.0001 Apo A-I 2.6 -0.2 2.8 (1.0, 4.6) 0.0024 Apo B/apo A-I ratio -46.3 -39.6 -6.7 (-8.9, -4.6) \< 0.0001 ITT:Intention to treat, CI:Confidence interval, LDL-C:Low-density lipoprotein cholesterol, TC:Total cholesterol, HDL-C:High-density lipoprotein cholesterol, TG:Triglycerides, Apo:Apolipoprotein, NS:Not statistically significant ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Cumulative percentage of patients to 1998 European LDL-C goal of\<3.0 mmol/L \[7\] by dose at 16 weeks. \p \< 0.05 rosuvastatin 10--40 mg vs atorvastatin 10--80 mg RSV: Rosuvastatin, ATV: Atorvastatin, LDL-C: Low-density lipoprotein cholesterol ::: ![](1475-2840-4-7-3) ::: During the 4-week fixed-dose period, significantly more patients on rosuvastatin 10 mg had reached the 1998 European LDL-C goal compared with patients on atorvastatin 10 mg (figure [4](#F4){ref-type="fig"}). When data from the fixed-dose period were re-analysed to the more stringent 2003 European LDL-C goal of \<2.5 mmol/L, 65% of patients receiving rosuvastatin 10 mg achieved goal compared with 33% of patients receiving atorvastatin 10 mg (p \< 0.001; figure [4](#F4){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Percentage of patients to 1998 and 2003 European LDL-C goals \[7,9\] at 4 weeks. \p \< 0.001 vs atorvastatin LDL-C: Low-density lipoprotein cholesterol ::: ![](1475-2840-4-7-4) ::: Rosuvastatin also reduced TC, non-HDL-C, LDL-C/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio significantly (p \< 0.0001) more than atorvastatin after 4 weeks of treatment (table [3](#T3){ref-type="table"}). Both treatments increased HDL-C and decreased TG from baseline to 4 weeks, but there were no statistically significant differences between the groups (table [3](#T3){ref-type="table"}). In addition, rosuvastatin significantly reduced levels of apo B and the apo B/apo A-I ratio, and increased apo A-I levels compared with atorvastatin (p ≤ 0.05) (table [3](#T3){ref-type="table"}). Similar significant effects on TC, non-HDL-C, apolipoproteins and lipid ratios were observed at 16 weeks (table [2](#T2){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Percentage change from baseline in lipid variables at 4 weeks (ITT population) ::: Variable Least-squares mean percentage change from baseline to 4 weeks Difference (95% CI) p-value ----------------------- ------------------------------------------------------------------- ------------------------- -------------------- ----------- LDL-C -47.6 -38.5^a^ -9.1 (-11.4, -6.7) \< 0.0001 TC -33.6 -27.9 -5.7 (-7.4, -4.0) \< 0.0001 HDL-C 4.4 2.6 1.8 (-0.5, 4.0) NS TG -19.2 -15.5 -3.7 (-9.5, 2.2) NS Non-HDL-C -42.6 -35.0 -7.6 (-9.6, -5.7) \< 0.0001 LDL-C/HDL-C ratio -49.3 -39.5 -9.8 (-12.2, -7.4) \< 0.0001 Non-HDL-C/HDL-C ratio -44.4 -35.9 -8.5 (-10.8, -6.2) \< 0.0001 TC/HDL-C ratio -35.8 -29.0 -6.7 (-8.7, -4.8) \< 0.0001 Apo B -42.9 -35.3 -7.6 (-9.7, -5.6) \< 0.0001 Apo A-I 2.6 0.8 1.8 (0, 3.5) 0.05 Apo B/apo A-I ratio -43.9 -35.4 -8.5 (-10.6, -6.4) \< 0.0001 ITT:Intention to treat, CI:Confidence interval, LDL-C:Low-density lipoprotein cholesterol, TC:Total cholesterol, HDL-C:High-density lipoprotein cholesterol, TG:Triglycerides, Apo:Apolipoprotein, NS:Not statistically significant ^a^n = 232 ::: There was no statistically significant difference in UAE rate from baseline to study end, or between the treatment groups, including those patients with baseline microalbuminuria (UAE \>20 μg/min). Safety ------ Both treatments were well tolerated, with overall incidences of adverse events being similar between the treatment groups (51% with rosuvastatin, 53% with atorvastatin). A total of 10 patients experienced serious adverse events (two in the rosuvastatin group, eight in the atorvastatin group), none of which were considered by the investigator to be related to study treatment. Ten patients discontinued because of adverse events, three in the rosuvastatin group and seven in the atorvastatin group. There were no cases of myopathy. Myalgia was reported by 3.4% of the patients in the study; none of the cases were associated with a clinically important elevation in CK (\>5 × ULN). Indeed, there were no clinically important elevations in CK in either group throughout the study and changes in CK were not related to dose of study medication or duration of treatment. The most frequent adverse events overall were nasopharyngitis, myalgia, and inadequately controlled diabetes mellitus (table [4](#T4){ref-type="table"}). There were no clinically relevant changes in ALAT or ASAT (\>3 × ULN). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Adverse events occurring in ≥ 3% of patients in any treatment group ::: Number (%) of patients with adverse event ------------------------------------------- ----------------------------------------------- ---------------------------- Adverse event Rosuvastatin (n = 233) Atorvastatin (n = 236) Nasopharyngitis 23 (9.9) 19 (8.1) Myalgia 13 (5.6) 7 (3.0) Inadequately controlled diabetes mellitus 14 (6.0) 11 (4.6) Constipation 9 (3.9) 6 (2.5) Headache 6 (2.6) 7 (3.0) Urinary tract infection 9 (3.9) 2 (0.9) Arthralgia 1 (0.4) 9 (3.8) ::: Discussion ========== Rosuvastatin provided significantly greater LDL-C reductions than atorvastatin, both at the initial dose of 10 mg and also when titrated over the dose range of 10--40 mg for rosuvastatin and 10--80 mg for atorvastatin. In addition, a significantly higher percentage of patients treated with rosuvastatin achieved the 1998 European LDL-C goal (\<3.0 mmol/L), both with the starting dose of 10 mg and after the period of dose titration. These results are consistent with the findings of studies in patients with hypercholesterolaemia that compared rosuvastatin with atorvastatin over 6 weeks \[[@B25]\], 8 weeks \[[@B31]\], 12 weeks \[[@B27],[@B28]\], and 52-week dose titration \[[@B28]\]. Patients with type 2 diabetes commonly have a highly atherogenic lipid profile including elevated LDL-C, increased TG and low HDL-C, which is associated with a high risk of developing cardiovascular disease \[[@B32],[@B33]\]. Statins are recognized as first-line therapy for cholesterol lowering, and their benefits have been shown to extend to patients with diabetes \[[@B18]-[@B22]\]. The present study was designed to reflect the treatment of diabetic dyslipidaemia in the clinical setting, in that statin treatment of patients was titrated upwards from the recommended starting dose to that required to achieve the 1998 European target of LDL-C \<3.0 mmol/L. The population of the present study was compared with patients with type 2 diabetes in the Swedish National Diabetes Registry and was found to be consistent in terms of baseline characteristics such as age, HbA~1c~, body mass index, percentage of smokers, blood pressure, and antidiabetic medication \[[@B34]\]. New European guidelines published in 2003 recommend a more stringent target (LDL-C \<2.5 mmol/L) \[[@B9]\] than that used when the present study was planned. Further analysis of the 4-week (fixed-dose) LDL-C data indicated that rosuvastatin 10 mg treated significantly more patients to the new 2003 European goal of \<2.5 mmol/L than atorvastatin 10 mg. As expected, the absolute percentages of patients achieving the more stringent 2003 goal were lower than the absolute percentages achieving the 1998 goal at 4 weeks, but the greater efficacy of rosuvastatin 10 mg compared with atorvastatin 10 mg remained the same. As more clinical trial evidence becomes available regarding the positive effects of intensive lipid lowering among patients with diabetes, it is likely that even more stringent LDL-C goals will be recommended. Indeed, National Cholesterol Education Program Adult Treatment Panel III recommendations were recently reviewed and a target of LDL-C \<70 mg/dL (1.8 mmol/L) was suggested as a therapeutic option for individuals considered to be at very high risk including those with both type 2 diabetes and established cardiovascular disease \[[@B35]\]. The availability of an agent that enables a large number of patients to achieve LDL-C goal at the starting dose is important given that many patients receiving lipid-lowering therapy fail to existing attain lipid targets due to a lack of dose titration and the use of less effective agents \[[@B36],[@B37]\]. A recent observational study, designed to reflect dyslipidaemia treatment in the clinical setting, evaluated the number of hyperlipidaemic patients with coronary heart disease or diabetes who achieved LDL-C goal with their initial statin dose and whether patients were dose titrated \[[@B38]\]. Less than half (48%) achieved LDL-C \<2.6 mmol/L with their initial dose and, of those who did not achieve goal, only 45% had their dose titrated. Dose titration increases costs and the need for follow-up, which, while necessary, can be time-consuming and inconvenient. The ability of rosuvastatin to enable greater proportions of patients to achieve LDL-C goal, with reduced requirement for dose titration is highly advantageous. The benefits of reaching treatment goals have been demonstrated in the Steno-2 study \[[@B39]\], in which patients with type 2 diabetes were randomised to receive conventional treatment or intensive multifactorial intervention to strict treatment goals (including TC\<4.5 mmol/L). LDL-C levels were reduced by 47% in those receiving intensive therapy, and the risk of both cardiovascular and microvascular events was reduced by approximately 50% compared with conventional treatment \[[@B39]\]. Rosuvastatin was also more effective than atorvastatin in reducing a range of other lipid variables, including TC, non-HDL-C, LDL-C/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio. Reductions in TC and/or TC/HDL-C are particularly relevant given that the Systemic Coronary Risk Evaluation (SCORE) system advocated in the new European guidelines uses these variables to estimate total risk \[[@B9]\]. In the present study, both treatments produced similar increases in HDL-C, which were lower than those observed previously. In the Measuring Effective Reductions in Cholesterol Using Rosuvastatin therapY (MERCURY I) study involving 3,161 patients with hypercholesterolaemia, 8 weeks\' treatment with rosuvastatin 10 mg increased HDL-C by 9.2% and this was significantly greater than atorvastatin 10 mg (6.8%) and atorvastatin 20 mg (5.7%) (p \< 0.01) \[[@B31]\]. In the current study, average baseline HDL-C values were higher than would be expected in the diabetic population, which could partly explain the relatively small increases in HDL-C compared with other studies \[[@B31],[@B40]\]. In the present study, the main apolipoprotein in HDL-C, apo A-I, was significantly increased by rosuvastatin compared with atorvastatin (p \< 0.05). In addition, significant reductions were also observed in apo B and apo B/apo A-I (p \< 0.0001). Changes in apolipoprotein levels may have important implications in the reduction of cardiovascular risk, since results from the Apolipoprotein-related Mortality Risk (AMORIS) study indicate that apo B, apo A-I and apo B/apo A-I are powerful predictors of cardiac events \[[@B41]\]. Taken together with the other changes to lipid variables, the findings of the present study indicate that a less atherogenic lipid profile was achieved with rosuvastatin. Treatment of diabetic dyslipidaemia may also reduce the incidence of microvascular disease including nephropathy \[[@B33]\]. Statins have been shown to have beneficial effects in diabetic nephropathy by reducing the rate of UAE \[[@B42],[@B43]\]. In the present study, statin treatment did not significantly alter the rate of UAE; however, this may reflect the fact that the treatment period was relatively short. Previously, a reduced UAE rate with statin therapy has been observed after at least 6 months\' treatment \[[@B42],[@B43]\]. Both treatments were similarly well tolerated, with no unexpected safety concerns, and tolerability was similar to that previously observed in non-diabetic patient populations \[[@B40],[@B44]\]. In conclusion, rosuvastatin was significantly more effective at reducing LDL-C and achieving European LDL-C goals both during the fixed-dose period and following dose titration than atorvastatin in patients with type 2 diabetes. Competing interests =================== This study was supported by AstraZeneca, Södertälje, Sweden, which provided study expenses and covered the article processing charge. Annica Siewert-Delle is an employee of AstraZeneca. Authors\' contributions ======================= Christian Berne and Annica Siewert-Delle participated in the design and coordination of the study and prepared the manuscript. Both authors read and approved the final manuscript. Acknowledgements ================ We gratefully acknowledge the investigators, their co-investigators and study co-ordinators, and the patients who participated in this trial. In addition to the authors of this publication, the following investigators participated in this trial: R Ekesbo, Dalby; L Fröberg, Höganäs; N Henningsen, Malmö; G Ilestam, Malmö; A Fahlbom, Malmö; J Nielsen, Skivarp; S Nordström, Knislinge; N Nörgaard, Höganäs; G Vatnaland, Ängelholm; A-C Knutsson, Ängelholm; J Wiuff, Oskarshamn; C Sjödin, Växjö; T Svensson, Växjö; G Tygesen, Lagan; O Benéus, Partille; L Benéus, Partille; L Lingetun, Falkenberg; I Wallin, Mölndal; J Alvång, Trollhättan; E Angesjö, Borås; P Hellke, Göteborg; L Nord, Stenungsund; C Andersson, Stenungsund; U Thorslund, Göteborg; M Öhberg, Skene; P E:son Jennersjö, Linköping; O Borgholst, Kungsör; B Cöster, Kristinehamn; G Holmberg, Karlstad; W Meyer, Köping; B Finger, Köping; R Baylis, Köping; E Sundequist-Stockhaus, Karlstad; P Sundin, Örebro; K Vetterskog, Västerås; J-E Andersson, Tyresö; G Widerström, Tyresö; A Bröijersen, Stockholm; B Thorsson, Stockholm; B Eriksson, Gustavsberg; H Noppa, Spånga; A Häggmark, Skärholmen; L Held, Skärholmen; H Salminen, Bagarmossen AB; P Nordström, Bagarmossen AB; R Zlatewa-Cuenca, Stockholm; L Hjelmaeus, Stockholm; E Hammarström, Stockholm; K Brismar, Stockholm; I Bäckström, Märsta; S Hellerstedt, Kungsbacka; L Håkansson, Norrtälje; A Lindh, Åkersberga; R-M Brinkeborn, Uppsala; T Lundmark, Kilafors; U Sundström, Kilafors; R Malmström, Sandviken; E Tönnesen, Uppsala; E Edén, Uppsala; K Åresund, Gävle; O Berglund, Umeå; K Henriksson, Krokom; L Lönneborg, Sundsbruk; M Mullaart, Sundsbruk; P Malm, Östersund; G Strömberg, Trehörningsjö; T Lindén, Västra Frölunda.	PubMed Central	2024-06-05T03:55:58.155173	2005-6-3	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166567/", "journal": "Cardiovasc Diabetol. 2005 Jun 3; 4:7", "authors": [ { "first": "Christian", "last": "Berne" }, { "first": "Annica", "last": "Siewert-Delle" } ] }
PMC1166568	The unbearable lightness of being \[a cardiac imaging specialist\] ================================================================== A Renaissance of cardiac imaging occurred in the 1980s \[[@B1]\]. New technologies allowed the non-invasive description of cardiac function, perfusion, and metabolism in a polychrome, three-dimensional, overwhelming fashion. Almost unlimited resources were devoted to patient care in the economic framework of the affluent society. At the beginning of the 1990s, The Renaissance made its transition into the splendid decadence of the Baroque. The increasing technological burden in clinical cardiology paradoxically did not bring a parallel increase in the quality of care but rather an increase in cost. The economic climate had changed; the illusion of unlimited economic resources had come to an end \[[@B2]\]. Keeping in mind that each test represents a cost, often a risk, and always a diagnostic hypothesis, we can agree that every unnecessary and unjustifiable test is one test too many. Small individual costs, risks, and wastes multiplied by billions of examinations per year represent an important population \[[@B3]\], society \[[@B4]\] and environmental \[[@B5]\] burden. Unfortunately, the appropriateness of cardiac imaging is usually extra-ordinarily low and there is little awareness among patients and physicians of the elementary physical basis, differential costs, radiological doses, and long term risks of different imaging modalities \[[@B6]\]. It is also well known that -- in the words of Bernard Lown -- \"technology in medicine is frequently untested scientifically, often applied without data relating to cost benefit, and driven by market forces rather than by patient needs.\" Bernard Lown, 2004 \[[@B7]\]. The cost of cardiac imaging =========================== \"Ten years ago, medical imaging wasn\'t even in the radar screen for most health insurers. In 2004, it\' s one of the highest cost items in a health plan\'s medical budget, and also one of the fastest growing\". (Atlantic info service newsletter, 2004) \[[@B8]\]. As an example, in U.S. during the year 2002, 7.8 million cardiac perfusion scans were performed, with a growth of 40% in the last 3 years \[[@B9]\]. Still in U.S., about 10 million CT scans were done in 1993, and about 60 millions in 2001 \[[@B10]\]. Booz Allen Hamilton projects that spending on diagnostic imaging could grow 28 % by 2005, with utilization growing by 9% per year. No doubt that any assessment of costs of medical imaging should also include the often unquestionable downstream benefits, leading to a reduction of overall costs due to more disabling disease prevented. For instance, ultrasound screening for abdominal aneurysms can reduce the risk of death by more than 50 % in men aged 65--74. When projected to 10 years, that screening costs 13.000 dollars per life year gained -- a highly favourable result in terms of cost-effectiveness \[[@B11]\]. However, it is beyond question that the explosion in imaging costs is also driven by a number of factors which not always improve the cost-effectiveness. High patient demand for the newest diagnostic tests, physicians eager to use the most effective technologies, and possibly even some providers who may boost utilization to help pay off investments in high-tech equipment. All too often physicians may not have adequate controls on the number of high cost imaging tests they are ordering. For a resting cardiac imaging test, being the average cost (not charges) of an echocardiogram equal to 1 (as a cost comparator), the cost of a CT is 3.1x, of a SPECT 3.27x, of a Cardiovascular Magnetic Resonance imaging 5.51x, of a PET 14.03x, and of a right and left heart catheterization 19.96x (Fig [1](#F1){ref-type="fig"}) \[[@B12]\]. For stress cardiac imaging, compared with the treadmill exercise test equal to 1 \[as a cost comparator\], the cost of stress echocardiography is 2.1x and of a stress SPECT scintigraphy is 5.7x \[[@B13]\]. Obviously, the evaluation of costs of medical imaging should also imply the entire clinical context (pre-test likelihood of disease, basic risk profile, target of stress testing, socio-economic characteristics of the health care system, etc.) to be taken into account. Some measures of precision and/or dispersion of each modality as compared to echocardiography may provide a more comprehensive information in the specific health care milieu where these tests are actually performed. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The average costs of CMR and other common cardiac imaging procedures when compared 2 D echocardiography (modified from Pennell, ref. 12). ::: ![](1476-7120-3-13-1) ::: The individual and social risks of imaging ========================================== Current protection standard and practices are based on the premise that any ionising radiation dose, no matter how small, can result in detrimental health effects \[[@B14]\]. These include long-term development of cancer and genetic damage \[[@B15]\]. For the purposes of radiation protection, the dose-response curve for radiation-induced cancer is assumed to be linear at low doses, with no minimum threshold \[[@B16]\]: (Fig. [2](#F2){ref-type="fig"}). The dose of 50 chest X rays (for example, a lung scintigraphy) corresponds to an extra-risk of cancer of about 1 in 20,000 exposed patients. The dose of 500 chest x rays (such as technetium sestamibi scan) corresponds to an extra-risk of about 1 in 2000 exposed patients. The dose of 1 000 chest x rays (associated with a Thallium scan) corresponds to an extra risk of cancer of about 1 in 1 000 exposed patients \[[@B17]\]. The radiation dose and risk associated with some common imaging examinations are expressed in Table [1](#T1){ref-type="table"} as equivalent dose of natural yearly background radiation, extra-risk of fatal cancer in the lifetime and lost life expectancy per exam \[[@B18]\]. Presented data refer to the best available estimates from the radiological Commission of Radiation Protection and conform to their suggested standards for communicating risk to patients. These estimations are a benchmark for the physicians and are incorporated in the European Commission guidelines for medical imaging. These small individual risks multiplied by billion examinations become significant population risk. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Presentation of cancer risk and radiation dose (in multiples of dose from a simple chest x rays) for some common radiological and nuclear medicine examinations (Modified from Picano E, ref. 37) ::: ![](1476-7120-3-13-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Radiation doses and estimated cancer risk from common radiological examinations and isotope scans ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Type of test Effective radiation dose (mSv) Equivalent period of natural background radiation Lifetime additional risk of cancer/examination Lost life expectancy Equivalent n. of chest x-rays --------------------- ------------------------------------ ------------------------------------------------------- ---------------------------------------------------- -------------------------- ----------------------------------- Chest radiograph 0.01 A few days Negligible risk 2 minutes 1 Skull radiograph 0.1 A few weeks Minimal risk (1 in 100,000 to 1 in1,000,000) 20 minutes 5 Lung isotope scan 1 A few months to a year Very low risk (1 in 10,000 to 1 in 100,000) 3 hrs 50 Cardiac gated study 10 A few years\ Low risk\ 2 days 500 (4 years) (1 in 2,000) Thallium scan 20 (8 years) (1 in 1,000) 4 days 1000 -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: Use of radiation for medical examinations and test is the largest manmade source of radiation exposure. The medical sources of radiation were about one fifth of the natural radiation in 1987 \[[@B18]\], close to one-half in 1993 \[[@B19]\], and almost 100% of natural radiation in 1997 \[[@B20]\], and the use of procedures with a high load of radiation continues to grow steadily \[[@B21]\]. This impressive amount of ionizing friendly fire translates into a significant population risk. Lifetime risk of developing cancer attributable to diagnostic X-rays is 0.6--3.2% in developed countries \[[@B22]\]. The numbers are striking, but underestimate the biological population burden of medical radiation for three reasons. First, they refer to the radiological volume of 10 years ago -- substantially lower than the current radiological volume. Second, they do not consider the practice of nuclear medicine, which adds a further 10 % to the global radiation burden. Third, in addition to the risk of cancer, one should consider the burden of teratogenesis. The risk of radiation -- induced damage passed onto the offspring is estimated to represent a fifth of the risk of fatal cancer \[[@B23]\]. Unawareness in imaging ====================== Several recent studies clearly prove that not only general practitioners but also cardiologists, orthopaedics and even radiologists and nuclear physicians usually ignore the dose and the risk of what they do: and the more they do, the more they tend to ignore \[[@B24]-[@B28]\]. Radiologists working in an academic US environment frequently underestimate of 100 to 500 times the dose of a common CT and 97 % of UK doctors underestimate of sixteen times the dose of a common CT chest scan. In 1 case out of 10, doctors believe that magnetic resonance employs ionising radiations, and 1 case out of 20 that ultrasound employs ionising radiation \[[@B25],[@B27]\]. The majority of doctors, and even of radiologists, is not aware of the oncogenic risk of common, high dose radiological examinations \[[@B26],[@B28]\]. The vast majority of cardiologists underestimates of 200 up to 1000 times the dose of a stress cardiac perfusion scintigraphy \[[@B28]\]: Figure [3](#F3){ref-type="fig"}. This unawareness generates inappropriateness, which is an endemic and pervasive disease in the world of cardiac imaging. We need more prudence, wisdom and responsibility in indicating and performing (cardiac) imaging tests. Prudence and responsibility should obviously be especially high for more expensive tests, greater for those exposing the patient to significant ionizing radiation \[[@B29]\], and greatest in special subset particularly vulnerable to the damaging effects of ionizing radiation, such as women in reproductive age \[[@B30]\] and children \[[@B31]-[@B33]\]. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### What cardiologists know about dose of a test they prescribe and/ or perform daily. 2 out of 3 physicians underestimate of 100 to 500 times the dose of a cardiac scintigraphy. The dose is equivalent to 500 chest x rays with technetium sestamibi, and to more than 1000 chest x rays with thallium scan (modified from Correia et al, ref 28) ::: ![](1476-7120-3-13-3) ::: From benefit to risk-benefit ============================ According to the International Commission of Radiological Protection \"Medical exposure is the only category in which large reductions in average dose are possible, and it is therefore highly desirable to reduce applications of medical radiation which are of no benefit to the patients and to minimise useless radiation in the course of medical examinations\" \[[@B34]\] In other words it is desirable to adopt a radiation sparing strategy not only for the physician and the patients but also for the society and the environment. Nuclear medicine and X-ray procedures are intended, as stated by the United Nations Annex on Medical Exposures, \"to provide doctors with diagnostic information and in principle conducted with the lowest practicable levels of patient dose to meet clinical objectives\" \[[@B5]\]. The clinical counterpart of this concept is that it is not enough that a test is marginally \"better\" than the other to justify its use: the extra-value should be proportional to the extra-cost and to the extra-risk. Both the physician and the patient should be well aware of the different individual risks and social costs posed by different diagnostic options. As recently stated by guidelines on fluoroscopically guided invasive cardiovascular procedures, \" the core principle governing the use of ionizing radiation is ALARA (as low as reasonably achievable). The ALARA principle recognizes that there is no magnitude of radiation exposure that is known to be completely safe. This principle confers a responsibility on all physicians to minimize the radiation injury hazard to their patients, to their professional staff, and to themselves\" \[[@B35]\]. If the information is comparable, every effort should be done to orient the patient towards non-ionizing testing. If it is true that \"local expertise and availability should guide the selection of imaging techniques\" \[[@B36]\], it is also true that the doses and risks associated with the different diagnostic options should be clearly spelled out to allow the patient and the prescriber to make an informed decision. This policy is encouraged by common sense \[[@B37]\], deontologic code \[[@B38]\], European Commission imaging guidelines \[[@B14]\] and the European law \[[@B39]\]. It propels to use a green technique whenever the information supplied is grossly comparable to the red one \[[@B40]\]. The basic concept underlying this cardiac imaging paradigm shift is obvious but is presently neglected. Medical images of heartbreaking beauty when considering the benefit, can be ambiguous when considering the cost-benefit relationship, and unacceptable when considering the risk-benefit. In this transition, we as physicians and imaging specialists, are the critical link. \"Health professionals involved in the processes of diagnosis and treatment are the critical link. Training them properly and ensuring intensive information exchange among them are, therefore, probably the most cost-effective ways of achieving patient safety\" \[[@B4]\]. (International Action Plan for the radiological Protection of patients).	PubMed Central	2024-06-05T03:55:58.157715	2005-5-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166568/", "journal": "Cardiovasc Ultrasound. 2005 May 25; 3:13", "authors": [ { "first": "Eugenio", "last": "Picano" } ] }
PMC1166569	Exercise training is an effective intervention that can partially restore the impaired aerobic capacity and musculoskeletal fitness (i.e. muscle strength) found in cardiac transplant recipients \[[@B1],[@B2]\]. However, previous reports have focused exclusively on the effects of exercise training in men. Therefore, the effect of exercise training on these outcomes in female cardiac transplant recipients is not known \[[@B2]-[@B7]\]. Importantly, a majority of female cardiac transplant recipients do not engage in regular physical activity leading to increased levels of fatigue, poor functional status and reduced exercise capacity \[[@B8]-[@B10]\]. Based on this rationale, the aim of this study is to examine the effect that exercise training has on improving aerobic endurance (i.e. distance walked in six-minutes) and lower extremity muscle strength in female cardiac transplant recipients. We hypothesized that exercise training would be a feasible and effective intervention to improve aerobic endurance and lower extremity strength in female cardiac transplant recipients. Methods ======= Subjects and procedures ----------------------- The participants for this study consisted of 20 (51 ± 11 years) clinically stable female cardiac transplant recipients who participated in the University of Alberta Post-Transplant Exercise Rehabilitation program between 1997 and 2003. All assessments and exercise training were performed in the Physical Therapy Department at the University of Alberta Hospital. Ethics approval for this study was obtained from the Biomedical Ethics Board at our University. Outcome Assessments ------------------- The six-minute walk test was performed in accordance with the American Thoracic Society guidelines \[[@B11]\]. In addition, the six-minute walk scores were compared with age-matched norms for healthy females published by Gibbons and associates \[[@B12]\]. Leg-press maximal strength testing was performed on a commercially available leg-press machine with the greatest weight lifted while adhering to strict technique being used as the maximal score. All assessments were repeated following the 12-week training program. Exercise Training Intervention ------------------------------ Exercise training consisted of supervised aerobic (cycling and/or treadmill exercise at an intensity between 12 to 14 on the BORG perceived exertion scale for 30 to 40 minutes/day including warm-up and cool-down, 5 days/week) and lower extremity strength training. Data analysis ------------- Statistical analysis was performed with a one-way analysis of variance. The alpha level was set \"a priori\" at p \< 0.05. Data are presented as mean ± SD. Results ======= Baseline testing was performed 37 ± 27 days after cardiac transplantation. Aerobic Endurance ----------------- At baseline, our participants\' aerobic endurance was 50% lower than age-matched predicted values (Figure [1](#F1){ref-type="fig"}). Twelve weeks of training resulted in a significant increase in aerobic endurance, however, it remained 22% lower than age-predicted values (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Effect of exercise training on the distance walked in six-minutes. \*, p \< 0.05 vs. pre-training; ^†^, p \< 0.05 vs. post-training. ::: ![](1468-6708-6-10-1) ::: Leg Press --------- Leg-press maximal strength increased by 64% after three months of training (pre: 48 ± 16 kg vs. post: 78 ± 27 kg, p \< 0.05). Discussion ========== This is the first study to examine the effect that combined aerobic and strength training have on improving aerobic endurance and musculoskeletal fitness in female cardiac transplant recipients. The main finding of this study is that combined aerobic and strength training is a feasible and effective intervention to partially restore female cardiac transplant recipients\' aerobic endurance and leg-press strength. Cole et al. \[[@B9]\] recently found that female cardiac transplant recipients\' aerobic capacity was 60% lower than age-predicted values. Consistent with this finding, our transplant recipients\' baseline aerobic endurance was 50% lower than age predicted values. Moreover, our participants pre-training leg-press strength was 36% lower than that found in age-matched male cardiac transplant recipients tested in our laboratory \[[@B13]\]. The mechanisms responsible for the impaired cardiovascular and musculoskeletal fitness are likely secondary to abnormalities in cardiac and skeletal muscle function associated with pre-transplant heart failure, post-transplant deconditioning, cardiac denervation or immunosuppressuion therapy \[[@B14]\]. Several research groups \[[@B2],[@B3],[@B13]\] have demonstrated that exercise training initiated in the early post-operative period is associated with an increase aerobic endurance \[[@B13]\], muscle mass \[[@B2]\], muscle strength \[[@B2],[@B13]\] and bone density \[[@B3]\] in male cardiac transplant recipients. This study extends previous investigations by demonstrating that 12 weeks of combined aerobic and strength training are associated with a significant and marked improvement in aerobic endurance and muscle strength in female cardiac transplant recipients. The mechanisms responsible for the improvement in aerobic endurance was not examined in this study, however, they may be due to favorable improvements in mitochondrial oxidative properties \[[@B15],[@B16]\] that increase arteriovenous oxygen difference during exertion as aerobic training does not alter exercise cardiac output in this population \[[@B4],[@B17]\]. The training mediated increase in leg-press strength that we found is likely secondary to the increase in muscle mass that occurs with aerobic \[[@B4]\] or combined aerobic and strength training \[[@B2]\]. The consequence of our training mediated improvement in cardiorespiratory and musculoskeletal fitness is that it may result in a favorable improvement in mortality. Specifically, Kavanagh et al. \[[@B18]\] reported that cardiac transplant recipients with the greatest training-mediated improvement in aerobic capacity and lean body mass had a lower mortality rate 12 years after cessation of the training program. A limitation of our investigation is that we did not have a non-exercise control group. However, our cardiac transplant recipients are required to participate in a supervised 12-week exercise program beginning as an inpatient and completed as an outpatient. Despite this limitation, the improvement in aerobic endurance and leg-press strength associated with our training program is similar to that found in male cardiac transplant recipients who participated in our outpatient exercise rehabilitation program \[[@B13]\]. Summary ======= A majority of female cardiac transplant recipients adhere to a sedentary lifestyle and as result their aerobic capacity is 60% lower than age-predicted values \[[@B9]\]. The effect that exercise training has on improving female cardiac transplant recipients\' aerobic endurance and lower extremity strength is not known. The primary finding of this investigation is that 12 weeks of aerobic and strength training is an effective intervention that can improve aerobic endurance and musculoskeletal fitness in recent female cardiac transplant recipients. Moreover, the improvement in aerobic endurance and leg-press strength is similar to that found in male cardiac transplant recipients after combined aerobic and strength training. In summary, female cardiac transplant recipients should be encouraged to perform aerobic and strength training to increase their aerobic endurance and musculoskeletal fitness. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MH. Conceived the study, performed data analysis and manuscript preparation. KR. Assisted with data collection and manuscript preparation. LF. Performed the exercise rehabilitation training and assisted with manuscript preparation. DK. Assisted with manuscript preparation. DW. Assisted with manuscript preparation. LJ. Assisted with manuscript preparation. WT. Assisted with manuscript preparation. All authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:58.159362	2005-5-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166569/", "journal": "Curr Control Trials Cardiovasc Med. 2005 May 26; 6(1):10", "authors": [ { "first": "Mark", "last": "Haykowsky" }, { "first": "Kenneth", "last": "Riess" }, { "first": "Linda", "last": "Figgures" }, { "first": "Daniel", "last": "Kim" }, { "first": "Darren", "last": "Warburton" }, { "first": "Lee", "last": "Jones" }, { "first": "Wayne", "last": "Tymchak" } ] }
PMC1166570	Background ========== Human skeletal muscles contain approximately 35% of total human body magnesium, which is an essential cofactor in a number of cell reactions. Magnesium ions influence the equilibria of many reactions involved in cellular bioenergetics by interacting with phosphorylated molecules and interfere with the kinetics of ion transport across plasma membranes \[[@B1]\]. There is considerable evidence that Mg^2+^is actively transported and regulated, although the mechanisms are still largely unknown \[[@B2]\]. In skeletal muscle variations of cytosolic pH, phosphocreatine (PCr) and inorganic phosphate (Pi) concentrations influence the complex multi-equilibrium system of the molecular species binding magnesium ions. As a consequence \[Mg^2+^\] changes considerably in different metabolic conditions such as rest, exercise and recovery, showing an increase matched by a decrease of intracellular pH during exercise and recovery \[[@B3]\]. We assessed the cytosolic pH and the \[Mg^2+^\] by ^31^P MRS at rest, during exercise and post-exercise recovery in the calf muscle of two patients with McArdle\'s disease with muscle glycogen phosphorylase deficiency (McArdle), and two brothers affected by Tarui\'s disease with muscle phosphofructokinase deficiency (PFK). These two type of glycogenosis, being characterized by almost absent activity of enzymes involved in glycogenolysis (McArdle) and glycolysis (PFK) pathways, show in general limited/absent production of intracellular lactic acid, depending on the degree of enzyme deficit \[[@B4],[@B5]\]. As consequence, patients with McArdle\'s and Tarui\'s disease, typically show a decrease or a lack of intracellular acidification during muscle exercise when studied by ^31^P MRS \[[@B6],[@B7]\]. We used these diseases as natural experimental models to study the pattern of free Mg^2+^during exercise and recovery in the absence of intracellular acidification to understand to what extent homeostasis of intracellular free Mg^2+^is linked to pH. Methods ======= Patients -------- We studied 4 patients: two unrelated males both aged 42, with myo-phosphorylase deficiency (named MCArdle I and II respectively) and two brothers aged 18 and 10 years with phosphofructokinase deficiency (named PFK I and II respectively), as detected by histochemical/biochemical analysis of muscle. Ten healthy volunteers (10 males age: 33 ± 15) were recruited as control subjects. Written informed consent was obtained from all subjects. Protocol -------- MR spectra were acquired on a General Electric 1.5 T Signa System whole-body scanner. Radiofrequency pulses at 25.866 MHz with a pulse width of 400 μs and a transmitter power of 0.5 kW were transmitted by a surface coil (20.5 cm diameter; General Electrics) and the resonance signals were collected by a 7.5 cm receiving coil. A data table of 1024 complex points was collected for each FID. The band width was 2 kHz. The delay between transmission and reception was 0.5 ms and the dwell time was 250 μs. The stimulation-response sequence was repeated every 5000 ms (TR = 5000 ms). Magnetic field homogeneity was optimized by shimming the ^1^H water spectrum (FWMH 0.25--0.35 ppm) The spectroscopic measurements were performed according to the quantification and quality assessment protocols defined by the EEC Concerted Research Project on \"Tissue Characterisation by MRS and MRI\", COMAC-BME II.1.3 \[[@B8]\]. Subjects lay supine with a 20.5/7.5 cm diameter transmitter/receiver surface coil centred on the maximal circumference of the right calf muscle. Muscle aerobic incremental exercise consisted of different levels of 1 minute each (12-FIDs) of plantar flexion against a pedal using a pneumatic ergometer \[[@B9]\]. All patients were asked to perform an exercise to reach a PCr depletion of about 50% at the end of exercise. Sixty-four FIDs at rest, and 12 FIDs for each level of work were averaged. During recovery 4-FIDs data blocks (20 s) were recorded for 60 s, while longer time blocks were collected thereafter. The area of each metabolite signal was fitted to a Lorentzian line shape using a time-domain fitting program AMARES/JMRUI\[[@B10]\], the PCr and Pi concentration were calculated by assuming a normal ATP concentration of 8 mM \[[@B11]\]. The cytosolic pH and \[Mg^2+^\] are calculated from the chemical shift of Pi and β-ATP respectively, both measured from the resonance of PCr, using an equation which takes into account the mutual influence between pH and \[Mg^2+^\] \[[@B3]\]. The simultaneous calculation of \[Mg^2+^\] and pH was performed by the specific software package MagicMC, that we developed and made available on the internet \[[@B12]\]. Results ======= ^31^P MRS spectra of human skeletal muscle typically show the peak signal of: phosphomonoesters (PME) which represents the phosphorylated monosaccharide intermediates of glycogenolysis, inorganic phosphate (Pi), phosphocreatine (PCr) and the three phosphate groups α, β, γ of ATP. Figure [1](#F1){ref-type="fig"} shows ^31^P MRS spectra acquired at the end of exercise in the calf muscle of McArdle and PFK patients compared to that of a control subject reaching a similar level of PCr depletion. End exercise spectra of both PFK patients show a marked increase in the PME peak. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### ^31^P MRS spectra of calf muscle during exercise.End-exercise ^31^P MRS spectra of calf muscle acquired in patients and in a control subject reaching a PCr depletion of about 50%. PFK patients showed a marked phosphomonoester (PME) accumulation, although to a different extent. ::: ![](1476-5918-4-7-1) ::: Table [1](#T1){ref-type="table"} reports the rest and end-exercise values of cytosolic \[Mg^2+^\] and pH in patients\' calf muscle, compared to the mean values obtained in a control group of ten subjects with a comparable end-exercise PCr depletion. At rest all patients showed a cytosolic pH not different from control values. Resting cytosolic \[Mg^2+^\] in PFK II patient was 0.45 mM, higher than the control value (0.32 + 0.04 mM) \[[@B3]\], while in PFK I and both McArdle\'s patients was normal. Both McArdle and PFK I patients reached a similar PCr depletion just above 50%, while PFK II patient stopped at a lower degree of PCr depletion. All patients displayed a lack of intracellular acidosis during exercise, showing pH values higher than controls at the end of exercise. Cytosolic free \[Mg^2+^\] at the end of exercise was lower in both PFK patients compared to control values. The variation of free \[Mg^2+^\] from rest to end-exercise (Δ\[Mg^2+^\]) was negative in both PFK patients and in McArdle II patient. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### \[Mg^2+^\] and pH values at rest and end-exercise ::: Rest End-Exercise --------------------------- --------------- --------------- ----------- --------------- --------------- ----------------- McArdle I 0.34 6.98 53.1% 0.39 7.06\* \+ 0.05 McArdle II 0.36 6.95 53.8% 0.32 7.07\* \- 0.04\* PFK I 0.38 6.95 54.1% 0.23\* 7.01\* \- 0.15\* PFK II 0.45\* 6.95 35.3% 0.23\* 7.01\* \- 0.22\* Controls Mean (n = 10) 0.31 6.96 47.7% 0.42 6.8 +0.11 S.D. 0.04 0.02 13.8% 0.06 0.09 0.07 control range \[min:max\] \[0.27:0.40\] \[6.94:7.00\] \[22:73\] \[0.35:0.54\] \[6.90:6.67\] \[+0.03:+0.24\] Rest and end-exercise values of cytosolic \[Mg^2+^\] and pH assessed by ^31^P MRS in patients calf muscle, compared to the mean values obtained in a control group of ten healthy subjects with a comparable end-exercise PCr depletion; Δ\[Mg^2+^\]: difference between end-exercise \[Mg^2+^\] and rest \[Mg^2+^\]; \%PCr: percentage of PCr depletion \* denotes values out of the range: control mean ± 2 SD. ::: Figure [2](#F2){ref-type="fig"} reports the patterns of cytosolic free \[Mg^2+^\] and pH obtained in patients during exercise (panel A and B) and recovery (panel C and D) compared with typical patterns from a healthy volunteer with comparable PCr depletion. During exercise and recovery the McArdle patients did not show any significant change (McArdle II) or small change (McArdle I) in free \[Mg^2+^\], while both PFK patients showed decreased free \[Mg^2+^\] during exercise. On the other hand, during recovery the pattern of free \[Mg^2+^\] was different in the two PFK patients, with PFK II showing a moderate increase during early recovery. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### \[Mg^2+^\] and pH patterns during exercise and recovery.Patterns of cytosolic free magnesium concentration and pH at rest, during exercise and recovery in patients compared with typical patterns from a healthy volunteer with comparable PCr depletion. (A): pattern of \[Mg^2+^\] during exercise; (C): pattern of \[Mg^2+^\] during recovery; (B): pH pattern during exercise; (D): pH pattern during recovery. ::: ![](1476-5918-4-7-2) ::: Figure [3](#F3){ref-type="fig"} report the PME pattern of the two PFK patients during exercise and recovery. PFK I patient shows a slower rate of both PME accumulation and recovery compared to PFK II patient. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### PME patterns of PFK patients during exercise and recovery.PFK I patient shows both a slower rate of PME accumulation during exercise and a slower recovery of PME after exercise compared to PFK II patient. PME signal comes from the phosphorylated monosaccharide intermediates of glycogenolysis. ::: ![](1476-5918-4-7-3) ::: Discussion ========== In the skeletal muscle variations of cytosolic pH, phosphocreatine and inorganic phosphate concentrations influence the complex multi-equilibrium system of the molecular species which bind magnesium ions. As a consequence free cytosolic \[Mg^2+^\] can change considerably in different metabolic conditions such as rest, exercise and recovery. It has been shown by ^31^P MRS that the increase of cytosolic free \[Mg^2+^\] occurring in skeletal muscle of healthy subjects during exercise and initial recovery is matched by a decrease in cytosolic pH, and the changes in cytosolic free \[Mg^2+^\] were mainly the result of the predominant effect of \[H^+^\]. \[[@B3]\]. This result was attributed to mechanisms of binding competition existing between Mg^2+^and H^+^towards the molecules negatively charged present in the cell cytosol \[[@B3]\]. However, the causal relationship between pH and \[Mg^2+^\] has not been proved yet, as it could be argued that muscular exercise per seelicits an increase in cytosolic free \[Mg^2+^\]. Therefore, we used patients with McArdle\'s and Tarui\'s disease as experimental models to study the pattern of \[Mg^2+^\] during exercise and recovery in the absence of intracellular acidification to understand to what extent homeostasis of intracellular free Mg^2+^is linked to pH. Due to the rare nature of these disorders we were able to enrol just two patients for both diseases and therefore we had to deal with a small sample size. The results show that the increase in cytosolic \[Mg^2+^\] occurring in skeletal muscle during exercise is actually the consequence of the increase of H^+^concentration and not of other mechanisms related to muscle contraction. In addition, we found that both PFK patients showed a reduction of \[Mg^2+^\] during exercise concomitant with the PME increase. The decrease of \[Mg^2+^\] also persisted during recovery in PFK I patient who displayed a slower PME recovery. The PME peak in the ^31^P MRS spectra corresponds to the phosphorylated monosaccharide intermediates of glycogenolysis. Therefore, due to the deficit of the phosphofructokinase activity in Tarui\'s disease, the PME accumulation shown by these patients is likely due to the increase of fructose- 6-phosphate, which represents an additional binding site for cytosolic Mg^2+^. As a consequence, we interpret the decrease of \[Mg^2+^\] concomitant with the PME increase as due to the binding of Mg^2+^to fructose-6-phosphate. A previous study (6) reported that the abnormal PME accumulation of PFK patients during exercise was accompanied by a subnormal Pi accumulation. This finding was interpreted as a result of the incorporation of free Pi into phosphorylated glycolytic intermediates. However, both our patients did not show any Pi trap into PME, since we found that the sum of PCr and Pi was constant for the whole exercise duration, while the total phosphates signal increased proportionally to PME increase. Therefore, the \[Mg^2+^\] decrease found in PFK patients cannot be ascribed to a diminished Pi build-up. Moreover, PFK II patient displayed a larger decrease of \[Mg^2+^\] from rest to end-exercise compared to PFK I patient. Accidentally, PFK II patient also had a smaller PCr breakdown, nevertheless, this cannot be the cause of the larger decrease of \[Mg^2+^\], as a lower \[PCr\] corresponds to lower calculated \[Mg^2+^\] (3). Conclusion ========== Our results show that: i\) free \[Mg^2+^\] is strongly linked to pH in skeletal muscle homeostasis as previously suggested by a study in healthy volunteers \[[@B3]\], and by computer simulation on a chemical model mimicking muscle cell cytosol \[[@B13]\]; ii\) the decrease of free \[Mg^2+^\] during exercise in both PFK patients suggests that \[Mg^2+^\] is influenced by the accumulation of fructose- 6-phosphate, an additional binding site for cytosolic Mg^2+^, as shown by the accumulation of the phosphomonoesters peak in the ^31^P MRS spectra of these patients. iii\) ^31^P MRS is a suitable tool for the in vivoassessment of free cytosolic \[Mg^2+^\] in human skeletal muscle during rest, exercise and recovery; Authors\' contributions ======================= EM participated in the study design, performed the post-processing and statistical analysis, RL and CT participated in the study design and coordinated the data collection, AM participated in the study design, BB participated in the coordination of the study, SI participated in the study design, coordinated the study, and drafted the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by Ricerca Fondamentale Orientata (ex quota 60%) and by Programmi di Ricerca Scientifica di Rilevante Interesse Nazionale -- Cofin (ex quota 40%).	PubMed Central	2024-06-05T03:55:58.160085	2005-6-6	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166570/", "journal": "Dyn Med. 2005 Jun 6; 4:7", "authors": [ { "first": "Emil", "last": "Malucelli" }, { "first": "Raffaele", "last": "Lodi" }, { "first": "Andrea", "last": "Martinuzzi" }, { "first": "Caterina", "last": "Tonon" }, { "first": "Bruno", "last": "Barbiroli" }, { "first": "Stefano", "last": "Iotti" } ] }
PMC1166571	Innate immune system ==================== The first line of immune defence mainly operates by detection of a broad range of molecular patterns foreign to mammalian tissues, called pathogen-associated molecular patterns that induce the activation of the innate immunity and inflammatory response \[[@B1]\]. The constitutive expression of a limited set of pattern recognition receptors by many cell types of the innate immunity does not require clonal expansion of specific cell populations. These germ cell-encoded proteins recognize microbial pathogens or ligands from damaged tissues based on shared molecular structures and induce host responses that localize the spread of infection and enhance systemic resistance to infection. Therefore, the expression of a limited number of highly active genes during the activation of innate immunity is able to induce rapid (minutes to hours) efficient defensive immune responses \[[@B2]\]. Several cell types contribute to innate immunity and the mononuclear phagocyte lineage plays a pivotal role in innate immunity. Monocytes, macrophages and their tissue-differentiated derivatives, such as microglia in the nervous system, express pattern recognition receptors, namely various scavenger and Toll-like receptors \[[@B3]\]. These receptors induce transmembrane signals that activate NF-kB and mitogen dependent protein kinase pathways \[[@B4]\]. Toll-like receptor activation also induces the expression of a wide number of genes encoding proteins, such as cytokines, with regulatory functions upon leukocyte activation and tissue inflammation \[[@B5]\]. Therefore, the capacity of each individual organism to regulate the activation of innate immunity and local inflammatory responses is crucial for initiating defensive action against pathogens, limiting tissue damage and enhancing fast recovery and tissue healing. The inflammatory response ========================= In response to cell injury elicited by trauma or infection the inflammatory response sets in, constituting a complex network of molecular and cellular interactions directed to facilitate a return to physiological homeostasis and tissue repair. The response is composed of both local events and a systemic activation mediated by cytokines. If tissue health is not restored or in response to stable low grade irritation, inflammation becomes a chronic condition that continuously damages the surrounding tissues. In fact, during chronic inflammatory immune responses, tissue injury and healing proceed simultaneously. The collateral damage caused by this type of inflammation usually accumulates slowly, sometimes asymptomatically for years but can eventually lead to severe tissue deterioration \[[@B6]\]. Cytokines are largely secreted molecules that act on the surrounding microenvironment by providing cell to cell signalling. Cytokines are components of a large, complex signalling network. The effects of cytokines on target cells may be inhibited or enhanced by other cytokines, hormones, and cytokine-receptor antagonists and circulating receptors. Tumor necrosis factor α (TNF-α), Interleukin-1 (IL-1) and IL-6 are the classical pro-inflammatory cytokines. Their ability to activate both local and systemic effects is well established. Locally, they contribute to the activation of the inflammatory cells and together with chemokines, which induce the expression of adhesion molecules, cause their local recruitment. When the causes of the inflammatory reaction are of a high intensity, the production of cytokines is increased and they are released in the circulation provoking the \"acute phase response\". On the other hand, \"inhibitory\" cytokines such as IL-10 damp down the activation of some effector functions of T lymphocytes and mononuclear phagocytes, by inhibiting the release of pro-inflammatory cytokines and therefore turning off the inflammatory processes \[[@B7],[@B8]\]. The acute phase response e.g. leukocytosis, fever, somnelence, anorexia and acute phase proteins synthesis, such as C reactive protein (CRP) in the liver, is a highly conserved inflammatory response which is rapidly activated by infections or trauma via pattern recognition molecules. Acute phase protein concentration rapidly increases after infection, and their production is controlled primarily by IL-6- and IL-1-type cytokines. The acute phase proteins provide enhanced protection against microorganisms and modify inflammatory responses by affecting cell trafficking and mediator release. The more traditional view of acute phase proteins has extended beyond that of opsonization of microorganisms. Some acute phase proteins have anti-inflammatory effects while others have important effects on leukocyte activation and trafficking \[[@B9]\]. CRP, named for its capacity to precipitate the somatic C-polysaccharide of Streptococcus pneumoniae, was the first acute-phase protein to be described, and is an exquisitely sensitive systemic marker of inflammation and tissue damage. It is a member of the pentraxin family of plasma proteins, which are part of the lectin fold superfamily of calcium-dependent ligand-binding and lectin (carbohydrate-binding) proteins. In healthy blood donors, the median concentration of CRP is 0.8 mg/l, but following an acute-phase stimulus, values may increase 10 000-fold. In an apparently healthy population the median baseline value is slightly higher and tends to increase with age with females showing slightly higher circulating concentrations. In most, but not all diseases, the circulating value of CRP reflects on-going inflammation much more accurately than do other biochemical parameters of inflammation, such as plasma viscosity or the erythrocyte sedimentation rate \[[@B10]\]. Innate immune system during ageing ================================== Age related changes in clonotypic immune system are well documented. For instance, involution of the thymus gland is an early feature of the immune reconfiguration and several age- related alterations of T or B cell compartments have been described \[[@B11]-[@B13]\]. Studies on age-associated changes of the innate immunity are not as advanced as those of the clonotypic immune system. However, investigations from aged mice showed a functional decline of monocytes and macrophages, a low expression level of Toll-like receptors from activated splenic and peritoneal macrophages and an altered secretion of several chemokines and cytokines \[[@B14],[@B15]\]. Aged macrophages also contribute to an impaired the proliferative response of activated peripheral T lymphocytes \[[@B16]\]. Aged phagocytes, such as macrophages and neutrophils, showed an impaired respiratory burst and reactive nitrogen intermediate production with a decreased ability to destroy pathogens. Moreover, aged dendritic cells were less efficient in activating both T and B cell populations and aged NK cells showed a decreased ability in killing tumour cells \[[@B17],[@B18]\]. However, not all immune activities are decreased during ageing. In fact, the \'in vitro\' production of pro-inflammatory cytokines, such as IL-1, IL-6 and TNF-α, by mitogen activated peripheral blood monuclear cells from elderly persons was higher than that from young donors \[[@B19]\]. Cytokine production may also be up-regulated \'in vivo\' in old subjects resulting in an abnormal elevation of pro-inflammatory cytokines during inflammatory responses \[[@B20]\]. An age- related increase of IL-6 levels has also been reported in plasma, serum and supernatants from healthy elderly and centenarians \[[@B19],[@B21],[@B22]\]. It is interesting to note that \'in vitro\' addition of IL-1 and TNF-α to fibroblasts induced an accelerated senescent phenotype which was rescued by anti-oxidant addition \[[@B23]\]. Inflammatory responses during ageing ==================================== A dramatic increase in mean life span and life expectancy, coupled with a significant reduction in early mortality, has lead to a large increase in the number of elderly people in modern societies. This demographic phenomenon has been paralleled by an epidemic of chronic disease usually associated with advancing age \[[@B24]\]. Most age-related diseases have complex aetiology and pathogenic mechanisms. The clinical diagnosis and therapy of these diseases requires a multidisciplinary medical approach with progressively increased costs. A body of experimental and clinical evidence suggests that the immune system is implicated, with a variable degree of importance, in almost all age-related or associated diseases. Both innate and the clonotypic immune system are usually involved in the pathogenesis of these chronic diseases \[[@B11],[@B25]\]. However, inflammatory responses appear to be the prevalent triggering mechanism driving tissue damage associated with different age-related diseases and the term \"Inflammaging\" has been coined to explain the underlining inflammatory changes common to most age-associated diseases \[[@B26]\]. This new term indicates that ageing is accompanied by an age-dependent up-regulation of the inflammatory response, due to the chronic antigenic stress which bombards the innate immune system thorough out life and potentially triggers the onset of inflammatory disease. Chronic inflammation is considered to be involved in the pathogenesis of all age-related diseases: Alzheimer\'s disease, atherosclerosis, diabetes, sarcopenia and cancer all have important inflammatory components. Inflammaging, i.e. the up-regulation of a variety of anti-stress responses at the cellular and molecular level, is the consequence of the body\'s ability to counteract and modulate the effects of a variety of stressors, which cause the accumulation of molecular and cellular scars \[[@B26],[@B27]\]. However, as recently discussed \[[@B28]\], a wide range of different aetiological factors is likely to contribute to increased low-grade inflammatory activity in elderly populations including a decreased production of sex steroids, smoking, subclinical disorders such as arteriosclerosis, asymptomatic bacteruria and a higher relative/absolute amount of fat tissue, which has been suggested to produce pro-inflammatory cytokines \[[@B29]\] (see below). Furthermore, increased levels of circulating inflammatory mediators may result from a constant, low-grade activation of cytokine-producing cells or a dysregulated cytokine response following stimulation \[[@B28]\], which does not readily become damped down. On the other hand, a recent hypothesis suggests that the reduction in lifetime exposure to infectious diseases and other sources of inflammation -- the cohort mechanism -- may also contribute to the historical decline in old-age mortality \[[@B30]\]. Recent studies have linked an individual\'s exposure to past infection to levels of chronic inflammation and to increased risk of heart attack, stroke, and cancer. For example, the risk of heart attack is correlated with serum levels of inflammatory proteins such as CRP \[[@B31]\]. Within individuals, CRP levels are also correlated with the number of seropositivities to common pathogens, suggestive of infection history \[[@B32]\]. Low-grade increases in levels of circulating TNF-α, IL-6, soluble IL-2 receptor (sIL-2R), and CRP and low levels of albumin and cholesterol, which also act as inflammatory markers, are strong predictors of all-cause mortality risk in several longitudinal studies of elderly cohorts. The effects of inflammatory mediators upon survival are independent of pre-existing morbidity and other traditional risk factors for death (smoking, blood pressure, physical exercise, total cholesterol, co-morbidity, body mass index, and intake of anti-inflammatory drugs) suggesting that cytokines may induce exaggerated pathological processes. Furthermore these molecules appear to be sensitive markers of pre-clinical disorders in elderly populations \[[@B28],[@B33]-[@B39]\]. In the following sections we will focus on the role of innate immunity and inflammation in the most common diseases of elderly in order to direct the reader\'s interest to these emerging topics in modern gerontology and geriatrics investigations. Frailty and sarcopenia ====================== Frailty has been defined as an age-related decline in lean body mass, decreased muscle strength, endurance, balance and walking performance, low activity and weight loss accompanied by a high risk of disability, falls, hospitalisation and mortality \[[@B28],[@B40]\]. It has been suggested that this syndrome reflects a metabolic imbalance caused by overproduction of catabolic cytokines and by the diminished availability or action of anabolic hormones, resulting from ageing itself and the presence of associated chronic conditions \[[@B28],[@B41]\]. Sarcopenia is obviously a central part of the frailty syndrome \[[@B28]\]. Sarcopenia is the loss of muscle mass and strength that occurs with normal ageing. In elderly, there is a decline in a variety of neural, hormonal, and environmental trophic signals to muscle. However, the most important endogenous cause may be the irreversible age-related loss of α-motor units in the central nervous system \[[@B42]-[@B44]\]. The most important environmental cause of sarcopenia is lack of physical activity which is an age-related phenomenon which shows universal decline with age in industrialized societies; obesity and type 2 diabetes being concomitant epidemics (see below). An emerging issue in the past few years for sarcopenia research has been whether there is also an increased catabolic signal, driven by systemic inflammation \[[@B44]\]. Systemic low-grade inflammation has been associated with decreased muscle mass as well as the development of functional disability in elderly populations \[[@B33],[@B44]-[@B47]\]. TNF-α has special effects that may contribute directly to sarcopenia \[[@B44]\], including increased basal energy expenditure, anorexia, loss of muscle and bone mass \'in vivo\' and association with wasting/cachexia in chronic inflammatory disorders. Consistent with these findings, muscle protein synthesis was inversely related to local levels of TNF-α protein in skeletal muscles in frail very old humans \[[@B48]\]. In a recent study of nursing home residents aged 85--96 years, systemic low-grade activation of the TNF-α system at baseline was inversely correlated to muscle strength after resistance training for 12 weeks, demonstrating that TNF-α could also be a limiting factor for training-induced improvement in muscle strength in very old people \[[@B49]\]. The role of IL-6 in sarcopenia is not clear. Epidemiological studies have reported that IL-6 is strongly associated with functional disability and loss of muscle mass but experimental studies have not been able to link IL-6 to sarcopenia \[[@B28],[@B44]\]. Human investigations have shown that inflammation impairs muscle strength in elderly people \[[@B50]\] while Insulin Growth Factor 1 and IL-6 plasma levels were synergistically related to disability and mortality in older women \[[@B51]\]. A complicating factor in untangling the cause-effect relationships underlying sarcopenia is the ability of IL-6, TNF-α, physical inactivity, abdominal obesity, and other factors to cause insulin resistance in the elderly \[[@B52]-[@B54]\]. Loss of control of the anabolic action of insulin on muscle would be another factor favouring sarcopenia, resulting in the elevation of one or more catabolic signals. Thus, the weight gain that occurs in most adults during middle age may actually predispose to sarcopenia as they age \[[@B44]\]. Obesity, the metabolic syndrome, type II diabetes ================================================= Western societies face a health care crisis of epidemic proportion, since the number of people with the metabolic syndrome is rising exponentially. This condition constitutes a major challenge for public health professionals in the field of preventive medicine and estimates suggest that more than 40 million US adults will be affected by the syndrome \[[@B55]\]. There is not known survival advantage of morbid obesity; on the other hand increased body fat is linked to high mortality \[[@B56]\]. In the past, humans have been plagued by famine when survival advantage would have been conferred by genes favouring an available energy source: the so called \'thrifty gene\' hypothesis. These genes led to genetic selection toward insulin resistance in the peripheral tissue in order to preserve glucose supply of the brain during starvation \[[@B57]\]. Human obesity shows a clear genetic component, which is usually polygenic and polymorphic variances in a number of \'thrifty\' genes could contribute to different susceptibilities to obesity and diabetes \[[@B58],[@B59]\]. Metabolic syndrome may be detected by 5 clinical diagnostic criteria 1) abdominal adiposity, 2) hypertriglyceridemia, 3) low high density lipoprotein, 4) hypertension, 5) fasting hyperglycaemia \[[@B55]\]. An involvement of inflammation with pathogenic mechanisms influencing the development of the metabolic syndrome has been suggested \[[@B55],[@B60],[@B61]\] and elevated CRP, IL-6 and TNF-α, associated with visceral adiposity, have been reported in this syndrome \[[@B61]\]. Adipocytes constitutively express the pro-inflammatory TNF-α \[[@B54]\] which decreases after weight loss \[[@B62]\]. Further work in this area has confirmed increased plasma concentrations of CRP, IL-6, and plasminogen activator inhibitor-1 (PAI-1) \[[@B63]-[@B65]\]. Therefore, a complex interplay may exist between inflammatory responses and general metabolism in atherosclerosis, cardiovascular disease, metabolic syndrome. It is becoming clear that the increased adipose tissue is not a simple reservoir for excess nutrients, but rather an active and dynamic organ capable of expressing several cytokines and other fat-derived peptides (FDP). Some FDP may have a role in the development of the metabolic syndrome but there is no evidence that these FDP are directly causing inflammation. It has been suggested that high levels of inflammatory factors are markers for obesity/abdominal obesity seen with aging, but some of these may not necessarily have a causative role in the development of inflammation \[[@B29]\]. Alternatively, a positive correlation was found between the body mass index and the percentage of resident macrophages, suggesting that fat tissue growth is associated with a recruitment of blood monocytes, responsible for cytokine production \[[@B66]\]. Insulin resistance is due to the reduced ability of peripheral tissues to properly respond to the activation induced by insulin. It is a key feature in the pathogenesis of type II diabetes and this condition may precede by 10--20 years the onset of hyperglycaemia and the clinical manifestation of the disease. Recent data suggests that a defect in insulin activation of glucose transportation in muscle cells could be induced by serine kinase cascade activation, which is down stream mediators of tissue inflammation factors \[[@B67]\]. Data from the Insulin Resistance Atherosclerosis Study showed that insulin resistance, as assessed by frequently sampled glucose tolerance tests, correlated with high blood levels of CRP, fibrinogen and PAI-1 and levels of these inflammatory factors were predictors of type II diabetes development \[[@B68]\]. Increased blood concentrations of TNF-α and IL-6 were associated with obesity and type II diabetes \[[@B69]\]. Finally, results of the population study from the European Prospective Investigation into Cancer and Nutrition Potsdam indicated a significant interaction between plasma IL-1β, IL-6 and type II diabetes development. In fact subjects with detectable IL-1β levels and increased levels of IL-6 showed an independently elevated risk of developing the disease \[[@B70]\]. Inflammation may predispose to a pre-diabetic state by increasing insulin resistance, since pre-diabetic subjects showed increased plasma levels of inflammatory proteins without primary defects of beta cell functions \[[@B71]\]. Sub-clinical inflammation was found significantly related to insulin resistance in a high risk group for diabetes, i.e. in subjects with positive family history of diabetes, obesity and hyper or dyslipoproteinemia \[[@B72]\]. Recent results from the INCHIANTI population study showed that subjects in the upper tertile of insulin resistance had increased serum levels of TNF-α, IL-1R antagonist and IL-6 and low levels of sIL-6R \[[@B73]\]. TNF-α causes an inhibition of auto-phosphorylation of tyrosine residues of the insulin receptor (IR) and an induction of serine phosphorylation of insulin receptor substrate-1, which in turn causes serine phosphorylation of the IR in adipocytes and inhibits tyrosine phosphorylation \[[@B74]\]. More recently, IL-6 has been shown to inhibit insulin signal transduction in hepatocytes \[[@B53]\]. Therefore, cytokines show relevant metabolic effects. Novel data have now appeared showing that the concomitant presence of the promoter polymorphisms of TNF-α and IL-6, linked to high production of these cytokines increases the risk of conversion to type 2 diabetes in obese subjects with impaired glucose tolerance response \[[@B75]\]. As discussed by Dandona et al. \[[@B69]\], two mechanisms might be involved in the pathogenesis of inflammation. Glucose and macronutrient intake causes oxidative stress and inflammatory changes. Chronic overnutrition (obesity) might thus be a pro-inflammatory state with oxidative stress. The increased concentrations of TNF-α and IL-6, associated with obesity and type 2 diabetes, might interfere with insulin action by suppressing insulin signal transduction, which in turn might promote inflammation. In fact, insulin reduces ROS generation by mononuclear cells, suppresses NADPH oxidase expression and intranuclear NF-kB binding, induces IkB expression and suppresses some inflammatory molecules \[[@B69],[@B76]\]. The well-known beneficial effects of caloric restriction on longevity in animal models by inducing reduced visceral fat mass might also induce a reduced secretion of multiple metabolically active factors, which are potentially responsible for the development of insulin resistance. This decrease in fat mass and its beneficial effects observed in ageing animal models might apply also to human ageing and its related pathology \[[@B77]\]. Cancer diseases =============== The majority of cancer occurs in subjects over the age of 65 years. Cancer rates increase sharply with age in both sexes: the incidence of cancer is 12--36 times higher in individuals aged 65 years or older compared with individuals aged 25--44 years, and 2--3 times more common than in persons aged 45--64 years. It is worth noting that 70% of deaths attributable to all cancer occur in men and in women aged 65 years or older whereas 35% cancer deaths in men and 46% of cancer deaths in women occur in those aged 75 years or older. The relationship between ageing and cancer is similar for most cancers, and it is well described by the multistage model. So, ageing might be considered not as a determinant of cancer per se, but as a surrogate marker of the duration of exposure to relevant carcinogenic factors \[[@B78]-[@B80]\]. Chronic inflammation induced by either biological, chemical, mechanical or physical injuries has been associated with increased incidence of cancer in different human tissues \[[@B81]\]. For instance, inflammatory bowel disease, ulcerative colitis and Crohn\'s disease are clinical conditions predisposing to cancer development of the large bowel or terminal ileum \[[@B81],[@B82]\]. Helicobacter pylori microorganisms are associated with atrophic gastritis, mucosal dysplasia, gastric adenocarcinoma and an unusual form of gastric lymphoma \[[@B81],[@B82]\]. Schistosome and trematode infections have been associated with cancers of the bladder and the biliary tracts \[[@B81]\]. The recognition of a role for inflammation in the natural history of a tumor has been known stretching from the mid-19th century \[[@B83]\]. The inflammatory microenvironment of tumors is characterised by the presence of host leucocytes both in the supporting stroma and in the tumor area \[[@B84]\]. The question whether the inflammatory infiltrate helps or hinders tumors is still open. In fact, inflammatory cells and cytokines found in tumors can contribute to tumor growth, progression, and immunosuppression as well mounting an effective host anti-tumor response. However, in most cases inflammation plays a role in the development of solid cancer. That is clearly demonstrated by a prospective, nested case-control study of a cohort of 22,887 adults followed for 11 years. A total of 172 colorectal cancer cases were identified. Up to 2 controls (n = 342) were selected for each case and matched by age, sex, race, and date of blood draw. Plasma CRP concentrations were higher among all colorectal cases combined compared with controls and the risk of colon cancer was higher in persons in the highest vs. lowest quartile of CRP \[[@B85]\]. Furthermore, in another recent study, a total of 174 patients considered to have undergone curative resection for cancer were studied. The results show that raised circulating concentrations of CRP, whether measured before or after operation, predict overall and cancer-specific survival in patients undergoing potentially curative surgery for colorectal cancer suggesting that the presence of a systemic inflammatory response predicts a poor outcome \[[@B86]\]. This data supports the hypothesis that inflammation is a risk factor for the development of some kind of solid tumors such as colon cancer. Cancer susceptibility and severity may also be associated with functional polymorphisms of cytokine genes involved in regulation of inflammation. In particular, as discussed in a recent review \[[@B79]\], a considerable body of data indicates that particular cytokine polymorphisms, especially those involving IL-6 and IL-10 genes, may influence susceptibility to, and in some cases prognosis in neoplastic diseases. It is intriguing that these two cytokines are involved in longevity \[[@B87]-[@B89]\]. Discrepant results might depend on confounding factors that affect case-control studies \[[@B90]\] or as discussed by Caruso et al. \[[@B79]\] might also depend on pleiotropic action of cytokines. It is paradigmatic in this respect that the favourable effect of high producer IL-10 genotype on hepatocarcinoma induced by HBV \[[@B91]\], may be due to the relative lack of control of HBV infection by immunosuppressive effects of high IL-10 levels. On the other hand, several and diverse mechanisms may link inflammation to cancer. 1\) Free radical production, derived by oxygen or NO metabolic pathways; the respiratory burst of leukocytes, and the arachidonic acid cascade activation are strongly implicated in DNA modification and protein damage which in turn promotes carcinogenesis \[[@B92]\]. 2\) High constitutive hyper-expression of NF-kB, an ubiquitous transcription factor with regulatory effect upon different inflammatory, apoptotic and oncogenic genes \[[@B82]\]. 3\) Alterations of the p53 gene induced by NO derived radicals promoting clonal expansion of aberrant or mutated cells \[[@B92]\]. 4\) Induction of angiogenesis by inflammatory factors favouring cancer progression \[[@B92]\]. 5\) Increased release of key pro-inflammatory factors and some cytokines, such as IL-1β, TNF-α and interferon (IFN)-γ, implicated in both the regulation of inflammation and the development of cancer \[[@B92],[@B93]\]. Atherosclerosis and cardiovascular diseases =========================================== Atherosclerosis and its complications are a major problem contributing to large sections of morbidity and mortality in old people. Cardiovascular disease is the leading worldwide cause of morbidity and death in Western societies \[[@B94]\]. However, our understanding of pathogenic mechanisms underling atherosclerosis and its complications is still incomplete, since more than half of patients with atherosclerosis do not show classical risk factors, such as hypercholesterolemia, hypertension, history of smoking, diabetes, obesity and sedentary life style \[[@B94],[@B95]\]. On the other hand, atherosclerosis, formerly considered an inocuous lipid storage disease, actually involves an ongoing inflammatory response (Figure [1](#F1){ref-type="fig"}). Recent advances in basic science have established a fundamental role for innate immunity in mediating all stages of this disease from initiation through progression and, ultimately, the thrombotic complications of atherosclerosis \[[@B96]-[@B98]\]. Clinical studies have shown that the emerging biology of inflammation in atherosclerosis applies directly to human patients. Elevation in markers of inflammation predicts outcomes of patients with acute coronary syndromes, independently of myocardial damage. In addition, low-grade chronic inflammation, as indicated by levels of the inflammatory marker CRP and cytokines, prospectively defines risk of atherosclerotic complications, thus adding to prognostic information provided by traditional risk factors. In fact, levels of CRP or IL-6 have been suggested as significant predictive risk factors for future development of cardiovascular events \[[@B31],[@B95],[@B99]-[@B103]\]. Increased levels of serum IL-1β have also been associated with high risk of congestive heart failure and angina pectoris \[[@B104]\]. Altered level of IL-1β is also suggested to be implicated in chronic inflammation underlining high blood pressure \[[@B105]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic representation of inflammatory mechanisms involved in pathogenesis of atherosclerosis and plaque formation. Monocytes and macrophages are the protagonists of atherosclerotic processes. ::: ![](1742-4933-2-8-1) ::: An initiating event is the accumulation of lipids in the vessel wall, which subsequently will become modified and triggers an inflammatory process. Low-density lipoproteins (LDL) are taken up by macrophages through scavenger receptors, leading to foam cells; the lipid deposits become oxidized forming pro-inflammatory lipid peroxides. Monocytes are attracted from the blood and differentiate into macrophages that take up the modified LDL and form lipid-laden foam cells, which is the first hallmark of atherosclerotic plaque development. Later on, inflammatory mediators increase, other immune cells are attracted, and smooth muscle cells are activated and become involved. More advanced stages of plaque development are characterized by increased deposition of extra cellular lipid cores, fibrous material, and often necrosis. Subsequently, these macrophages are further activated, leading to the production of a wide range of cytokines and growth factors. Myocardial infarction may occur as a result of erosion or uneven thinning and rupture of the fibrous cap, often at the shoulders of the lesion where macrophages enter, accumulate and are activated and where apoptosis may occur \[[@B7],[@B96]-[@B98]\]. Because genetic traits contribute significantly to the risk of coronary heart disease \[[@B106]\], a number of studies have now addressed the hypothesis that allelic variations in genes of innate immunity may increase the risk of this disease \[[@B107],[@B108]\]. Differences in the genetic regulation of inflammatory processes might partially explain why some people, but not others, develop the disease and why some develop a greater inflammatory response than others. Accordingly, common gene polymorphisms controlling high production of inflammatory molecules have been associated with atherosclerosis and a good control of inflammation might play a protective role against atherosclerosis \[[@B107]-[@B110]\]. Recently, a combination of alleles in different inflammatory genes has been associated with increased risk of developing acute myocardial infarction in elderly men \[[@B111]\]. In addition to innate immunity, clonotypic immunity plays also a role in atherosclerosis. Recent findings support the hypothesis that a crucial component of atherosclerosis is represented by T cell-mediated immune responses that are inappropriate in terms of time of onset, intensity, and target \[[@B112]\]. On the other hand, the most direct evidence for the critical role for T cells, IFN-γ, and IFN-γ-driven molecules in atherosclerosis is provided by mice with combined deficiencies of apolipoprotein E (apoE) and the IFN-γ receptor, in which the development of atheroma is significantly reduced in comparison to mice with only apoE deficiency, whereas exogenous IFN-γ enhances atherosclerosis \[[@B113],[@B114]\]. In contrast, the role of B cells has remained unclear. B-cells are not always present and may be protective. In fact, recent studies suggest that this cell type may inhibitthe development of vascular pathology in models of atherosclerosisand restenosis \[[@B115]\]. Finally, Wick et al. \[[@B116]\] have developed an autoimmune-inflammatoryconcept of atherogenesis based on experimental and clinical models. They argue that atherosclerosis may be the price humans pay for pre-existing immunity to microbial or autologous heat shock protein (HSP) 60. In summary, their autoimmune hypothesis of atherogenesis postulates that HSP60 expression is induced in arterial endothelial cells as a response to the action of stress factors, notably the classical atherosclerotic risk factors. Autologous HSP60 epitopes exposed on the surface of stressed endothelial cells are recognized by either pre-existing humoral and cellular anti-microbial HSP60 immune reactions or by invoking bona fideautoimmunity based on exposure of altered autologous HSP60 epitopes. The authors accept the importance of well-established atherosclerotic risk factors during atherogenesis, but assign a new role for them in the earliest stages of the disease, viz. acting as stressors \[[@B116]\]. Brain degenerative diseases: Alzheimer\'s disease ================================================= Alzheimer\'s disease (AD), a heterogeneous and progressive neurodegenerative disease which in Western societies mainly accounts for clinical dementia, is expected in the USA to rise from 4.6 today to 16 millions cases in 2050 \[[@B117]\]. Neuro-pathological hallmarks of AD are neuronal and synapsis loss, extracellular amyloid deposits (neuritic plaques) and intracellular deposition of degenerate filaments (neurofibrillary tangles) \[[@B118]\]. Major clinical manifestations of the disease are memory loss and cognitive impairment \[[@B119]\]. Inflammation clearly occurs in pathologically vulnerable regions of the AD brain, and it does so with the full complexity of local peripheral inflammatory responses. In the periphery, degenerating tissue and the deposition of highly insoluble abnormal materials are classical stimulants of inflammation. Likewise, in the AD brain damaged neurons and neuritis, highly insoluble Aβ42 peptide deposits and neurofibrillary tangles provide obvious stimuli for inflammation. Senile plaques in AD brains are associated with reactive astrocytes and activated microglial cells; cytokines and acute phase proteins are also overexpressed in microglia and astrocytes surrounding neuropathological lesions in AD brains. Inflammatory factors, such as cytokines, chemokines, complement components and acute phase proteins co-localize as secondary components in neuritic or senile plaques, or are over-produced in AD brains. Finally, activated microglia surrounds senile plaques and areas of neurodegeneration \[[@B120],[@B121]\]. There is accumulating evidence that Aβ peptide may promote or exacerbate inflammation by inducing glial cells to release immune mediators. Moreover, microglial and astroglial cells surrounding mature plaques in AD brains have been found to express activation markers. Enriched populations of human microglial cells isolated from mixed cell cultures prepared from embryonic human telencephalon tissues are able to express constitutively mRNA transcripts for cytokines and chemokines and treatment with pro-inflammatory stimuli as lipopolysaccharide or Aβ peptide leads to increased expression of mRNA levels of these inflammatory molecules \[[@B122]\]. The role of inflammation is further emphasized by a number of epidemiological studies demonstrating that the long-term use of nonsteroidal anti-inflammatory drugs may protect against AD. There are now a several published observational studies demonstrating that people who are known to be taking anti-inflammatory drugs considerably reduce their odds of developing AD and population studies have confirmed this negative association \[[@B123]\]. However, alternative hypothesises have been proposed. In particular, this effect has been hypothesised as relating to the ability of these drugs to inhibit angiogenesis. In fact, the brain endothelium secretes the precursor substrate for the β-amyloid plaque and a neurotoxic peptide that selectively kills cortical neurons. So, antiangiogenic drugs targeting the abnormal brain endothelial cell might be able to prevent and treat this disease \[[@B124]\] The long-term prospective association between dementia and the well known inflammation marker CRP was evaluated in a cohort of Japanese American men. These subjects were seen in the second examination of the Honolulu Heart Program (1968--1970) and subsequently were re-examined 25 years later for dementia in the Honolulu-Asia Aging Study (1991--1996). In a random subsample of 1,050 Honolulu-Asia Aging Study cases and noncases, high-sensitivity CRP concentrations were measured from serum taken at the second examination; dementia was assessed in a clinical examination that included neuroimaging and neuropsychological testing and was evaluated using international criteria. Compared with men in the lowest quartile (\<0.34 mg/L) of high-sensitivity CRP, men in the upper three quartiles had a 3-fold significantly increased risk for all dementias, mainly Alzheimer\'s disease and vascular dementia. These data support the view that inflammatory markers may reflect not only peripheral disease, but also cerebral disease mechanisms related to dementia, and that these processes are measurable long before clinical symptoms appear \[[@B125]\]. On the other hand, several other investigations have shown increased blood levels of some cytokines, such as IL-1β and IL-6, and acute phase proteins α-1-antichymotrypsin, (ACT) in patients with clinical AD \[[@B126]-[@B129]\]. Therefore, altered immune responses in the brain and the peripheral blood appeared to be associated with the disease. Finally, plasma levels of ACT also correlated with the degree of cognitive impairment in AD patients from a case-control study \[[@B126]\] suggesting that peripheral markers of inflammation or impaired immune responses could be used for monitoring the progression of the disease. Elevated levels of IL-6 in both brain homogenates and peripheral blood from AD patients have also been reported \[[@B130]\]. These findings suggest that an important, but still largely unknown, interplay between brain and peripheral immune responses may exist in the disease. In conclusion, the brain lesions associated with AD, which are referred to as neurofibrillary tangles and senile plaques, are characterized by the presence of a broad spectrum of inflammatory mediators, produced by resident brain cells, including neurons. Although secondary to the fundamental pathology caused by the presence of tangles and plaques, there is strong evidence that inflammation exacerbates the neuronal loss. Accordingly, several reports have appeared indicating that the risk of AD is substantially influenced by several polymorphisms in the promoter region, and other untranslated regions, of genes encoding inflammatory mediators. Alleles that favour increased expression of the inflammatory mediators or alleles that favour decreased expression of anti-inflammatory mediators are more frequent in patients with AD than in controls. The polymorphisms are fairly common in the general population, so there is a strong likelihood that any given individual will inherit one or more of the high-risk alleles \[[@B121],[@B127],[@B129]-[@B135]\] (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Alzheimer\'s disease: amyloid deposition is one the main pathogenetic mechanism. Accumulation of Aβ peptide may be caused by 1) gene mutations (PS1, PS2 and APP human mutations in familial Alzheimer\'s disease) 2) genotype (and/or phenotype) favoring unbalanced inflammatory responses (pro-inflammatory genotype/anti-inflammatory genotype). ::: ![](1742-4933-2-8-2) ::: Conclusion ========== In this presentation we have presented evidence linking innate immunity with the pathogenesis of age-related chronic degenerative diseases. Inflammation may be a pivotal and common background shared by the majority of these pathological conditions. Although evidence already presented has focused on innate immunity, clonotypic immunity has support as an emerging mechanism related to the pathogenesis of these diseases \[[@B116]\]. The evidence that a decline in mortality exists when comparing Swedish cohorts born in the middle of the 18^th^century to those in the 20^th^suggests important links between mortality in old age and in early life. Cohort mortality during childhood has been linked to cohort mortality in old age, implying that early exposure to infection is important in determining and imprinting the cohort morbidity phenotype \[[@B30]\]. Thus, chronic inflammatory mechanisms carry the imprint of early-life infections into later life morbidity and mortality. Evidence presented suggests that serological indicators of infection and inflammation in present day populations are related to vascular disease and other morbidities of ageing. Such inflammatory responses can be induced by invading pathogens, as well as by trauma or internal tissue injury. Thus, adaptive responses to short term infections or injury can become maladaptive as life progresses -a double-edged sword that evolutionary biologists refer to as antagonistic pleiotropy \[[@B136]\]. The process of life for the individual is the struggle to preserve its biological and immunological integrity. However, the preservation of the integrity of the organism comes with the price of responsiveness to systemic inflammation \[[@B137]\] which must be finely tuned otherwise dysregulation becomes a damaging accompanynt. With ageing, the reason why the innate immune system becomes over-activated is not clear, but increased exposure to infectious agents or cumulative damage to tissues could spark the change. Inflammation is not \'per se\' a negative phenomenon: it is the response of the immune system to pathogenic viruses or bacteria. Thorough out evolution, man has been set to live about 40 or 50 years but in today\'s world the immune system is active for several decades compared with the past centuries. A long period of activitation may lead to chronic inflammation which inexorably damages several/all organs and is the phenotype linked to both ageing and chronic disease. In cardiovascular disease, conventional risk factors remain important, but differential baselines in inflammatory status may explain why cholesterol levels seem not always directly associated with cardiovascular disease \[[@B22],[@B109]\]. Low-grade inflammation is also associated with parameters such as obesity, smoking, and physical inactivity, so inflammatory mediators constitute a link between life style factors, infections and physiological changes in the process of ageing on the one hand, and risk factors for age-related diseases on the other \[[@B28]\]. This immune response also depends on the genetic background of individuals. In fact, emerging evidence suggests that polymorphic alleles of inflammatory cytokines, involved in high cytokine production, are related to \'unsuccessful\' ageing as noted previously with respect to atherosclerosis and AD. On the other hand, controlling inflammatory status may enhance individual chance of achieving \'successful\' ageing. So, major findings reporting a relationship between cytokine polymorphisms and longevity suggest that those individuals who are genetically predisposed to produce low levels of inflammatory cytokines or high levels of anti-inflammatory cytokines may have an increased capacity to reach the extreme limit of human life-span \[[@B87]-[@B89],[@B109],[@B138],[@B139]\]. In other words, age-related diseases are \"the price we pay\" for an active immune system that defends us in youth but may harm us later on \[[@B139],[@B140]\]. Such data support the notion that antagonistic pleiotropy \[[@B136]\] plays a relevant role in diseases and longevity. Pro-inflammatory genotypes may therefore be both friends and foes. They are an important and necessary part of the normal host responses to pathogens, but the overproduction of inflammatory molecules may aggravate immune-inflammatory-related disease and contribute to earlier death. An immune system evolved to control pathogens is likely to favour a highly charged pro-inflammatory response programmed to resist fatal infections with avoidance of early mortality. However, persons with low responder genotypes may risk earlier mortality in response to serious infection but if they survive may respond less aggressively to age-related disease development. Such conditions might result in an increased chance of longevity in an environment with reduced pathogenic antigen load and/or adequate medical treatment \[[@B139],[@B141],[@B142]\]. Further studies on this field will open the way for new diagnostic approaches for early diagnosis of relevant pre-clinical states of age-related diseases. It may be possible, before clinical manifestations appear, that anti-inflammatory or other treatments might play a decisive role in preventing or significantly retarding the manifestation of the disease. Other studies focused on clarifying the specific contribution of each immune factor to a given disease will also contribute to the discovery of new drugs and/or innovative intervention protocols specific for these diseases related to ageing. Acknowledgements ================ Original work of Authors was supported by grants from the Italian Ministry of Education, University and Research, excofin 40% and 60%, to FL, GC, GCR, DL, CF and CC and FIRB to CF and CC, CURA financial support to FL is also acknowledged. Funds from EU 5FP T-CIA project to CF and Ministry of Health to GCR and CF are also acknowledged. The collaboration between thee \"Gruppo di Studio sull\'immunosenescenza\" coordinated by Prof. C. Caruso and INRCA was enhanced by a cooperation contract (Longevity and elderly disability biological markers).	PubMed Central	2024-06-05T03:55:58.161543	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166571/", "journal": "Immun Ageing. 2005 May 18; 2:8", "authors": [ { "first": "Federico", "last": "Licastro" }, { "first": "Giuseppina", "last": "Candore" }, { "first": "Domenico", "last": "Lio" }, { "first": "Elisa", "last": "Porcellini" }, { "first": "Giuseppina", "last": "Colonna-Romano" }, { "first": "Claudio", "last": "Franceschi" }, { "first": "Calogero", "last": "Caruso" } ] }
PMC1166572	Introduction ============ In nursing rabbits, sexual receptivity and fertility achieved after artificial insemination is depressed during the period of lactation, presumably through a hormonal antagonism between prolactin and gonadotropin release \[[@B1]-[@B3]\]. Several studies have demonstrated that separation of the doe from its litter for short periods of time, prior to artificial insemination, is very effective in stimulating ovarian activity in the mother \[[@B4]-[@B6]\], with endocrine changes that may explain the activation of ovarian function \[[@B7]-[@B9]\] Thus, separation of the does from their litters could be an effective procedure to augment breeding efficiency under farm conditions \[[@B5]-[@B7]\]. It must be noted that the published studies on the endocrine changes taking place in the doe after litter separation have been performed at single time points in the daily cycle, usually in the morning, which is an important drawback in view of the circadian nature of the secretion of the pituitary hormones involved \[[@B10],[@B11]\]. Indeed a number of circadian functions have been examined in rabbits \[[@B12]-[@B17]\], but there is no information on 24-h pattern of hormone release. This prompted us to undertake the present study whose aim was to examine the effect of litter separation for 48 h on 24-h changes in plasma prolactin (PRL), follicle stimulating hormone (FSH) and luteinizing hormone (LH) levels of the doe. Specifically, we sought to answer two questions: (i) did suppression of a major neuroendocrine and circadian stimulus like the stimulation of nipples by the lactating pup affect the 24-h changes in gonadotropin and prolactin release?; (ii) could the changes in circulating hormone levels be related to augmentation of breeding efficiency found after litter separation from the doe? Materials and methods ===================== Animals ------- The study was performed in 84 multiparous, lactating Californian x New Zealand White crossbreed female doe rabbits. Animals were housed in the research facilities of the Animal Production Department, Universidad Politécnica de Madrid, under controlled light-dark cycles (LD 16:8, light on at 8:00 h), housed in individual metal cages, fed at libitum using a commercial pellet diet (Lab Rabbit Chow, Purina Mills, Torrejón de Ardoz, Madrid, Spain) and having access to tap water ad libitum. The study was performed according to the CEE Council Directives (86/609, 1986) for the care of experimental animals. Groups of 6--7 female lactating rabbits were maintained with their litters or separated from them for 48 h, starting at different times (i.e., at 09:00, 13:00, 17:00, 21:00, 01:00 or 05:00 h). Ninety five percent of the does suckled the pups between 03:30 and 04:30 h during the dark period, as has been previously described \[[@B16]\]. On day 11 post-partum, the does were killed by decapitation at 6 different time points throughout a 24-hour cycle starting at 0900 h. Blood was collected from the cervical wound and the plasma was separated to measure prolactin and gonadotropin concentration. Hormone assays -------------- Plasma prolactin, FSH and LH levels were measured in duplicate samples by specific RIA methods \[[@B18]\] using AFP-991086, AFP-472176 and AFP-3120489 antibodies for prolactin, FSH and LH respectively, supplied by the National Institute of Health (NIH, Bethesda, MD, USA) and Dr. A. F. Parlow (Harbour-UCLA Medical Center, CA, USA). Hormones were labeled with ^125^I by the Chloramine T-method \[[@B19]\]. The antibody titers used were 1:62,500 for prolactin, 1:45,000 for FSH and 1:250,000 for LH assays, respectively. The volume of plasma used was 10 μl (prolactin assay), 75 μl (FSH assay) and 100 μl (LH assay). Staphylococcus aureus(prepared by the Department of Plant Physiology, U.A.M., Madrid, Spain) was used to precipitate the bound fraction \[[@B18]\]. The assays were previously validated in our laboratory \[[@B18]\]. All samples were measured in the same assay run to avoid inter-assay variations. The limits of detection for prolactin, FSH and LH were 0.125, 0.48 and 0.05 ng/mL respectively. The intra-assay coefficient of variation, calculated from a pool of plasma measured ten times in the same assay, was \< 5%. Statistical analysis -------------------- After determining that the homogeneity-of-variance assumption was tenable and that the distribution appeared unimodal and nonskewed, the statistical analysis of results was performed by a two-way factorial analysis of variance (ANOVA). Generally, the factorial ANOVA included assessment of the group effect (i.e. the occurrence of differences in mean values between control and separated groups), of time of day effects (the occurrence of daily changes) and of the interaction between the two factors (separation and time, from which inference about differences in timing and amplitude could be obtained). A one-way ANOVA followed by Student-Newman-Keuls\' test was then employed to show which time points were significantly different within each experimental group to define the existence of peaks. A Student\'s t test was performed to assess differences between the experimental groups at particular time intervals. P values lower than 0.05 were considered evidence for statistical significance. Results ======= Figure [1](#F1){ref-type="fig"} depicts the plasma prolactin levels through a 24-h cycle in control does and in does separated from their litters for 48 h. Analyzed as a main factor in a factorial ANOVA, significant time of day changes occurred (F = 35.4, p \< 0.00001) with absence of any significant effect of litter separation. Both in control and separated does, plasma level of prolactin changed throughout the day showing two maxima, at 0500 -- 0900 h and at 1700 -- 2100 h, respectively (p \< 0.001, Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### 24-h changes in plasma prolactin levels in female lactating rabbits. Groups of 6--7 animals maintained with their litters or separated from them for 48 h were killed by decapitation on day 11 post-partum, at 6 different time points throughout a 24-h cycle, starting at 0900 h. The dark bar indicates scotophase duration. Results are the means ± SEM. Letters indicate the existence of significant differences between time points in each group after a one-way ANOVA followed by a Student-Newman-Keuls\' test, as follows: ^a^p \< 0.05 vs. 09:00 and 17:00 h, ^b^p \< 0.01 vs. 09:00 h, p \< 0.05 vs 05:00 h. For further statistical analysis, see text. ::: ![](1740-3391-3-9-1) ::: Figure [2](#F2){ref-type="fig"} shows the 24-h changes in plasma FSH concentration in control does and in does separated from their litters for 48 h. Analyzed as a main factor in a factorial ANOVA, litter separation augmented FSH levels by 37 % (F = 104.6, p \< 0.00001). A significant effect of time of day and a significant interaction \"time of day x litter separation\" were found (F = 41.1 and 23.3, p \< 0.00001, respectively), i.e., a single maximum in the first half of the light period was seen in controls whereas two maxima, at 0500 -- 0900 h and at 1700 -- 2100 h, respectively, were found after litter separation. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### 24-h changes in plasma FSH levels in female lactating rabbits. Groups of 6--7 animals maintained with their litters or separated from thrm for 48 h were killed by decapitation on day 11 post-partum, at 6 different time points throughout a 24-h cycle, starting at 0900 h. The dark bar indicates scotophase duration. Results are the means ± SEM. Asterisks indicate significance differences with control at that particular time interval (Student\'s t test, \* p \< 0.05, \\ p \< 0.01). Letters indicate the existence of significant differences between time points in each group after a one-way ANOVA followed by a Student-Newman-Keuls\' test, as follows: ^a^p \< 0.01 vs. all remaining groups, ^b^p \< 0.05 vs. 17:00 h, ^c^p \< 0.01, vs 09:00, 13:00, 21:00 and 05:00 h, p \< 0.05 vs. 17:00 h. For further statistical analysis, see text. ::: ![](1740-3391-3-9-2) ::: Figure [3](#F3){ref-type="fig"} displays the 24-h changes in plasma LH concentrations. A significant effect of litter separation and time of day was apparent (F = 16.1, p \< 0.0001 and F = 2.51, p \< 0.03, factorial ANOVA). Litter separation brought about a small albeit significant 16% increase in mean circulating LH values. As shown by the significant interaction \"time of day x litter separation\" found (F = 29.4, p\< 0.00001), litter separation disrupted the plasma LH rhythm by phase-shifting its maximum by 12 h, from 1300 h in controls to 0100 h in separated mothers (Figure [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### 24-h changes in plasma LH levels in female lactating rabbits. Groups of 6--7 animals maintained with their litters or separated from them for 48 h were killed by decapitation on day 11 post-partum, at 6 different time points throughout a 24-h cycle, starting at 0900 h. The dark bar indicates scotophase duration. Results are the means ± SEM. Asterisks indicate significance differences with control at that particular time interval (Student\'s t test, \* p \< 0.05, \\ p \< 0.01). Letters indicate the existence of significant differences between time points in each group after a one-way ANOVA followed by a Student-Newman-Keuls\' test, as follows: ^a^p \< 0.05 vs 21:00, 01:00 and 05:00 h, ^b^p \< 0.05, vs 05:00 h, ^c^p \< 0.01 vs. 09:00, 21:00, 01:00 and 05:00 h, ^d^p \< 0.05 vs. 21:00, 01:00 and 05:00 h. For further statistical analysis, see text. ::: ![](1740-3391-3-9-3) ::: Discussion ========== The questions posed in the Introduction may now be answered. First, our results indicate that the 24 h patterns of plasma FSH and LH, but not of prolactin, changed significantly in nursing rabbits after litter separation for 48 h. Second, litter separation disrupted the 24-h rhythmicity of plasma FSH and LH concentration and caused a moderate increase in their concentration (when assessed as the mean 24-h values). Previous studies in the rabbit showed that litter separation for short periods of time augmented sexual receptivity and fertility of the doe \[[@B4]-[@B9]\]. Thus, the changes in gonadotropins reported herein could be a reflection of the same mechanisms involved in the massive release of gonadotropin releasing hormone (GnRH), LH and FSH triggered by mating or artificial insemination in litter-separated mothers. The rabbit exhibits an unusual form of maternal care, with a single and very short visit (3--5 min) every day to nurse \[[@B20]\]. This daily nursing visit of the doe is extremely regular, with some individuals showing a day-to-day variability of only a few minutes. Estrogen, androgen, progesterone and prolactin promote the onset of this behavior in does \[[@B21]\] while its maintenance relies on stimuli from the litter (i.e., maternal responsiveness is altered or abolished by prevention of mother/young contact at parturition or during early lactation). From a number of studies on the distribution of estrogen, androgen and prolactin receptors, quantification of expression of immediate-early genes, and lesions of structures of the olfactory system, it was concluded that rabbits rely on the same hormonal and extrahormonal factors that stimulate maternal behavior in other mammals except for the very peculiar circadian nursing pattern that is unique to rabbits \[[@B22]-[@B24]\]. Since the early observations by McNeilly and Friesen \[[@B25]\] it is known that postpartum blood levels of prolactin are similar in lactating and postpartum nonlactating females. Such an observation was confirmed in the present study in which plasma prolactin levels, measured at six time intervals in a 24-h cycle (the closest to nursing time was at 05:00 h), were essentially similar in control and litter-separated does. In lactating females, suckling evoked an immediate increase (3- to 5-fold) in circulating prolactin levels, an effect mimicked by the tactile stimulation of the teats \[[@B25]\]. Likewise, in hares, prolactin levels increased significantly during lactation only after suckling stimuli \[[@B26]\]. It must be noted that, in contrast to rabbits, plasma prolactin levels are significantly changed by nursing in most species, the suckling stimulus being an effective masking signal for the 24-h release of prolactin \[[@B27],[@B28]\]. This does not occur in the doe, the circadian changes of plasma prolactin levels remaining essentially unchanged after litter separation (presumably because of the very short nursing period). This suggests that the circadian secretion of prolactin and the prolactin response to physical stimulation of the nipples are independent phenomena that occur throughout the nursing period. Previous reports using single sampling procedures \[[@B8]\] indicated that litter separation decreased thr doe\'s prolactin levels and did not affect FSH. Discrepancies are possibly dependent on the sampling frequency and time of day examined. Collectively, the results underline the importance of performing 24-h studies to have a more precise picture of the hormonal changes. Litter separation disrupted the 24-h rhythmicity of both FSH and LH significantly. McNeilly \[[@B29]\] suggested that a reduction of plasma LH levels found during the light period could be coupled to an increase of pulsatile pattern of hypothalamic GnRH release. In the present study, the does exhibited, after litter separation, an inverse 24-h pattern of LH release with the lowest values during the light phase of daily photoperiod. Presumably, the disrupted 24-h rhythmicity of LH (and FSH) are linked to the greater mating or artificial insemination-induced release of LH and FSH found in does separated from their litters. In summary, the present study demonstrates the existence of 24-h variations in circulating prolactin, LH and FSH levels in nursing does. Litter separation for short periods of time, an effective procedure to stimulate ovarian activity prior to artificial insemination, markedly influences the secretory patterns of FSH and LH, a finding that can be related to the higher reflex ovulatory response to mating or artificial insemination observed in does separated from their pups. The specific value of the present study in terms of augmenting breeding efficiency should be further explored.	PubMed Central	2024-06-05T03:55:58.166008	2005-6-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166572/", "journal": "J Circadian Rhythms. 2005 Jun 2; 3:9", "authors": [ { "first": "Pilar", "last": "Cano" }, { "first": "Vanessa", "last": "Jiménez" }, { "first": "Maria P", "last": "Álvarez" }, { "first": "Mario", "last": "Alvariño" }, { "first": "Daniel P", "last": "Cardinali" }, { "first": "Ana I", "last": "Esquifino" } ] }
PMC1166573	Background ========== The residents of the Arab countries in the Persian Gulf region have become more sedentary and have dramatically changed their diet over the last two decades. They consume more fat, meat, sugar, rice and wheat flour than before \[[@B1]-[@B4]\]. This has resulted in a rise in obesity, diabetes, and cardiovascular disease prevalence. There is a need to study the relation between diet and chronic diseases in this population, but there is no customized instrument to do so. The Food Frequency Questionnaire (FFQ) is the most commonly used method to assess diet in relation to chronic disease \[[@B5]\]. Based on a complete search of the literature (using Medline medical subject heading and text words) and personal communication with nutritionists in United Arab Emirates and Kuwait Universities, we decided to develop a SFFQ and accompanying food composition database for UAE and Kuwait. Subjects and methods ==================== Population ---------- Our sample consisted of 326 apparently healthy persons of Arab origin, between the ages of 18 and 65 years, living in UAE and Kuwait in 2003--2004. Of these 126 provided information using the 24-hour dietary recall, and 200 participated in the testing of the long-FFQ in the SFFQ development. Development of the semi-quantitative food frequency questionnaire ----------------------------------------------------------------- To develop the SFFQ we went through the following steps: construction of a food list, definition of portion sizes, and assignment of frequency of consumption, pilot test of long-FFQ and assembling the selected food list into SFFQ. 24-hour Dietary Recall (DR) --------------------------- Qualified nutritionists interviewed the participants. The nutritionists asked the participants what they ate in the previous 24-hours in direct chronological order from the first foods in the morning to the last foods before breakfast on the day of the interview. To standardize data collection, we prepared a manual of procedures for the interviewers. The interviewers used a food atlas modified from \"Food portion sizes: A photographic Atlas\" \[[@B6]\] containing coloured pictures of 8 different portion sizes of foods commonly eaten in UAE and Kuwait. We constructed a long food list from information we obtained from the 24-hour dietary recalls, and supplemented it with foods derived from popular cookbooks, and suggestions of experienced local dieticians. We obtained the range of portion sizes from the 24-hour dietary recall and cookbooks. For items such as eggs we considered one of those items as a portion. For fruits such as banana or orange, a medium size was considered to be a \"unit\" of the fruit. For foods with non-unitary measures, such as grapes, we considered 1/2 cup to be one unit. For other foods one unit of a utensil that was mostly reported by respondents was assumed as one portion size. To get an estimate of the usual portion we used the mode of the portion size distribution for each food reported. We formatted the SFFQ based on the pattern of the Harvard FFQ. We organized the main foods by the traditional food groups into seven food categories: 1. Milk, milk products and fats (including milk with different amounts of fat, Labnah, etc.), 2. vegetables (fresh or cooked), 3. fruit, 4. meat, eggs and meat products (including meat organs), 5. cereal and cereal product, 6. beverages 7. sweet and baked goods (Faaloodah, baklava, etc.) and nuts like almond, and hamoose. To distinguish the fibre content of the diet and the quality of carbohydrates, we differentiated between the types of consumed cereal and cereal products (white or wholemeal bread, whole grain). In the last section of the FFQ we asked about type of cooking oils and fats and also, minerals including calcium, iron, zinc and multivitamins. We used 9 categories to assess frequency of intake varying from \"never or less than once a month to 6 or more times per day. For each food item, participants indicated their average frequency of consumption over the past year of a specified serving size by checking 1 of the 9 frequency categories. For foods that contain an extremely high amount of a particular nutrient but are used infrequently, such as liver, we re-categorized the options for frequency of intake. For instance, we used eliminated options of higher intake (once per day) but distinguished between never and less than once a month at the lower end. We will compute the daily intake based on the midpoint of the reported frequency category for each food item; for example we will take a response of \"2--4/week\" to be 3/7 or 0.43 times/day. Although most fruits and vegetables are available year round in UAE and Kuwait, their intakes may vary depending upon cost and cultural preferences of foods by season. For this reason we designed specific questions for fruit and vegetable consumption by season. We determined the length of the season from local experts who used their experience. We pilot tested the long-FFQ among 200 participants from the same populations (but not those who participated in the 24 hr DR). The objectives were to determine the completeness of the food list and to shorten the food list by deleting foods that were not commonly consumed. Based on the analysis of the pilot we completed the food list and created SFFQ. ### Food Composition Database We will use the food composition database to convert intakes of foods into nutrients. We constructed one nutrient database for UAE and Kuwait, as most foods are similar in both countries. We extracted the nutrient contents of the food items from Table SR17 of the USDA food database, which is available online, as the starting point to establish the database <http://www.nal.usda.gov/fnic/foodcomp/Data/SR17/sr17.html>. From SR17 we chose those varieties of food items which are not very specific to a region and are more representative, for example; for apple we chose \"apples, raw, with skin (NDB No:09003)\" or for orange, we chose \"oranges, raw -- all commercial varieties (NDB No: 09200)\". We used Statistical analyses SPSS 10.1 for windows (SPSS Inc, Chicago IL) for all analyses. Results ======= There were 326 participants with different occupations in Kuwait and UAE in our study. The populations were similar in UAE and Kuwait with respect to age, sex distribution, and BMI, but Kuwaiti women had higher education levels than those in UAE (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). No significant differences were noted between female and male respondents with respect to BMI (P = 0.4, 0.5 for UAE and Kuwait, respectively). However, in UAE, women had a higher mean BMI than men while in Kuwait men were heavier than women. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of UAE participants ::: Men (N = 35) Women (N = 91) Overall (N = 126) --------------------- ------------------ -------------------- ----------------------- Age 43.0 ± 11 34.3 ± 10 37.0 ± 11 BMI (kg/m^2^) 27.4 ± 4 28.7 ± 7 28.3 ± 7 Education None 22.7% 10% 13% Primary school 9.0% 8% 9% Secondary school 23% 30% 31% Trade School 32% 3% 2% University 23% 49% 45% Income (Dirham) \<5000 12.5% 9% 11% 5001--10000 12.5% 63% 60% 10001--15000 19% 24% 22% \>15000 12.5% 4% 6% ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Demographic characteristics of Kuwaiti participants ::: Men (N = 56) Women (N = 145) Overall (N = 201) ------------------ ------------------ --------------------- ----------------------- Age 45.6 ± 14.0 37.3 ± 12 39.6 ± 13.3 BMI (kg/m^2^) 26.6 ± 7 25.9 ± 5 26.0 ± 6 Education None 0% 0% 0% Primary school 0% 5% 4% Secondary school 32% 18% 21% Trade School 5% 4% 4% University 64% 72% 70% Income \<500 68% 31% 61% 501--1000 32% 54% 36% \>1000 0% 15% 3% ::: Daily food intake ----------------- The estimated daily intakes of seven major food groups among Kuwaiti and UAE men and women are shown in Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}. On average the participants in UAE reported eating 3.4 servings/d of fruits and 3.1 servings/d of vegetables versus 2.8 servings/d of fruits and 3.2 servings/d of vegetables in Kuwait. Cereals are an important staple in the diet of both countries and the participants reported eating cereals 4.8 times/d in UAE and 5.3 times/d in Kuwait. All participants reported consuming cereals at least once per day. Meat was consumed nearly two times/day in both countries and among the meat group, poultry was consumed more often than red meat or fish. The mean intake of dairy products was 2.2/day in UAE and 3.4/day in Kuwait. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Average daily intake of main foods estimated by long-FFQ reported by UAE participants (Men = 24, Women = 76) ::: Sex Foods Min Max Mean Std. Deviation --------- ----------------------------- --------- --------- ---------- -------------------- Women Fruits 0.2 8.5 3.0 1.7 Vegetables 0.4 7.9 3.0 1.5 Dairy products 0.0 5.2 2.1 1.0 Meat 0.6 4.2 1.7 0.7 Cereals and cereal products 0.7 10.7 4.7 2.2 Beverages 0.0 9.5 3.4 2.1 Baked goods 0.0 3.4 0.5 0.5 Men Fruits 1.8 8.8 4.8 2.0 Vegetables 0.3 6.6 3.7 1.6 Dairy products 0.4 7.0 2.5 1.3 Meat 0.6 4.4 2.3 0.9 Cereals and cereal products 1.5 6.6 5.0 1.2 Beverages 1.3 8.5 5.0 1.9 Baked goods 0.0 2.1 0.8 0.6 Overall Fruits 0.2 8.8 3.4 1.9 Vegetables 0.2 7.9 3.1 1.6 Dairy products 0.2 7.1 2.2 1.1 Meat 0.6 4.4 1.9 0.8 Cereals and cereal products 0.7 10.8 4.8 2.0 Beverages 0.0 9.5 3.8 2.2 Baked goods 0.0 3.4 0.6 0.5 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Average daily intake of main foods estimated by long-FFQ reported by Kuwaiti participants (Men = 22, Women = 78) ::: Sex Foods Min Max Mean Std. Deviation --------- ----------------------------- --------- --------- ---------- -------------------- Women Fruits 0.4 8.1 2.8 1.7 Vegetables 0.0 10.5 3.3 2.2 Dairy products 0.0 9.7 3.3 2.3 Meat 0.2 9.9 1.8 1.3 Cereals and cereal products 0.7 20.1 4.9 3.2 Beverages 0.0 11.1 2.9 2.3 Baked goods 0.0 5.0 1.0 1.0 Men Fruits 0.5 7.3 3.1 1.8 Vegetables 0.3 6.5 3.0 1.9 Dairy products 0.4 6.6 3.5 1.7 Meat 0.4 6.6 2.5 1.6 Cereals and cereal products 2.3 12.8 6.5 2.4 Beverages 0.0 9.3 4.0 2.7 Baked goods 0.0 5.7 1.2 1.5 Overall Fruits 0.4 8 2.8 1.7 Vegetables 0.00 10.5 3.2 2.1 Dairy products 0.00 9.7 3.4 2.3 Meat 0.2 9.9 1.9 1.4 Cereals and cereal products 0.7 20.1 5.3 3.1 Beverages 0.0 5.7 1.0 1.1 Baked goods 0.0 11.0 3.1 2.4 ::: Frequency of consumed food -------------------------- Tables 6 and 7 show the frequency of consumption of some foods in UAE and Kuwait. About 32% of respondents in UAE and 26% of Kuwaiti respondents reported that they had at least one glass of milk daily on average in the past year. Sixty-eight percent of UAE participants and 48% of Kuwaiti participants reported consuming rice at least once per day. 67% and 51% of people (UAE and Kuwaiti respectively) ate an egg at least 2 times/week. Overall in UAE, 86% of participants did not eat chicken with skin while in Kuwait 42% of participants did not eat chicken with skin. Among fruits, apples, oranges and bananas were consumed very frequently. Nutrient database ----------------- To convert local mixed dishes to nutrients, we created a new nutrient database appropriate for local foods (Table [5](#T5){ref-type="table"}). Therefore, a local nutritionist collected 2 recipes from each low, middle and high-income family. We supplemented these recipes from the \"Food composition: Kuwaiti composite dishes\" \[[@B7]\] and other popular cookbooks. The average amount of ingredients from those recipes was used to create a base recipe for the nutrient database. We matched ingredients of the recipes with the appropriate food items in the USDA database to obtain nutrient content, taking cooking method into account. The SFFQ for UAE is in Appendix I and the SFFQ for Kuwait is in Appendix II. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Nutrient composition per serving (100 g) of some commonly eaten foods in UAE and Kuwait ::: Vitamins (mg) ------------------------------ ----- ------ ------ ------ ------- ------- ------ ------------------- ------ ------- ------- ------ ------ ------ ----- [Mixed dish]{.underline} Kofta 173 3.1 11.1 15.0 0.6 63.1 14.1 0.09 0.26 1.7 12.6 0.1 23.7 128 1.6 Qouzi 206 15.0 11.2 10.5 0.6 5.1 36.7 0.1 0.1 0.9 2.3 0.3 18.0 90.3 1.4 Marga Laham 95 7.1 4.0 7.5 0.8 19.0 4.6 0.07 0.2 0.7 7.0 0.04 7.1 65.4 0.8 Jereesh 226 29.7 7.0 12.9 5.2 13.9 22.7 0.2 0.2 0.7 3.5 0.4 21.1 252 2.1 [Sweets]{.underline} Balalett 210 35 4.7 6.6 0.1 Trace 10 0.01 0.06 Trace Trace 0.5 28.5 105 1.5 Elba 264 30.6 12.6 11.3 Trace 206 25 0.1 0.1 1.3 1.1 0.6 522 356 2.1 ::: Discussion ========== In this paper, we have described the development of a semi-quantitative FFQ and food composition database for the Arab population in UAE and Kuwait. The goals of diet assessment in epidemiologic studies are to obtain a measure of usual rather than current diet, and rank people by intake, in contrast to clinical settings where current absolute intake is more important. The FFQ has been developed with these purposes in mind and has become the standard method to collect dietary data in studies of chronic disease all over the world. We opted to use a semi-quantitative FFQ, which estimated food intake in categories rather than the exact frequency, because it has been shown that there is minimal loss of information in estimating nutrient intakes using food intake categories \[[@B8]\]. We also asked the participants about intakes of pre-specified portion sizes rather than asking them to estimate their regular portion size. Correlations for nutrient intake calculated using the FFQ with and without taking portion sizes into account were over 0.9 \[[@B5]\]. The advantage of using categories to estimate food intake and pre-specified portion sizes is that the SFFQ becomes easier to administer, and likely, more reliable. We did not attempt to make a comprehensive list of foods to include in the SFFQ. Rather, we kept items in the SFFQ if they were nutrient rich, consumed frequently and discriminated intakes between individuals. The other criterion we considered together with the nutrient content (including caloric value) was the presence of other substances of interest, for instance, caffeine. Most FFQs have between 100--150 items \[[@B9]\] and our SFFQ has 153 (UAE) and 152 (Kuwait) items. Increasing the number of items in the FFQ has been shown to increase over-reporting \[[@B10]\]. To estimate nutrient intake from SFFQ, there is a need for a food composition table listing the average nutrient content of foods contained in the SFFQ. To obtain nutrient intake we multiplied the average nutrient content of a specified portion of food listed in the food composition table by the average frequency of intake reported in the SFFQ. The food composition table can be a substantial source of variation in the estimation of nutrients using the SFFQ. As no nutritional database has ever been gathered in UAE or Kuwait, we used the US Department of Agriculture nutrient database as our standard to estimate nutrient content. The advantages of this approach are: First, the USDA food composition database is probably the most comprehensive in the world. For example, there are 26 categories of spinach including different types of spinach, raw spinach, and spinach cooked in a variety of ways \[[@B11]\], allowing us to choose the most appropriate one. Second, the nutrient estimation assays have been done in a standardized manner. Third, it has the largest number of nutrients reported. Fourth, the USDA food composition database is continually updated. Last, UAE and Kuwait import foods from all around the world and a mixture of food items from different regions are available in any market. For mixed dishes that were not listed in the USDA database, we calculated nutrient intake by analyzing recipes. Moreover, there are nearly 150 food composition tables in use around the world and their values are primarily based on USDA \[[@B12]-[@B14]\], and even European countries include nutrient information from USDA in their food composition tables \[[@B15],[@B16]\]. Finally, similar approaches have been taken by other investigators in Israel, \[[@B17]\] and Costa Rica \[[@B18]\]. A limitation of this study is that the age groups represented in the UAE and Kuwait sample are mostly \<50 years for both males and females. Thus, the overall impression in dietary habits is biased towards this younger group. For example, the consumption of rice as well as dates might be underestimated. The way to make it more accurate is of course to repeat it (validate). Another limitation of this study is that most participants from both countries were women and some foods, which men may eat, may be underestimated. However, a nutritionist with experience in those countries reviewed the food lists to ensure their completeness. Conclusion ========== The validated questionnaire and food composition database will not only be useful tools for our own study, but they will also be assets that other researchers in the region can use or adapt to suit their needs. We have enclosed two SFFQs in this article so other researchers in the field of public health can use this comprehensive SFFQ. We are currently evaluating SFFQs and the validated SFFQ will be available online for all public health researchers in the region. List of Abbreviations ===================== SFFQ Semi-Quantitative Food Frequency Questionnaire 24 hr DR 24 hour Dietary Recall UAE United Arab Emirates PURE Prospective Urban and Rural Epidemiologic SPSS Statistical Packages for Social Sciences Authors\' contributions ======================= MD Participated in design of study, coordinated and performed statistical analysis, drafted the manuscript AM Participated in design of study, performed statistical analysis, helped to draft the manuscript NH Coordinated study in Kuwait, helped to draft the manuscript AY Facilitated data collection in UAE, helped to draft the manuscript FN Coordinated study in UAE SY Participated in design of study, helped to draft the manuscript All authors read and approved the final manuscript. Acknowledgements ================ The authors acknowledge the contribution of Mrs. Sumathy Rangarajan who supported the fieldwork during the data collection phase and are grateful to the nutritionists who collected the dietary data in Kuwait and UAE and Mrs. Pam Mackie for secretarial support. We are grateful to the participants of the study for their cooperation. We also wish to thank Dr. Yusuf (Director of Population Health Research Institute) for all his support and guidance. We also acknowledge the \"Sheikh Hamdan bin Rashid Al Makhtoum Award for Medical Sciences\"for their Grant no.MGR-16/2001--2002 for the UAE portion of the study.	PubMed Central	2024-06-05T03:55:58.167288	2005-5-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166573/", "journal": "Nutr J. 2005 May 27; 4:18", "authors": [ { "first": "Mahshid", "last": "Dehghan" }, { "first": "Nawal", "last": "Al Hamad" }, { "first": "AfzalHussein", "last": "Yusufali" }, { "first": "Fathimunissa", "last": "Nusrath" }, { "first": "Salim", "last": "Yusuf" }, { "first": "Anwar T", "last": "Merchant" } ] }
PMC1166574	Background ========== Apoptosis is a type of programmed cell death and is a natural phenomenon occurring when the cells are subjected to stress such as DNA damage, death signals or lack of growth factor. Apoptotic stimuli allow an intracellular cascade of signals such as the caspases, a family of cysteine proteases implicated in the cleavage of a number of important proteins which results in cell disassembly and cell death, phagocytosis and removal of cell debris by immune cells. Apoptosis plays an important role during embryo implantation in rodents where morphological characteristics of apoptosis are observed in endometrial epithelial cells at the embryo implantation site \[[@B1]-[@B3]\]. Moreover, this phenomenon also occurs during late pregnancy, especially during regression of the decidua basalis (DB) in the rat endometrium \[[@B4],[@B5]\]. Two decidual zones are formed during pregnancy: the primary decidual zone on the antimesometrial side of the uterus and the secondary decidual zone (or antimesometrial decidua) which is formed following expansion of the primary decidual zone \[[@B5],[@B6]\]. The secondary decidual zone eventually transforms stromal cells in the mesometrial region to form the DB that regresses following day 14 of pregnancy \[[@B7]\]. Whether the phenomenon of growing size of embryo is a cause or correlation to the increase of apoptosis remains to be elucidated. The first members of the transforming growth factor-beta (TGF-β) superfamily were identified on the basis of their ability to induce a transformed phenotype of certain cells in culture \[[@B8]\]. They are now known as multifunctional polypeptides involved in the regulation of cell proliferation and differentiation, immunoregulation, angiogenesis and the regulation of extracellular matrix \[[@B9],[@B10]\]. They act via cellular signalling through Smads and Phosphorylated-Smads (P-Smads), the active form of Smads. Those proteins are translocated to the nucleus and activate transcription factors which in turn activate caspases and other regulation proteins \[[@B11],[@B12]\]. Another characteristic of TGF-β is its capacity to induce apoptosis in several cell types \[[@B7],[@B13]\]; in fact, TGF-β was shown to have a pro-apoptotic function mediated by caspases \[[@B14]-[@B16]\]. Genes encoding the three isoforms are localized on different chromosomes and the isoforms molecular weights are slightly different: 15, 12.5 and 12 KDa for TGF-β1, β2 and β3 respectively; they share 80 % sequence identity and are produced in latent forms which are activated into a 112-amino acid mature peptide \[[@B17]\]. Multiplicity of TGF-β isoforms and sequence conservation within each form through evolution suggests important specific roles. Moreover, it has been demonstrated that TGF-β1, -β2 and -β3 are differently expressed in the mouse uterus \[[@B18]\] and porcine conceptus-maternal interface \[[@B19],[@B20]\]. The uterus is a hormone-dependent organ and it is subject to an abundant amount of cellular proliferation and cell death. Studies have shown that apoptosis is increased in the rat endometrium during implantation and regression of the decidual basalis in the rat \[[@B1],[@B2],[@B21]\]. The mRNA for TGF-β1 has been shown to be present within the uterus during rat pregnancy and was localized to the luminal and glandular epithelial cells during early and late pregnancy \[[@B22]\]. TGF-β1 and -β2 mRNAs were also found in the mouse uterus during pregnancy \[[@B23]-[@B25]\]. Expression of TGF-β2 and TGF-β3 mRNAs were also shown to be expressed in the mouse periimplantation uterus \[[@B18]\]. A study showed that TGF-β1 and -β2 treatments on cultured endometrial rat stromal cells induced apoptosis \[[@B7]\]. These studies suggest that TGF-β isoforms might be involved specifically in the control of apoptosis in the uterus during pregnancy. However, the mechanisms involved in the control of apoptosis in the endometrium during pregnancy are poorly documented. Studies suggesting the importance of TGF-β isoforms at specific time of pregnancy were done at the mRNA level and it is important to determine their presence at the protein level. Thus, the aim of this study was to determine the expression and the developmental expression of TGF-β1, β2 and β3 proteins in the rat uterus throughout pregnancy and to further determine in vitro, the effect of TGF-β in determining decidual cell fate. We found that the three isoforms of TGF-β were expressed and regulated differently in epithelial and stromal endometrial cells during rat pregnancy and showed using primary decidual cell cultures the involvement of TGF-β in the regulation of programmed cell death. Moreover, the present study shows that TGF-β signals through Smad2, which coincide with XIAP and Akt down regulation and induction of apoptosis. Methods ======= Reagents -------- TGF-β1 (sc-146, lot \# F262, 200 μg/ml), TGF-β2 (sc-90, lot \# B202, 200 μg/ml) and TGF-β3 (sc-82, lot \# A222, 200 μg/ml) polyclonal antibodies were purchased from Santa Cruz Biotechnology, Inc (Santa Cruz, CA, USA). CDC47/MCM7 antibody was obtained from Medicorp (Montréal, QC, Canada). Phospho-Akt (Ser 473), Akt, XIAP, Cleaved caspase-3, and Phospho-Smad2 (Ser 465 / 467) antibodies were obtained from Cell Signaling Technology (Beverly, MA, USA). The Keratin 8/18 antibody used to determine cell culture purity was donated by Dr Monique Cadrin (Univ. of Québec at Trois-Rivières, QC, Canada). Anti-Smad2/3 antibody was purchased from Calbiochem (San Diego, CA, USA). Vectastain ABC Kit for rabbit IgG was purchased from Vector Laboratories Inc. (Burlingame, CA, USA). In SituCell Death detection kit (TUNEL), POD and DAB substrate was obtained from Roche (Laval, QC, Canada). TGF-β1 recombinant protein was purchased from Biosource (Cat \# PHG9104, lot \# 16865-01S, 5 μg, diluted at 50 μg/ml, QC, Canada). Animals ------- Sprague-Dawley female rats, 200--225 g, were obtained from Charles River Laboratories Canada. Animals were maintained on standard chow and water, which were available ad libitum, in animal facilities illuminated between 6:00 h and 20:00 h. All procedures were performed in accordance with guidelines of the Canadian Council on Animal Care for the handling and training of laboratory animals and the Good Health and Animal Care Committee of the Université du Québec à Trois-Rivières. Male and female mice were mated overnight and confirmation of pregnancy was determined by vaginal smears and/or the presence of a vaginal plug (day 1). Rats were killed on day 2, 4, 5, 6, 8, 10, 12, 14, 16, 18 and 20 of pregnancy at 10:00 h in the morning and at 18:00 h for days 5.5 and 6.5. Six to 8 different rats were used for each time of pregnancy. Uteri were collected and fixed for immunohistochemical staining (IHC) and apoptotic cell death detection by \[TdT\]-mediated deoxyuridinetriphosphate nick end-labeling (TUNEL) or endometrial protein extracts collected for Western blot analysis. Rat pretreatments and decidual endometrial stromal cell culture --------------------------------------------------------------- A total of 10 rats were ovariectomized and then allowed to recover from surgery for a minimum of 10 days. They were pre-treated with physiological doses of estradiol (1,3,5(10)-Estratriene-3,17β-diol, Sigma-aldrich) and progesterone (Laboratoire Mat, PQ) to induce decidualization as described previously \[[@B26]\]: 1) 0.2 ug estradiol injection per day for three days (in the morning, day -2,-1 and 0); 2) On the third day (day 0 of pseudopregnancy), another injection in the afternoon of estradiol (0.2 μg) and progesterone (1 mg) was performed; 3) No treatment for 2 days (day 1 and 2 of pseudopregnancy); 4) Injections of estradiol (0.1 μg) and progesterone (4 mg) for three days (day 3, 4 and 5 of pseudopregnancy); 5) Another injection of estradiol (0.1 μg) in the afternoon on day 7 (day 4 of pseudopregnancy); 6) Rats were killed on day 8 (day 5 of pseudopregnancy). All endometrial stromal cells collected for cultures were recovered from rats treated with the protocol described above. Uteri were removed and horns taken and immerged in HBSS solution containing HEPES (20 mM), penicillin (100 units/ml), streptomycin (100 μg/ml) and fungizone (1,25 μl/ml) (Invitrogen, ON, Canada). Further manipulations were performed in a sterile environment. The uterine horns were transferred into a sterile petri containing HBSS, slit longitudinally and immersed in trypsin type I solution (0.3%) (Roche Diagnostics, QC, Canada) in HBSS and agitated for 60 minutes at room temperature. Uterine horns were then vortexed at maximum for 5 sec and supernatant containing epithelial cells was discarded. Uterine horns were washed three times with 2.5 ml of HBSS and immersed in a HBSS solution containing trypsin type I (0.03%), DNAse I (0.016%) and collagenase type II (0.064%) for 15 minutes at 37°C in a water bath. Uterine horns were then vortexed at maximum for 5 sec. The supernatant containing stromal cells was transferred into a sterile falcon tube containing 150 μl of FBS D.C (Dextran-Charcoal extracted). Uterine horns were washed two times with 2.5 ml of HBSS and the supernatant was mixed with stromal cells. Uterine horns were discarded and stromal cells were centrifuged at 1000 g for 5 minutes. Cells were washed two times with HBSS and centrifuged. The supernatant was discarded and cells diluted with DMEM-F12 (Ph 7.1) (Invitrogen, ON, Canada) containing 2.438 g/L NaHCO~3~, 10% FBS D.C. and gentamycine 50 μg/ml. Cells were incubated at 37°C in an atmosphere of 5% CO~2~. Cells were plated in 6-well plates (Corning plates) at a density of 50% (4 × 10^5^cells per well). The medium was changed two hours after the first incubation in order to eliminate epithelial cell contamination from stromal cell cultures. The purity of stromal cells was more than 97%: cell culture contamination with epithelial cells was evaluated by cellular morphology and immunofluorescence using a Keratin 8/18 antibody. Three to 5 days after plating (more than 90% of confluency reached), cells were treated for 24 hours in the presence or absence of increasing doses of TGF-β recombinant protein. Total proteins from treated cell cultures were extracted using TRIZOL (Invitrogen, ON, Canada). For Western blot analyses, 15 μg of total protein was used for each analysis. Immunohistochemical staining ---------------------------- The uterus was fixed in 4% paraformaldehyde solution and embedded in paraffin. Tissue sections 7 μm thick were mounted on polylysine-coated slides, deparaffinized, rehydrated, and then heated in 10 mM citrate buffer (pH 6) containing triton X-100 (Sigma-Aldrich) 0.1% (v/v). After two washes with PBS, slides were then incubated with 0.3 % hydrogen peroxide in methanol for 30 min to quench endogenous peroxidase activity. After washing with PBS, tissues were incubated with blocking serum (Vectastain ABC Kit) at room temperature for 1 h. Then, a primary antibody diluted in blocking serum (TGF-β1, β2 or β3; 1:50 dilution or CDC47/MCM7; 1:100 dilution) was added to the slides and incubated at 4°C overnight in a humidified chamber. After washing 5 min. in PBS, tissue sections were incubated for 30 min. with 3 μg/ml biotinylated antibody (anti-rabbit or anti-mouse). Subsequently, slides were washed with PBS and incubated with avidin-biotin complex reagent containing horseradish peroxidase for 30 min. Slides were washed with PBS for 5 min and color development was achieved using DAB substrate. The tissue sections were counterstained with haematoxylin. Negative controls were performed using the same protocol but substituting the primary antibody with normal rabbit IgG (Vector Laboratories Inc., Burlingame, CA, USA). Immunofluorescence ------------------ Tissues were prepared as described in the immunohistochemical section. Cleaved caspase-3 antibody was diluted 1:100 in blocking serum and slides were incubated at 4°C overnight. After washing twice for 5 min. in PBS, tissue sections were incubated for 30 min. at room temperature with 2 mg/ml Alexa Fluor 488 donkey anti-rabbit (1:50). Subsequently, slides were washed with PBS and mounted. Negative controls were performed using the same protocol but substituting the primary antibody with normal rabbit IgG. Sections were examined using an OlympusBX60 microscope equipped with a Coolsnap-pro CF digital camera (Carsen Group, ON, Canada). TdT-mediated deoxyuridinetriphosphate nick end-labeling (TUNEL) --------------------------------------------------------------- Tissue sections were deparaffinized, rehydrated and rinsed with PBS. They were incubated with proteinase K (20 μg/ml) for 30 min. at room temperature. Slides were washed twice with PBS, the endogenous peroxidase was quenched with 0.3 % hydrogen peroxide in methanol for 30 min. The slides were rinsed and incubated with 10 mM citrate solution for two min on ice. Then, tissue sections were rinsed with PBS and incubated with TdT labelling reaction (In SituCell Death Detection, POD) for 30 min at 37°C in humidified environment. Slides were washed three times in PBS and tissue sections were blocked with 3% BSA for 20 min. at room temperature. Converter-POD solution was added to the slides and incubated for 30 min. at 37°C in humidified environment. Slides were washed for 5 min. in PBS, colour development was achieved using DAB substrate and counterstained with haematoxylin. Negative controls were performed using the same protocol without TdT enzyme. Protein extraction and Western analysis --------------------------------------- Protein homogenates from pregnant endometrium were isolated according to a protocol previously described \[[@B27]\]. Briefly, uteri from Day 2 to Day 20 pregnant rats were rapidly excisedand placed in ice-cold saline until dissected. Uteri were carefully laid on a glass plate and placed on the stage of a dissecting microscope. In early pregnancy (Day 2 to 5.5), total endometrium was scraped using a microscope glass and collected. Uteri from Day 6 to 10 the placenta and decidua were at an early stage of differentiation and could not be reliably separated. For this reason, DB dissectedfrom animals between these days of pregnancy contain some chorioallantoic cells, but antimesometrial decidua, choriovitelline tissues, fetus, and myometrium were removed. Even though we carefully dissected DB from these tissues, it is a possibility that a contamination with some antimesometrial decidual cells that regress to form the deciduas caspularis (DC) would occur. This is an important fact that we need to take into consideration. In uteri collected from Day 12 to 20 pregnant rats, DB were isolated by gently separating the placenta and myometrial regions with 23-gauge needles. Additionally, the DB began to regress on Day 14 and became too thin to reliablydissect after Day 17. The protocol for DB isolation was described previously by Ogle and George \[[@B28]\]. Endometrial cells from pregnant animals were homogenized using a pipette in RIPA lysis buffer (PBS 1× pH 7.4; 1% Nonidet P-40; 0.5% Sodium deoxycholate; 0.1% SDS; Protease Inhibitor Cocktail Tablets (Roche Diagnostics Canada, PQ)). Homogenates were centrifuged (12,000 × g for 20 min at 4°C) to remove insoluble material. The supernatant was recovered and stored at -20°C pending analysis. Protein content was determined with the Bio-Rad DC Protein Assay. Protein extracts (50 μg) were heated at 94°C for 3 min, resolved by 10% SDS-PAGE and electrotransferred to nitrocellulose membranes using a semidry transfer (Bio-Rad, Mississauga, ON). The membranes were then blocked 2 h at room temperature with PBS containing 5 % milk powder, then incubated with anti TGF-β 1-2-3 1:1000 ; P-Smad2 (Ser 465 / 467) 1:1000 and Smad 2/3 1:1000 and subsequently with horseradish peroxidase-conjugated anti-rabbit or anti-mouse secondary antibody (1:3000; room temperature for 45 min). All membranes were stripped with Restore western blot stripping buffer (Pierce, \# 21059, lot \# FH71541), reprobed with an antibody specific to β-actin which was used as an internal standard. Peroxidase activity was visualized with the Super signal^®^West Femto maximum sensitivity substrate (Pierce, Arlington Heights, IL, USA), according to the manufacturer\'s instructions. Signal was visualized using the Biochemi Imaging System (UVP, CA, USA). Densitometrical analyses were performed (protein of interest and β-actin) using the GelDoc 2000 and the Quantity One software (Bio-Rad, Mississauga, ON, Canada). Results are expressed as a ratio (protein of interest/β-actin) to correct for loading for each endometrial sample. Hoechst and trypan blue exclusion staining ------------------------------------------ Following TGF-β treatment, both floating and attached cells were resuspended in PBS containing Hoechst 33258 for 24 hours at 4°C or resuspended in trypan blue solution (0,4%) for 5 minutes. Hoechst nuclear staining was viewed and photographed using a Olympus BX60 fluorescence microscope and a Coolsnap-Pro CF digital Camera (Carsen Group, ON, Canada). Cells with typical apoptotic nuclear morphology (nuclear shrinkage, condensation and fragmentation) were identified and counted using randomly selected fields on numbered photographic slides, of which the \"counter\" was not aware of the treatment, so as to avoid experimental bias. A minimum of 200 cells per treatment group were counted in each experiment. For trypan blue exclusion test, blue cells were counted under a regular microscope and were counted as non-living cells. Statistical analysis -------------------- Western analyses of pregnant animals were repeated six to eight times (6 to 8 different endometrial extract per day of pregnancy from 6 to 8 different rats). Endometrial extracts from each rat were assessed individually. Western analyses of cultured decidual cells were repeated 5 times for each TGF-β dose (for each culture experiment, decidual cells were recovered from a pool of ten different ovariectomized/treated rats). Results subjected to statistical analyses were expressed as mean ± SEM. Data were subjected to one-way ANOVA (PRISM software version 4.0; GraphPad, San Diego, CA). Differences between experimental groups were determined by the Tukey\'s test. Results ======= Apoptosis expression during pregnancy ------------------------------------- In order to confirm the presence of apoptosis in DB, the presence of the activated form of caspase-3 was measured by Western analysis (Fig. [1A](#F1){ref-type="fig"}) and immunofluorescence (Fig. [1B](#F1){ref-type="fig"}) using a day 14 pregnant uterine section. TUNEL measurement was also performed using a day 14 pregnant uterine section (Fig. [1C](#F1){ref-type="fig"}). As demonstrated by Western analysis, Fig. [1A](#F1){ref-type="fig"} clearly demonstrates that apoptosis was present in the endometrium at day 14: the 17 KDa cleaved caspase-3 fragment was significantly and gradually increased from day 8 and was maximal at day 14. Cleaved caspase-3 fragment was observed in the cytoplasm of apoptotic cells as demonstrated by immunofluorescence (Fig. [1B](#F1){ref-type="fig"}) and TUNEL positive cells were also found in the in the endometrium at day 14 (Fig. [1C](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Detection of apoptosis in pregnant endometrial tissues as demonstrated by Western and TUNEL analyses. A) Apoptosis as determined by Western analysis of cleaved caspase-3 (one blot presented out of 6). β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis of the 17 KDa cleaved fragment. B) Cleaved caspase-3 immunofluorescence at day 14 of pregnancy. Magnification: control: 10× ; inset and Day 14: 40×. C) TUNEL analysis at day 14 of pregnancy. B) and C) One representative section is presented out of 6 experiments. C-: negative control. Magnification: control: 10× ; inset and Day 14: 40×. Arrows indicate positive staining. ::: ![](1477-7827-3-20-1) ::: Expression of TGF-β1, β2 and β3 in the rat uterus during pregnancy ------------------------------------------------------------------ To document the presence and expression of TGF-β proteins in the uterus throughout pregnancy, IHC and Western analyses were performed on uterine sections and lysate of pregnant rats respectively. It was important to document the presence of TGF-β proteins since the information found in the literature shows principally mRNA expression of different TGF-β isoforms in specific periods of pregnancy rather than in the whole gestation period. Western blot analyses shown in figure [2](#F2){ref-type="fig"} demonstrate that TGF-β1 and β2 are both expressed in a similar pattern. Their expression is increased following implantation (days 5.5 to 6.5) and is maximal during regression of the DB (day 14; p \< 0.01). However, the localization of the expression of those two isoforms during pregnancy is slightly different (Fig. [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"} respectively): immunohistochemicals analysis confirms that during early pregnancy the signal is found in both epithelial and stromal cells, but during late pregnancy TGF-β1 is expressed mainly in stromal cells while TGF-β2 is located in epithelial cells. It is interesting to see that both TGF-β1 and β2 isoforms are clearly present in the epithelial cells and in stroma at the time of implantation (day 5.5) all around the uterine lumen surrounding the implanting conceptus. On the opposite, TGF-β3 has not been found during early pregnancy (Fig. [2](#F2){ref-type="fig"} and [5](#F5){ref-type="fig"}). However, TGF-β3 was increased and present at the time of DB regression (days 12 to 16; p \< 0.001) suggesting that its action might be limited to decidual regression during pregnancy. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### TGF-β1, -β2 and -β3 expressions in rat endometrium during rat pregnancy. Total endometrial proteins were collected at different days of pregnancy. β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis. Data represent the mean ± SEM of six independent experiments (six different rats). C+: positive control (day 14 endometrial protein extract). ::: ![](1477-7827-3-20-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Immunohistochemistry of TGF-β1 in rat endometrium during pregnancy. IHC shown are from one representative experiment and were repeated 6 times using 6 different uterine sections from 6 different rats per day of pregnancy. Representative days of pregnancy are presented (A: day 4; B: day 5.5; C: day 6.5; D: day 10; E: day 14; F: negative control in which primary antibody was absent). le: luminal epithelium; em: embryo; s: stroma. Magnification: 40×. ::: ![](1477-7827-3-20-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Immunohistochemistry of TGF-β2 in rat endometrium during pregnancy. IHC shown are from one representative experiment and were repeated 6 times using 6 different uterine sections from 6 different rats per day of pregnancy. Representative days of pregnancy are presented (A: day 4; B: day 5.5; C: day 6.5; D: day 10; E: day 14; F: negative control in which primary antibody was absent). le: luminal epithelium; em: embryo; s: stroma. Magnification: 40×. ::: ![](1477-7827-3-20-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Immunohistochemistry of TGF-β3 in rat endometrium during pregnancy. IHC shown are from one representative experiment and were repeated 6 times using 6 different uterine sections from 6 different rats per of day pregnancy. Representative days of pregnancy are presented (A: day 4; B: day 5.5; C: day 6.5; D: day 10; E: day 14; F: negative control in which primary antibody was absent). le: luminal epithelium; em: embryo; s: stroma. Magnification: 40×. ::: ![](1477-7827-3-20-5) ::: Expression of Smad2 and Phospho-Smad2 during pregnancy ------------------------------------------------------ TGF-β signal transduction is mediated intracellularly by Smad proteins, including the R-Smads (receptor regulated Smads including Smad2), the I-Smads (inhibitor Smads), and the Co-Smad (common Smad). The triggering event in Smad activation is the type I receptor-dependent sequential phosphorylation of the two C-terminal serine residues in Smads \[[@B29]\]. Therefore, phosphorylation of the C-terminus of receptor-activated Smads, particularly Smad2, is crucial for initiation of the TGF-β signaling. Western blot analyses of Smad2 and phospho-Smad2 on endometrial cell lysate from pregnant rats were carried out to confirm that TGF-β isoforms present in the endometrium during pregnancy have an activity on those cells. The results demonstrate that Smad2 expression is regulated throughout pregnancy (Figure [6](#F6){ref-type="fig"}). Its expression is high from day 5 to 10 and is dramatically reduced from day 12 to the end of pregnancy (p \< 0.05). However the levels of phospho-Smad2, the activated form, are high on days 12 and 14 and are gradually reduced to the end of pregnancy. The presence of high levels of phospho-Smad2 correlates with the high expression of the three TGF-β isoforms and the presence of apoptosis, with an exception on day 2 of pregnancy (Figure [6](#F6){ref-type="fig"}). Although high levels of phospho-Smad2 are observed, levels of TGF-β1 and -2 isoforms are higher when compared to days 4 and 5 and caspase-3 cleaved fragment is also present. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Smad2 and phospho-Smad2 (P-Smad2) expressions in rat endometrium during pregnancy. Total endometrial proteins were collected at different days of pregnancy. β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis. Data represent the mean ± SEM of six independent experiments (six different rats). ::: ![](1477-7827-3-20-6) ::: TGF-β action in vitroon decidual endometrial stromal cell fate ---------------------------------------------------------------- As demonstrated previously by Moulton in 1994 \[[@B7]\], TGF-β1 induces DNA cleavage in rat decidual cells. To confirm that TGF-β1 is responsible for the induction of apoptosis, three different techniques were used to measure apoptosis (Figure [7](#F7){ref-type="fig"}). Hoechst nuclear staining (Figure [7A](#F7){ref-type="fig"}) and TUNEL analysis (Figure [7C](#F7){ref-type="fig"}) clearly demonstrated that TGF-β1 induced apoptosis in a dose-dependent manner (p \< 0.0001). Apoptosis was increased to 20% at 1 ng/ml of TGF-β1 and up to 30% at 10 ng/ml. Trypan Blue exclusion staining assay was used to test for viability and cell death; even though this test is not an apoptotic specific test, it shows a direct effect of TGF-β1 on cell survival and viability. A dose-dependent increase of cell death was observed in response to TGF-β1 treatment (Figure [7B](#F7){ref-type="fig"}). ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Effect of TGF-β1 (ng/ml) on cell survival in cultured rat endometrial cells as demonstrated by Hoechst staining, TUNEL and trypan blue exclusion analyses. A) Apoptosis as determined by Hoechst nuclear staining. Data represent the mean ± SEM of six independent experiments. All doses are significantly different from control (p \< 0.001). B) Cell viability as determined by trypan blue exclusion assay. Data represent the mean ± SEM of six independent experiments. 1, 5 and 10 ng/ml TGF-β1 doses are significantly different from control (p \< 0.0001). C) Apoptosis as determined by TUNEL assay. Representative fields are presented out of 6 experiments. Arrows indicate positive staining. ::: ![](1477-7827-3-20-7) ::: Effect of TFG-β on Akt, P-Akt, CDC-47 and XIAP expression in vitro on decidual endometrial stromal cells -------------------------------------------------------------------------------------------------------- To further determine how TGF-β might act at the intra-cellular level to induce apoptosis in decidual cells, experiments were carried out to determine the possible interaction of TGF-β and the PI3K/Akt survival pathway. Recent studies have suggested and demonstrated that TGF-β directly acts with Smad3 to regulate the sensitivity to TGF-β induced apoptosis \[[@B30],[@B31]\] and another study showed that TGF-β exerts a largely inhibitory effect on basal meningioma proliferation possibly through Smad 2/3 \[[@B32]\]. Since Smad3 is directly involved in TGF-β signaling, a similar mechanism might be involved in the control of decidual cell fate. Another study has shown that Akt activity might in turn be affected by the presence or absence of inhibitor of apoptosis proteins such as XIAP \[[@B33],[@B34]\]. Figure [8](#F8){ref-type="fig"} shows that Phospho-Smad2, the activated form, is significantly increased in response to TGF-β1. However, the concentration-response relationship in term of Smad phosphorylation appears to be biphasic, with 1 ng/ml stimulating, while 10 ng/ml having a significant effect compared to control (p \< 0.05) but lower than the 1 ng/ml dose. There was no significant difference observed in term of Smad2 protein expression in response to TGF-β1. As demonstrated in Figure [8](#F8){ref-type="fig"}, the proliferation marker CDC-47 was significantly decreased in response to TGF-β1 treatment. Figure [9](#F9){ref-type="fig"} demonstrates that Phospho-Akt, the active form of Akt, was highly expressed in control cells, indicating that this pathway is active and important in cell proliferation and cell survival. However, in the presence of TGF-β1, Phospho-Akt was significantly decreased suggesting an interaction of TGF-β and the PI3K/Akt survival pathway. Total Akt protein expression was not influenced by TGF-β1 treatment. Moreover, XIAP protein expression, a well known inhibitor of apoptosis protein, was significantly reduced in response to increasing doses of TGF-β. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Expression of Smad2, P-Smad2 and CDC47 in cultured rat endometrial cells in vitroas demonstrated by Western blot analyses in response to TGF-β1 (ng/ml). β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis and are the mean ± SEM of four independent experiments. \Significantly different from control (p \< 0.05). ::: ![](1477-7827-3-20-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Expression of Akt, P-Akt and XIAP in cultured rat endometrial cells in vitroas demonstrated by Western blot analyses in response to TGF-β1(νγ/μλ). β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis and are the mean ± SEM of four independent experiments. \Significantly different from control (p \< 0.05). ::: ![](1477-7827-3-20-9) ::: Discussion ========== Transforming growth factor-β isoforms have been known to be expressed and regulated differently in several types of tissues such as ovine uterus \[[@B35]\], human colon carcinoma \[[@B36]\] and at the porcine conceptus-maternal interface \[[@B19]\]. They are also recognized as pro-apoptotic factors in many cell types including fetal rat hepatocytes \[[@B14]\] and human leiomyoma smooth muscle cells \[[@B37]\]. However, little is known about the role of TGF-β isoforms in the rat uterus during pregnancy. The aim of the present study was to investigate the pro-apoptotic functions of TGF-β isoforms in rat uterus during pregnancy and also to determine more precisely the different patterns of expression of those isoforms in pregnant endometrium. The results presented in this study demonstrate that, as observed in other tissues, TGF-β isoforms (β1, β2 and β3) are differently regulated in rat uterus throughout pregnancy. It is already known that apoptosis is induced during embryo implantation and decidualization in rodents \[[@B1],[@B2],[@B7]\]. Our results confirm that apoptosis is induced in the pregnant rat uterus especially during regression of the decidua basalis. Caspases are well known executioners of apoptosis \[[@B15],[@B38]\]. The highest concentration of cleaved caspase-3 protein, the activated form of caspase-3, was found on day 14 of pregnancy, at the time of DB regression. The presence of active caspase-3 shows that this pathway might be important to induce cleavage of critical survival proteins and to further stimulate apoptosis during the regression of DB. However, caspase-3 activation was weakly observed at the time of implantation and this might be explained by the fact that, although uterine epithelium undergoes degeneration in the presence of embryo \[[@B4]-[@B6]\], only some epithelial cells undergo apoptosis during this process at the implantation site. Western blot and IHC analyses might not be sensitive enough techniques to detect the presence of small levels of cleaved caspase-3 in epithelial cells at implantation. It is also a possibility that degeneration of the epithelium may also involve non-apoptotic pathways. The logical pursuit of this study was to determine the expression of TGF-β isoforms during those critical stages of pregnancy. The present results are consistent with previous work published in regards to TGF-β isoforms in other systems and in mouse uterus \[[@B18],[@B22],[@B39]\], where expression and regulation is different for each isoform. TGF-β1 was already known to induce apoptosis in human \[[@B40]\] and rat \[[@B7]\] endometrial stromal cells. In the present study, TGF-β1 and TGF-β2 were found throughout pregnancy and were particularly strong during apoptotic phases such as regression of the DB. On the other hand, using both IHC and Western analyses, TGF-β3 protein was undetectable in the early stages of gestation, suggesting that TGF-β3 may not be required in the regulation of cell death during embryo implantation and early pregnancy. This result supports a study performed by Das et al. \[[@B18]\] showing that TGF-β3 mRNA was absent during early pregnancy. Nevertheless, the three mammalian forms of TGF-β are strongly expressed during regression of the DB suggesting they could all play an important role during the regulation of programmed cell death to induced regression of DB. Those three isoforms are also expressed differently in distinct cells of the same tissues; similarities have been observed in human normal and malignant prostate epithelial cells \[[@B39]\] as TGF-β2 and TGF-β3 are more expressed in epithelial cells than in stromal cells during late pregnancy. The strong expression of TGF-β1 during rat early pregnancy (day 5.5 to 6.5) may be explained by the fact that this isoform could be necessary to initiate embryo implantation processes during the periimplantation period, a similar situation that is observed during trophoblast invasion at the time of human embryo implantation \[[@B41]\]. The next logical step of the present study was to determine if the TGF-β isoforms present during pregnancy were active. A lot of information can be found in many physiological systems regarding TGF-β and its cellular receptors \[[@B8],[@B10]\] and Smad proteins responsible for its intracellular signal transduction \[[@B42]-[@B44]\]. Recent studies have suggested and demonstrated that TGF-β directly acts with Smad3 to regulate the sensitivity to TGF-β induced apoptosis \[[@B30],[@B31]\] and another study showed that TGF-β exerts a largely inhibitory effect on basal meningioma proliferation possibly through Smad 2/3 \[[@B32]\]. Since Smad3 is directly involved in TGF-β signaling, a similar mechanism might be involved in the control of decidual cell fate. Since it is well known that TGF-β signals through Smads proteins, Western blot and IHC analyses were performed on pregnant endometrial cell lysates to measure Smad2 and phospho-Smad2 (the activated form) to test this hypothesis. The results showed that Smad2 protein was increased and stronger at the time of embryo implantation and that Smad2 phosphorylation was gradually increased during regression of the DB, which correlates with the presence of TGF-β isoforms. These results suggest that TGF-β isoforms present during these two critical stages of pregnancy might act through Smads proteins to induce apoptosis or to induce other genes known to be regulated by TGF-βs. Smad2 protein is relatively low during regression of the DB compared to early stages of pregnancy. This result is supported by a recent study showing that increased Smad2 expression is probably caused by the invasion of the trophoblast resulting in the formation of the first decidual zone \[[@B45]\]. However, Smad2 phosphorylation is not as much increased during early pregnancy as compared to late pregnancy. It is again a possibility that, since only epithelial cells undergo apoptosis at the time of embryo implantation, the possible increase of Smad2 phosphorylation induced by TGF-β might only be observed in a small number of cells which was undetectable with the techniques used. To better understand the effect of TGF-β at the cellular level, decidual cell cultures were used to further investigate the interaction of TGF-β and the PI 3-K/Akt survival pathway. Although a recent study showed the possibility that cells obtained following artificial decidualization might be different from those obtained from pregnant animals \[[@B46]\], this model is an excellent alternative to obtain sufficient material to test the role of TGF-β and to correlates the data with the physiologic situation. It has been demonstrated recently that Akt phosphorylation is directly induced by 17β-estradiol in the ovariectomized rat uterus indicating that sex steroids have an important influence on endometrial cell fate \[[@B47]\]. Very recent studies have shown that TGF-β might directly block Akt activity through Smad activation \[[@B30]\]. Thus, stromal decidual cells were treated with different doses of TGF-β1 to determine if Smad activation might in turn block Akt survival pathway to induce cell death. TGF-β1 induced Smad2 phosphorylation in cultured endometrial stromal cell in vitroand triggered apoptosis in a dose-dependent manner, which was accompanied by a reduction of cell proliferation, confirming TGF-β as an apoptotic factor in rat decidual endometrial stromal cells. In response to TGF-β1, Akt phosphorylation was significantly decreased indicating that Akt activity inhibition might be an important mechanism for TGF-β-induced apoptosis in this model. Other studies have shown that Akt activity might in turn be affected by the presence or absence of inhibitor of apoptosis proteins such as XIAP \[[@B33],[@B34]\] and cIAP-1 \[[@B48]\]. The results support the hypothesis that TGF-β action via Smads not only acts at the transcriptional level to induce production of apoptotic factors in decidual cells but also at the protein level to block activation of survival factors. Akt phosphorylation was recently shown to be regulated by XIAP, a well known inhibitor of apoptosis protein, in human ovarian epithelial surface cells and in rat granulosa cells \[[@B33],[@B34]\]. The present results show a possible interaction (direct or indirect) between TGF-β1 and XIAP protein expression. TGF-β1 reduced XIAP expression in vitroin a dose dependent manner. Recent studies have demonstrated that XIAP can act as a cofactor in the regulation of gene expression induced by TGF-β and is independent of Smad4 \[[@B14],[@B49]\]. Because TGF-β1 reduced Akt phosphorylation and that XIAP was shown to induce Akt phosphorylation \[[@B34]\], it is possible that TGF-β action in this case might be through Smad2 activation and XIAP gene expression which in turn would act on Akt phosphorylation. Further experiments will be necessary to have a better understanding of the interactions between TGF-β and Akt survival pathway. In particular, how specifically Smads act on XIAP expression and how XIAP acts on the PI 3-K/Akt survival pathway activity. Other factors such as Smac/Diablo, a XIAP intracellular inhibitor, which was demonstrated as being regulated by 17β-estradiol in the rat during estrous cycle \[[@B50]\], might also be a putative candidate for regulation of TGF-β activity in decidual cells. Conclusion ========== In conclusion, this study demonstrates that the three isoforms of TGF-β are differently localized and regulated in endometrial cells of pregnant rats, particularly at the time of implantation and regression of the DB. The present study also showed that TGF-β plays an important role in the control of cell survival and cell death and that it may interacts with the PI 3-K/Akt survival pathway through Smad activation to allow apoptosis induction. Further experiments will be necessary to understand more precisely the effect of other Smads and/or co-Smads proteins during TGF-β-induced apoptosis. Other investigations will also be required to better understand the specific roles of TGF-β2 and TGF-β3 at the intracellular and molecular level in vitroto determine how and if these isoforms control cell survival through Smad signal transducers. Authors\' contributions ======================= CS drafted the paper. CS, PLC, GFF, VL and MCD performed the experiments. EA conceived the study, participated in its design and coordination, and wrote final version of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work has been supported by a grant from NSERC (238501-01). Eric Asselin is a chercheur-boursier from the Fond de la Recherche en Santé du Québec (FRSQ) and New Investigator of the Canadian Institute of Health Research of Canada (CIHR). Marie-Claude Déry is recipient of a FRSQ and NSERC studentships. We are grateful to Mrs Rollande Caron for the contribution of her precious time and expertise to our projects. We thank also Mrs Sophie Parent and Mrs Daphne Efford for reviewing the manuscript.	PubMed Central	2024-06-05T03:55:58.170260	2005-5-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166574/", "journal": "Reprod Biol Endocrinol. 2005 May 31; 3:20", "authors": [ { "first": "Carl", "last": "Shooner" }, { "first": "Pierre-Luc", "last": "Caron" }, { "first": "Guylaine", "last": "Fréchette-Frigon" }, { "first": "Valérie", "last": "Leblanc" }, { "first": "Marie-Claude", "last": "Déry" }, { "first": "Eric", "last": "Asselin" } ] }
PMC1166575	Background ========== The vertical transmission of human immunodeficiency virus type 1 (HIV-1) accounts for more than 90% of all HIV-1 infections in children. HIV-1 infected pregnant women can transmit the virus to their infants during all stages of their pregnancy, including prepartum (trans-placental passage), intrapartum (exposure of infants\' skin and mucous membranes to contaminated maternal blood and vaginal secretions) and post-partum (via breast milk) at an estimated rate of 30% \[[@B1]-[@B4]\]. However, the rate of vertical transmission can be reduced by antiretroviral therapy during pregnancy. The risk of vertical transmission increases with several parameters, including advanced maternal disease status, low maternal CD4 cell count, high maternal viral load, recent infection of the mother, prolonged exposure of infant to ruptured membranes during parturition, and higher viral heterogeneity in the mother \[[@B5]-[@B8]\]. Viral heterogeneity is one of the classical means by which HIV-1 evades the host immune system. The heterogeneity of HIV-1 is attributed to the error-prone reverse transcriptase (RT) enzyme, which is responsible for converting the single stranded viral genomic RNA to double-stranded DNA that integrates into the host chromosome. As reverse transcription is the first step of the viral replication cycle \[[@B9]\], errors made at this stage ensures propagation of the erroneously copied genome to form the quasi-species of HIV-1 found in the infected individuals. These quasi-species infect other uninfected target cells and the cycle of error-prone reverse transcription continues. We have previously demonstrated that HIV-1 sequences from transmitting mothers (mothers who transmitted HIV-1 to their infants) were more heterogeneous compared with HIV-1 sequences from non-transmitting mothers (mothers who failed to transmit HIV-1 to their infants) \[[@B10]\]. This finding further suggests that the reverse transcription step that is responsible for generation of viral heterogeneity, may also play an important role in vertical transmission. The RT gene is unique in that it is also exposed to the same mutating effects of the RT enzyme as other part of the HIV-1 genome. Therefore, we sought to examine HIV-1 RT sequences from five infected mother-infant pairs following perinatal transmission. The HIV-1 RT shows significant sequence and structural similarity to other viral reverse transcriptases as well as viral and bacterial RNA polymerases \[[@B11]-[@B13]\]. HIV-1 RT is a heterodimeric protein comprising of two subunits, 66 kDa and 51 kDa. It is encoded as a Gag-Pol precursor, Pr160^gag-pol^, which is cleaved by viral protease to yield the Gag protein and the viral polymerase which codes for RT \[[@B9],[@B14]\]. The larger subunit (p66) of the heterodimer acts as an RNA-dependant DNA polymerase, a DNA-dependant DNA polymerase and has RNase H activity associated with the C-terminus \[[@B15],[@B16]\], whereas the p51 subunit lacks the C-terminus RNase H activity, is folded differently from the p66 subunit and is thus inactive \[[@B17]-[@B20]\]. The p66 is folded to form a structure similar to a right hand with palm, finger and thumb subdomains \[[@B21]-[@B23]\] that are connected to the RNase H by the \"connexion\" subdomain \[[@B22],[@B24],[@B25]\]. Each domain has several secondary structural elements which are critical for primer binding, template binding \[[@B14],[@B22],[@B23],[@B26],[@B27]\] and nucleotide recruitment \[[@B28]\]. More specifically, the aspartate residues at position 110, 185 and 186 are believed to be the active sites of the polymerase and are located in the palm subdomain at the bottom of the DNA binding cleft \[[@B14],[@B16],[@B20],[@B28],[@B29]\]. Mutations in this subdomain and the active site abolish the enzymatic activity of HIV-1 RT \[[@B2],[@B19],[@B22],[@B30]-[@B32]\] and alter viral replication, which may also affect HIV-1 mother-to-infant transmission. In this study, we characterized the HIV-1 RT quasi-species from five mother-infant pairs following vertical transmission, including a mother with infected twin infants. We show that the open reading frame of the RT gene was highly conserved in the sequences from five mother-infant pairs. In addition, there was a low degree of heterogeneity and high conservation of functional domains essential for RT activity. These findings may be helpful in the understanding of the molecular mechanisms of HIV-1 vertical transmission. Results ======= Patient population and sample collection ---------------------------------------- Blood samples were collected from five HIV-1-infected mother-infant pairs following perinatal transmission, including samples from a set of twins (IH1 and IH2) in the case of mother H. The demographic, clinical and laboratory findings on these mother-infant pairs are summarized in Table [1](#T1){ref-type="table"}. The Human Subjects Committee of the University of Arizona, and the Institutional Review Board of the Children\'s Hospital Medical Centre, Cincinnati Ohio, approved this study. Written informed consent was obtained for participation in the study from mothers of infected mother-infant pairs. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic, Clinical, and Laboratory Parameters of HIV-1 Infected Mother-Infant Pairs ::: Patient Age Sex CD4+ cells/mm3 Length of infection ^a^ Antiviral drug Clinical Evaluation ^b^ --------- --------- ----- ---------------- ------------------------- ---------------- ---------------------------------------------- MB 28 yr 509 11 mo None Asymptomatic IB 4.75 mo M 1942 4.75 mo None Asymptomatic, P1A MC 23 yr 818 1 yr6 mo None Asymptomatic IC 14 mo F 772 14 mo ZDV Symptomatic AIDS;P2A,D1,3,F MD 31 yr 480 2 yr6 mo None Asymptomatic ID 28 mo M 46 28 mo ddC^c^ Symptomatic AIDS, P2AB,F; failed ZDV therapy MF 23 yr 692 2 yr10 mo None Asymptomatic IF 1 wk M 2953 1 wk ZDV Asymptomatic,P1A MH 33 yr 538 5 mo None Asymptomatic IHT1 7 mo F 3157 7 mo ACTG152 Hepatosplenomeglay lymphadenopathy IHT2 7 mo F 2176 7 mo ACTG152 Hepatosplenomegaly lymphadenopathy M: mother; I: infant. ^a^Length of infection: The closest time of infection that we could document was the first positive HIV-1 serology date or the first visit of the patient to the AIDS treatment Center, where all the HIV-1 positive patients were referred to as soon as an HIV-1 test was positive. Therefore, these dates may not reflect the exact dates of infection. ^b^Evaluation for infants is based on CDC criteria, ^c^ddC, Zalcitibine ::: Phylogenetic analysis of RT sequences of mother-infant isolates --------------------------------------------------------------- We first performed multiple independent polymerase chain reaction (PCR) amplifications from peripheral mononuclear cells (PBMC) DNA of five mother-infant pairs and obtained 10 to 14 clones from each patient followed by nucleotide sequencing of these clones. We then performed the phylogenetic analysis by constructing a neighbor-joining tree of the 132 RT sequences from these mother-infant pairs, including the set of twins from mother H and the reference strain NL4-3, as shown in Figure [1](#F1){ref-type="fig"}. A model of evolution was optimized for the entire nucleotide sequence data set using the approach outlined by Huelsenbeck and Crandall \[[@B33]\]. The model of choice was incorporated into PAUP \[[@B34]\] to estimate a neighbor-joining tree and the tree was bootstrapped 1000 times to ensure fidelity. The phylogenetic tree demonstrated that the RT sequences from five mother-infant pairs were well discriminated in separate clusters and that the mother and infant sequences were generally separated in distinct subclusters. However, there was some intermingling between mother and infant sequences in pair C. Furthermore, the formation of separate subclusters of RT sequences from twins of mother H suggests that the there was probably compartmentalization of HIV-1 in the two fetuses causing independent evolution. We also compared our mother-infant pairs\' RT sequences with the RT sequences of several clades present in the HIV databases and found that our RT sequences grouped with clade or subtype B sequences (not shown). The data on phylogenetic analysis indicate that the epidemiologically linked mother-infant sequences are closer to each other than epidemiologically unlinked sequences and that there was no PCR cross contamination. It is important to note that the mother-infant pairs grouped in the same subtree, even when some of the infants\' ages were more than 2 to 3 years, suggesting that the epidemiological relationships are maintained in mother-infant pairs no matter how long the infection in the infants has progressed. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Phylogenetic analysis of HIV-1 RT of 132 RT sequences from five mother-infant pairs, including B, C, D, F and H. The neighbor-joining tree is based on the distance calculated between the nucleotide sequences from the five mother-infant pairs. Each terminal node represents one RT gene sequence. The numbers on the branch points indicate the percent occurrence of branches over 1,000 bootstrap resamplings of the data set. The sequences from each mother formed distinct clusters and are well discriminated and in confined subtrees, indicating that the variants from the same mother-infant pair are closer to each other than to other sequences and that there was no PCR cross-contamination. These data were strongly supported by the high bootstrap values indicated on the branch points. ::: ![](1742-4690-2-36-1) ::: Coding potential of RT gene sequences ------------------------------------- The multiple sequence alignments of the deduced amino acid sequences of HIV-1 RT genes from five mother-infant pairs, B, C, D, F, mother H and her twin infants IH1 and IH2 are shown in Figures [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"}, [6](#F6){ref-type="fig"}, and [7](#F7){ref-type="fig"}, respectively. These sequences were aligned with consensus subtype B RT sequence (CON B). We found that 115 of the 132 sequences analyzed contained a complete RT open reading frame (ORF), with an 87.2% frequency of intact RT open reading frames thus indicating that the coding potential of the RT ORF was maintained in most of the sequences in 1680 bp sequenced. Moreover, the infected mothers\' sequences showed a frequency of 85.5% of intact RT ORF while infants demonstrated a frequency of 88.5%. Several clones in mother-infant pair B and mother H were found to be defective due to a single nucleotide substitution, insertion or deletion resulting either in frame-shift or stop codons. The RT sequences also displayed patient and pair specific amino acid sequence patterns. Several amino acid motifs changes were observed in majority of the mother-infant pairs\' sequences, including a glutamic acid (E) or proline (P) at position 122, an arginine (R) at 277, and a threonine (T) or serine (S) at 376 and 400. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother-infant pair B involved in vertical transmission. In the alignment, the top sequence is the consensus RT sequence of subtype or clade B (CON B) to which mother-infant pair-B RT sequences are aligned. In mother-infant pair B sequences, each line refers to a clone identified by a clone number with M referring to mothers and I referring to infants. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT activity are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother-infant pair C in reference to consensus subtype B (CON B) RT sequence. In the alignment, the top sequence is CON B RT sequence and the bottom sequences are mother-infant pair C sequences (M refers to mother sequences and I to sequences). The number of clones sequenced is represented with clone numbers. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Spanning arrowheads indicated above the alignment shows relevant amino acid motifs and domains essential for RT function. ::: ![](1742-4690-2-36-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother-infant pair D. The patient sequences are aligned in reference to consensus RT sequence of HIV-1 subtype or clade B (CON B) at the top. In the mother-infant pair sequences, each line refers to a clone identified by a clone number with M referring to mother and I to infants. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT activity are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase gene from mother-infant pair F. In the alignment, the top sequence (CON B) is the consensus subtype B RT sequence and the bottom sequences are from mother-infant pair F sequences (M stands for mother sequences and I for infant sequences and the number of clones for mother and infant are indicated by clone number). The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT functions are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother H, who had given birth to infected twins, H1 and H2 (alignment shown in Figure 7). In the mother H sequences, each line refers to a clone identified by a clone number with M referring to mother. The mother sequences are aligned in reference to consensus RT sequence of HIV-1 subtype or clade B (CON B) shown at the top. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Spanning arrowheads indicated above the alignment shows relevant amino acid motifs and domains required for RT activity. ::: ![](1742-4690-2-36-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase gene (RT) from infected twin infants, H1 and H2 of mother H (alignment shown in Figure 6). In the alignment, the top sequence is the consensus subtype B RT sequence (CON B) and the bottom sequences are of infants H1 and H2 represented by I and clone numbers. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT activity are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-7) ::: Variability of RT gene sequences in mother-infant isolates ---------------------------------------------------------- The degree of genetic variability of RT sequences, measured as nucleotide and amino acid distances based on pairwise comparison (as described in Methods), was determined for the five mother-infant pairs\' sequences, and is shown in Table [2](#T2){ref-type="table"}. The nucleotide sequences of RT within mothers (mothers B, C, D, F and H) differed by 0.80, 1.76, 1.37, 1.21 and 2.90% (median values), respectively, ranging from 0 to 3.46%. The variability in the infant sets (infants B, C, D, F, H1 and H2) was similar to the mother sequences and differed by 0.80, 1.49, 1.37, 1.31, 0.64 and 1.24% (median values), respectively, ranging from 0 to 2.21%. Interestingly, the variability between epidemiologically linked mother and infant sets (pairs B, C, D, F and H) was also on the same order of 1.05, 1.7. 1.74, 1.22 and 1.45 (median values) respectively, ranging from 0 to 4.48%. Moreover, the amino acid sequence variability of RT within mothers (mothers B, C, D, F and H) differed by 1.26, 2.81, 1.98, 1.26 and 2.27% (median values), respectively, ranging from 0 to 5.51%. The variability within infants (infants B, C, D, F, H1 and H2) differed by 1.44, 2.35, 1.80, 1.62, 1.44 and 1.62% (median values), ranging from 0 to 4.57%, and between mother-infant pairs (pairs B, C, D, F and H) by 1.44, 2.90, 2.53, 1.44 and 2.17% (median values), ranging from 0 to 6.47%, respectively. We also determined sequence variability between epidemiologically unlinked individuals and found that the nucleotide distances ranged from 0 to 9.1% (median 5.4%) and amino acid from 0 to 12.4% (median 6.34%). The variability in general was lower between epidemiologically linked mother-infant pairs\' sequences than epidemiologically unlinked individuals, suggesting that epidemiologically linked mother-infant pair sequences are closer to each other. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distances in the RT sequences within mother sets, within infant sets, and betweenmother-infant pairs ::: Nucleotide distances -------------------------- -------------------- -------------------------------- ---------- ----------- --------- ---------- ---------- ----------- --------- ---------- ---------- Within mothers Within infants Between mother and infants Pair Min Med Max Pair Min Med Max Pair Min Med Max MB 0.0 0.80 2.10 IB 0.0 0.80 1.30 B 0.0 1.05 2.05 MC 0.0 1.76 3.46 IC 0.0 1.49 2.17 C 0.0 1.70 3.26 MD 0.0 1.37 2.21 ID 0.0 1.37 2.21 D 0.0 1.74 4.48 MF 0.0 1.21 1.54 IF 0.0 1.31 2.93 F 0.0 1.22 2.08 MH 0.0 2.90 2.60 IH1 0.0 0.64 1.34 H 0.0 1.45 3.30 IH2 0.0 1.24 1.75 Total 0.0 1.34 3.46 Total 0.0 1.48 2.21 Total 0.0 1.32 4.48 Amino acid distances Within mothers Within infants Between mother and infants Pair Min Med Max Pair Min Med Max Pair Min Med Max MB 0.0 1.26 4.61 IB 0.0 1.44 2.72 B 0.0 1.44 4.57 MC 0.0 2.81 5.51 IC 0.0 2.35 4.01 C 0.0 2.90 5.51 MD 0.0 1.98 3.83 ID 0.0 1.80 4.57 D 0.0 2.53 6.47 MF 0.0 1.26 2.35 IF 0.0 1.62 3.09 F 0.0 1.44 3.09 MH 0.0 2.27 3.09 IH1 0.0 1.44 2.17 H 0.0 2.17 6.27 IH2 0.0 1.62 2.72 Total 0.0 1.52 5.51 Total 0.0 1.42 4.57 Total 0.0 2.90 6.47 M: mother; I: infant. Min: Minimum; Med: Median; Max: Maximum. Totals were calculated for all pairs together ::: We also investigated if the low variability of RT sequences seen in our mother-infant pair isolates is due to errors made by LA Taq polymerase used in our study. We did not find any errors made by the LA Taq polymerase when we used a known sequence of HIV-1 NL 4--3 for PCR amplification and DNA sequencing of the RT gene. Dynamics of HIV-1 RT gene evolution in mother-infant isolates ------------------------------------------------------------- The maximum likelihood estimates and chi square tests performed by Modeltest 3.06 \[[@B35]\] suggested different models of evolution for each patient sample. The estimates of genetic diversity of RT sequences from the five mother-infant pairs were determined by using the Watterson model, assuming segregating sites and the Coalesce method assuming a constant population size. The estimates of genetic diversity shown as theta values (estimated as nucleotide substitutions per site per generation) are shown in Table [3](#T3){ref-type="table"}. The levels of genetic diversity among infected mothers and infants, as estimated by Watterson method, ranged from 0.012 to 0.025 and 0.009 to 0.021, respectively. Similar results were obtained when the mother-infant pair populations were analyzed by the Coelesce method, with the values ranging from 0.020 to 0.058 in mothers and from 0.016 to 0.060 in infants. These data suggest that the mother and infant populations evolved very slowly and at similar rates. The differences observed in the estimates of genetic diversity between and mothers and infants sequences are not statistically significant. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Estimates of genetic diversity of HIV-1 RT within mother sets and infant sets ::: MOTHERS INFANTS ------------- ------------- ----------- ----------- ----------- -------- ----------- ----------- N θ~w~ θ~c~ θ~w~ θ~c~ Mother B 12 0.015 0.038 Infant B 12 0.014 0.033 Mother C 12 0.025 0.058 Infant C 13 0.021 0.060 Mother D 11 0.017 0.042 Infant D 10 0.019 0.040 Mother F 14 0.012 0.029 Infant F 12 0.018 0.053 Mother H 14 0.020 0.020 Infant H1 11 0.009 0.016 Infant H2 11 0.015 0.044 Totals 63 0.018 0.037 69 0.016 0.041 N -- number of RT clones sequenced. θ~w~-- genetic diversity as calculated by the Watterson method; θ~c~-- genetic diversity as calculated by the Coelesce method. Totals were indicated as an average of all values. ::: Rates of accumulation of nonsynonymous and synonymous substitutions ------------------------------------------------------------------- Selection pressure on the RT gene was estimated as a ratio of accumulation of non-synonymous to non-synonymous substitutions using the Nielsen and Yang model \[[@B36]\] as implemented in codeML \[[@B37]\]. Although there are several models to predict the rate of positive selection, most of these models assume that all sites in a sequence are under the same selection pressure with the same underlying dN/dS ratio \[[@B38]\]. As substitutions of critical regions of a protein can lead to deleterious mutations, it is unrealistic to make assumptions about equal degree of selection throughout the protein. In cases where positive selection is operating on proteins, it has been shown that only a limited number of amino acids may be responsible for adaptive evolution. In such a case, methods that estimate dN/dS ratios over an entire sequence may fail to detect positive selection even when it exists \[[@B39]\]. The codeML method uses the codon as a unit of evolution as opposed to a nucleotide, and thus allows us to estimate the percentage of positions that are being positively selected instead of averaging the rates of positive selection over the entire gene \[[@B39]\]. This method also provides the percentage of mutations that are conserved, neutral or positively selected based on dN/dS values of 0, 1 or \> 1, respectively. The dN/dS values as well as the proportions of each site category estimated using the Nielsen and Yang model are shown in Table [4](#T4){ref-type="table"}. As described in the methods, a dN/dS value of greater than 1 suggests positive selection. The percentage of the substitutions being positively selected is shown in column p3. Except for viral populations in infants C and F, all isolated populations were associated with dN/dS ratio \>1, indicating positive selection. In case of infants C and F, there was no positive selection on the mutations and most of the substitutions were neutral. All mothers generally displayed a higher proportion of positively selected p3 sites as compared to the infants. Although the dN/dS values for infant H1 and H2 seem higher than mother H, closer observation shows that the percentage of sites undergoing positive selection is higher in the mother than in the twin infants. Table [4](#T4){ref-type="table"} shows that in mothers, over half the sites (66.6%) belong to the conserved p1 category, whereas the frequency of neutral and positively selected sites was equally distributed. This is in contrast to the viral population from the infants where the conserved site category (p1) had a frequency of only 36.5% and close to half the sites (55.7%) belongs to the neutral p2 category. Statistical analysis revealed that only the proportion of the neutral p2 category was significantly different between mothers\' and infants\' sequence viral populations (p \< 0.05). This is signified by the case that all the sites in Infant F belonged to the p2 category. Higher proportion of p2 sites in infants have also been shown in the nef gene product in these same mother infant pairs \[[@B40]\]. The variable (positively selected) sites (p3) in the mothers\' sequences were associated with dN/dS ratios that ranged from 2.34 to 8.9, with viral sequence populations from three mothers (MD, MF, MH) that displayed a dN/dS ratio of below three. This is in contrast to the infants\' viral populations that were either associated with a dN/dS of below 1, indicating no directional selection (IC and IF), a dN/dS ratio between 3 and 4 (IB and ID) or a very high dN/dS ratio as found in the sequences isolated from the twins H1 and H2. This analysis showed that the RT gene in both the mothers and infants is under positive selection pressure. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### dN/dS values in HIV-1 RT sequences within mother sets and within infant sets. ::: MOTHER INFANT ------------ ------------ ---------- ---------- ---------- ----------- ----------- -------- ---------- ---------- --------- ----------- N P1 P2 P3 dN/dS N P1 P2 P3 dN/dS Mother B 12 53 18.8 27 8.9 Infant B 12 41 42 16 3.31 Mother C 12 55.5 43 1.3 6.09 Infant C 13 0 81.2 18.8 0.01 Mother D 11 70.6 5.7 23.6 2.52 Infant D 10 74.8 19.2 5.9 4.44 Mother F 14 81.7 7.8 10.4 2.67 Infant F 12 0 100 0 0.001 Mother H 14 72 0 27 2.34 Infant H1 11 47 50 2.8 14.04 Infant H2 11 56 42 0.6 16.58 Totals 66.5 15.1 18.4 4.50 69 36.5 55.7 7.8 6.39 N -- number of RT clones sequenced.; P1 = proportion of conserved codons as a percent; P2 = proportion of neutral codons as a percent; P3 = proportion of positively selected codons as a percent. dN/dS = ratio of synonymous to non-synonymous at P3 sites. Totals were calculates as an average of all values. ::: Analysis of functional domains of RT in mother-infant pairs ----------------------------------------------------------- HIV-1 RT is a heterodimeric protein comprising of two subunits, p66 and p51. The larger subunit of the heterodimer acts as an RNA-dependant DNA polymerase, a DNA-dependant DNA polymerase and an RNase H that is associated with the C-terminus \[[@B15],[@B16]\]. The p66 is folded to form a structure similar to the right hand with palm, finger and thumb subdomains \[[@B21],[@B23],[@B32]\] that are connected to the RNase H by the \"connexion\" subdomain \[[@B22],[@B24],[@B25]\]. Each domain has several secondary structural elements, which are critical for primer binding, template binding \[[@B14],[@B22],[@B23],[@B26],[@B27],[@B41]\] and nucleotide recruitment \[[@B28]\]. The active sites of the polymerase comprise of aspartic acid (D) residues at positions 110, 185 and 186, which are located in the palm subdomain at the bottom of the DNA binding cleft \[[@B22],[@B23]\]. Mutations of these aspartic acid residues abrogates the polymerase activity of RT \[[@B22],[@B23],[@B29],[@B32]\]. These aspartate residues of the RT active site were conserved within the five mother-infant pairs RT sequences. Furthermore, the D185 and D186 that form a part of an essential highly conserved YMDD \[[@B32],[@B42],[@B43]\] motif involved in binding to the 3\'OH of the primer strand \[[@B14],[@B26]\], were highly conserved in our mother-infant pairs\' RT sequences (Figures [2](#F2){ref-type="fig"} to [7](#F7){ref-type="fig"}). The amino acids at positions 73--90 that constitute the template grip required for positioning and binding the RT template near the active site of the RT \[[@B23]\], were also conserved in most of our RT sequences. The primer grip responsible for primer binding extends from amino acids 227 to 235 \[[@B22],[@B23]\] and these amino acids were also conserved in the mother-infant RT sequences. The K263, K353 and R358 that form salt bridges with the phosphate groups \[[@B14],[@B21],[@B22],[@B30],[@B44]\] of the template and primer were found to be conserved in most of the RT sequences analyzed. The thumb subdomain of RT is comprised of two anti-parallel α helices, αH and αI, which bind to the opposite strand of dsDNA. The αH also directly inserts into the minor groove of the DNA \[[@B14],[@B22],[@B41]\]. Both these helices were generally conserved in our mother-infant RT sequences. The connexion subdomain that links the RT to the RNase H and forms the floor of the template binding cleft \[[@B22],[@B24],[@B25],[@B42]\], showed some substitutions, including V293I, A376S and A400T in our mother-infant RT sequences. Mutations at positions H361 and Y501 reduces RNase H activity \[[@B24]\]. Examination of the five mother-infant pairs\' sequences revealed that these two positions were intact in all RT sequences (Figures [2](#F2){ref-type="fig"} to [7](#F7){ref-type="fig"}). Furthermore, the RNase H active sites contain four acidic amino acid residues, D443, E478, D498 and D549 \[[@B22],[@B24],[@B25],[@B41],[@B42]\], which were highly conserved in our mother-infant pairs sequences. In addition, several substitutions were seen in regions of RT that are not known to have critical function. The relevance of these changes is not known. Mutations associated with anti-retroviral drug resistance --------------------------------------------------------- Several naturally occurring mutations in the polgene in treatment-naïve patients have been reported \[[@B45],[@B46]\], although most of these mutations are not seen in our RT gene sequences. In addition, these mutations found in treatment-naïve patients were usually seen in non-subtype B infections and our patient population was from subtype B infected individuals. These changes were usually in amino acids where the mutations did not actually confer nucleoside reverse transcriptase inhibitor (NRTI) drug resistance but were accessory mutations \[[@B46]-[@B48]\]. Several amino acid changes in RT seen in patients undergoing NRTI therapy are selected primarily with zidovudine (ZDV) treatment. These mutations referred to as thymidine analog mutations (TAMs) include M41L, D67N, K70R, L210N, T215Y/F and K219Q \[[@B47],[@B49]\]. Since most of our infected mothers were treatment naïve but infants were actively on ZDV therapy or on other drugs (Table [1](#T1){ref-type="table"}), we examined the RT sequences for ZDV resistant mutations (Figure [2](#F2){ref-type="fig"}). Several TAMs associated with drug resistance were observed in our infants C and D who were either on prolonged or failed ZDV therapy. These mutations included M41L in three clones from infant C and two clones in infant D, D67N and K70R in five clones from infant C, L210W in one clone from infant D and T215F in seven clones from infant D and K219Q in four clones from infant C and D. In addition, one clone from infant C had all the above mutations, indicating significant resistance to ZDV \[[@B46],[@B50]\]. Although Mother C was not on any antiretroviral therapy two clones had TAMs at M41L and K219Q positions, suggesting that these mutations were naturally occurring. It is interesting to note that the infant of this mother yielded several clones with these two mutations. An R211K mutation known as an accessory mutation associated with NRTI resistance \[[@B46]\] was also observed in all mother-infant pair H clones. Immunologically relevant mutations in the CTL epitopes of RT ------------------------------------------------------------ The cytotoxic T lymphocyte (CTL) responses have been shown to exert significant immune pressure during HIV-1 infection. Strong CTL responses are maintained in long-term nonprogressors and these responses correlate with decrease in viral load \[[@B51]-[@B55]\]. It has been shown that transmitting mothers have larger numbers of CTL escape variants as compared to non-transmitting mothers \[[@B56]\], emphasizing that CTL escape variants may become a part of circulating virus that influences vertical transmission \[[@B56],[@B57]\]. Several regions in the RT gene have been shown to elicit strong CTL responses during HIV-1 infection. The CTL eptitope, TVLDVGDAY, between amino acid positions 107--115 <http://www.hiv.lanl.gov/content/immunology/ctl_search>, is highly conserved among known HIV-1 isolates \[[@B57]\]. This epitope contains the amino acid D110 which is part of the RT active site. This epitope was highly conserved in most of the mother-infant RT clones sequenced (Fig. [2](#F2){ref-type="fig"}). Another motif, TAFTIPSI, between amino acid positions 128--135 is an HLA-B51 restricted epitope <http://www.hiv.lanl.gov/content/immunology/ctl_search>. This epitope is present in the palm region consisting of positions A129 and I135 as anchor residues \[[@B57]\]. This motif was mostly conserved in the RT sequences of the five mother-infant pairs analyzed. In addition, I135T mutation decreases CTL response but increasing concentration of mutant peptide re-establishes appropriate responses \[[@B57]\]. The I135T mutation was seen in several of our mother-infant pair\'s D sequences. The next motif AIFQSSMTK from amino acid positions 158--166, comprising of I159, F160, K166 anchor residues and recognized by several HLA types, is conserved among known HIV-1 isolates and believed to be associated with vertical transmission \[[@B56],[@B57]\]. Our mother-infant pairs\' RT sequences showed conservation in this motif. Another CTL epitope YPGIKVRQL from positions 271--279 has been reported to be conserved in transmitting mothers and infants with several natural occurring variants \[[@B56]\], was also found to be conserved in our mother-infant pairs\' RT sequences. In addition, a P272H mutation that causes significant loss of CTL response for this epitope \[[@B56]\] was not seen in any of the RT clones analyzed. Discussion ========== In this study, we show for the first time that reverse transcriptase open reading frames from five mother-infant pairs following perinatal transmission were maintained with a frequency of 87.2%. The functional domains required for reverse transcriptase activity in HIV-1 replication were highly conserved in most of the mother-infants sequences. We also demonstrate a low degree of sequence variability and estimates of genetic diversity for reverse transcriptase genes after mother-to-infant transmission. However, epidemiologically unlinked individual\'s sequences were more heterogeneous than epidemiologically linked mother-infant pair\'s sequences. Several motifs in reverse transcriptase responsible for primer and template binding and positioning and motifs involved in nucleotide recruitment were conserved in all mother-infant pairs\' sequences. The data we show here are comparable to those of our previously analyzed conserved genes, including gagP17MA, vif, vpr, tatand nef\[[@B58]-[@B62]\]. Our findings suggest that an intact and functional reverse transcriptase open reading frame is essential for HIV-1 replication in mothers and their infants and low degree of viral heterogeneity is maintained following vertical transmission. The RT open reading frame was maintained in 115 of the 132 sequences (1680 base pairs sequenced), whereas 17 sequences contained stop codons (Figure [2](#F2){ref-type="fig"}). The frequency of conservation in five mother-infant pairs was found to be 87.2%. The comparison of the RT sequences with those of other conserved genes from HIV-1 infected mother-infant pairs showed comparable frequency of conversation, including gagp17 (86.2%), vif(89.8%), vpr(92.1%), tat(90.9%), nef(86.2%) and vpu(90.12%). There was no significant correlation between the conservation of RT open reading frame and disease progression in mothers and infants \[[@B63]-[@B65]\]. Several amino acid motifs were found to be a signature characteristic of each mother-infant pair, even in older infants where infection has progressed for more than 3 years. Phylogenetic analysis of the RT sequences revealed that the five mother-infant pairs were well discriminated, separated and confined within subtrees (Fig. [1](#F1){ref-type="fig"}), indicating that the epidemiologically linked mother-infant pairs were closer to each other and that there was no PCR product cross-contamination \[[@B66],[@B67]\]. In addition, most of the mother and infant sequences of the same pair formed separate subclusters, with little intermingling between sequences of mother and infant in some pairs. In some mother-infant pairs, minor variants of the mothers seem to be predominating in the infants, which was also seen in our previous V3 region analysis \[[@B68]\]. We also observed intermingling of sequences in mother-H and her infected twins, indicating that different mother\'s variants were transmitted to the twins. With respect to viral heterogeneity, there was a low degree of genetic variability in the RT sequences from mother-infant pairs estimated by several methods. Similar levels of genetic diversity were seen in other conserved genes of the same mother-infant pairs, including gag, vif, vprand tat\[[@B59]-[@B61],[@B69]\]. The low degree of genetic variability was observed in RT sequences of mothers and maintained in the infants following transmission, suggesting the essential nature of this gene in viral pathogenesis. It is important to note that the mother-infant pairs retained the same epidemiological relationship, even when some of the infant\'s age was more than 2 to 3 years. We believe this is an important finding that the epidemiological relationships as well as certain signature sequence motifs are maintained in mother-infant pairs or transmitter-recipient partners no matter how long the infection has progressed. This information may be critical in terms of vaccine development. Examining the motifs of the deduced amino acid sequences of the RT gene from five mother-infant pairs, we found that the essential motifs required for RT activity were mostly conserved in our mother-infant pairs\' sequences (Figure [2](#F2){ref-type="fig"}). The sites essential for primer binding, template binding, positioning of template and primer, which are located in α-Helix H and α-Helix I \[[@B22],[@B23]\], were are all conserved in RT sequences (Figure [2](#F2){ref-type="fig"}). Specifically, the amino acids involved in recruitment of nucleotides during reverse transcription \[[@B28]\] were mostly conserved. The active sites of the polymerase are located in the palm subdomain at the bottom of the DNA binding cleft comprising of aspartic acid (D) residues at positions 110, 185 and 186 were conserved within the five mother-infant pairs\' RT sequences. Furthermore, the D185 and D186 also form a part of an essential YMDD motif, which is highly conserved in known HIV-1 isolates \[[@B14],[@B22],[@B23],[@B26],[@B32],[@B43]\], was also conserved in our mother-infant pairs\' RT sequences analyzed. Some of the amino acids of the connexion subdomain that are critical for RNase H activity and replication \[[@B9],[@B24],[@B25]\] are conserved in our RT sequences with several substitutions of compatible nature, including V293I, K358R, A376S, and A390T. These substitutions were located in the regions of the connexion that forms the base of the binding cleft. It is possible that such mutations in the binding cleft may change the size of the cleft and affect fidelity of the reverse transcriptase without affecting the active site. Further assessment also shows that our RT sequences harbor mutations in the connexion and RNase H subdomains that are not at the critical sites required for RT activity. The implications of these mutations can be studied by performing the biological characterization of these RT clones in the context of HIV-1 replication. It would be interesting to determine whether the degree of genetic variability and conservation of RT functional domains in non-transmitting mothers and compare their sequences with the data presented here. Nonetheless, the data described here suggest that functional domains of the RT enzyme, including reverse transcriptase, DNA polymerase and RNase H, were highly conserved in our five mother-infant pair sequences. In terms of CTL epitopes in the RT gene, Wilson et al., have shown that the transmitting mothers have larger numbers of CTL escape variants as compared to non-transmitting mothers but the transmitted viruses carrying epitopes are not escape variants \[[@B56]\]. It is possible that the CTL responses studied are tissue specific and a representation of peripheral blood, and the virus and the CTL variants in the placenta, birth canal, and breast milk are different \[[@B70]\]. In addition, there is evidence suggesting that Nef and Pol specific CTLs found in breast milk showed no detectable responses in peripheral blood. Although several previously defined CTL motifs in the RT gene \[[@B56],[@B57]\] were conserved in our RT sequences, other mutations that either abrogated or improved the CTL responses \[[@B56],[@B57]\] were not seen in our sequences. The possibilities exist that the mutants observed in the CTL epitopes in our study may contribute to differential responses in a tissue specific manner and thus influence vertical transmission. While antiretroviral treatment during pregnancy has reduced the risk of vertical transmission in the United States, HIV-1 infection in children, as a result of perinatal transmission, is still increasing rapidly in developing countries. There is a global need of better preventive strategies of HIV-1 vertical transmission. If we characterize the properties of the transmitted viruses, we can then develop interventions against the properties of the transmitted viruses. We have already shown that the minor genotypes with R5 phenotypes are transmitted from mothers to infants and are initially maintained in the infants with the same properties \[[@B71]\]. Additional data on the properties of HIV-1 from mothers and infants following perinatal transmission presented in this study may aid in a better understanding of the molecular mechanisms of vertical transmission and development of effective strategies for prevention and control of HIV-1 infection in children. Conclusion ========== We have demonstrated that an intact and functional RT gene was maintained in infected mother-infant pairs following perinatal transmission. In addition, there was a lower degree of viral heterogeneity and estimates of genetic diversity in epidemiologically linked mother-infant pairs compared with epidemiologically unlinked individuals. Several amino acid motifs were found as a signature sequences in each mother-infant pair. We also found that the functional motifs of RT responsible for reverse transcription, DNA polymerization and RNase H were highly conserved in mother-infant RT sequences. These findings support the notion that RT is essential for HIV-1 replication in mothers and their infected infants. Methods ======= PCR amplification, cloning and nucleotide sequencing ---------------------------------------------------- Peripheral blood mononuclear cells (PBMCs) were isolated by a single step Ficoll-Hypaque procedure (Pharmacia-LKB) from whole blood samples of HIV-1-infected mother-infant pairs. DNA was isolated as described previously \[[@B68]\]. The HIV-1 RT gene was amplified by a two-step PCR method, first using outer primers RT1 (5 GTACAGTATTAGTAGGACCTACACCTGTC, 2470 to 2498, sense) and RT2 (5\'AAAATCACTAGCCATTGCTCTCCAATTAC, 4307 to 4279, antisense) and then with nested primers RT3 (5\'TGGAAGAAATCTGTTGACTCAGATTGG, 2507 to 2533, sense) and RT4, (5\'TTCTCATGTTCTTGGGCCTTATCT, 4270 to 4244, antisense). Equal amounts of PBMC DNA (approximately 25 to 50 copies from each patient) as determined by end-point dilution was subjected to multiple (5 to 8) independent PCRs to obtain clones that were sequenced and analyzed. PCRs were performed according the modified procedure of Ahmad et al., \[[@B68]\] in a 25 μl reaction mixture containing 2.5 μl of 10X PCR buffer (100 mM Tris-HCL, pH 8.3, 100 mM KCl, 0.02% Tween 20), 2.5 mM MgCl~2~, 400 μM each of dATP, dCTP, dGTP and dTTP, 0.2 to 1.0 μM of each of outer primers, and 2.5 U of TaKaRa LA Taqpolymerase (TaKaRa Biomedicals, Shiga, Japan). The reactions were carried out at 94°C for 30s, 45°C for 45s and 72°C for 3 min for 35 cycles, with the last cycle allowing for seven minutes of additional polymerization. After the first round of PCR, 4μl of the first-PCR product was used for nested PCR, using inner primers and same reagents at 94°C for 30s, 52°C for 45s and 72°C for 3 min for 35 cycles. We used negative control with each PCR amplification and a known HIV-1 DNA, pNL4-3, to assess errors generated by the LA Taqpolymerase. To avoid contamination, all samples, reagents and PCR products were stored separately and dispensed in a separate room free of all DNA used in the lab. The PCR products were then visualized on a 1% agarose gel, excised ad extracted by using a QIAquick Gel Extraction kit (Qiagen Inc.). These DNAs were cloned into the TA cloning system (pCR 2.1-TOPO vector, Invitrogen Inc.) and transformed into chemically competent TOP10 cells (Invitrogen Inc.). The white colonies were screened for correct size inserts and 10 to 14 clones from each patient obtained from multiple independent PCRs were initially manually sequenced and then sequenced using University of Arizona Biotechnology Center automated system. Sequence analysis ----------------- The nucleotide sequences of HIV-1 RT gene (approximately 1680 bp) from five mother-infant pairs were analyzed with the Wisconsin package 10.1 version of the Genetics Computer group (GCG) and were translated to corresponding deduced amino acid sequences (560 amino acids). A multiple sequence alignment was performed for the nucleotide and amino acid sequences with a reference HIV-1 consensus clade or subtype B RT sequences with a gap-opening penalty of 10 and a gap extension penalty of 5 using Clustal X. The transitions were not weighted and the amino acids were scored using a BLOSUM matrix. A model of evolution was optimized for the entire nucleotide sequence data set using the approach outlined by Huelsenbeck and Crandall \[[@B33]\]. Likelihood scores for different models of evolution were calculated using PAUP \[[@B34]\] and a chi square test was performed by Modeltest 3.06 \[[@B34],[@B35],[@B40],[@B72]\]. Using the Model test and Akaike Information Criterion \[[@B72]\], all the null hypotheses were rejected except a GTR+G model. The five rate categories were as follows: R (A-C) = 2.962, R (A-G) = 10.5176, R (A-T) 1.3663, R (C-G) = 0.6563, R (C-T) 12.5484, R (G-T) = 1. A gamma distribution with the shape parameter (α) of the distribution estimated from the data matrix via maximum likelihood was used to account for the rate of heterogeneity. This shape parameter α was = 0.7775. The model of choice was incorporated into PAUP \[[@B34]\] to estimate a neighbor-joining tree and the tree was bootstrapped 1000 times to ensure fidelity. Models to represent patterns of evolution of variants of each patient population were identified and were used to estimate corrected pairwise nucleotide distances using PAUP \[[@B34]\]. Amino acid distances were also estimated using the Jukes-Cantor model with the Wisconsin package 10.1 of GCG. The minimum, median and maximum nucleotide and amino acid distances for each patient and linked patient pairs were calculated from these data (Table [2](#T2){ref-type="table"}). To analyze the evolutionary processes acting on the RT gene, we estimated the ratio of non-synonymous (dN) to synonymous (dS) substitutions by a maximum likelihood model using codeML, a part of the PAML \[[@B37]\] package. The Nielsen and Yang \[[@B36]\] model considers the codon instead of the nucleotide as the unit of evolution and incorporates three distinct categories of sites. Every mutation is three times more likely to cause a nonsynonymous than a synononymous substitution and codeML accounts for this bias. The first category p1 represents the sites that are conserved and invariable where dN/dS = 0. The second category p2 represents neutral sites where dN/dS = 1 and represents sites at which the dN and the dS are fixed at the same rate. The third category p3 represents sites that are under positive selection where the dN have a higher rate of fixation than dS proportionally and dN/dS \>1. The dynamics of HIV-1 evolution was assessed using techniques of population genetics. In population genetics, genetic diversity is defined as θ = 2N~ei~μ, where N~ei~is the inbreeding effective population size and μ is the per nucleotide mutation rate per generation. The Watterson model based on segregating sites and the Kuhner model assuming constant population size were used to estimate differences in genetic diversity, using the program Coalesce, <http://inbio.byu.edu/faculty/kac/crandall_lab> which is part of the Lamarc software package. The tree files and the data matrixes from PAUP were used to estimate θ values as a measure of genetic diversity. ### Nucleotide sequence accession numbers The sequences have been submitted to GenBank with accession numbers [AY560388](AY560388) to [AY560528](AY560528). Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= VS carried out the PCR, cloning, and sequencing. VS and TH performed the sequence analysis by computer programs. VS and NA participated in the experimental design, data interpretation and writing of the manuscript. All the authors read and approved the final manuscript. Acknowledgements ================ This work was supported by grants to NA from the National Institute of Allergy and Infectious Disease (AI 40378, AI 40378-06) and the Arizona Disease Control Research Commission (ADCRC-7002, 8001). We thank Raymond C. Baker, Children\'s Hospital Medical Center, Cincinnati, Ohio and Ziad M. Shehab Department of Pediatrics, University of Arizona College of Medicine for providing HIV-1-infected mother-infant pairs blood samples. We thank members of Ahmad Lab, including Tiffany Davis and Kamlesh Patel for their help in cloning of the RT genes and Rajesh Ramakrishnan, Roshni Mehta and Brian Wellensiek for critically reading this manuscript and providing helpful suggestions.	PubMed Central	2024-06-05T03:55:58.173935	2005-5-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166575/", "journal": "Retrovirology. 2005 May 26; 2:36", "authors": [ { "first": "Vasudha", "last": "Sundaravaradan" }, { "first": "Tobias", "last": "Hahn" }, { "first": "Nafees", "last": "Ahmad" } ] }
PMC1166576	Findings ======== Two independent screens identified TRIM5 as a potent retrovirus restriction element that targets select viruses after entry into primate cells \[[@B1],[@B2]\]. The biochemical basis for specificity of restriction is only evident in cells of the owl monkey where HIV-1 CA is recognized by the C-terminal cyclophilin domain that is unique to the TRIM5 orthologue found in this genus \[[@B2]-[@B4]\]. In all other primates, including humans and macaques, potent CA-specific restriction is conferred by the TRIM5α isoform \[[@B1],[@B5]-[@B9]\], which possesses a C-terminal SPRY domain \[[@B10]\]. The mechanism by which TRIM5α selects retroviruses bearing particular CAs for restriction is unknown, though the TRIM5α SPRY domain is required for restriction and variation in SPRY amino acid residues determines the CA-specificity of given TRIM5α orthologues \[[@B9],[@B11]-[@B13]\]. Conventional biochemical and two-hybid experiments failed to detect an interaction between TRIM5α and CA (SS and JL, unpublished data). The observation that non-infectious virus-like particles saturate TRIM5α-mediated restriction \[[@B14]\], but only if the particles bear a mature core from a restriction-sensitive virus \[[@B15],[@B16]\] suggests that the TRIM5α SPRY domain recognizes a complex structure unique to the core of susceptible virions. Consistent with this model, expression within target cells of gag, gag-pol, or gagfragments encoding CA, CA-NC, or ubiquitin-CA-NC fusions, failed to block restriction activity (David Sayah and JL, unpublished data). Retrovirion cores can be liberated from the viral membrane envelope by detergent \[[@B17]\]. HIV-1 virion cores were prepared with several different detergents and mixed with recombinant TRIM5 orthologues. After TRIM5 enrichment by affinity chromatography, CA associated with owl monkey TRIMCypA, as reported with other methods \[[@B3],[@B4]\], but not with the equally potent HIV-1 restriction factor rhesus macaque TRIM5α (SS and JL, unpublished data). We then selected murine leukemia virus (MLV) for study because, relative to HIV-1, MLV CA remains tightly associated with viral reverse transcription (RT) and preintegration complexes \[[@B18],[@B19]\]. MLV strains bearing an arginine at CA residue 110 (so-called N-MLV) are highly susceptible to restriction by human TRIM5α whereas MLV virions bearing glutamate in this position (B-MLV) are completely resistant to restriction \[[@B5]-[@B8]\]. VSV G-pseudotyped N- and B-tropic MLV virions were produced as previously described \[[@B20]\] and, after normalization on non-restrictive Mus dunnicells, N-MLV was roughly 100-fold less infectious than B-MLV on HeLa cells (Figure [1A](#F1){ref-type="fig"}). Full-length human TRIM5α was then produced as a GST-fusion protein in 293T cells and mixed with purified N-MLV virions. CAp30, the major MLV core protein constituent, associated with TRIM5α (Figure [1B](#F1){ref-type="fig"}). CAp30 from B-MLV virions did not associate with TRIM5α (Figure [1B](#F1){ref-type="fig"}) demonstrating that TRIM5α binding was specific for restriction-sensitive CA. CAp30 did not associate with TRIM5 lacking the SPRY domain (Figure [1B](#F1){ref-type="fig"}), indicating that the SPRY-domain is required for CA-recognition. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Human TRIM5α binds CA from restricted MLV virions. (A) HeLa cells were infected with VSV G-pseudotyped, N- and B-tropic MLV-GFP vectors after normalization for RT activity and infectivity on non-restrictive Mus dunnitail fibroblasts. The percentage of infected (GFP-positive) cells was determined by flow cytometry. (B) 293T cells were transfected with plasmids encoding glutathione S-transferase (GST) fusions with full-length TRIM5α or with TRIM5 lacking the SPRY domain. Cells were lysed (50 mM Tris pH 8.0, 150 mM NaCl, 1% NP-40, 0.1% SDS) and mixed for 2 hrs at 4°C with virions (N-MLV or B-MLV) that had been concentrated by acceleration through 25% sucrose. GST fusions and associated proteins were enriched on glutathione-sepharose beads and immunoblotted with goat anti-MLV CA antibody (CA pull-out), or anti-GST antibody (bottom panel). Unbound CA remaining in the binding reaction was probed with anti-MLV CA antibody (CA input). TRIM5 protein domains fused to GST are indicated schematically on the bottom left: RF, ring finger; BB, B box; CC, coiled-coil. ::: ![](1742-4690-2-40-1) ::: Retroviral restriction specificity thus seems to be determined by TRIM5α binding to CA in a process that requires the SPRY domain. The fact that TRIM5α recognized retroviral CA presented by detergent-stripped virion cores, but not free CA protein, suggests that the SPRY domain recognizes a complex surface of multimerized CA. Once cores of restriction-sensitive viruses are singled out by the SPRY domain, TRIM5α blocks retroviral RT \[[@B1]\] by a mechanism that awaits elucidation. Our findings bring us one step closer to understanding how the potent antiviral activity of TRIM5α might be harnessed to block HIV-1 infection in people. List of abbreviations ===================== HIV-1, human immunodeficiency virus; MLV, murine leukemia virus; TRIM, tripartite motif protein; RT, reverse transcriptase; CA, retroviral capsid protein; GST, glutathione S-transferase; RF, ring finger domain; BB, B box domain; CC, coiled-coil domain. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= SS and JL conceived the experiments and wrote the manuscript. SS performed the laboratory work. Both authors read and approved the final manuscript. Acknowledgements ================ This work was supported by NIH grant AI 36199 to J.L.	PubMed Central	2024-06-05T03:55:58.179481	2005-6-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166576/", "journal": "Retrovirology. 2005 Jun 20; 2:40", "authors": [ { "first": "Sarah", "last": "Sebastian" }, { "first": "Jeremy", "last": "Luban" } ] }
PMC1166577	Introduction ============ Several chronic lung diseases are associated with an exaggerated accumulation and activity of neutrophils locally in the human bronchoalveolar space \[[@B1]\]. These lung diseases include acute, severe asthma, chronic bronchitis and chronic obstructive pulmonary disease (COPD) \[[@B1]\]. Neutrophils may play a pathogenic role in these diseases through the release of free oxygen radicals and by producing enzymes that increase the local proteolytic burden. Thus, via these compounds, neutrophils may contribute to pathogenic hallmarks such as non-specific bronchial hyper-responsiveness and hypersecretion, as well as to epithelial damage and tissue remodelling \[[@B1]\]. There is evidence that, among the proteolytic enzymes in the lungs, matrix metalloproteinase (MMP)-9 plays an important pathogenic role and this is most likely due to its capacity to cleave structural proteins such as collagens and elastin \[[@B2]\]. In support of this, local MMP-9 is increased in asthma and in COPD \[[@B3]-[@B8]\]. It is believed that neutrophils constitute an important source of MMP-9 in the lungs, even though macrophages, bronchial epithelial cells and fibroblasts may also produce MMP-9 under certain conditions \[[@B2]\]. Recent studies on mice in vivoand on human cells in vitroare compatible with activated T-lymphocytes contributing to the local accumulation of neutrophils in the bronchoalveolar space through the release of the cytokine interleukin (IL)-17A. However it remains unknown whether lymphocytes producing IL-17A actually reside within the bronchoalveolar space of humans \[[@B1]\]. In the bronchoalveolar space of mice and rats, local administration of recombinant IL-17A protein increases the number of neutrophils substantially \[[@B1],[@B9]\]. Systemic pre-treatment with a neutralizing anti-IL-17A antibody attenuates endotoxin-induced neutrophil accumulation in the same compartment of mice \[[@B10]\]. It is likely that IL-17A causes this neutrophil accumulation in part indirectly through the induced production of neutrophil-mobilizing cytokines, such as granulocyte-macrophage colony-stimulating factor (GMC-SF), IL-6 and IL-8, in local cells, including bronchial epithelial cells, fibroblasts and smooth muscle cells \[[@B1],[@B10]\]. In addition, there are now studies on mice and rats indicating that IL-17A may control the local proteolytic burden in the bronchoalveolar space in vivothrough its impact on neutrophil accumulation \[[@B11],[@B12]\]. We recently showed that organic dust in a swine confinement, a strong stimulus of neutrophilic inflammation, causes a substantial increase in the concentration of free, soluble IL-17A protein in the human bronchoalveolar space \[[@B13]\]. Interestingly, this increase in local IL-17A protein is associated not only with an accumulation of neutrophils but also with an accumulation of lymphocytes. The current study was undertaken to determine whether these lymphocytes can produce IL-17A in humans. The study was also undertaken to address whether production of IL-17A is associated with an increased local proteolytic burden from neutrophils. Exposure to organic dust in a swine confinement was utilised as a pro-inflammatory stimulus and intracellular expression of MMP-9 was targeted as a sign of proteolytic burden in the bronchoalveolar space of human healthy volunteers. Methods ======= Study design ------------ The study protocol was approved by the Ethics Committee at Karolinska Institute (Diary N. KI 95:347). Six healthy, non-smoking volunteers gave their written consent to participate in the study after receiving both written and oral information. All the participants were non-atopic, non-asthmatic as determined by history and questionnaire. They were exposed to organic dust while weighing pigs in a swine confinement (ie. a barn) during three hours, as previously described \[[@B14]-[@B16]\]. The procedure incorporates exposure to aerosolised organic dust, with a concentration of approximately 24 mg dust and 1.2 μg endotoxin per m^3^, as assessed in a recent study \[[@B13]\]. Normal lung function and normal airway responsiveness were ascertained in all subjects prior to exposure. The subjects underwent bronchoscopy with bronchoalveolar lavage (BAL-see below) 2 weeks before and 24 hours after the exposure. Lung function ------------- Spirometry was performed in accordance with the criteria of the American Thoracic Society using a low-resistance rolling-seal spirometer (OHIO model 840; Airco, Madison, WI, USA) \[[@B17]\]. Swedish reference values were used \[[@B18],[@B19]\]. Bronchial responsiveness ------------------------ The methacholine provocation test was conducted prior to exposure to ascertain a normal airway function, as previously described \[[@B20]\]. Briefly, after the inhalation of the diluent (vehicle) alone, doubling concentrations of methacholine were given starting at 0,5 and ending at 32 mg/ml. The test was stopped either at a FEV~1~decrease of 20%, compared to the value registered after the inhalation of the diluent, or after inhalation of the highest methacholine provocation (32 mg/ml). The cumulative dose causing a 20% decrease of FEV~1~(PD~20~FEV~1~) was calculated. Bronchoalveolar lavage ---------------------- Bronchoscopy and bronchoalveolar lavage were performed according to standard procedures previously described \[[@B13]\]. In short, after pre-medication with morphine-scopolamine, a flexible fibreoptic bronchoscope (Olympus Type 4B2; Olympus Optical Co. Ltd., Shinjukuku, Tokyo, Japan) was inserted through the nose under local anaesthesia with lidocaine (Xylocain R, Astra, Södertälje, Sweden). The bronchoscope was wedged in a middle lobe bronchus and a total of 250 ml of sterile saline at 37°C was instilled in 5 aliquots of 50 ml. After each instillation, the BAL suspension was gently aspirated and collected in a siliconized plastic bottle kept on ice. After sterile filtration (70 μm filter), the BAL suspension was centrifuged (Hereus Omnifuge 300 g, 10 min, +4°C). The cell-free supernatant was kept frozen (-80°C) until analysis. The cell pellet was resuspended (phosphate-buffered saline \[PBS\]: 5 ml). Viability of cells was assessed using trypan blue exclusion and the total cell number (ie. concentration) was determined for each sample in a Bürker chamber. BAL cytospin samples were prepared and stained (May-Grünwald-Giemsa) and differential cell counts were performed using light microscopy (Zeiss Axiopaln 2, Carl Zeiss, Jena, Germany). Messenger RNA for IL-17A ------------------------ Remaining cells from the resuspended pellet were centrifuged (see above) and the supernatant was removed. The cell pellet was re-suspended (PBS: 50 μl plus RNAlater from Ambion Inc, Austin, USA: 250 μl per 3 × 10^6^cells), transferred into microtubes, (3 × 10^6^cells per tube) and incubated (24 hrs, +4°C). The RNeasy minikit was used according to the manufacturer\'s protocol with the following modification: The eluted RNA was treated with Dnase I (Rnase-Free Dnase set: 30 min, 37°C). After this, the steps from adjusting binding conditions to eluting RNA were repeated. The eluted total RNA was measured for quantity and purity using spectrophotometry (SpectraMaxPlus^®^Molecular Devices Corporation Sunnyvale, CA, USA; absorbance at 260 and 260/280 nm respectively) and then stored frozen (-83°C) until purification of Total RNA. The RT-PCR ELISA technique was employed to quantify relative changes in the IL-17A mRNA gene transcript. A one-step RT-PCR was utilized with a Gene Amp PCR system 2400 (Perkin Elmer, Wellesley, MA, USA) for amplification. Each RT-PCR of 50 μl was conducted using 100 ng of total cellular RNA and 30 pmol of each primer, 10 U RNase inhibitor (recombinant RNasin, Promega Corporation, Madison, WI, USA), 200 μM PCR DIG Labelling mix (20 μM dATP, dGTP, dCTP each plus 19 μM dTTP plus 1 μM DIG-dUTP), 5 mM DDT, 10 μl RT-PCR reaction buffer (1.5 mM Mg+), 1 μl Titan enzyme mix (AMV reverse transcriptase+Taq DNA polymerase+Pwo DNA polymerase) or 0,5 μl Expand high fidelity PCR system (Taq DNA polymerase+Pwo DNA polymerase) for DNA controls (all reagents from Roche Diagnostics). Reverse transcription was performed (30 min, 50°C). The IL-17A and the house-keeping gene HPRT were subsequently annealed (55°C). IL-17A and HPRT were then amplified (35 and 30 cycles, respectively) and the DIG labelled PCR product was stored frozen (-20°C) before the detection step. RT-PCRs were performed in duplicate. Gene sequences were accessed from NCBI Database and accession number used was for human (h) HPRT V00530 and for hIL-17 U32659. Scandinavian Gene Synthesis AB (Köping, Sweden) provided the oligonucleotides. The gene primer sequences used for RT-PCR and ELISA detection: hHPRT Sense (5\') 5\'CGT CGT GAT TAG TGA TGA TGA AC3\' Antisense (3\') 5\'GCA AAG TCT GCA TTG TTT TGC CA3\' Internal probe 5\'GAG GCC ATC ACA TTG TAG CCC TCT GTG3\' hIL-17 Sense (5\') 5\' GTG AAG GCA GGA ATC ACA ATC 3\' Antisense (5\') 5\' ACC AGG ATC TCT TGC TGG AT 3\' Internal probe 5\' CAG AGT TCA TGT GGT AGT CCA CGT TCC CA 3\' The DIG labelled PCR products were denaturated and hybridized with the biotinylated internal probe specifically designed to hybridize with each gene PCR product, and immobilized on streptavidin coated microtiter plates (3 hrs, 42°C), utilising the PCR ELISA DIG-detection kit (Roche). After washing, the bound PCR-products were incubated (30 min, 37°C) with an anti-DIG-horseradish peroxidase conjugated antibody followed by reaction (20 min) with the substrate 2,2\'-azino-di(3-ethyl benzthiazoline sulfonate) (ABTS). The absorbance was then measured using spectrophotometry (see above, at 405/492 nm) in an ELISA plate reader (Labsystems Multiscan Multisoft, Vanda, Finland). The expression of transcripts for IL-17 mRNA was normalized to the expression of HPRT transcripts and shown as percent of the house-keeping gene (% HPRT). In parallel with tested samples, control samples for PCR, probe specificity, DNA contamination, hybridisation and sample dilution were run. Immunocytochemistry (ICC) ------------------------- ### General procedure BAL cells were fixed with formaldehyde (2%, 30 min, on ice) and washed twice in buffer prior to making cytospin preparations. Air-dried samples were stored frozen (-80°C) until further use. After thawing, samples were treated with donkey serum (10%) to avoid unspecific binding. Endogenous biotin was blocked using the Biotin Blocking System (DAKO corporation, Glostrup, Denmark). ### IL-17A immunoreactivity (IR) The intracellular expression of IL-17A protein was assessed utilising cytospin slides incubated with a polyclonal goat anti-human IL-17A antibody (R&D: 1 hr). As secondary antibody, a biotinylated F(ab\')~2~fragment donkey anti-goat IgG (Jackson ImmunoResearch laboratories Inc) was used, followed by alkaline phosphatase-conjugated streptavidin (DAKO). All the solutions above were supplemented with saponin (Sigma: 0.1 %), to permeabilise cell membrane. Bound antibodies were visualised with Vector Red Alkaline Substrate Kit (Vector Laboratories, Inc. Burlingame, CA, USA). Mayer\'s Hematoxylin (Sigma) was used for counterstaining. ### MMP-9 immunoreactivity The intracellular expression of MMP-9 protein was assessed utilising cytospin slides incubated with a polyclonal goat anti-human MMP-9 (R&D: 1 hr). The secondary antibody and the detection system were the same as for IL-17A (see above). Human neutrophilic lipocalin ---------------------------- The concentration of the neutrophil-specific activity marker human neutrophilic lipocalin (HNL) was measured in cell-free BAL fluid using a solid phase, double-ligand radioimmunoassay as described elsewhere \[[@B21]\]. Data analyses ------------- Data are presented as median (range) values unless otherwise stated. The analysis of differences between measurements prior to and after the exposure, were conducted utilizing the sign test, assuming binominal distribution. To avoid falsely positive conclusions, the significance level at each test was set to 0.05/3 (ie. 0.0167), since three key variables (IL-17A mRNA, IL-17 IR and MMP9 IR) were compared in the current material. nrefers to number of human subjects. Results ======= Clinical characteristics prior to exposure ------------------------------------------ Four males and 2 females with a median age of 22 (19--24) were included in the study. There were no apparent gender-related differences in pulmonary function of the subjects. FEV~1~was 96 (92 -- 105) % of predicted value and vital capacity 98 (90--102) % predicted value. All subjects had normal bronchial responsiveness to metacholine (PD~20~FEV~1~for metacholine: 5.3 \[1.7--15\] mg). Cells in BAL fluid ------------------ The BAL recovery volume was similar before (195 \[170--234\] ml) and after (196 \[190--212\] ml) the exposure. The exposure caused more than a two-fold increase in the total BAL cell number, more than a five-fold increase in the number of lymphocytes and more than a ten-fold increase in the number of neutrophils.(Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Concentration of BAL cells. BAL cells before and after exposure to organic dust in a swine confinement. Data are presented as 1 × 10^6^/L median (range). ::: Cell type Before After --------------- ------------------- --------------------- Total cells 71,3 (57,9--80,9) 165,1 (98,2--365,6) Lymphocytes 4,4 (1,2--10.0) 23,1 (9,8--61,6) Neutrophils 2,3 (0,8--3,6) 23,1 (6,6--167,3) ::: Messenger RNA for IL-17A ------------------------ Before the exposure, 4 out of 6 persons displayed no detectable levels of mRNA for IL-17A, whereas all six subjects displayed detectable and increased levels after the exposure (fig. [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Messenger RNA for IL-17A in BAL cells. Messenger RNA for IL-17A in BAL cells (percentage of the house-keeping gene HPRT), measured 2 weeks before and 24 hours after exposure to organic dust in a swine confinement. Data are shown as individual (rhombs) plus median (bold horizontal lines) values. \: p \< 0,0167; n = 6 ::: ![](1465-9921-6-44-1) ::: IL-17A immunoreactivity ----------------------- IL-17A immunoreactivity was mainly detected in BAL lymphocytes (fig. [2](#F2){ref-type="fig"}). In addition, the percentage of lymphocytes positive for IL-17A immunoreactivity (fig. [3](#F3){ref-type="fig"}) tended to increase in 5 out of 6 subjects after the exposure. In contrast, a much lower fraction of macrophages expressed a weak signal for IL-17A immunoreactivity and there was no pronounced increase in this signal after the exposure (data not shown). BAL neutrophils did not display IL-17A immunoreactivity (data not shown) ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Immunocytochemical detection of IL-17A protein in BAL lymphocytes. BAL lymphocytes expressing IL-17A IR (brown arrows) after exposure to organic dust in a swine confinement, detected using immunocytochemistry. ::: ![](1465-9921-6-44-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### BAL lymphocytes expressing IL-17A protein. Percentage of BAL lymphocytes expressing IL-17A IR before and after the exposure to organic dust in a swine confinement. Data are shown as individual (rhombs) plus median (bold horizontal lines) values. NS: p \> 0,0167; n = 6 ::: ![](1465-9921-6-44-3) ::: MMP-9 immunoreactivity ---------------------- Neutrophils were the prevailing cell type in BAL fluid expressing MMP-9 immunoreactivity (fig. [4](#F4){ref-type="fig"}) and this signal was substantially increased after exposure (fig. [5](#F5){ref-type="fig"}). However, the fraction of neutrophils (% all neutrophils) expressing MMP-9 immunoreactivity was not increased after exposure (data not shown). Very few macrophages expressed MMP-9 immunoreactivity and this signal was consistently weak (data not shown). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Immunocytochemical detection of MMP-9 protein in BAL neutrophils. Abundance of BAL neutrophils positive for MMP-9 IR (purple arrows) after exposure to organic dust in a swine confinement. ::: ![](1465-9921-6-44-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### BAL neutrophils expressing MMP-9 protein. BAL neutrophils expressing MMP-9 IR (percent of total cells) before and after exposure to organic dust in a swine confinement. Data are shown as individual (rhombs) plus median (bold horizontal lines) values. \: p \< 0,0167; n = 5. ::: ![](1465-9921-6-44-5) ::: HNL --- The exposure did not cause any substantial increase in the concentration of free, soluble HNL in the BAL fluid, even though there was a weak average trend in this direction (data not shown). Furthermore, the amount of HNL per neutrophil tended to be decreased after exposure (data not shown). Discussion ========== The current study demonstrates that, in previously healthy human volunteers, exposure to organic dust substantially increases the messenger RNA for IL-17A in BAL cells. Our study also demonstrates that lymphocytes constitute the predominant cell type that expresses intracellular IL-17A protein among BAL cells. The demonstrated increase in IL-17A mRNA is associated with an increased absolute number of MMP-9 expressing BAL neutrophils at the same time point. The aerosolised organic dust in a swine confinement, that was utilised as a pro-inflammatory stimulus in our study, is a more complex stimulus than endotoxin, because of its mixed content of grain dust, ammonia, fungi, Gram-positive and Gram-negative bacteria \[[@B15]\]. It has previously been shown that exposure to this type of organic dust causes severe inflammation in the bronchoalveolar space of humans \[[@B13],[@B15],[@B16]\]. Our current study now adds to these previous studies novel evidence of de novosynthesis of IL-17A by local cells within the bronchoalveolar space of humans \[[@B1]\]. Furthermore, our study forwards lymphocytes as a local source of IL-17A protein in the human bronchoalveolar space, because the intracellular immunoreactivity for this specific protein was found mainly in certain BAL lymphocytes. Noteworthy, even though this immunoreactivity should be regarded as a qualitative assessment primarily, it was increased eight-fold by average after exposure with only one subject lacking a detectable increase. Clearly, this is compatible with lymphocytes accounting for production of IL-17A protein in the bronchoalveolar space of humans. Indeed, our current results are supported by our previous demonstration of an increase in the final consequence of the increased transcription and intracellular loading of protein, namely in free, soluble IL-17A protein in cell-free BAL fluid \[[@B13]\]. It is also noteworthy that in our current study, certain but not all lymphocytes in the human bronchoalveolar space express IL-17A protein after exposure to organic dust. This observation is compatible with previous studies on activated mouse CD4^+^and CD8^+^lymphocytes from BAL and spleen as well as human CD4^+^and CD8^+^lymphocytes from blood \[[@B1],[@B22]-[@B28]\]. Taken together, these observations warrant new specific cellular targets for additional analysis in future studies on human airways. We assessed intracellular immunoreactivity for MMP-9 as a marker of proteolytic burden, both before and after exposure to organic dust. Even though some of the BAL macrophages displayed a very weak expression of MMP-9 immunoreactivity, as has previously been reported in a different inflammatory setting \[[@B29]\], we found the most pronounced increase in and expression of MMP-9 immunoreactivity consistently among BAL neutrophils. This observation fully supports that neutrophils can constitute an important source of MMP-9 in the human bronchoalveolar space in vivo, at least after exposure to organic dust. At the same time point after the exposure, the MMP-9 expression in neutrophils was associated with an increase in mRNA for IL-17A, even though the fraction (percentage) of neutrophils expressing MMP-9 was not increased. This is compatible with IL-17A indirectly contributing to the local proteolytic load, mainly through its increasing effect on the number of neutrophils within the human bronchoalveolar space; a type of mechanism supported by previous studies on the bronchoalveolar space of mice in vivo\[[@B9],[@B12]\]. We also measured the concentration of free, soluble HNL protein in BAL fluid, as an assessment of neutrophil-specific activity \[[@B30]\]. However, we detected only a weak average trend towards a modest increase in the HNL concentration after exposure to organic dust and this increase did not even correspond to the substantial increase in neutrophil number; the amount of HNL per neutrophil was actually decreased after the exposure. Furthermore some subjects displayed even a decrease in the total HNL concentration after exposure. Moreover, separate and preliminary assessments of the myeloperoxidase (MPO) concentration in BAL fluid displayed a weak increase not corresponding to the increase in neutrophil number (data not shown). Similar findings on MPO and neutrophils have previously been published \[[@B31]\]. Taken together, all these findings imply that, in terms of MPO and HNL, there is actually no true increase in the average activity of each accumulated neutrophil in the human bronchoalveolar space after exposure to organic dust. Again, from a mechanistical point of view, this finding in the bronchoalveolar space of humans is supported by our recently published findings on the effect of recombinant IL-17A in the bronchoalveolar space of mice in vivo\[[@B12]\]. This IL-17A induces an increase of MMP-9 protein and in corresponding gelatinase activity, that is due to an increased number of neutrophils rather than an increased amount of MMP-9 per neutrophil \[[@B12]\]. In conclusion, supported by our recent study showing a corresponding increase in free, soluble IL-17A protein \[[@B13]\], the current study adds novel evidence for organic dust inducing de novosynthesis of IL-17A protein in a subset of local lymphocytes within the human bronchoalveolar space. Our current study on humans also adds evidence compatible with induced de novosynthesis of IL-17A being associated with an increased proteolytic burden due to a local accumulation of neutrophils rather than an increased activity in each of these inflammatory cells. Interventional studies will be required to determine whether targeting IL-17A is therapeutically beneficial in lung diseases characterised by excessive accumulation of neutrophils. Authors\' contributions ======================= LP, KL, AL -- study design and coordination SI, LP, KL, PV -- exposure in swine farm, laboratory work SI, AL -- data analysis and interpretation of results SI, LP, KL, PV, AL -- preparation and revision of the manuscript Acknowledgements ================ The present study was financially supported by Göteborg University, the Karolinska Institute, the Swedish Heart-Lung Foundation, the Swedish Medical Research Council (K2002-74X-09048-13A), the Vårdal Foundation and, finally, by Glaxo-Wellcome, Sweden. No support was obtained from the tobacco industry. The excellent technical support by Carina Malmhäll, B Sci, and Margareta Sjöstrand, PhD, is gratefully acknowledged. The support with statistical analysis by Associate Professor Kerstin Wiklander, Department of Mathematics, the Chalmer\'s University of Technology in Göteborg is also acknowledged.	PubMed Central	2024-06-05T03:55:58.180495	2005-5-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166577/", "journal": "Respir Res. 2005 May 25; 6(1):44", "authors": [ { "first": "Stefan", "last": "Ivanov" }, { "first": "Lena", "last": "Palmberg" }, { "first": "Per", "last": "Venge" }, { "first": "Kjell", "last": "Larsson" }, { "first": "Anders", "last": "Lindén" } ] }
PMC1166578	Introduction ============ Transcription, the first step of information flow from DNA, is regulated by sequence specific DNA-protein interactions. The regulation depends on the presence of cis-acting elements. The best examples of cis-acting elements are promoters. Other well-known examples in bacteria include the Shine Dalgarno (SD) sequence, sigma 32 binding site, LexA binding site, etc. In bacteria, promoters recognized by sigma factors initiate transcription. The responses of an organism to various stimuli are mediated by changes in gene expression patterns. These changes are initiated by promoter-sigma factor interactions and regulated by other cis-acting elements. Thus, families of co-regulated genes are under the control of the same promoter. Though core promoters are small words (6--8 bases), certain changes that are permissible in promoter sequences have little or no effect on their activity. This means that a few closely-related sequences, in the right context, can function as promoters. Identifying promoters is a challenging yet rewarding problem; challenging because promoters can differ subtly in sequence and still retain function, and rewarding because it can shed light on an organism\'s life style. Computational approaches are required since experimental methods for identifying promoters are not applicable on a genome-wide scale. In most instances, computational identification or prediction of promoters involves model-based searches. The model is, by and large, derived from prior data. Techniques using artificial neural networks \[[@B1]\] or genetic programming methodologies \[[@B2]\] are also used, and require prior experimental data. Using prior data for identifying new candidates is also known as dictionary-based searching. Databases of experimentally verified cis-acting elements are available for promoter prediction \[[@B3]\] through dictionary-based approaches. These approaches are biased towards the best-characterized promoter in the initial dataset, though non-redundant data sets have been used recently \[[@B4]\]. A paucity of experimental data can compromise the efficiency of these methods. The success of dictionary-based methods is directly dependent on the relatedness of the database to the query. It has also been observed that, while using dictionary-based methods, taking into account the local genomic landscape for generating Markov profiles improved the prediction quality in eukaryotes \[[@B5]\]. Another method that has been applied to both simpler and larger genomes is the comparative genome analysis method. It is observed that functional regions, albeit non-coding, are conserved across species and genera. Analyses of this kind have been used for yeast \[[@B6],[@B7]\], higher eukaryotes \[[@B8],[@B7]\] and bacterial regulons \[[@B9],[@B7]\]. In Saccharomyces cerevisiae, the distribution of certain words across the genome is non-random. For example, some words appear to be preferred in regions upstream \[[@B10]\] or downstream \[[@B11]\] of genes. Analyses showed that such words occurring preferentially near the genes represent functional elements. Though non-random usage of k-sized words in bacterial genomes has been documented \[[@B12],[@B13]\] in genomic contigs, studies have not focused on the upstream regions of prokaryotic genes. We have developed a method that uses preferential occurrence of k-sized words within specific (gene-proximal) regions in a given genome to predict cis-acting elements. This method does not use a dictionary or database for initiating searches. The method can be applied to any genome of which the gene co-ordinates are known. Its advantage is that there is no extrapolation of data. This allows unique families of cis-acting elements for a given genome to be determined. Inter-genome comparison can establish the functionality of conserved words across genera. The results of oligonucleotide profiling as applied to the genome of E.coliK12 \[[@B14]\] are presented. Comparative analyses of the resultant oligonucleotide profiles show that a subset of preferred hexanucleotides in E.coli-K12 is conserved across two other genomes, those of Salmonella typhiand Yersinia pestis\[[@B15],[@B16]\]. We suggest a function for the ubiquitous hexanucleotides that are preferentially present in -100 regions and are neither single-base relatives of TATAAT or AGGA nor CTAG-containing, and we propose a novel model for bacterial promoters. Results and Discussion ====================== The results of oligonucleotide profiling, as performed for E.coliK12 genome, are discussed. The word size was restricted to six. For higher word sizes the word occurrence frequency was low. Smaller words were not used since the intra-word Markov dependencies, if any, are statistically invalid \[[@B17]\]. Word occurrences were analyzed in four contiguous sequence sets, F4 through F1 (Fig. [1a](#F1){ref-type="fig"}),. A threshold of 200% (two-fold increase in occurrence over the genomic average) was set to identify signals for cis-acting elements. The average occurrence of a random hexanucleotide in a sequence set is 4.6% of its genomic total and the standard deviation is 0.573. It can be seen that a two-fold increase (9.2%) is more than six times the standard deviation (σ) above the average. Any hexanucleotide that had at least 9.2% of its overall genomic occurrence within any of the four fragments analyzed was termed \"enriched\" in that respective region. Such enrichment was more pronounced in the gene-proximal regions (-1 to -100 region) than in the distal regions (-300 to -400). In the three random sequence sets (controls), only once did we find enrichment (a CTAG-containing element). Fig. [1a](#F1){ref-type="fig"} schematically illustrates this procedure. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### \(A) A schematic representation of the procedure used for profiling, incorporating the definition of the four fragments F1, F2, F3 and F4 used in this study. (B) Comparison of the occurrence distribution in the random control (series 1), F4 (series 2), F3 (series 3), F2 (series 2) and F1 (series 4). (C) Number of words whose occurrence is greater than μ+Nσ, where N is on the x axis. (D) Distribution of the three classes of oligonucleotides in the four fragments: TATAAT for class 1, AGGAGG for class 2 and AAAAAA for class 3. ::: ![](1742-4682-2-20-1) ::: The preferential occurrences of hexanucleotides within the controls and the fragments under study are contrasted in Table [1](#T1){ref-type="table"}. The distributions of hexanucleotide occurrence in control 1 and fragments (F1-F4) are shown in Fig. [1b](#F1){ref-type="fig"}, while Fig. [1c](#F1){ref-type="fig"} shows the number of hexanucleotides with frequencies N × (σ) more than average. The units on the X-axis are N (N times σ) and 200%. ::: {#T1 .table-wrap} Table 1 Threshold(C)/Region(R) \%\>μ+2 \%\>μ+3 \%\>μ+4 \%\>200 ------------------------ --------- --------- --------- --------- Control1 184 42 18 1 Control2 155 30 9 0 Control3 164 46 16 0 F1 229 66 24 7 F2 341 116 44 0 F3 662 387 236 14 F4 1112 834 634 190 ::: The method retrieved 183 hexanucleotides that were enriched in the -100 region. These included the Pribnow box (TATAAT), SD consensus (AGGA), the LexA binding site (CTATNN), sigma 32 binding site one-base relative (CCCTAT) and CTAG-containing regulatory elements \[Supplementary Information 1\]. The CTAG-containing elements are known to be core repressor binding regions in the Trp, Met and MalPQ operons and the treA gene \[[@B18]-[@B20]\]. They occur at high frequency near the rRNA gene clusters \[[@B12]\]. However, in the rest of the genome, we find their distribution to be roughly uniform (data not shown). Certain trends are apparent in the usage of enriched oligonucleotides by bacterial genomes. The occurrence of some oligonucleotides increases gradually with proximity to genes (class I oligonucleotides), while others (class II oligonucleotides) peak near the genes. A third class comprises non-specifically preferred oligonucleotides (Class III oligonucleotides). Class I Oligonucleotide ----------------------- Bacteria are expected to have limited number of promoter elements and to have them near genes. The Pribnow box in E.coliis a representative promoter. The overall frequency of the Pribnow box is lower than the genomic average (1067 occurrences as against the genomic average of \~1400). Here, we analyze: the occurrence of the Pribnow box and its single base substitution relatives, the distribution of the Pribnow box within the -100 region, and the position-dependency of other bases on the Pribnow box in its vicinity. For this analysis, Pribnow box occurrences in the -100 region alone were taken into account for four strains of E.coli. ### Occurrence of Pribnow box This analysis shows that the occurrence of the Pribnow box increases gradually as one goes closer to genes. Furthermore, seven of its one-base substitution relatives figure in the enriched list \[Table [2](#T2){ref-type="table"}\]. Most of these one-base relatives show a gradual but definite increase in their occurrence as we move nearer the genes \[Table [2](#T2){ref-type="table"}\]. This gives an idea as to how an element that has a function similar to the Pribnow box would behave in other genomes. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Occurrence of single base relatives of TATAAT in E.coligenome. F1:-301 to -400; F2:-201 to -300; F3: -101 to -200; F4: -1 to -100. Those elements that are enriched (\> = 200%) are marked by an asterisk in the last column. ::: Hex Total Occ. F4 F3 F2 F1 Occ. % in F1 ENR -------- ------------ ----- ----- ----- ----- -------------- ----- TCTAAT 595 29 30 47 46 7.731092 TAGAAT 507 25 23 36 71 14.00394 \* TATAGT 681 19 38 30 87 12.77533 \* TATAAG 879 45 32 67 102 11.6041 \* TACAAT 903 42 56 71 109 12.07087 \* TATAAC 1590 65 80 94 121 7.610063 TATGAT 1539 63 111 94 127 8.252112 TATACT 620 39 40 62 131 21.12903 \* CATAAT 2448 90 106 115 132 5.392157 TATAAT 1036 58 68 78 134 12.93436 \* TGTAAT 1870 74 80 129 140 7.486631 TATCAT 2082 101 96 128 164 7.877041 TATATT 1943 87 130 145 168 8.646423 TATTAT 2280 118 124 155 178 7.807018 GATAAT 3735 125 140 172 201 5.381526 TATAAA 2304 98 144 177 236 10.24306 \* TTTAAT 3671 182 205 219 249 6.782893 TAAAAT 2947 142 159 204 268 9.093994 AATAAT 4132 188 201 287 287 6.945789 ::: ### Distribution of Pribnow box Analyses show that the maximal number of strong minimal promoters occur within the -100 region and that the Pribnow box prefers the -30 to -70 position, centering around -40 \[Fig. [2a](#F2){ref-type="fig"}\]. The report by Collado-Vides et al. shows that \~80% of the 800 genes analyzed have their promoters in the -100 region. In fact, the highest concentration of promoters that they report is at the -40 region \[[@B21]\], which we corroborate. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Addressed promoter model. (A) Occurrence distribution of TATAAT, AGGAGG and AAAAAA within the -100 region using a 30-base window: -1 to -30, -10 to -40, -20 to -50, \..., -70 to -100. (B) A schematic comparison of the classical and the addressed promoter models. Blue peaks represent the canonical promoter. Red background (where present) represents the address. ::: ![](1742-4682-2-20-2) ::: ### Markov dependency analysis of sequences surrounding Pribnow box Markovian analysis of TATAAT-containing sequences (within the -100 region) was done for E.coli. For analysis, such sequences were taken from all four E.coligenomes (K12, O157:H7, EDL933 and CFT073) to improve statistical significance (TATAAT occurred only 128 times in the -100 region of the K12 genome). The results showed that TTGACA is preferred between positions -32 and -27. Further, it was seen that, with G at -14, the occurrence of TTGACA decreased, (All corresponding data points are highlighted in the Supplementary Information 2 file.) This has been reported by analysis of experimentally characterized promoters \[[@B22]\]. These correlations validate the results of oligonucleotide profiling with respect to the sigma 70 binding site. Class II Oligonucleotides ------------------------- AGGA- (SD consensus) and CTAG-containing hexanucleotides belong to this class. Unlike the Class I oligonucleotides, Class II oligonucleotides show a steep increase in occurrence in the -100 region. This is expected in the case of the Shine-Dalgarno sequence (AGGA), since it should lie within 30 base pairs upstream of the ORF start site (owing to geometric constraints imposed by the ribosomal complex). Another example of this class is the tetranucleotide CTAG, representing all the hexanucleotides that contain it. CTAG kinks DNA when bound by proteins \[[@B23]\], making it a likely candidate for a regulatory site. CTAG also has low genomic frequency, uniform distribution and a preference for the -100 region. This might imply a global regulatory function. Class III Oligonucleotides -------------------------- Certain oligonucleotides not only have a more than average genomic frequency but are also more common in the -100 region. Many of these are A/T rich oligonucleotides, which are known to bend DNA when present in stretches \[[@B24]\]. The presence of such A/T repeat elements upstream \[[@B25]\] and downstream \[[@B26]\] of the canonical promoter is necessary. They are evidently not stand-alone signals. We propose that they are facilitator elements that are necessary but not sufficient for promoter recognition and function. The set of such oligonucleotides that were readily distinguished as facilitators is given, along with their distribution, in Supplementary Information 3. They occur preferentially up to -100 and beyond. We find this significant since a recent report shows that DNA of size 90 base pairs can bend upon itself in a sequence-dependent manner \[[@B24]\]. Though all 64 A/T containing hexanucleotides were found to occur more frequently than the genomic average, only 18 of them were enriched in the -100 region. Thus, the increased occurrence of Class III hexanucleotides is not an artifact of increased base frequency. It transpires that the genome increases the bending capacity of the -100 region by preferential usage of certain oligonucleotides. The occurrence of hexanucleotides representing each of the three classes is shown in Fig. [1d](#F1){ref-type="fig"}. TATAAT is used to represent class I, AGGA-containing hexanucleotides to represent class II and AAAAAA to represent class III. Protein Binding Capacity of the -100 region: Evidence from NDB -------------------------------------------------------------- We analyzed the occurrence of enriched hexanucleotides in a protein-bound state in the NDB database \[[@B27]\]. Of the \~130 hexanucleotides that are neither TATAAT-related (1 base substitution oligonucleotides) nor AGGA- or CTAG-containing, 112 have at least one occurrence in the database, bound to proteins \[data not shown\]. Most of them occurred more than once in the database in a protein-bound state. These results show the propensity of the genome to increase the protein-interacting capacity of the -100 region and hence increase the activity of this region. Dependency Analysis ------------------- A position-specific probability matrix (PSPM) was created for enriched oligonucleotides that were not TATAAT related or AGGA/CTAG-containing. This matrix was used to determine the tendency of hexanucleotides to assume specific consensus words within the -100 region of the genes. Secondary matrices were derived by anchoring the first base in the PSPM. The consensus words derived from these matrices are given in Supplementary Information 4. For each secondary matrix, two more character states were chosen for anchoring on the basis of their prominence,. The results show a strong preference for tetra-A signals, TATA-containing signals and GGA-containing signals. Inter-genome comparison of hexanucleotide usage profiles -------------------------------------------------------- Conservation of DNA sequence across genomes has been established as a pointer to functionality. This method has been used to identify regulatory regions in Saccharomyces\[[@B6]\] by sequence comparison among different species. We see that the logic extends beyond conservation of sequences and patterns to that of oligonucleotide profiles. We have compared the profile of enriched hexanucleotides between E.coli, Salmonella entericaand Yersinia pestisto test its validity. The E.coliand Salmonellaprofiles shared 110 enriched oligonucleotides out of 160 in Salmonella typhi. Yersinia pestis, whose profile had 97 enriched oligonucleotides, shared 66 of them with E.coli. Of those that were conserved across genomes, the AGGA-containing and CTAG-containing hexanucleotides, TATAAT, and the LexA binding site were prominent (Supplementary Information 5). While conservation of hexanucleotides usage implies functionality, the converse may not be true and might reflect unique regulatory / facilitator elements for each genome. Role of facilitator elements in promoter identification and the Addressed Promoter Model ---------------------------------------------------------------------------------------- Classical promoters in bacteria are sigma factor binding sites. The sequence that is known to bind to sigma factor with maximal affinity in vitro is taken to be the strongest promoter. DNA footprinting experiments do not allow us to assess the importance of the surrounding sequences. It is clear from the profiles that the strongest promoters have limited occurrence in the genome. Most genes are controlled by sigma 70 in E.coli\[[@B28]\], and only \~12% of the overall strong consensus occur in a region where they are maximally effective \[[@B21]\]. The question to be addressed is how a sigma factor (Sigma 70 in this case) can distinguish the promoter from non-specific promoter-like signals (degenerate -10 and -35 like signals in non-functional places in the genome). The sigma factor could not read every one of the possible signal combinations since this would result in enormous loss of time in bacterial genomes. In larger genomes, given the small size and degeneracy of the promoters, it is possible that the sigma factor would recognize a false signal on most occasions. To account for the efficiency of promoter recognition in the organism, we propose the addressed promoter model, where the sigma factor binding element is an information-dense peak (specific information) within a plateau of moderate information density (different but related words). The peak and the plateau together constitute the promoter. The plateau is formed by class III oligonucleotides that have the capacity to bend DNA. The facilitators are an integral part of the promoter. The presence of facilitators, which occur in greater frequencies around the core promoter, will serve as addresses for the core promoter. These addresses act as homing segments that allow the transcription factor to recognize the core promoter and bind to it. This model immediately suggests a way of identifying cis-acting regions in eukaryotes, where greater genome sizes and more degeneracy are seen. The extension of this logic would be to view enhancers and other regulatory regions in large eukaryotic genomes as local landscapes rather than as sequence motifs. While the protein binding sites would still be sequence motifs, their occurrence in a particular landscape may prove to be the determining factor for their activity. This accords with the observation of Huang et. al. \[[@B5]\] that local genomic landscape information affects the prediction quality of promoter elements. To illustrate this model, we have analyzed the distribution of one representative element from each of the three classes. The distribution was studied in a 30-base sliding window with a 10-base pitch. The representative elements are TATAAT (Class I), AGGAGG (Class II) and AAAAAA (Class III). The distribution is shown in Fig. [2a](#F2){ref-type="fig"}. It can be seen that AAAAAA forms a plateau around the TATAAT peak. The classical model and the addressed promoter model are contrasted in Fig. [2b](#F2){ref-type="fig"}. Conclusion ========== This method for identifying regulatory regions in DNA is powerful. Its strength is its ability to use the genomic sequence as a control. This obviates the need for data extrapolation from related genomes. The method can identify functional elements that can be experimentally characterized. Application of this method to the E.coliK12 genome reveals the presence of at least three classes of cis-acting elements. The occurrence, distribution and dependencies of these elements have been analyzed. Most of the profile data correlate with existing experimental evidence. The canonical sigma70 promoter has been analyzed in further detail in four E.coligenomes. The information derived from E.coliK12 using this method suggests that the functionality of a promoter is determined not only by the sequence of the core promoter element but also by its local milieu. We note that the occurrence of proposed facilitator elements extends just beyond the length known for DNA to bend upon itself (90 bp) and this, together with other reports about AT-rich tracts in the vicinity of the canonical promoter, suggests that the sigma factor recognizes a promoter more efficiently if it is present in the \"address\" region. This immediately explains why the transcription process is efficient in spite of the degeneracy that the promoter exhibits. We see that the occurrence of facilitators is not an artifact of increased base frequencies. The occurrence of many of the enriched hexanucleotides as protein-bound DNA complexes in the NDB database is indicative of their protein-interacting ability. This reflects on the protein binding capacity of the gene proximal regions in E.coliK12. The limitation of this method is its inability to pick up rare regulatory elements. In small genomes the method is known to give false positives, and in degraded genomes it picks up false negatives. In such cases, comparative analysis with related genomes will give valuable information. Methods ======= Sequence Extraction ------------------- Published genome sequences from the NCBI database <http://www.ncbi.nlm.nih.gov>(.fna file) were used. The start sites of genes given in the annotation file (.ptt file) were used for extracting upstream sequences of all the genes. Upstream sequences were taken only from their respective strands (+ strand for + genes and vice versa) because of the directionality of promoters. Four such fragments were taken from upstream of each gene, viz. -1 to -100, -101 to -200, -201 to -300 and -301 to -400. The distance between any two genes was not given importance because of the possibility that regulatory and transcriptional start sites may be present in the coding region of the preceding gene. Profiling --------- For every gene in the E. coliK12 genome, four contiguous DNA fragments from the corresponding strand were extracted. The length of each fragment was 100 bases. The fragments were named F4 through F1, where F1 is the gene-proximal fragment. There are 4311 genes in E.coli. Four sequence sets, one each for F1, F2, F3 and F4, were created for all the genes. Each of these sequence sets covers approximately 4.6% of the genome. Occurrence of all hexanucleotides was counted on both strands of the genome and the four upstream-sequence sets. The Compseq program from the EMBOSS \[[@B29]\] suite was used for this purpose. Any word that was non-functional was expected to be distributed equally across the sequence sets. Thus, for a non-functional word in the upstream context, we expected approximately 4.6% of its genomic occurrence in any of the sequence sets. Since cis-acting elements are gene-proximal, we expected their occurrence to be higher in F1 than elsewhere. We set a threshold (T) of 200% in word frequency to identify signals. Given a standard deviation of 0.56, it is apparent that a 200% increase (9.2% of genomic occurrence) is more than 6σ, which is significant. Words whose frequency in a given sequence set was 9.2% or more were termed \"enriched\" in the corresponding fragment. All analyses were carried out using Perl 5.6.1<http://www.perl.com> scripts on a Mandrake Linux 9.1 platform. The complete dataset is available in an in-house MySql <http://www.mysql.org>-based server. Markov Dependency Analysis -------------------------- We analyzed the character-state probabilities of all the words (137 words) for which a function could not be assigned. For this, we created a position-specific probability matrix (PSPM). The PSPM was derived from a position-specific frequency matrix (PSFM), which is defined as follows. For a word size of L, a PSFM is a 4 × L matrix M, where each element M~i,j~\[i ∈ {A,T,G,C} and j ∈ {1,2, \... L}\] is the number of times the character state i occurs at position j. In this case, L = 6. If S is the sum of all occurrences of words, then the PSPM is related to the PSFM as given below: PSPM = (1/S) × PSFM Such a matrix was used to derive consensus words preferred in the -100 region. From the PSPM, four sub-matrices were derived by anchoring the various character states (A, C, G, and T) at the first position. Further dependencies were analyzed by subsequent anchoring of two more positions, based on their prominence in the sub-PSPMs, to their representative character states. Markov Analysis for TATAAT-dependent Signals -------------------------------------------- For each occurrence of the Pribnow box within the -100 region, the preceding 50-base region was extracted. The PSPM was created for the sequence set as described above, where the value of L is 50. Different profiles were created by anchoring the base profile at all positions with all four bases. This was used to analyze the dependency of upstream signals on TATAAT. This analysis was done on a sequence set collated from all the four strains of E.coli. Authors\' contributions ======================= KS gave the core idea for oligonucleotide profiling, analysis of occurrence and proposed the model. ASNS worked with KS in profiling and analyzing the statistical significance of results, and KrS worked with KS in analyzing the distribution of words in gene proximal regions. GM was involved in analysis of results and critically analyzing the manuscript. PG is the group leader. Acknowledgements ================ The authors would like to thank Ms. Anishetty for discussions, and to acknowledge the financial support given by Council for Scientific and Industrial Research, Government of India and Department of Biotechnology, Government of India through the BTIS programme. We also extend our thanks to the developers of EMBOSS for making it available free of cost. We wish to acknowledge the contribution of the Free Software Foundation, MySQL, PERL community and others for making valuable software available free.	PubMed Central	2024-06-05T03:55:58.182968	2005-5-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166578/", "journal": "Theor Biol Med Model. 2005 May 31; 2:20", "authors": [ { "first": "Karthikeyan", "last": "Sivaraman" }, { "first": "Aswin Sai Narain", "last": "Seshasayee" }, { "first": "Krishnakumar", "last": "Swaminathan" }, { "first": "Geetha", "last": "Muthukumaran" }, { "first": "Gautam", "last": "Pennathur" } ] }
PMC1166579	Background ========== Colorectal cancer remains a leading cause of cancer-related mortality and morbidity in the Western world. In the UK it is the third commonest cancer, after lung and breast, with more than 35,000 new cases diagnosed each year and the second commonest cause of cancer-related mortality with approximately 16,000 deaths per year (CancerStats: Large Bowel -- UK, Cancer Research UK 2003, <http://www.cancerresearchuk.org>\[[@B1]\]). Complete surgical resection is the cornerstone of curative treatment. Other treatment modalities, such as chemotherapy and radiotherapy, administered before or after surgery, are used to counteract disease progression particularly in Stage III and IV disease \[[@B2],[@B3]\]. Despite continuing technical advances in the use of these techniques the improvement in overall five-year survival rate has been limited (5--6%). This fact, along with the high proportion of patients who present with tumours not amenable to surgery, drives the search for additional modalities of anti-cancer treatment. Immunotherapy, the use of the adoptive immune system to target cancer cells, holds much promise in this regard. The concept of immune surveillance, whereby the host immune system detects and removes cancerous cells, arose from the observation that immunodeficiency predisposes to the development of malignancies. Furthermore, it has been proposed that rare reports of solid tumour regression are the result of an anti-tumour immune response. Early work suggested that such anti-tumour responses were impotent against colorectal cancer \[[@B4]-[@B6]\]. However, evidence that colorectal malignancies may generate tumour specific antigens (TSA\'s) has encouraged considerable efforts to develop means of harnessing the host immune system to combat colorectal cancer. The use of immunotherapeutic approaches in colorectal cancer has evolved substantially (not reviewed here) and now focuses on the use of tumour specific antigens. This involves either passive immune therapy with antibodies targeted directly to tumour cells or by active immune therapy via vaccination with tumour cells, tumour cell lysates, peptides, carbohydrates, gene constructs encoding proteins, or anti-idiotypic antibodies that mimic TSA\'s. Despite some successes these techniques are still in their infancy and many hurdles remain, largely because our understanding of the biology of host-tumour interactions in colorectal cancer remains elementary. In the last ten years a distinct subset of colorectal cancer, with characteristics of widespread microsatellite instability, has emerged as a group in which immune responses may be pronounced. Clearly, if this type of colorectal cancer does indeed have particular immunogenic features, its study should provide vital advances in our understanding of colorectal tumour immunology. Here we review the literature in relation to putative immune responses in colorectal cancers with microsatellite instability. Microsatellite instability (MSI) -------------------------------- It is now clear that colorectal cancers may arise by a number of distinct molecular pathways. The \"classical pathway\" characterised by chromosomal instability (CSI) and a series of genetic events that drives the progression from normal colon through adenomatous polyp to invasive cancer accounts for 50--60% of cancers \[[@B7]\]. However, this is but one of many alternate pathways and one of these, involving mismatch repair dysfunction, leads to colorectal cancer characterised by microsatellite instability. High-degree microsatellite instability (MSI-H) is a feature of most colorectal cancers arising as part of the hereditary non-polyposis colorectal cancer (HNPCC) syndrome and also 15--20% of sporadic colorectal cancers \[[@B8]\]. Microsatellites are short segments of repetitive DNA bases found throughout the genome but mainly located in intronic (non-coding) segments. Due to their repetitive nature they are prone to insertion and deletion mutations during DNA replication, which commonly introduces such errors at a rate of 1 in 100,000 base pairs copied. The majority (99%) of these errors are detected and corrected by the inherent proof reading capabilities of DNA polymerase enzymes. A small proportion of mismatches that are missed by this mechanism are normally identified and corrected by a group of proteins collectively termed the mismatch repair (MMR) system \[[@B9]\]. Dysfunction of the MMR system allows the accumulation of replication errors and leads to microsatellite instability \[[@B10]\], defined as \"a change of any length due to either insertion or deletion mutations of repeating units in a microsatellite\" \[[@B11]\]. Mismatch repair dysfunction --------------------------- The molecular mechanisms that underlie MMR dysfunction are different in familial and sporadic colorectal cancers. In HNPCC colorectal cancers a germline mutation within an individual mismatch repair gene is inherited and a subsequent \"second hit\" \[Knudson\'s \'two-hit\' hypothesis \[[@B12]\]\] in the normal allele leads to mismatch repair deficiency. The commonly affected genes are hMLH1, hMSH2 and to a lesser extent PMS2, and their involvement predicts the development of extensive microsatellite instability \[[@B13]\]. By contrast, cancer cells with isolated mutations in hMSH3 and hMSH6 continue to maintain some mismatch repair function \[[@B14]-[@B16]\]. In sporadic MSI colorectal cancers genetic alterations in MMR genes are relatively rare. Instead epigenetic modulation of MMR genes, particularly hMLH1, such as hypermethylation of the promoter region leads to silencing and MMR deficiency \[[@B17],[@B18]\]. Hypermethylation is characteristic of a subset of colorectal cancers designated the CpG island methylator phenotype (CIMP+) that is relatively common in right sided tumours \[[@B19]\]. Sporadic MSI-H colorectal cancer appears to represent a subset of the methylator phenotype in which global hypermethylation silences MMR genes. These cancers have biological properties quite distinct from that of HNPCC tumours, in which MMR dysfunction is the sole abnormality, whereas widespread changes in methylation targets the expression of numerous other genes not involved in the MMR pathway \[[@B20]\]. All colorectal cancers display some degree of microsatellite instability if enough markers are studied \[[@B21]\]. In 1997 an International Collaborative Group laid down guidelines for the identification and assessment of microsatellite instability \[[@B22]\]. They classified tumours into microsatellite stable (MSS) and two groups of microsatellite unstable cancers, discriminated by their degree of instability: high (MSI-H) or low (MSI-L). Two mononucleotide microsatellite markers (BAT-25 and BAT-26) and three dinucleotide markers (D2S123, D5S346 and D17S250) were chosen as a panel to be studied for assignation of MSI status. Tumours displaying instability in two or more markers were designated MSI-H, instability in only one marker was designated MSI-L and if no markers were affected the tumour was designated MSS. There is now a growing consensus that the MSI-H clinico-pathological phenotype is characterised particularly by instability in the mononucleotide markers \[[@B21]\]. Colorectal cancers with MSI-H clearly demonstrate distinct biological behaviour whereas MSI-L cancers are less well characterised and are often grouped with cancers that show no instability (MSS) \[[@B23]\]. The focus of this review is the clinico-pathological phenotype characteristic of sporadic MSI-H colorectal cancer. ### MSI-H colorectal cancer Sporadic MSI cancers do not harbour some of the characteristic features or gene mutations associated with CSI cancers, such as aneuploidy and p53mutation. A number of human genes implicated in pathways regulating growth control and apoptosis, for example IGF-IIR, TGFβ-RIIand BAX, contain microsatellite repeats within the coding regions of their transcribed sequences. More recently the BRAFgene has been noted to be commonly mutated in sporadic MSI-H, as well as in serrated polyps, suggesting that serrated polyps are likely precursors of these cancers \[[@B20]\]. It has been suggested that instability of such sequences plays a role in the progression of disease, presumably by altering gene activity \[[@B8],[@B24]\]. Altered regulation of all or some of these genes may be involved in cancer development but precise mechanisms are yet to be identified. Colorectal cancers with MSI-H also have distinct clinico-pathological characteristics. Sporadic MSI-H tumours are associated with the right side of the colon and display marked lymphocyte infiltration into tumour epithelium, stroma, peri-tumoral cuffs and Crohn\'s type reactions \[[@B25]\]. Sporadic MSI-H colorectal cancers have also specifically been associated with medullary-type poor differentiation, high mucin content and reduced lymph node involvement \[[@B20],[@B25]\]. ### MSI-H and improved survival Early reports on microsatellite instability noted an association of MSI with improved patient survival \[[@B26],[@B27]\]. Some small studies have failed to reproduce this finding \[[@B28]-[@B30]\] but generally the association of MSI-H with improved survival in sporadic cancers has been confirmed in larger studies \[[@B31]-[@B33]\]. Many further studies have followed that confirm improved prognosis for patients with sporadic MSI-H colorectal cancer, particularly in lymph node positive cancers, and the link to improved survival has now been confirmed in a systematic review \[[@B34]\]. One explanation offered for the improved survival of MSI-H colorectal cancer patients has been that these cancers may be more sensitive to adjuvant chemotherapy. Indeed early retrospective analyses suggested that cancers with high-degree microsatellite instability had improved response to 5-Fluorouracil (5FU)-based adjuvant chemotherapy \[[@B35]-[@B37]\]. However, these data were retrospectively derived from non-randomised groups and the effects of small numbers, younger age of patients receiving chemotherapy, and different responses of right and left-sided tumours might have confounded the findings. In contrast, two studies have reporting retrospective data derived from randomised control trials of chemotherapy conclude that 5-FU regimes confer no benefit in sporadic MSI-H colorectal cancers and may even be detrimental \[[@B38],[@B39]\]. There is clearly a need for robust prospective randomised controlled trials to evaluate the effects of 5-FU, and other agents, on survival with respect to MSI status. An alternative explanation for the improvement in patient survival is one of enhanced host immune response. An immunohistochemical study noted that the improved survival of patients with MSI-H was associated with the higher frequency of activated tumour infiltrating lymphocytes in these cancers \[[@B40]\]. Consequently, it has been suggested that these lymphocytes may actually represent a host immune response that contributes to improved survival and subsequent work has confirmed a possible link \[[@B41]\]. The hypothesis that colorectal cancers with MSI-H are more immunogenic than microsatellite stable cancers is very attractive, as study of these cancers may provide crucial insights into colorectal cancer immunology. The further elucidation of immune responses to colorectal cancer in vivois crucial to the continuing development of immunotherapy. ### MSI-H and tumour immunogenicity Current concepts in solid tumour immunology identify certain features that are fundamental to the stimulation of an anti-tumour response \[Reviewed in \[[@B42]\]\]. Successful immune response requires a cancer to possess antigens that can be recognised as non-self during immune surveillance. In MSI-H colorectal cancer the accumulation of errors in microsatellites, that reside in gene exons, leads to the generation of a large number of abnormal peptides due to frameshift mutations: the so-called \"mutator phenotype\" \[[@B43]\]. Clearly, such an array of abnormal peptides might represent a pool of TSA\'s that render MSI-H tumours inherently more detectable by the host immune system. Studies on abnormal peptides characteristic of MSI-H have demonstrated their potential for immune recognition. A peptide generated by frameshift mutations in TGFβRII, characteristic of the mutator phenotype, has been used to pulse dendritic cells, which in turn induced activated cytotoxic (CD8+) T cells \[[@B44]\]. One cytotoxic T cell clone demonstrated specific anti-tumour activity against a cell line expressing this peptide in in vitroconditions. The same group have also identified MHC class-II restricted epitopes by induction of helper (CD4+) T cells from the peripheral blood mononuclear cells of normal volunteers and MSI-H colorectal cancer patients \[[@B45]\]. They also demonstrated that tumour infiltrating lymphocytes derived from an HNPCC MSI-H cancer recognised this frameshift-mutation-derived peptide but unusually the infiltrate in this tumour was predominantly CD4+ T cells. More recently, another group have identified several other frameshift-mutation-derived peptides as potential tumour specific antigens (TSA\'s) by using SEREX (serological analysis of recombinant cDNA expression cloning) \[[@B46]\]. They also detected IgG antibodies specific to these peptides only in the MSI-H cancer patients. In an HNPCC patient they demonstrated an antibody specific to a mutated CDX2 peptide that was found in the tumour tissue of the patient and this antibody was not seen in any of the other patients. Interestingly, the antibody disappeared seven years after curative resection. These observations lend significant support to the notion that MSI-H TSA\'s are effectively recognised by the host immune response and an immune response is induced, but more work is needed to confirm that findings in HNPCC patients are mirrored in sporadic MSI-H cancers. Tumour specific antigens must be appropriately presented to the host immune system if a specific immune response is to occur. This normally requires the presentation of antigen in conjunction with the appropriate Major Histocompatibility Complex (MHC) molecules that interact with the T cell receptor (TCR): Class I MHC for presentation to cytotoxic (CD8+) T cells and Class II MHC for presentation to helper (CD4+) T cells \[[@B47]\]. It has been argued that MSI-H colorectal cancers are incapable of eliciting cytotoxic responses in vivodue to the truncation of β~2~-microglobulin that is characteristic of these tumours \[[@B48]\]. The β~2~-microglobulin protein is an essential co-factor for MHC Class I and its defective function would render effective presentation of antigens impossible. However, whilst this defect negates antigen presentation by the tumour itself there is now considerable evidence that the direct presentation of antigen by a tumour may be relatively unimportant. Instead, the concept of \"cross-priming\" in which APC\'s such as dendritic cells, and to a lesser extent macrophages, pick up antigens released by dead tumour cells and subsequently present them to T cells may be more important in vivo\[[@B49],[@B50]\]. Thus, β~2~-microglobulin mutations may be of relatively little importance. Instead a pool of abnormal peptides generated by the mutator phenotype may be recognised and harvested as TSA\'s by immature dendritic cells. On acquisition of non-self antigens these dendritic cells mature into professional APC\'s, with a natural propensity to appropriately present antigen and co-stimulate T lymphocytes \[[@B50]\]. Antigen presenting cells may obviate the need for direct presentation by tumour cells and cross-priming may involve the use of HLA Class II machinery. Immunohistochemical studies have demonstrated that HLA Class II expression is increased in MSI-H colorectal cancer compared to MSS cancers, although this was not associated with improved survival \[[@B51]\]. Additionally, molecular analyses using single subtraction hybridisation \[[@B52]\] and oligonucleotide microarrays \[[@B53]\] have also confirmed increased Class II mRNA levels in MSI-H colorectal cancer. These findings support the in vitrostudies of tumour infiltrating lymphocytes in MSI-H colorectal cancer, mentioned earlier, that demonstrated HLA Class II restricted T helper cell activity and the presence of dendritic cells \[[@B44],[@B45]\]. Either of these two populations may provide an alternative pathway by which cytotoxic T cells may be recruited and stimulated indirectly, but more work is needed to ascertain whether cytotoxic populations themselves are capable of recognising TSA\'s. It has recently become evident that appropriate co-stimulation is pivotal to effective antigen presentation \[[@B54]\]. Co-stimulation refers to the interaction between surface molecules on the antigen-presenting cell (APC), such as CD80 (B7.1) and CD86 (B7.2), and surface molecules on the T cells (CTLA-4, OX40, CD40-L). If these molecules do not bind each other during antigen-presentation then co-stimulation fails and the T cells become anergic, i.e. ineffective and tolerogenic \[[@B54]\]. A microarray analysis comparing the gene expression profiles of MSI-H colorectal cancers to MSS counterparts demonstrated increased signal intensity of CD80 and CD86 in the former group \[[@B53]\] but further studies to validate these findings are still required. After appropriate antigen-presentation and co-stimulation lymphocytes are activated in a tumour-specific response \[[@B49],[@B54]\]. They may then migrate to the tumour, infiltrate it and release mediators that can induce tumour cell death \[[@B47]\]. ### MSI-H colorectal cancer and lymphocyte infiltration For many years the presence of a pronounced infiltrate of lymphocytes in colorectal cancer has been associated with improved prognosis \[[@B55],[@B56]\] but conclusions have been confounded by inconsistencies. Lymphocytes associated with colorectal cancer can be sub-divided into those that invade tumour epithelium (intra-epithelial lymphocytes, IEL\'s), those confined to the stroma or those aggregated around the tumour (peri-tumoral lymphocytes). Previous studies on the significance of tumour infiltrating lymphocytes have been confusing due to variability in the definition of the subset being studied. More recently it has been shown that the presence of IEL\'s is positively associated with an improved survival in colorectal cancer in general \[[@B57]\] and specifically MSI-H colorectal cancer \[[@B41]\]. It has been suggested that they IEL\'s may represent evidence of an immune response to tumour antigens. In a study specific to MSI-H colorectal cancer, Dolcetti and colleagues demonstrated increased immunostaining for IEL\'s when compared to MSS cancers \[[@B58]\]. These lymphocytes were shown to be predominantly CD8+ in keeping with a cytotoxic lymphocytic response and these findings have been confirmed by other groups using immunohistochemistry and RT-PCR \[[@B41],[@B59]\]. However, debate continues as to whether these lymphocytes are of the αβT cell receptor lineage that are involved in tumour infiltration from the circulation or whether they are derived from γδ T cells that normally reside in the gastrointestinal tract \[[@B60]\]. The αβT cell phenotype represents lymphocytes that are targeting the tumour after antigen priming in lymphoid tissue, whereas the γδ T cells might represent a proliferation of lymphocytes resident in tumour epithelium. These characteristics of the lymphocyte infiltrates need further clarification. One immunohistochemical study, involving relatively small numbers and two groups that differed considerably in size (n = 17 v n = 7), has noted the increased presence of CD8+ CD103+ (α^E^β~7~+) lymphocytes in MSI-H colorectal cancer compared to MSS cancers \[[@B61]\]. The presence of CD103+ cells was much higher in tumour tissue than in adjacent colon in both groups and the authors\' postulate that this integrin subunit, which binds to E-Cadherin, has a role in migration of lymphocytes from stroma to tumour epithelium. They suggest that the differences in lymphocyte infiltration between the two groups are likely to be due to increased expression of CD103 by local lymphocytes in MSI-H cancer that promotes infiltration into the tumour. They also postulate that local mediators released by the tumour, such as TGFβ-I rather than a systemic immune response, leads to CD103 upregulation, but they provide no evidence to support this. The expression of CD103 may well facilitate lymphocyte infiltration from stroma to tumour, but whether the difference in IEL numbers between MSI-H and MSS is due to tumour-derived mediators or an immune response is unclear from this study \[[@B61]\]. The increased expression of cytolytic mediators by IEL\'s in MSI-H colorectal cancer suggests that these cytolytic T-cells are indeed activated. Increased expression of perforin, granzyme B and granulysin in MSI-H colorectal cancers \[[@B53],[@B59],[@B62]\] suggests that IEL infiltrates are attempting to induce tumour cell death. This is lent further credence by the finding that the activation marker IL2Rα is also expressed more significantly in these tumours \[[@B59]\]. No single activation marker has been shown to correlate with survival in studies to date but this may well due to the complexities of immune responses. The process of lymphocyte infiltration and anti-tumour activity is also influenced by the differential activity of helper (CD4+) T lymphocyte subsets and local cytokine profiles. Immune responses are promoted by the Th1 subset of CD4+ lymphocytes and their associated cytokines \[Interferon (IFN)-γ, Tumour necrosis factor (TNF)-α, Interleukin (IL)-2 and IL-12). By contrast the Th2 subset of CD4+ lymphocytes and cytokines such as IL-10 promote tolerance to unrecognised antigens and prevent successful antigen-specific responses. A microarray analysis performed in our laboratory has demonstrated the upregulation of several pro-inflammatory cytokines in MSI-H colorectal cancer \[[@B53]\]. As well as the common players this analysis has identified potentially crucial roles for IL-18 and IL-15, both of which are capable of co-ordinating innate and acquired responses \[[@B63],[@B64]\]. Clearly, activity of these interleukins should promote the likelihood of a successful immune response. This microarray analysis also demonstrated that the heat shock protein family is up-regulated in MSI-H colorectal cancer \[[@B53]\]. This is highly significant in the context of previous work by Srivastava and others that has demonstrated the role of heat shock proteins in innate and antigen-specific immune responses \[[@B65],[@B66]\]. Heat shock proteins act as chaperones that are actively involved in regulating the folding of newly synthesised proteins \[[@B67]\]. Indeed their name derives from the observation that these proteins are over-expressed in response to heat shock, and other stresses, that lead to native protein unfolding. The native fold of a protein is encoded by its amino acid sequence and it is therefore attractive to suggest that cancers with high-degree microsatellite instability, that generate large numbers of abnormal proteins with consequent misfolding, might excite a heat shock protein response. Heat shock proteins are used as natural adjuvants in immunotherapy trials and their up-regulation in MSI-H tumours may reflect a similar function in naturally occurring immune responses \[[@B68],[@B69]\]. Our microarray findings have been validated using quantitative real-time RT-PCR across several genes of interest and demonstrate clear differences in the tumour biology of MSI-H and MSS cancers at the mRNA level. Tumour escape and counter-attack -------------------------------- To counteract immune surveillance solid cancers may employ mechanisms to evade the immune system. Several possible methods of immune escape have been identified in colorectal cancer, such as the down-regulation of MHC proteins and co-stimulatory molecules \[[@B70],[@B71]\]. The available evidence suggests that MSI-H colorectal cancers demonstrate higher levels of MHC Class II machinery than MSS counterparts, although the inefficacy of HLA Class I molecules in MSI-H tumours has been proposed as a means of tumour escape \[[@B48]\]. The significance of this in light of the potential for immune cross-priming is controversial, as discussed earlier. Alternative mechanisms of immune escape identified in colorectal cancer include the release of mediators that suppress the function of tumour infiltrating lymphocytes: so-called \"tumour counter-attack\". In the past in vitrostudies of lymphocytes infiltrating colorectal cancer have failed to demonstrate specific anti-tumour activity even though analysis of their receptor usage strongly suggested antigen-specificity \[[@B72],[@B73]\]. These studies did not discriminate tumours according to microsatellite status but one explanation offered for the findings was that the presence of anti-lymphocyte agents, released by the tumour itself, neutralized the activity of tumour infiltrating lymphocytes. The fatty acid synthase (Fas)/Fas-Ligand system has been implicated as a pathway of colorectal cancer counter-attack \[[@B74]\]. Fas-L released by colorectal cancers may bind to Fas receptors on infiltrating lymphocytes and induce their apoptosis. Infiltrating lymphocytes may release Fas-L themselves but colorectal cancers evade the detrimental effects of this by uncoupling their Fas receptors from the intra-cellular machinery that mediates cell death. An immunohistochemical study examining membrane-bound Fas-L expression found that, contrary to expectation, Fas-L expression is not increased in MSS cancers and thus does not explain the less prominent lymphocyte infiltrates seen in these cancers \[[@B75]\]. The authors conclude that some other, as yet unidentified, mechanism of counter-attack must exist in these cancers to account for the reduced infiltrate in MSS colorectal cancer rather than attributing the difference to variable degrees of immunogenicity. MSI-H colorectal cancer and apoptosis ------------------------------------- Dolcetti and colleagues showed that MSI-H colorectal cancers stained significantly more heavily for markers of apoptotic cell death \[[@B76]\]. Subsequent follow-up analysis of the patients from whom the tumours had been removed showed that increased apoptosis counts correlated well with improved survival \[[@B40]\]. However, Michael-Robinson and co-workers also studied apoptosis in these cancers using TUNEL and though they confirmed the presence of more apoptotic cells, they found no co-localisation between these cells and infiltrating lymphocytes \[[@B77]\]. They concluded therefore that increased T cell infiltrates were not likely to be responsible for more marked tumour cell death. However, it may be argued that the release cytokine mediators of target cell death by infiltrating lymphocytes may render the need for direct co-localisation obsolete. In our opinion as MSI-H colorectal cancer does appear to be infiltrated by an activated cytotoxic T cell population and the tumours show high levels of apoptosis the two are likely to be linked but we acknowledge the need for further clarification. An alternative explanation for the increased apoptosis observed in MSI-H colorectal cancer has been the theory of \"effete malignancy\". This concept proposes that the mutator phenotype affects peptides or pathways involved in cell processes fundamental to life. Thus, the accumulation of mutations adversely influences the tumour cells\' viability and hence, leads to apoptotic cell death \[[@B78]\]. The anti-apoptotic bcl-2 protein has been noted to be under-expressed in MSI-H colorectal cancers \[[@B79]\] but the pro-apoptotic gene BAX also appears to be a common target for mutation in the MSI pathway \[[@B80]\], an event that may promote clonal expansion and tumourigenesis. These observations confirm the complex role of apoptosis in tumour development and progression. The apoptosis observed in these cancers, whether induced by effete malignancy or an immune response, is of significance for another reason. The death of tumour cells releases intra-cellular proteins into the tumour micro-environment. This would be a vital step in the promotion of TSA release, recognition and presentation to the host T lymphocytes. It might also release heat shock proteins, upregulated in MSI-H colorectal canecr, which may promote both innate and antigen-specific arms of the immune response. The importance of dendritic cells in immune responses has already been mentioned \[[@B49]\] and immunohistochemical analysis of colorectal cancer has demonstrated the presence of dendritic cells at tumour margins \[[@B81]\]. The presence of this population has not been evaluated in MSI-H cancers specifically but the presence of dendritic cell infiltrates has been associated with reduced tumour progression in colorectal cancer \[[@B82]\]. Conclusion ========== Clearly, the mutator phenotype may influence the function of several cellular processes and these may include those that are designed to nullify the host response. Equally it has been suggested that it is the effect of microsatellite instability on other cellular processes, unrelated to tumour immunology, which leads to the survival benefit observed in MSI-H colorectal cancer. Effete malignancy may, for instance, limit a tumour\'s ability grow and survive \[[@B78]\]. Alternatively processes critical to tumour metastasis may be adversely affected by the mutator phenotype. The down-regulation of VEGF in MSI-H colorectal cancer is a one example of how these tumours may suffer reduced metastatic potential \[[@B83]\]. Certainly many of the biological characteristics of MSI-H, such as diploidy, wild-type p53 expression, activating β-catenin mutations and the CIMP+ phenotype, have all been associated with improved prognosis \[[@B84]\]. However, the effects of all these factors and that of an immune response may be additive, rather than exclusive and they may all contribute to the improvement in outcome. There is now an accumulation of evidence that promotes the argument that MSI-H colorectal cancer is subject to a marked, possibly antigen-specific, immune response. We acknowledge that this may not be the sole determinant of improved prognosis although it seems likely to be a major influence. We believe that MSI-H may be a natural paradigm of host-tumour interactions in immunogenic colorectal cancers and as such further studies are required to clarify the nature of immune responses in these tumours. The use of heat shock proteins and abnormal peptides, derived from frameshift mutations in coding microsatellites, already hold promise in advancing the field of immunotherapy \[[@B85],[@B86]\]. Ultimately the demonstration of specific anti-tumour activity in cytotoxic lymphocytes from patients with MSI-H colorectal would provide the best evidence in support of such a response. Such studies should enhance our understanding of the immune responses in MSI-H tumours and the tumour immunology of colorectal cancer in general. This may be crucial to the advance of efforts in the field of immunotherapy. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AB searched and evaluated the literature and drafted the manuscript. SAB and SD revised the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This study was funded by the Bowel and Cancer Research (Registered charity no. 328667, 4th Floor Alexandra Wing, The Royal London Hospital, Whitechapel, London E1 1BB, UK) the funding source had no influence over the content or conclusions of this manuscript.	PubMed Central	2024-06-05T03:55:58.185575	2005-5-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166579/", "journal": "World J Surg Oncol. 2005 May 12; 3:26", "authors": [ { "first": "Ayan", "last": "Banerjea" }, { "first": "Stephen A", "last": "Bustin" }, { "first": "Sina", "last": "Dorudi" } ] }
PMC1166580	Epidemiology of hepatocellular carcinoma ======================================== Background ---------- Hepatocellular carcinoma (HCC) is one of the most common malignant tumors, representing more than 5% of all cancers. The estimated annual number of cases exceeds 500,000 \[[@B1]\], with a mean annual incidence of around 3--4% \[[@B2]\]. In terms of relative frequencies, HCC ranks as the fifth most common cancer in the world, it is also the fifth among men and eighth among women; it is the second among cancers of the digestive tract after stomach cancer \[[@B3]\]. The incidence of HCC varies considerably with the geographic area because of differences in the major causative factors. The geographic areas at the highest risk are located in Eastern Asia, with age-adjusted incidence rates (AAIR) ranging from 27.6 to 36.6 per 100,000 in men; Middle Africa (AAIR 20.8--31.1/100,000) and some Western African countries (30--48/100,000). The geographic areas at lowest risk are Northern Europe, Australia, New Zealand and the Caucasian populations of North and Latin America (AAIR 1.5--3.0). In Southern Europe, AAIR is around 10 per 100,000 in men \[[@B3]\]. The most powerful risk factor for development of HCC is the existence of liver cirrhosis, regardless of its etiology \[[@B4]\]. Among cirrhotics, viral infection and high alcohol intake are associated with the highest risk \[[@B5]-[@B8]\]. Of the primary hepatitis viruses, only hepatitis B and C viruses cause HCC \[[@B9]\]. Hepatitis A and E viruses do not produce long-term pathological sequelae. Although hepatitis D virus (HDV) always occurs as co-infection with hepatitis B virus and leads to severe acute or chronic hepatic disease, there is controversy whether it increases the carcinogenic potential \[[@B10],[@B11]\]. Risk factors for development of HCC ----------------------------------- ### Hepatitis B Hepatitis B virus (HBV) infection is a major public health problem. It is estimated that two billion people have been infected worldwide and 360 million suffer from chronic HBV infection \[[@B12]\]. Over 520,000 die each year, 50,000 from acute hepatitis B, 470,000 from cirrhosis and liver cancer. In South-East Asia hepatitis B is mostly acquired perinatally from an infected mother. In sub-Saharan Africa, it is mostly acquired in early childhood by horizontal infection, whereas in Northwestern Europe, North America and Australia infection is mainly through sexual contact or needle sharing among injecting drug users, with a peak incidence in the 15--25 age group \[[@B12]\]. Infection acquired perinatally and in early childhood is usually asymptomatic, becoming chronic in 90 and 30% of cases, respectively. In adults, infection resolves in \>95% with loss of serum HBsAg and the appearance of anti-HBs. Chronic infection is characterized by the persistence of HBsAg for more than 6 months. Acute hepatitis B usually results in complete recovery with little if any risk of HCC. In cases with persistent HBV infection, HBV is one of the most important risk factors for HCC. Chronic HBV infection presents as one of three potentially successive phases: immune tolerant, immune activeand low- or non-replicative. In the immune tolerant phase, serum HBsAg and HBeAg are detectable, serum HBV DNA levels are high, serum aminotransferases are normal or minimally elevated. In the immune active phase, serum HBV DNA levels decrease and serum aminotransferase levels increase. Flares of aminotransferases may be observed, in some patients these flares are followed by HBeAg-anti-HBe seroconversion. Following this conversion, in the low- or non-replicative phasethe HBV replication persists but at a very low level suppressed by the host immune response. HBV DNA in serum is undetectable by conventional, non-PCR based techniques. This phase is also called the \'inactive carrier state\'. It may lead to resolution of HBV infection where HBsAg becomes undetectable and anti-HBs is detected, anti-HBc staying positive as sign of contact with the virus. Recently it has been reported that HBV DNA can persist in the serum and liver tissue even after negativation of HBsAg \[[@B13]\]. Recent advances in molecular technology have allowed the isolation of HBV variants that either cannot produce HBeAg or produce it less efficiently, based on precore stop codon mutation and mutations in the core promoter region respectively. In patients with HBV variants, progressive liver damage occurs in parallel with relatively high levels of viremia. In perinatally infected people, the immunotolerant phase lasts till the age of 15--35 years, after which hepatitis flares may occur, leading eventually to viral remission. In patients infected during later childhood or adulthood, there is no immunotolerant phase. Most studies on the risk of developing HCC in chronic HBV infection have been performed in the Far East. Here, most patients acquired the HBV infection as newborn infants \[[@B14]\]. It has been noted that the probability of acquiring HCC increases with severity of liver disease. The annual risk of HCC is 0.5% for asymptomatic HBsAg carriers and 0.8% for patients with chronic hepatitis B \[[@B15]\]. Patients with HBV-cirrhosis have a 1000 times higher risk of developing HCC compared to a HBsAg negative control group \[[@B16]\]. The incidence of HCC in compensated cirrhosis due to HBV from Asia was 2.7%. In Japan, the mean interval between the time of initial infection with HBV and the occurrence of HCC is 50 years. As most people here are infected at birth, HBV related liver cirrhosis usually develops in patients in their 40\'s and HCC in their 50\'s \[[@B17]\]. Few adequate studies have been performed in the West to address the issue of the incidence of HCC in persons who are positive for HBsAg. Most of the studies in Western countries have included small numbers of HBsAg positive patients and/or have not specifically analyzed the group of HBsAg carriers. There is also lack of uniformity in the timing of initiation of follow-up monitoring. In a cohort of 350 Western European patients with compensated cirrhosis followed for a mean period of 6 years, the 5-year cumulative incidence of HCC was 6% \[[@B18],[@B19]\]. The incidence was 2.2% in a series of 179 untreated Caucasian patients \[[@B19],[@B20]\]. In a retrospective analysis of cirrhotic European patients with HBV infection, the 5-year incidence of HCC was 9% irrespective of HBeAg or HBV DNA status at the time of diagnosis of cirrhosis \[[@B21]\]. The hepatitis B replication status seems to play an important role in determining the risk of development of HCC \[[@B22]-[@B24]\]. A recent study found that whereas the relative risk of HCC among men with HBsAg alone was 9.6 compared to those without HBsAg, the risk increased to 60.2 when they were positive for both HBsAg and HBeAg \[[@B23]\]. Another analysis showed that the level of HBV DNA is a prognostic marker for HBV-related HCC and that HCC patients with a less favorable course appear to either clear the virus poorly or to have a greater level of virus production \[[@B24]\]. It was recently demonstrated that positivity for anti-HBc alone in absence of HBsAg and anti-HCV is not rare in Japanese patients with HCC, which may indicate that HBV virus might be involved in so-called non-B HCC \[[@B25]\]. The entire nucleotide sequences of HBV genomes have been classified into 8 genotypes (A-H), with predominance of genotypes A and D in Western countries, and B and C in Southeast Asia and the Far East \[[@B26]-[@B29]\]. Several studies from the Far East evaluated the association between distinct genotypes and severity of liver disease. Genotype C was shown to be associated with the development of liver cirrhosis and HCC in Taiwan \[[@B30]\], China \[[@B31]\] and Japan \[[@B32]\], whereas genotype B was shown rarely to be associated with the development of HCC in China and Japan. In contrast, in Taiwan genotype B is the predominant type in patients with HCC who are younger than 35 years \[[@B30]\]. Another study from Taiwan showed that patients with genotype C had a greater tumor recurrence rate after curative resection of HCC compared with those with genotype B \[[@B33]\]. It was also shown that the likelihood of presence of T1762/A1764 mutations in the basal core promoter parallels the progression of liver disease, and that this mutation is found more frequent in HBV genotype C than B patients \[[@B34]\]. PreS deletions were shown to be more frequent in patients with HBV genotype C, and associated with more advanced disease such as liver cirrhosis and hepatocellular carcinoma \[[@B35]\]. ### Hepatitis C Hepatitis C is also a major public health problem. There are more than 170 million people infected worldwide \[[@B36]\]. Approximately 80% of HCV infected patients develop chronic hepatitis C. About 20% of these patients will develop severe chronic hepatitis C and cirrhosis, which becomes detectable in the second and third decade after infection. The natural history of chronic hepatitis C infection is characterized by a predominantly asymptomatic course and a variable clinical outcome. For these reasons it is difficult to define the rate of progression to cirrhosis and HCC. The risk of cirrhosis in chronic hepatitis C is less than 10% in women infected at a young age and \>30% in men infected after the age of 40 over a 20 year period \[[@B37],[@B38]\]. Five prospective studies from Europe and the US have shown that during the first 10--15 years after initial infection, liver cancer is a rare occurrence \[[@B39]-[@B43]\]. In patients with hepatitis C, there is an increased risk of HCC coinciding with the establishment of cirrhosis with yearly incidence between 3--8% \[[@B6],[@B7],[@B44]-[@B47]\]. In Japan, the mean interval between infection and development of HCC is 30 years \[[@B48]\]. A study from the US shows a long time lag (mean 28 years, range 8--42) between transfusion-associated hepatitis and development of HCC \[[@B49]\]. There is conflicting information on the relationship between HCV genotype and progression to HCC in longitudinal studies. It is suggested by some authors that genotype 1b (most prevalent in Europe and Japan) is associated with a higher incidence of HCC than infection with other genotypes \[[@B50],[@B51]\]. In other studies, however, this was not observed \[[@B52],[@B53]\]. ### Coinfection of HBV and HCV Both HBV and HCV are transmitted parenterally and coinfection is not uncommon in intravenous drug users and in countries with a high prevalence of HBV \[[@B54]\]. Coinfection of HBV and HCV seems to result in more severe liver disease than either infection alone \[[@B55]\]. The risk of developing HCC in subjects with both infections has been investigated in a meta-analysis of 32 epidemiological studies between 1993 and 1997 \[[@B56]\]. The odds ratio for development of HCC in HBsAg positive, anti-HCV/HCV RNA negative subjects was 20.4; in HBsAg negative, anti-HCV/HCV RNA positive subjects 23.6; and subjects positive for both markers 135. These data suggest a more than additive but less than multiplicative effect of HBV and HCV coinfection on the relative risk for HCC. The viruses may act through common as well as different pathways in the carcinogenic process. It has been reported that HBV DNA is still present after seroconversion of HBsAg in patients with hepatitis B. Several reports indicate that prior HBV infection, characterized by presence of anti-HBc, affects the development of HCC in patients infected with HCV \[[@B57]-[@B59]\]. Given these data, in patients with chronic HCV infection, serologic markers of past HBV infection should be checked, not just HBsAg. Other authors, however, were not able to document any adverse event of occult HBV infection on the clinicopathologic course of chronic HCV infection \[[@B60]\]. In case of coinfection with HBV (whether active or past), a more aggressive surveillance to detect early HCC could be suggested \[[@B61]\]. However, to date screening and surveillance programs have not demonstrated a significant survival benefit. In view of the role of HBV as cofactor in the development of HCV related cirrhosis and HCC, vaccination of patients with chronic hepatitis C against HBV has been advocated with the presumption of avoiding additional liver injury \[[@B62],[@B63]\]. ### Coinfection of HBV and HDV Verme et al\[[@B11]\] suggested that HBsAg positive patients with HDV superinfection develop cirrhosis and HCC at an earlier stage (mean age 48 year) than HBsAg carriers without HDV infection (mean age 62 years). ### Coinfection of HBV and HCV with HIV Coinfection of HBV and HCV with HIV is common because these diseases share the same routes of transmission. Recently a series of HCC in HIV-HCV coinfected patients was published, indicating an unusually rapid development of HCC in these patients \[[@B64]\]. This is not surprising, as chronic hepatitis C is more aggressive in HIV positive subjects, leading to cirrhosis and end-stage liver disease in a shorter period of time \[[@B65]\]. ### Coinfection of HCV and S. mansoni An Egyptian study showed that Schistosoma infection increased the risk of HCC, only in the presence of HCV, whereas isolated S. mansoniinfection does not \[[@B66]\]. ### Role of alcohol consumption in HBV or HCV infection Reports suggest that HBV and ethanol act synergistically to promote HCC \[[@B67],[@B68]\]. Habitual heavy drinking was reported to be a significant risk factor for HCC in patients with HCV-related liver cirrhosis by multiple logistic regression analysis \[[@B57]\]. A recent study showed synergism between alcohol drinking and HBV or HCV infection, with approximately a twofold increase in the odds ratio for each hepatitis virus infection for drinkers\' \> 60 g/d, with a more than additive but less than multiplicative risk \[[@B69]\]. Although two case-control studies did not show a relationship of alcohol consumption with the occurrence of HCC \[[@B70],[@B71]\], another case-control study found a positive interaction between HBsAg positivity and HCV RNA positivity and heavy alcohol intake in the development of HCC \[[@B72]\]. Furthermore, Hassan et al. \[[@B73]\] showed synergistic interaction (more than additive) between heavy alcohol consumption ≥ 80 ml/d and chronic HBV or HCV infection (odds ratio 53.9) and insulin or non-insulin dependent diabetes mellitus (odds ratio 9.9). Incidence of HBV- and HCV-related HCC worldwide ----------------------------------------------- Chronic hepatitis B and C infection are responsible for the great majority of cases of HCC worldwide \[[@B9]\]. They also account for the peculiar geographical distribution of the tumor. The relative frequencies of HBV and/or HCV related HCC in the world is illustrated in Table [1](#T1){ref-type="table"}\[[@B17],[@B72],[@B74]-[@B93]\]. The worldwide incidence of HCC varies and is predominantly related to the regional prevalence of chronic viral hepatitis and its associated chronic liver disease and cirrhosis. Aflatoxin intake has a role in the genesis of HCC only in patients who have pre-existing chronic hepatitis B \[[@B84]\]. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Relative frequencies of HBV and HCV related HCC in the world ::: Author \[reference\] Country Era Sample size HBsAg + (%) Anti-HCV + (%) HBsAg/anti HCV + (%) Other (%) --------------------------- -------------- ---------- ----------------- ----------------- -------------------- -------------------------- --------------- Chen, 1990 \[74\] Taiwan NR 66 35 (53.0) 15 (22.7) 7 (10.6) 9 (13.6) Chuang, 1991 \[75\] Taiwan NR 128 87 (68.0) 13 (10.1) 12 (9.4) 16 (12.5) Lee, 1992 \[76\] Taiwan NR 326 233 (71.5) 31 (9.5) 10 (3.1) 52 (15.9) Jeng, 1991 \[77\] Taiwan NR 129 62 (48.1) 29 (22.5) 19 (14.7) 19 (14.7) Leung, 1992 \[78\] Hong Kong 1986--90 424 341 (80.3) 16 (3.8) 15 (4.0) 52 (12.3) Nishioka, 1990 \[79\] Japan NR 180 64 (35.6) 80 (44.4) 11 (6.1) 25 (13.9) Saito, 1990 \[80\] Japan NR 253 49 (19.4) 136 (53.8) 2 (0.8) 66 (26.1) Kiyosawa, 1990 \[17\] Japan 1958--89 83 19 (22.9) 51 (61.4) 10 (12.0) 3 (3.6) Hassan, 2001 \[81\] Egypt 1995--96 33 5 (15.2) 25 (75.8) NR NR Kew, 1990 \[82\] South Africa NR 380 137 (36.1) 63 (16.6) 47 (12.4) 127 (33.4) Yu, 1990 \[83\] USA 1984--89 58 22 (37.9) 36 (62.1) NR NR Di Bisceglie, 1991 \[84\] USA 1987--88 99 7 (7) 12 (12) 1 (1) 79 (79) Hadziyannis, 1995 \[85\] Greece 1991--92 65 33 (50.8) 5 (7.6) 3 (4.5) 23 (38.3) Colombo, 1989 \[86\] Italy 1975--88 132 19 (14.4) 64 (48.5) 22 (16.7) 27 (20.5) Levrero, 1991 \[87\] Italy 1980--88 167 38 (22.8) 82 (49.1) 15 (9.0) 32 (19.2) Simonetti, 1992 \[88\] Italy 1982--88 212 15 (7.1) 133 (62.7) 18 (8.5) 46 (21.7) Donato, 1997 \[72\] Italy 1995--96 172 37 (21.5) 65 (37.8) 4 (2.3) 66 (38.4) Stroffolini, 1998 \[89\] Italy 1996--97 1083 125 (11.5) 771 (71.2) 55 (5.1) 132 (12.2) Bruix, 1989 \[90\] Spain NR 96 4 (4.2) 67 (69.8) 5 (5.2) 20 (20.8) Nalpas, 1991 \[91\] France 1982--89 55 3 (5.5) 28 (50.9) 9 (16.3) 15 (27.3) Van Roey, 2000 \[92\] Belgium 90s 154 37 (24.0) 62 (40.0) NR 55 (36.0) Haydon, 1997 \[93\] UK 1985--94 80 13 (16.3) 22 (27.5) 2 (2.5) 43 (53.8) NR: not reported; Bold: predominant cause ::: In the Far East and sub-Saharan Africa, where HBV is highly endemic, HBV is the main cause of HCC. In areas with an intermediate rate of liver tumors such as Southern Europe, Egypt and Japan, HCV is the predominant cause of HCC. Here HCC is mostly discovered at an older age in patients with longstanding cirrhosis due to HCV. In regions with a low incidence of HCC such as Northern Europe and the United States, HCC related to HCV or HBV infection are found in a minority of cases and the tumor is often related to other factors such as alcoholic liver disease. In these low endemic areas, HCC is usually discovered at an older age in patients with longstanding cirrhosis due to alcohol abuse \[[@B72]\]. In France, ethanol is still the leading cause of cirrhosis and was responsible for 60% of all HCC causes during the last decade \[[@B8]\]. Time trends in the incidence of HCC ----------------------------------- An important epidemiological fact is the rising incidence of HCC in developed countries during the last two decades \[[@B79],[@B89],[@B95],[@B99]\](Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Time trends on the incidence of HCC in the world ::: Author \[reference\] Country Number/100,000 era 1 Number/100,000 era 2 ------------------------------- ------------- -------------------------- -------------------------- El Serag, 1999 \[95\] USA 1976--80: 1.4 1991--95: 2.4 El Serag, 2000 \[96\] USA 1993--95: 2.3 1996--98: 7.0 Benhamiche, 1998 \[97\] (men) France 1976--79: 7.5 1992--95: 10.2 Stroffolini, 1998 \[89\] Italy 1969: 4.8 1994: 10.9 Law, 2000 \[98\] (men) Australia 1983--85: 2.1 1995--96: 4.0 Nishioka, 1991 \[79\] Japan 1968--77: 9.5 1984--85: 16.0 Yoshizawa, 2002 \[99\] Japan 1980: ca 10 2000: ca 40 ::: In [Japan]{.underline}, the HCC-related mortality rate has sharply increased since 1975 from 10/100,000 to almost 40/100,000 in 2000 \[[@B99]\]. An analysis of the Shinshu University Hospital (Japan) showed a change in etiology of the HCC \[[@B100]\]. Whereas in the 1971--1980 decade, hepatitis B was the predominant cause of HCC, in the 1991--1995 period hepatitis C was largely predominant (Table [3](#T3){ref-type="table"}). However, the total numbers of yearly deaths because of HCC in HBsAg carriers\' stays constant, approximately 10% in the survey conducted in 1995. The rapid increase of mortality due to HCC in Japan is mainly attributable (ca 80%) to persistent infection with HCV \[[@B99]\]. The hepatitis C epidemic in Japan originated due to intravenous drug use by the young generation after World War II during the late 40s and early 50s. It spread in the general population due to remunerated blood donors. Abrogation of paid blood donation in 1968, exclusion of blood units contaminated with HBV in 1973 and HCV in 1989 decreased the risk of posttransfusion hepatitis from \> 50% in the 60s to almost zero at present. The incidence of HCV in Japan is decreasing. As the interval between the time of the initial infection with the hepatitis C virus and the development of HCC is 30 years \[[@B79]\], the growing incidence of HCC in Japan is expected to reach a plateau around the year 2015, and then to decrease \[[@B99]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Changing causes of HCC in Japan, 1971--95 ::: Author \[reference\] Era Sample size HBsAg + (%) Anti-HCV + (%) HBsAg/anti HCV + (%) Other (%) -------------------------- ---------- ----------------- ----------------- -------------------- -------------------------- --------------- Kiyosawa, 1992 \[100\] 1971--80 112 60 (54%) 38 (34%) 5 (4%) 9 (8%) 1981--90 267 82 (31%) 159 (59%) 4 (2%) 22 (8%) 1991--95 162 21 (13%) 126 (78%) 5 (3%) 10 (6%) Bold: predominant cause ::: Also in [Italy]{.underline} the mortality rate of HCC is rising \[[@B89]\] from 4.8/100,000 in 1969 to 10.9/100,000 in 1994, reflecting the large cohort of subjects infected with HCV through iatrogenic route during the 50s and 60s when glass syringes were commonly used for medical treatment. Likewise in [Australia, France and the United States of America (US)]{.underline} the HCC mortality is increasing, most probably because people infected with HCV have grown old and reach the cancer-bearing age \[[@B95]-[@B98]\]. In the [US]{.underline}, an increase of about 80% in the incidence of HCC over the past 20--30 years is described, it is estimated that approximately 15,000 new cases occur each year. Also in [France]{.underline} the incidence of HCC is steadily and markedly increased, the estimated number being about 4,000 per year \[[@B101]\]. Although the prevalence of HCV is declining in developed countries because of the decline in incidence in the 90s, the number of persons infected for ≥ 20 years is expected to increase substantially before peaking in 2015 \[[@B102]\]. Analysis of long-term serial HCV samples from the US and Japan suggest that HCV was introduced into the US population around 100 years ago and widely disseminated in the 1960s. In contrast, HCV was introduced in Japan \> 100 years ago and widely disseminated in the 1930s and 40s. The HCV genotype 1b population in Japan started to decrease around 1995 whereas HCV genotype 1a in the US is still growing exponentially. It is predicted that an increased HCC prevalence will occur in the US over the next two to three decades \[[@B103]\]. The reasons advocated for explaining the increased incidence of HCC are the increased rate of HCV infection and an improvement of the clinical management of cirrhotic patients. Enhancing the survival of patients with advanced cirrhosis leads to an increased incidence of HCC. In fact, a decade ago, most of the deaths in cirrhotic patients were due to digestive hemorrhage or bacterial infections, two conditions that are now efficiently prevented and cured \[[@B104]\]. Therefore, HCC has become the leading cause of death in patients with cirrhosis. Screening tests for HCC in patients with chronic viral hepatitis ---------------------------------------------------------------- Despite knowledge of the risk factors for HCC, screening of HCC is controversial, as there have been no randomized controlled studies demonstrating the efficacy of screening for HCC. As HCC mostly occurs in patients with cirrhosis, or at least advanced fibrosis, most studies have been performed in these patients at risk. The most frequently used tests have been serum alfa-fetoprotein (AFP) and hepatic ultrasound (US). There is one non randomized prospective cohort study suggesting that HCC was detected earlier and was more often resectable in patients who had twice yearly screening with serum AFP and hepatic US than in patients who had usual care \[[@B105]\]. Twenty-four studies, which included patients with chronic hepatitis B or C or both, addressed the sensitivities and specificities of screening tests \[[@B106]\]. Serum AFP for detection of HCC was evaluated in 19 studies. They were relatively consistent in showing that the sensitivity of serum AFP for detecting HCC increases from very low levels to moderately high levels of 60 to 80% as the threshold value decreased from 400 to 10 ng/mL, with corresponding specificity decreasing from 100 to 70--90%. A threshold between 10 and 19 ng/mL seems most appropriate as sensitivity usually is moderately high (45 to 100%), with a specificity of 70 to 90%. It has been shown that AFP is not always specific for HCC and titers can increase with flares of active hepatitis \[[@B107]\]. Seven studies evaluated screening with US, reporting high specificity of 95--100%, but variable sensitivity, varying from 11--99% \[[@B94]\]. A surveillance study combining US and AFP in 1,125 patients with HCV, HBV or both, reported a sensitivity of 100% when using a serum AFP \> 10 ng/mL together with US, compared with a sensitivity of 75% using only AFP \> 10 ng/mL and a sensitivity of 87% when using US alone \[[@B108]\]. Computed tomography and magnetic resonance imaging have a high sensitivity and specificity in detecting HCC, but are too expensive to be used in surveillance \[[@B1]\]. The surveillance intervals studied varied from 3 to 12 months. In a study of patients with hepatitis B, the most rapidly growing tumor increased from 1 to 3 cm in 5 months \[[@B109]\]. The ideal time for re-screening has not been identified. Some investigators suggest a 4--5 month interval, others have suggested that a 6-month interval may be most appropriate \[[@B109],[@B110]\]. It is suggested that in case of concomitant HBV and HCV infection serum AFP levels should be obtained every 3 months, and that persistent AFP levels should prompt an aggressive imaging search for HCC \[[@B61]\]. It can be concluded that screening patients with advanced chronic hepatitis B or C with AFP and US may improve detection of HCC, but further studies are needed whether screening improves clinical outcomes. Pathogenesis of hepatitis B and C-induced hepatocellular carcinoma ================================================================== Introduction ------------ Epidemiologic data indicate that chronic hepatitis B and C are independent risk factors for development of HCC \[[@B7],[@B16]\]. Furthermore, animal models confirm the oncogenic potential of HBV and HCV in the liver: transgenic mice for hepatitis B and C \[[@B110],[@B111]\], and natural models such as the woodchuck infected with the woodchuck hepatitis virus, a hepadnavirus closely related to the HBV \[[@B112]\]. Carcinogenesis is believed to be a multistage process, occurring through a sequence of steps termed initiation, promotionand progression. This process evolves over several or many years. Tumor initiationbegins in cells through mutations induced by exposure to carcinogens. DNA changes, maintained during successive cell divisions, activation of oncogenes and inactivation of suppressor genes lead to dysregulation of the cell division and to immortalization \[[@B113]\]. Tumor-initiated cells have a decreased responsiveness to both intercellular and intracellular signals that maintain normal cellular architecture and regulate homeostatic growth. Tumor promotionresults in a further selective clonal expansion of initiated cells. During tumor progression, pre-malignant cells continue to develop progressive phenotypic changes and genomic instability (dysplasia), culminating as overt carcinoma \[[@B115]\]. More than 80% of HCC originate in cirrhotic livers. Macronodules (macroregenerative nodules and adenomatous hyperplasia), irregular hepatocyte regeneration, and some hyperplastic foci are considered as precancerous \[[@B116]-[@B119]\]. Large cell dysplasia and small cell dysplasia are considered to be risk factors for development of HCC \[[@B120]-[@B122]\]. HBV and HCV can be implicated in the development of HCC in an indirect way, through induction of inflammation, necrosis and chronic hepatocellular regeneration, or directly by means of viral proteins or, in the case of HBV, by creating insertional mutations by integration in the genome of the hepatocyte. Indirect carcinogenicity of HBV and HCV --------------------------------------- In most patients with chronic hepatitis B and/or C the occurrence of HCC is preceded by a process of longstanding inflammation. It is probable that malignant transformation is related to continuous or recurring cycles of hepatocyte necrosis and regeneration \[[@B123]\]. The resulting accelerated cell turnover rate may act as a tumor promotor by increasing the probability of spontaneous mutations or damage to DNA by exogenous factors. The accelerated rate of cell division leaves less time for altered DNA to be repaired before the cell divides again, resulting in transmission of altered DNA to the daughter cells. In this way a series of mutations may accumulate in individual cells over time. This process can lead to focal uncontrolled liver cell growth and eventual malignant cell transformation \[[@B115],[@B124]\]. Another mechanism of induction of malignant transformation is the generation of mutagenic reactive oxygen species as a result of the inflammatory process, such as nitric oxide (NO), superoxide anion (O~2~^-^), hydroxyl radical (OH•) and hydrogen peroxide (H~2~O~2~) \[[@B124]\]. Evidence for a causal role for chronic necro-inflammation is provided by transgenic mice into which HBV preS/S genes have been introduced. These mice overproduce pre S1 protein that accumulates in the endoplasmatic reticulum of hepatocytes, producing severe and prolonged injury to these cells, initiating a response characterized by inflammation, regenerative hyperplasia and transcriptional deregulation that progresses ultimately to neoplasia \[[@B125]\]. Patterns of gene expression in cirrhosis and hepatocellular carcinoma have recently been shown to be of value in predicting prognosis. Kim et alcould identify, using the complementary DNA microarray, a 273-gene signature that distinguished high risk types of cirrhosis (hepatitis B, hepatitis C, hereditary hemochromatosis) from low risk types (autoimmune hepatitis, PBC, alcoholic liver diseases) \[[@B126]\]. The same 273-gene signature was present in samples from patients with proven HCC. A subset of 30 genes was most significantly altered in both the high risk types of cirrhosis and the HCC patients. The TACSTD1, a gene associated with HCC development in other studies, is a lead gene in this gene signature. Lee et alcould identify a limited number of genes that accurately predicted survival in a series of 91 HCC patients \[[@B127]\]. The genes involved are implicated in cell proliferation and apoptosis, but also in ubiquitination and histone modification. Delpuech et alidentified distinct patterns of gene expression according to the viral aetiology \[[@B128]\]. Finally, Hann et alcould demonstrate the presence of antibodies to differentially expressed genes in hepatitis B and C, and this appeared to be linked with decreased survival \[[@B129]\]. These discoveries not only increase our insight in hepatocarcinogenesis, but may ultimately lead to the development of clinically valuable preneoplastic and prognostic blood markers. Direct carcinogenicity of HBV and HCV ------------------------------------- ### Hepatitis B A significant proportion of HBV-related HCCs arise in an otherwise normal liver, implicating that the virus can also be directly oncogenic \[[@B124]\]. It has been demonstrated that HBV integrates into the DNA of the host cells. This integration may dysregulate the control mechanisms on the cell cycle by chromosomal abnormalities, production of viral proteins or alteration of human genes and proto-oncogenes. It is, however, controversial whether viral integration plays an important role in the process leading to development of HCC. The hepadnaviral integration process appears to involve recombination mechanisms that do not preserve the viral genome sequence. Thus it is impossible for the viral integrant to function as a template for subsequent virus replication. Several studies suggest that DNA integration sites are at random and that integration occurs at random times during the course of a chronic viral infection \[[@B130],[@B131]\]. HBV integration can be present in chronically infected liver tissue without evidence of HCC \[[@B132]\]. Non-neoplastic hepatocytes may have a similar pattern of rearrangement of viral sequences following integration into human DNA. Chromosomal DNA instability --------------------------- Several studies have shown that HBV DNA integration enhances chromosomal instability. In many hepatic tumors large inverted duplication insertions, translocations and micro- and macrochromosomal deletions have been associated with HBV insertion \[[@B133]-[@B136]\]. These changes can result in loss of important cellular genes, sometimes involving tumor-suppressor genes and other genes involved in the regulation of regeneration and growth processes. ### Trans-activation of cellular genes HBV DNA may induce malignant transformation in another way. Mammalian hepadnaviruses contain a gene (the HBX gene), of which the protein (HBX protein) can trans-activate several cellular promotors and upregulate their expression of different cellular and viral genes \[[@B137],[@B138]\]. Integrated HBX, even when truncated, frequently encodes functionally active trans-activator proteins \[[@B139]\]. This protein has been shown to transform mouse fetal hepatocytes into a full malignant phenotype \[[@B140]\]. There are studies in transgenic mice with the HBX gene that developed multifocal areas of altered hepatocytes, adenomas and HCCs \[[@B110]\]. In contrast to mammalian hepadnaviruses associated with HCC, avian hepadnaviruses such as the duck hepatitis virus or heron hepatitis virus, lack the HBX gene and are not associated with HCC \[[@B123]\]. A gene that may be affected by the HBX gene is the p53 tumor suppression gene. This gene has been shown to play an important role in hepatocarcinogenesis. It is considered to negatively regulate the cell cycle. The HBX protein has been shown to complex p53 protein and to inhibit its function \[[@B141],[@B142]\]. In a transgenic mouse model it was shown that HCC development correlates with p53 binding to HBX \[[@B143]\]. ### Oncogenes It has been proposed that HBV acts as an insertion mutagen by integrating into the host genome and activating the cellular proto-oncogenes c-myc, rasand c-fos\[[@B144]\]. The preS2/S gene is integrated in most HCCs associated with HBV. When 3\'-truncated it generates a truncated protein that is oncogenic by trans-activating proto-oncogenes c-mycand c-fos\[[@B145]\]. ### Growth factors Growth factors and their receptors function as positive or negative modulators of cell proliferation and differentiation. Insulin-like growth factor-II and transforming growth factor-β expression correlate with HBX protein expression in animal models \[[@B146],[@B147]\], suggesting trans-activation of these growth factors facilitating tumor formation. ### Role of PreS mutations PreS deletion mutants accelerate the storage of large envelope proteins in hepatocyte cytoplasm, which could induce cytotoxic effects toward the development of end-stage liver disease \[[@B148]\]. The accumulation of large envelope protein can activate cellular promoters by inducing endoplasmic reticulum stress \[[@B149]\]. Furthermore, pre-S1 sequences can stimulate the transcription of transforming growth factor α (TGFα). Coexpression of TGFα and HBsAg could accelerate hepatocellular carcinogenesis by stimulation of hepatocyte proliferation \[[@B150]\]. ### Allelic loss of chromosome 4q Allelic loss of chromosome 4q is one of the most frequent genetic aberrations found in HCC. It was found to be associated with HBV-related hepatocarcinogenesis, probably by inactivation of a putative tumor suppressor gene included in it \[[@B151]\]. Hepatitis C ----------- In contrast to HBV, HCV is an RNA virus that lacks a reverse-transcriptase enzyme and cannot integrate into the host genome. Thus, insertional mutagenesis can be excluded as a pathogenic mechanism for the development of HCC associated with chronic HCV infection. The molecular pathogenetic mechanisms by which HCV contributes to cell transformation remain unclear. One possibility is that the development of HCC is simply related to chronic necro-inflammatory liver disease. Overall, 97% of patients with HCV markers and HCC have cirrhosis \[[@B152],[@B153]\], and most of the remainder develop HCC in the presence of chronic hepatitis. An alternative mechanism of HCV-induced hepatocarcinogenesis may be that HCV has a direct oncogenic action. Viral replication might cause inappropriate expression of two growth factors that may be implicated in hepatic carcinogenesis: transforming growth factor-α and insulin-like growth factor II \[[@B154],[@B155]\]. The non-structural HCV protein NS3has both protease and helicase activity. HCV may therefore induce genomic instability and favor mutations through its helicase activity \[[@B156]\]. The protein also has an activity similar to protein kinase A, and could disturb cellular homeostasis \[[@B157]\]. The HCV envelope protein E2and the non-structural protein NS5Ainhibit RNA-dependent protein kinase, key mediator of the antiviral, antiproliferative and anti-oncogenic effect of interferon \[[@B158]-[@B160]\]. The HCV core proteinhas characteristics that imply that this protein could function as a gene-regulator \[[@B161],[@B162]\]. The presence of the protein in transgenic mice can induce HCC \[[@B111]\]. After mutation, the HCV core protein can also inhibit tumor suppressor genes such as p53, as has been demonstrated in hepatic oncogenesis \[[@B163]-[@B165]\]. It has recently been shown that the HCV core protein induces nuclear factor κB (NF-κB), thereby suppressing TNF-α-induced apoptosis \[[@B166]\]. This anti-apoptosis may be a mechanism by which HCV leads to viral persistence and possibly to hepatocarcinogenesis. Prevention of hepatocellular carcinoma caused by viral hepatitis ---------------------------------------------------------------- ### Primary prevention The most effective tool to prevent HCC is avoidance of the risk factors such as viral infection by HBV or HCV. Any action diminishing the potential transmission of contaminated blood products (uncontrolled blood transfusion, needle sharing, invasive procedures without proper health standards) will decrease the likelihood of viral spread. The major advance has come from the availability of an effective vaccine that protects against HBV. In 1969, Taiwan was an hyperendemic area of HBV infection with a high rate of HBsAg positivity, 19% of the population being infected before the fourth decade of life. In 1976, HBsAg prevalence was \> 80% in HCC in Taiwan \[[@B167]\]. In 1984 a program to control cirrhosis and HCC began. All neonates born to HBsAg positive mothers were given hepatitis B vaccine in order to counter perinatal infection. In 1986 all neonates were included in the program. As a consequence, there was a decrease in HBsAg positivity in six-year-olds from 10.6% in 1983--1984 to 0.8% in 1993--1994. There was a parallel decline in incidence of childhood HCC (6--14 years old), in the cohort born between 1980 and 1984. The incidence of liver cancer in children between 6 and 14 years old decreased to zero for children born in 1986 and 1987 \[[@B168]\]. The decline of HCC in children after universal vaccination can be considered as an early indicator of the effectiveness of vaccination in reducing the rate of HCC. Since the incidence of HCC in Taiwan peaks in the sixth decade of life, it may take 40 years or longer to see an overall decrease in the rate of HCC as a result of the vaccination program. Vaccination against HBV should become a health priority together with the promotion of adequate health standards. Unfortunately, there is no vaccine against HCV. Up to now, the only effective method to prevent its transmission is the avoidance of contamination with infective blood products. Prevention of HCC in patients with previously acquired risk =========================================================== Introduction ------------ Chronic viral carriage is one of the main risk factors for the development of HCC. Effective antiviral treatments have been developed in recent years and this has changed the management of viral infection. Interferon-alfa is still considered the reference therapy for HBeAg positive chronic hepatitis B. However, its efficacy is limited, with seroconversion from anti-HBe negative to anti-HBe positive in up to 40%. Only \<10% of patients become HBsAg negative \[[@B169]\]. Other possible treatments are antiviral drugs such as lamivudine and adefovir dipivoxil \[[@B12]\]. For the treatment of chronic hepatitis C, interferon-alfa monotherapy yielded only limited response. Combination with ribavirin led to a significant increase in sustained viral response to about 40% in treatment-naïve patients \[[@B170],[@B171]\]. Recently, the combination of peginterferon-alfa and ribavirin improved the sustained viral response rate to nearly 60% in treatment-naïve patients \[[@B172],[@B173]\], and is now considered the reference treatment. It is under debate whether interferon-alfa-based treatments are effective in declining the incidence of HCC in chronic hepatitis B and C. Anti-oncogenic effects of interferon-alfa ----------------------------------------- HCC prevention by interferon-alfa might be the result of several direct or indirect mechanisms. Interferon has an antiproliferative and pro-apoptotic effect \[[@B174]\]. Interferon inhibits the expression of the c-myconcogene and induces the expression of anti-proliferative factors and tumor suppressor genes \[[@B175]-[@B177]\]. In experimental animal models, the anti-neoplastic potential of interferon was demonstrated in already established tumors. In a transgenic mouse model it was demonstrated that early and prolonged administration of interferon diminished the severity of preneoplastic lesions and slowed down the development of HCC \[[@B178]\]. Interferon-alfa also could indirectly reduce the oncogenic risk by inhibition of synthesis of viral proteins which potentially dysregulate the cell cycle, and by enhancing the immune system eliminating not only infected hepatocytes but also initiated or fully malignant cells. Furthermore, interferon-alfa has an antifibrotic and anti-angiogenetic effect, which could also have an influence on tumor development \[[@B179]\]. Interferon and antiviral treatment ---------------------------------- ### Noncirrhotics In patients with chronic hepatitis B, clearance of the HBeAg after treatment with interferon-alfa is associated with improved clinical outcome in terms of survival and development of complications of cirrhosis \[[@B180]\]. Another study confirmed these results and showed a reduction of incidence of HCC in the responders \[[@B181]\]. As most of these patients were non-cirrhotics at entry of the study, the prophylactic effect of interferon on development of HCC can be explained by prevention of cirrhosis development. In Chinese patients with chronic hepatitis B infection, however, interferon-alfa was of no long-term benefit in inducing HBeAg conversion, or in the prevention of HCC and other cirrhosis-related complications \[[@B182]\]. ### Cirrhotics Seven studies investigated the possible effect of interferon treatment on development of HCC in patients with already established cirrhosis \[[@B183]-[@B189]\] (Table [4](#T4){ref-type="table"}). A meta-analysis was performed on these studies \[[@B190]\]. Interferon seemingly decreased the rate of HCC in all trials, while a significant difference was observed in 2 studies \[[@B183],[@B186]\]. Virologic response was strongly associated with reduced risk for HCC in the studies of Oon \[[@B183]\] and Mazzella \[[@B184]\], suggesting that arrest of viral replication is a critical factor. Subgroup analysis in relation to ethnic origin of patients (European, Oriental) showed no preventive effect of interferon on the development of HCC in the European patients \[[@B190]\]. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Studies of treatment with interferon-α for prevention of HCC in patients with hepatitis B-related cirrhosis ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Author \[reference\] Country Type of study Interferon regimen (duration in weeks) Follow-up (range) in months Sample size Rate of HCC (n/n) Significance -------------------------- ----------------------------------------- ------------------- -------------------------------------------- --------------------------------- ----------------- ---------------------------------- ------------------ Oon, 1992 \[183\] Singapore NRCT, P 10 MU daily, 10 days/month (12) 12 (12--60) T 600\ T: 0/600 (0%)\ Significant C 180 C: 10/180 (5.6%) Mazzella, 1996 \[184\] Italy NRCT, P 10 MU tiw (26) 49 (12--119) T 34\ T: 2/34 (5.9%)\ Not significant C 28 C: 4/28 (14.3%) Fattovich, 1997 \[185\] Europe NRCT, P ≥ 300 MU (12--52) 84 (80--92) T 40\ T: 3/40 (7.5%)\ Not significant C 50 C: 4/50 (8.0%) Ikeda, 1998 \[186\] Japan NRCT, P 12 MU/wk (26) 84 (6--168) T 94\ T: 10/94 (10.6%)\ Significant C 219 C: 51/219 (23.3%) IHCSG, 1998 \[187\] Argentina, Germany, Italy, Saudi Arabia NRCT, P 9--30 MU/wk for 3--30 months (36--250) T 49\ T: 8/49 (16.3%) C: 18/97 (18.6%) Not significant C 97 Benvegnù, 1998 \[188\] Italy NRCT, P 6--10 MU (20--26) 72 T 10\ T: 0/10 (0%)\ Not significant C 18 C: 4/18 (22.2%) Di Marco, 1999 \[189\] Italy NRCT, P 655 MU 93 (6--180) T 26\ T: 2/26 (7.7%)\ NR C 60 C: 6/60 (10%) ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- NRCT: non-randomized controlled trial P: prospective T: treated C: controls MU: million units NR: not reported ::: It should be noted that the studies are very heterogeneous and that none of them were randomized controlled trials, so that the results should be interpreted with caution. A recent study showed a significant reduction of the risk of HCC in patients with chronic hepatitis B and advanced fibrosis or cirrhosis, treated with lamivudine for a maximum of five years, compared to placebo \[[@B191]\]. Interferon treatment in HCV patients and HCC prevention ------------------------------------------------------- ### Noncirrhotics Three studies assessed whether interferon treatment prevents the development of HCC in noncirrhotic patients with chronic hepatitis C \[[@B192]-[@B194]\] comprising 3,798 noncirrhotic patients treated with interferon-alfa monotherapy. Pooled together, the incidence of HCC was 60/2,532 (2.37%) in sustained virological responders and 76/1,266 (5.29%) in nonresponders. In a study of 291 noncirrhotic patients with chronic hepatitis C who were nonresponders to interferon therapy and followed for 6--117 months after therapy, the incidence of HCC was significantly lower in patients who received \> 500 MU of interferon. Patients with a transient response (i.e. relapse after end of treatment) had a significant lower rate of HCC development (4/166 = 2.4%) than nonresponders (12/125 = 9.6%) \[[@B195]\]. This anti-oncogenic benefit can presumably be explained by an arrest or slowing down of the cirrhogenic process. ### Cirrhotics The findings of 13 studies of interferon treatment and development of HCC in HCV-infected patients with compensated cirrhosis are summarized in Table [5](#T5){ref-type="table"}\[[@B45],[@B184]-[@B186],[@B194],[@B196]-[@B204]\]. Only 3 studies were randomized \[[@B199],[@B201],[@B202],[@B204]\], the remainders were observational cohort studies. Statistical combination of data is not possible because of different definitions of response (biochemical, virological), different dose schedules for interferon and different duration of follow-up. All studies showed a lower risk for development of HCC in the interferon-treated patients, suggesting that interferon may prevent HCC in compensated cirrhosis caused by hepatitis C. The overall result was largely influenced by three Japanese studies \[[@B194],[@B198],[@B201],[@B202]\], which had the highest incidence of HCC in untreated patients (5--6% per year). This may be explained by intensiveness of the screening programs, but also by genetic, environmental or viral factors. Four European studies failed to document a significant reduction in risk of developing HCC \[[@B45],[@B196],[@B199]\]. In the studies of Fattovich et al\[[@B196]\] and Bruno et al\[[@B45]\], interferon-alfa treatment showed a decrease in incidence of HCC in univariate analysis. However, this was not present in multivariate analysis. In the study of Fattovich \[[@B196]\], a very low natural incidence of HCC was observed, rendering difficult to show a significant decrease. The prospective randomized controlled trial of Valla et al\[[@B199]\] also failed to show a significant effect of interferon treatment on the development of HCC. However, the number of patients in this study was limited and the follow-up relatively short. Also a recently published randomized controlled study from Italy comprising 51 interferon-treated and 71 untreated patients with compensated hepatitis C-cirrhosis, failed to demonstrate any reduced risk in development of HCC after a mean follow-up of 96.5 months \[[@B204]\]. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Studies on treatment with interferon-α for prevention of HCC in patients with HCV-related cirrhosis ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Author \[reference\] Country Type of study Interferon regimen (duration in weeks) Follow-up (range) in months Sample size Rate of HCC (n/n) Significance ----------------------------------- ----------------------------------------- ------------------- -------------------------------------------- --------------------------------- ----------------- ----------------------- ------------------ Mazzella, 1996 \[184\] Italy NRCT, P 3 MU tiw (52) 32 (12--71) T 193\ T: 5/193 (2.6%)\ Significant C 91 C: 9/91 (9.9%) Fattovich, 1997 \[196\] Europe NRCT, P ≥ 200 MU 60 (1--153) T 193\ T: 7/193 (3.6%)\ Not significant C 136 C: 16/136 (11.8%) Bruno, 1997 \[45\] Italy NRCT, P 6 MU tiw (26) 68 (60--84) T 82\ T: 6/82 (7.3%)\ Not significant C 81 C: 14/81 (17.3%) Serfaty, 1998 \[197\] France NRCT, P 3 MU tiw (48) 40 (6--72) T 59\ T: 2/59 (3.4%)\ Significant C 44 C: 9/44 (20.1%) IHCSG, 1998 \[187\] Argentina, Germany, Italy, Saudi Arabia NRCT, R 9--30 MU/wk (3--30 months) (36--250) T 232\ T: 2/232 (0.9%)\ Significant C 259 C: 48/259 (18.5%) Imai, 1998 \[198\] Japan NRCT, R 480 MU (26) 48 (3--65) T 32\ T: 8/32 (25%)\ Significant C 20 C: 7/20 (35%) Benvegnù, 1998 \[188\] Italy NRCT, P 3--6 MU tiw (26--52) 72 T 75\ T: 4/75 (5.3%)\ Significant C 77 C: 20/77 (26.0%) Valla, 1999 \[199\] France RCT 3 MU TIW (48) 40 (37--53) T 47\ T: 5/47 (10.6%)\ Not significant C 52 C: 9/52 (17.3%) Yoshida, 1999 \[194\] Japan NRCT, R 480 MU (23) 52 T 230\ T: 33/230 (14.3%)\ NR C 107 C: 29/107 (27.1%) Okanoue, 1999 \[200\] Japan NRCT, R 3--10 MU qd or tiw (16--24) 1--7 years T 40\ T: 7/40 (17.5%)\ NR C 55 C: 22/55 (40.0%) Nishiguchi, 1995/2001 \[201,202\] Japan RCT 6 MU tiw (12--24) 104 (31--110) T 45\ T: 12/45 (26.7%)\ Significant C 45 C: 33/45 (73.3%) Gramenzi, 2001 \[203\] Italy RCT, P 741 MU 72 T 72\ T: 6/72 (8.3%)\ Significant C 72 C: 19/72 (26.4%) Testino, 2002 \[204\] Italy RCT 3 MU tiw (52) 96.5 ± 18 T 51\ T: 15.51 (29.4%)\ Not significant C 71 C: 24/71 (33.8%) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- NRCT: non-randomized controlled trial RCT: randomized controlled trial P: prospective R: retrospective NR: not reported T: treated C: controls MU: million units ::: In most studies, virological and/or biochemical response are associated with a lower risk of development of HCC, which is less clear in nonresponders. In the study of Imai et al. \[[@B198]\], patients with sustained biochemical response after interferon therapy were at low risk for development of HCC (risk ratio versus controls 0.06; 0.95 in nonresponders). Also in the study of Mazzella \[[@B184]\], a statistically significant effect of interferon treatment was demonstrated when biochemical responders were compared with controls but not when compared with nonresponders. In the study of Benvegnù et al\[[@B188]\], the beneficial effect of interferon treatment on development of HCC was independent of the type of response. In the study of Yoshida et al\[[@B194]\] the risk for HCC was reduced especially among patients with sustained virological but also merely biochemical response that tested positive for HCV RNA. Okanoue et al\[[@B200]\] studied 1,148 patients with chronic hepatitis C treated with interferon-alfa, 40 of them having cirrhosis (fibrosis stage F4). They were followed for 1--7 years after therapy. The cumulative incidence of HCC was significantly decreased in sustained biochemical responders, compared to nonresponders and transient responders, in patients with stage F2 fibrosis, but not in the more advanced stages F3 and F4. In the study of Testino et al\[[@B204]\] HCC did also develop in sustained biochemical responders. Tanaka et al\[[@B205]\], however, demonstrated in 55 patients with HCV-cirrhosis that long-term administration of interferon prevented HCC in those with biochemical and virological response, whereas HCC only appeared in nonresponders. The mechanisms by which an interferon treatment might reduce the risk of HCC development in cirrhosis caused by HCV independent of virological response remains speculative. Maintenance of serum transaminases at low levels may protect against the development of HCC as hepatocyte necrosis, cell damage and increase in hepatocyte replication result in increased DNA damage, influencing hepatocarcinogenesis. Other possible mechanisms for prevention of HCC are the direct and indirect effects of interferon. It is, however, perplexing that only 6 or 12 months of therapy can produce this benefit without virological response. Because of potential biases in the published trials it is premature to advocate the use of interferon as established therapy in HCV infected patients with cirrhosis to prevent HCC. Prospective randomized controlled trials should reproduce the findings in large numbers of patients before a definitive conclusion on the long term effects of interferon in HCV cirrhosis can be established. It must also be realized that a sustained virological response to interferon-alfa monotherapy can be obtained only in 0--8% of patients with cirrhosis \[[@B206]-[@B208]\]. New treatments are now available for chronic hepatitis C, which are more performant in difficultly to treat cases as patients with cirrhosis. The combination of interferon-alfa and ribavirin results in a sustained virologic response in up to 25% of cirrhotics due to hepatitis C \[[@B207]\]. A sustained virologic response of 32% was reported after peginterferon-alfa-2a monotherapy \[[@B207]\] and of 43% after combination of peginterferon-alfa2a and ribavirin \[[@B173]\]. It should be investigated in prospective trials, taking into account the sustained virological and biochemical responses if these more performant treatment regimens will also influence favorably the incidence of HCC, as no data on the long-term effects of these treatments are available up to now. Secondary prevention -------------------- A few sudies focus on the possible role of intereron in the secondary prevention of HCC recurrence in patients with chronic hepatitis B and C after curative resection or ablation. Ikeda et al\[[@B209]\] showed that interferon prevented HCC recurrence after complete resection or ablation of the primary tumor depending on the clearance of HCV viremia. Kubo et al\[[@B210]\] reported a decreased recurrence after surgical resection independent of clearance of HCV or normalization of serum ALT. Another study demonstrated prevention of HCC recurrence after medical ablation therapy for primary tumors in hepatitis B but not in hepatitis C patients by the use of interferon-alfa \[[@B211]\]. Conclusions =========== Chronic hepatitis B and C, mostly in the cirrhotic stage, are responsible for the majority of the hepatocellular carcinomas worldwide. The rising incidence in HCC in developed countries during the last two decades is due to the increasing rate of hepatitis C infection and improvement of the clinical management of cirrhosis. Vaccination against hepatitis B seems to protect against the development of HCC. In patients with chronic hepatitis B or C, interferon alpha treatment in a noncirrhotic stage is protective for HCC development in responders, probably by prevention of cirrhosis development. When cirrhosis is already present, the protective affect is less clear. Further prospective long-term studies should be performed on the new treatments for chronic hepatitis B and C. Some studies also suggested a favourable effect of interferon alpha in the prevention of HCC recurrence in patients with chronic hepatitis B and C after curative resection or ablation. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= PPM participated in the literature search and was responsible for the redaction of the paper. SMF participated in the redaction of the manuscript and critical review of the paper. JLV participated in the literature search and finalizing of the lay-out of the paper.	PubMed Central	2024-06-05T03:55:58.188809	2005-5-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166580/", "journal": "World J Surg Oncol. 2005 May 20; 3:27", "authors": [ { "first": "Peter P", "last": "Michielsen" }, { "first": "Sven M", "last": "Francque" }, { "first": "Jurgen L", "last": "van Dongen" } ] }
PMC1166581	Background ========== Castleman\'s disease(CD) is a rare lymphoid disorder where pathogenesis is a lymphoid tissue hyperplasia related to chronic herpes virus infection. It has been described in nearly every lymph node basin since it first description by B. Castleman in 1956 \[[@B1],[@B2]\]. Two basic pathologic types of this disease could be encountered: the hyaline vascular (HV) and the plasma cell (PC) types. The first tends to be localized in one lymph node and asymptomatic; the second has a more aggressive course and tends to be multifocal with systemic manifestations. The authors\' present two cases of localized Castleman\'s disease arising in the peri-renal space and review previous reports abdominal and retroperitoneal CD. A review of literature on Castleman\'s disease pathogenesis, clinical and radiological characteristics as well as its treatment is also included in this manuscript. Case presentation ================= Case 1 ------ A 33-years-old woman with no significant past medical history complained of abdominal right upper quadrant discomfort associated with an history of weight lost (8 kg over 2 months). Physical examination revealed a right upper quadrant mass on deep palpation. Routine hematology and blood biochemistry were normal. The patient was HIV1-2 negative. Chest and abdominal roentgenograms were considered normal. Abdominal ultrasonography (USG) revealed a large hypoechogenic mass, with regular border in the right anterior peri-renal space. Computed tomography (CT) scan showed a 10 × 8 cm mass with regular contour, containing small calcifications, which strongly enhanced with vascular contrast. The lesion was in contact with the right kidney and ureter. Surgery was planned with a preoperative diagnosis of malignant retroperitoneal tumor versus lymph node hyperplasia. Through a right transverse incision, after mobilization of the duodeno-pancreatic bloc, a tumor was found in contact of the right kidney, ureter and caval vein. While the possibility of malignancy could not be neglected, the mass was dissected en-blocwith wide margin in peri-renal fat. To allow free surgical margin clinically a segment of the right ureter as well as inferior pole of kidney were also excised en bloc. A right nephrectomy was finally performed latter on as the ureteral defect could not be repaired. Histopathological examination of resected specimen revealed localized Castleman\'s disease of the hyaline vascular type. Patient had smooth postoperative recovery and is free of disease more than 6 years after resection. Case 2 ------ A 25-year-old woman with no significant past medical history, presented with post-prandial epigastric discomfort evolving over 2 years and post-prandial vomiting since 1 month. The patient reported 2.5 kg weight loss over 2 months. Physical examination revealed a left para-renal mass on deep palpation. Blood analyses were not relevant. CEA and CA19-9 were in the normal range. The patient was HIV1-2 negative. Chest and abdominal X-ray films were considered normal. An upper abdominal barium follow through was considered normal. Abdominal ultrasonography revealed a 6 cm diameter hypoechogenic mass in the left peri-renal space. CT scan showed a 6 × 7 cm mass, containing multiple small calcifications, which was highly hypervascular and regular in shape (figure [1](#F1){ref-type="fig"}). An arteriography confirmed the presence of hypervascularity with flushing of the mass (figure [2](#F2){ref-type="fig"}). Surgery was planned with a preoperative diagnosis of malignant retroperitoneal tumor. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Computed tomography scanner showing a large left-pararenal mass (Hyalin-vascular type of Castleman\'s disease). Note the presence of microcalcification (white spot) within the mass. ::: ![](1477-7819-3-33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Arteriography showing the presence of a hypervascular mass (Hyalin-vascular type of Castleman\'s disease) with rapid flushing of the tumor. The feeding vessels originated from the aorta and left renal artery. ::: ![](1477-7819-3-33-2) ::: At laparotomy a 7 cm diameter mass was found in the left anterior peri-renal space, just inferior to the renal artery. The lesion was completely excised with what seems, clinically, to be a capsule. No organ resection was needed. Pathologic diagnosis was localized Castleman\'s disease of the hyaline vascular type. Patient had had a simple postoperative period, except of persisting lumbar pain, attributed to a small inferior renal infarct (confirmed by CT scan). The patient is free of disease 5-years after resection. Discussion ========== Castleman\'s disease (CD), also known as angiofollicular lymph node hyperplasia, was first reported by Symmers in 1921 \[[@B3]\]. This pathology was characterized by B. Castleman in 1956 as a benign lymph node hyperplasia resembling a thymoma \[[@B1],[@B2]\]. Keller et alidentified two pathologic types of CD in 1972 \[[@B4]\]. First the hyaline vascular (HV) type which present as a pathological and extensively hypertrophied lymph node. Macroscopically it appears as an encapsulated homogenous mass with an orange-yellowish color. Microscopically, it is characterized by giant lymph follicles centered on a central vessel with marked hyalinization. Follicles are surrounded by circumferentially arranged layer, in an onion skin feature, of small polyclonal B-lymphocytes. These pathologic lymph nodes present a strong hypertrophied vascular arborescence \[[@B4]-[@B8]\]. The plasma cell (PC) type has the same macroscopic aspect as HV type, but contains much more mature polyclonal plasma cells with a less marked hyalinization and vascularization \[[@B4],[@B6]-[@B8]\]. It been shown that this two types histology are not always clearly separated and that mixed HV-PC types can also occur \[[@B7]\]. The histology of PC type is not specific of systemic CD and can be found in autoimmune disease, AIDS and in lymph nodes draining carcinoma, so it is imperious to exclude this condition before diagnosing CD of PC type \[[@B5],[@B6]\]. This implies that serologic testing for HIV should be performed whenever a diagnosis of CD is contemplated\[[@B5],[@B6]\]. The etiology of CD is related to chronic Human Herpes virus 8 (HSV8) \[[@B7]\] as HSV8 has been found in lymphoid cells in case of systemic form, or PC type, of Castleman\'s disease \[[@B9]\]. Its nature is not neoplastic as confirm by the fact that the lesion are made of a polyclonal proliferation \[[@B6],[@B7]\]. It\'s seems that CD is the result of a chronic low grade inflammatory process triggered by latent infection with HSV8, which leads to lymphoid system hyperplasia \[[@B5]-[@B7]\]. Human herpes virus 8 (HSV8), also called Kaposi\'s sarcoma-associated herpes virus (KHSV) is the initiator of this chronic inflammation by establishing a chronic or latent infection in lymph nodes \[[@B7]\]. Chronic infection by HSV8 stimulates secretion of IL-6 which in turn induces a hyperplasic reaction of the lymphoid system \[[@B6],[@B7]\]. While this lymphoid hyperplasia could be contained in one lymph node as in the localized form, which is mainly of HV type, it could also be generalized as in the systemic, or multifocal, form which is the predominant form for the PC type \[[@B6]\]. The patient\'s immunological status seems to play a major role in the development of these two forms. While localized form is encountered mainly in immunocompetent patients, the systemic form is found in patient with AIDS or other immunodeppression related either to immunosuppression or pathological state \[[@B6]\]. The localized form of CD arises predominantly in the mediastinum, where it was first described by B. Castleman \[[@B1]\]. However, it can be found in the neck, abdomen, axilla, inguinal region and in virtually all lymph node area \[[@B5],[@B7],[@B8],[@B10],[@B11]\]. Even non-nodal tissue could be involved, as is has been described in: lung, pancreas, breast, adrenal gland, muscle and other extremely rare locations \[[@B4],[@B7],[@B10],[@B11]\]. Testa et al\[[@B11]\] have reported the location of 315 cases of localized CD, 65% were in the mediastinum, 16% in the neck, 12% in the abdomen, 3% in the axilla and 4% in diverse locations. A literature review of the abdominal and retroperitoneal case of localized HV type of CD has been done. In 1992, 54 abdominal and retroperitoneal cases were reviewed by Seco et al\[[@B5]\]. Now 195 cases of localized CD have been reported, in the world literature, arising in the abdomen and retroperitoneum. Of these 195 cases, 122 (63%) were in the retroperitoneum and 73 (37%) in the abdominal cavity (Table [1](#T1){ref-type="table"}). Of the 122 lesions localized in the retroperitoneum, 24 (20%) were in the peri-renal region, as were our cases. Nearly all this lesions were derived from lymph node tissue, but 5 of these 195 cases (2%) seem to have originating in extra lymphoid organ. Three pancreatic, and one each of splenic and adrenal CD have been described \[[@B58]-[@B60],[@B65],[@B80]\]. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Abdominal and retroperitoneal case of localized Castleman\'s disease (HV type) ::: *Location* *Number of cases* *Authors* ---------------------- ----------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Retroperitoneum 97 Seco et al\[5\], Bapat et al\[12\], Yamakita et al\[13\], Morishita et al\[14\], Genoni et al\[15\], Ng et al\[16\], Baikovas et al\[17\], Johnson et al\[18\], Martino et al\[19\], Guglielmi et al\[20\], Ziv et al\[21\], Gheysens et al\[9\], Halvic et al\[22\], Herrada et al\[23\], Furuhata et al\[24\], Ebine et al\[25\], Sadamoto et al\[26\], Gravalos et al. \[27\], Iwamoto et al\[28\], Sanna et al\[29\], Singletary et al\[30\], Curciacrello et al\[31\], Gonzalez Sanchez et al\[32\], Schutz et al\[33\], Perez Garcia et al\[34\], Irsutti et al\[35\], Parez et al\[36\], Buchanan et al\[37\] Peri-renal 24 Ebisno et al. \[38\], Inoue et al. \[39\], Takihara et al. \[40\],,Feudis et al\[41\], Barret et al. \[42\], Okada et al\[43\], Present two cases Mesentery 27 Seco et al. \[5\], Barki et al. \[44\], Hung et al. \[45\], Schroff et al. \[46\], Makipernaa et al. \[47\], De Heer-Groen et al\[48\], Parez et al\[36\], Neerhout et al\[49\], Powel et al\[50\], Burke et al\[51\] Greater omentum 3 Volta et al\[52\], Kiguchi et al\[53\] Gastric 2 Kiguchi et al\[53\], Yebra et al\[54\] Peri-pancreatic 5 Kiguchi et al\[53\], Rotman et al\[55\], Brossard et al\[56\], Inoue et al\[39\], Erkan et al\[57\] Pancreatic 3 Chaulin et al\[58\], Corbisier et al\[59\], Lepke et al\[60\] Porta hepatitis 5 Rahmouni et al\[61\], Farkas et al\[62\], Peck et al\[63\], Cirillo et al\[64\] Adrenal gland 1 Debatin et al\[65\] Pelvis 27 Seco et al\[5\], Latte et al\[66\], Daley et al\[67\], Boxer et al. \[68\], Tsukamoto et al\[69\], Luburich et al\[70\], Isik et al\[71\], Ylinen et al\[72\], Schwartz et al\[73\], Mondal et al\[74\], Kiguchi et al\[53\], Calvo Villas et al\[75\], Mac Donald et al\[76\], Fields et al\[77\], Kkasantikul et al\[78\], Halvic et al\[22\], Murphy et al\[79\] Spleen 1 Taura et al\[80\] *Total of cases* *195* ::: The clinical presentations of CD differ greatly between the localized and the systemic forms (Table [2](#T2){ref-type="table"}). The first appears in young generally healthy patients and cause few symptoms \[[@B6],[@B7]\]. Abdominal and retroperitoneal locations, as were our cases, can be associated with mass effect symptoms related to compression of adjacent organs. This could present as: post-prandial discomfort, anorexia, vomiting, weight loss, urinary retention and abdominal or lumbar pain \[[@B80]\]. Systemic, or multifocal, CD is associated with systemic disturbance as anemia, increased erythrocyte sedimentation rate, polyclonal hypergammaglobulinemia, hypoalbuminemia and thrombocytopenia which all can be associated with a specific symptoms \[[@B6]-[@B8]\]. The clinical picture includes asthenia, fever, weight loss, generalized lymphadenopathy, hepatomegaly, splenomegaly, peripheric edema, pleural effusion, impaired renal function and sometimes polyneuropathy \[[@B6]-[@B8]\]. Rarely POEMS syndrome (Polyneuropathy, Organomegaly, Endocrinopathy, M protein and Skin change) or amyloidosis may be associated to systemic CD \[[@B6]-[@B8]\]. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Clinical forms of Castleman\'s disease ::: Localized form Multifocal form ---------------------------- ------------------------------------------------- --------------------------------------------------------- Mean age (years) 3^rd^decade 6^th^decade Clinical signs Incidental mass effect Systemic symptoms Localization Mediastinum, cervical or abdominal, etc\... Multifocal, mostly peripheric lymph nodes Histologic type HV, rarely HV-PC PC Treatment Surgical resection Corticosteroids, chimiotherapy, radiotherapy Prognosis Excellent: 100% survival at 5 years Poor: median survival of 30 months Recurrence after treatment Extremely rare, related to incomplete resection Nearly always Association Rarely lymphoma Frequent: AIDS, Kaposi\'s sarcoma, lymphoma and myeloma Table adapted according to references: \[5, 6, 7, 11\] ::: Radiographic characteristics of CD are non specific, but some features could help to suspect the diagnosis (Table [3](#T3){ref-type="table"}) \[[@B77],[@B80]\]. Plain radiographic finding includes a mass effect and in nearly 30% of localized form calcifications harboring a radial arrangement or star-shaped calcifications which is said to be characteristic of CD \[[@B77]\]. Ultrasonography (US) usually demonstrates a hypoechogenic and homogenous mass with quite clear delimitation \[[@B8],[@B30],[@B77]\]. US can show central areas of sharp acoustic shadowing due to calcification \[[@B30]\]. CT scan show a solid, homogenous and well delimited mass which enhance with vascular contrast as a results of hypervascularity \[[@B5],[@B30],[@B77],[@B80]\]. It can also show star-shaped microcalcifications which are quite specific on pre-contrast images \[[@B5],[@B27],[@B77]\]. Post-contrast IV study, while demonstrating dense enhancement of the mass, could demonstrate a central stellate scar \[[@B30],[@B80]\]. Angiography shows a strongly hypervascular lesion, which present a dense and homogenous flush during the capillary phase \[[@B5],[@B6],[@B8],[@B40],[@B53],[@B77],[@B80]\]. This flush begins in periphery to become diffuse whiting the mass and is specific for the HV type of CD \[[@B77],[@B80]\]. It can also demonstrate hypertrophied feeding vessel, an useful information when resection is plan \[[@B8],[@B53],[@B77],[@B80]\]. Magnetic resonance imaging (MRI) characteristics of CD are: hypodense mass on the T1 weighted study and hyperdense lesion on the T2 weighted image sometimes with star-shaped calcifications \[[@B60],[@B77],[@B80]\]. Gadolinium injection produces an enhancement which appear in periphery to become diffuse similarly to the flush observed during angiography \[[@B77],[@B80]\]. All these radiological finding are not specific but some like the star-shaped calcifications and the type of hypervascularisation are quite specific and should alert clinician to the possibility of CD. In summary, in front of an abdominal or retroperitoneal mass which is well delimited, homogenous and harbor star-shaped calcifications and hypervascularisation associated with typical flush, the diagnosis of CD should be strongly suspected \[[@B53],[@B77],[@B80]\]. The differential radiological diagnosis is mainly malignant neoplasm because of the hypervascularity \[[@B40],[@B77]\], and the fact that 80% of the retroperitoneal tumor are malignant \[[@B40],[@B81],[@B82]\]. The major tumors found in the retroperitoneum are soft tissue sarcoma (liposarcoma, fibrosarcoma, leiomyosarcoma, neurofibrosarcoma, undifferentiated and rabdomyosarcoma) which are frequently heterogeneous mass and show necrosis on CT scanner; vascular tumor (hemangiosarcoma and lymphangiosarcoma) which are cystic and of liquid density on USG and CT scanner; and the lymphoma which generally presents as multiple adenopathy and homogeneous mass on CT scanner \[[@B30],[@B83]-[@B86]\]. Urological tumors like seminoma, prostatic cancer and teratoma tends to give rise to metastatic disease in the retroperitoneum in the form of adenopathy which are generally multiple \[[@B86]\]. While a preoperative diagnosis of CD is difficult to obtain, fine needle biopsy is not a definitive tools because of there low specificity and the differential diagnosis with lymphoma is impossible by this approach \[[@B77],[@B88]\]. In addition to be non-specific and rarely yielding enough useful tissue, needle biopsy is associated with tumoral seeding with reported frequency of 1/40 000 to 1/1 000 biopsy \[[@B83],[@B89]\]. Thus when surgery is planned, indication for fine needle biopsy should be carefully deliberated, while an open biopsy could always be done during surgery \[[@B83]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Radiological characteristics of Castleman\'s disease ::: Non specific signs Specific signs --------------- -------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------- Radiography calcification. Star-shaped calcification. Echography Hypoechogenic and homogenous mass. Central areas of acoustic shadowing (calcification). CT scanner Tissue density, Homogenous and well delimited mass. Contrast enhancement beginning in periphery. Star-shaped microcalcifications Star-shaped scar on post-contrast study. Arteriography Hypervascularity with hypertrophied feeding vessel. Flush beginning in periphery to become diffuse and homogenous during capillary phase. MRI Hypodense on T1 and hyperdense on T2. Contrast enhancement beginning in periphery. Star-shaped microcalcification or star-shaped hypodense signal on T2. Table adapted according to references: \[5,9,27,30,34,42,48,65,\] ::: Treatment of CD differs between localized and multifocal, or systemic, forms. The standard therapy of localized form is surgical excision which is curative when resection is complete and en-bloc. No recurrences have been reported after total excision in the literature as it was the case with our cases \[[@B5]-[@B7],[@B11],[@B38],[@B40],[@B53],[@B82],[@B90]\]. Because these lesions are highly vascularised, embolization before surgery could be helpful to minimize blood lost during surgery \[[@B5],[@B53]\]. The problem that is faced during resection is to resolve the differential diagnose between a malignant pathology and CD. Macroscopically it is nearly impossible because CD lesions harbor dense fibrous adherences to adjacent organ and hypervascularization typically seen in malignant pathology \[[@B5],[@B38],[@B82],[@B90]\]. For this reason peroperative diagnosis by open biopsy is helpful, and necessary in case of CD suspicion. It enables one to avoid extensive resection and especially resection of nearby organ for this benign disorder which does not invade adjacent organ even in case of tight contact \[[@B38],[@B82],[@B83],[@B90]\]. The five years survival after resection is nearly 100% for the localized form \[[@B6]\]. Recurrences have rarely been reported generally when excision was incomplete \[[@B6],[@B38],[@B82],[@B90]\]. For the systemic form no curative therapies have been found yet. Corticotherapy, immunosuppressive drugs, chemotherapy and radiotherapy have been tried without any convincing results \[[@B5]-[@B7],[@B91],[@B92]\]. The prognosis of this form is poor with a median survival of 30 months \[[@B6]\]. However recently, the reports of improvement and prolonged survival after splenectomy have been reported with steroid therapy and chemotherapy this could change the prognosis of the systemic form in the future \[[@B91],[@B92]\]. Of importance from the surgical point of view is that in these cases the remission was obtained after splenectomy only \[[@B91]\]. The systemic, PC type, and extremely rarely the localized, HV type, forms are associated with malignant disorder such as Kaposi\'s sarcoma, malignant lymphoma and myeloma. This association is stronger in HIV positive patients with CD, for example Kaposi\'s sarcoma is associated to 13% of case of PC type CD in HIV negative and in 75% of case when HIV positive patients. The association with Kaposi\'s sarcoma could be explained by the fact that HSV8 is cause in the pathogenesis of these two disorders \[[@B7]\]. These neoplasms can appear as late as 8 years after the diagnosis of CD, so a long term follow-up is required in patient in whom the diagnosis of CD has been established \[[@B6]\]. Conclusion ========== Castleman\'s disease (CD) is a rare lymphoid disorder, where etiology is related to Human herpes virus 8. CD could present in two forms: the localized (hyaline vascular type) and the systemic (plasma cell type). The localized hyaline vascular form has a unique indolent lymph node hyperplasia; which can be found in the abdomen, retroperitoneum or any lymph node basin; as a solitary mass. Localized CD is radiologically nearly undistinguishable from malignant neoplasms but some characteristics are quite specific, like the type of hypervascularization and the star-shape microcalcifications. A good preoperative work-up and an open biopsy during surgery, for abdominal and retroperitoneal mass if no diagnosis has been establish, can help to avoid extensive resection when facing this benign disorder. Complete surgical excision is curative; recurrences have only been described after incomplete resection. The prognosis is excellent with a five years survival of nearly 100%. List Of Abbreviations ===================== CD = Castleman\'s Disease HV = Hyaline vascular type (of Castleman\'s disease) PC = Plasma cell type (of Castleman\'s Disease) CT scan = Computed tomography scanner US = Ultrasonography Competing Interests =================== No competing interests have to be reported for this work and manuscript by any of the authors or institution. Authors\' Contributions ======================= PB: Study design, literature review, patient\'s follow-up, medical charts review, and manuscript. GC: Literature research and review. GZ: Medical charts review. FR: Literature review and manuscript. OH: Patient\'s surgical management, manuscript review. PhM: Manuscript review. Acknowledgements ================ Patients consent was obtained for publication of these case reports.	PubMed Central	2024-06-05T03:55:58.194734	2005-6-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166581/", "journal": "World J Surg Oncol. 2005 Jun 7; 3:33", "authors": [ { "first": "Pascal", "last": "Bucher" }, { "first": "Gilles", "last": "Chassot" }, { "first": "Guillaume", "last": "Zufferey" }, { "first": "Frederic", "last": "Ris" }, { "first": "Olivier", "last": "Huber" }, { "first": "Philippe", "last": "Morel" } ] }
PMC1167557	Like SARS, Ebola, and other emerging infectious diseases, antibiotic resistance in bacteria may have a zoonotic origin \[[@pmed-0020232-b1]\]. Evidence suggests that antibiotic use in agriculture has contributed to antibiotic resistance in the pathogenic bacteria of humans, but the chain from cause to effect is long and complicated. Antibiotic use clearly selects for antibiotic resistance, but how far do these effects extend beyond the population where antibiotics are used? Antibiotics and antibiotic-resistant bacteria (ARB) are found in the air and soil around farms, in surface and ground water, in wild animal populations, and on retail meat and poultry \[[@pmed-0020232-b2]\]. ARB are carried into the kitchen on contaminated meat and poultry, where other foods are cross-contaminated because of common unsafe handling practices \[[@pmed-0020232-b10],[@pmed-0020232-b11]\]. Following ingestion, bacteria occasionally survive the formidable but imperfect gastric barrier, and colonize the gut \[[@pmed-0020232-b12]\]. Patterns of colonization (asymptomatic carriage) and infection (symptomatic carriage) in human populations provide additional evidence that ARB occasionally move from animals to humans \[[@pmed-0020232-b13],[@pmed-0020232-b14]\]. The strongest evidence comes from the history of the use of antibiotics for growth promotion in Europe. After first Denmark and then the European Union banned the use of antibiotics for growth promotion, prevalence of resistant bacteria declined in farm animals, in retail meat and poultry, and within the general human population \[[@pmed-0020232-b8],[@pmed-0020232-b15]\]. Despite the evidence linking bacterial antibiotic resistance on farms to resistance in humans, the impact of agricultural antibiotic use remains controversial \[[@pmed-0020232-b16]\] and poorly quantified. This is partly because of the complex of population-level processes underlying the between-species ("heterospecific") and within-species, host-to-host ("horizontal") spread of ARB. To emerge as human pathogens, new strains of ARB must (1) evolve, originating from mutations or gene transfer; (2) spread, usually horizontally among humans or animals, but occasionally heterospecifically; and (3) cause disease. All three of these steps are complex and imperfectly understood. The emergence of a new type of resistance is a highly random event, which can\'t be predicted accurately, and may involve multiple steps that preclude perfect understanding even after the fact. Spread is equally complicated and may obscure the origins of resistance. In some cases, emergence of resistance in one bacterial species is a consequence of the emergence and spread in another species, followed by the transfer of resistance genes from one bacterial species to another. Because of the underlying complexity, mathematical models are necessary to develop theory---a qualitative understanding of the underlying epidemiological processes \[[@pmed-0020232-b20]\]. Theory helps researchers organize facts, identify missing information, design surveillance, and analyze data \[[@pmed-0020232-b26]\]. Horizontal Transmission {#s2} ======================= Theory clearly shows that the impact of agricultural antibiotic use depends on whether resistant bacteria have high, low, or intermediate horizontal transmission rates in human populations \[[@pmed-0020232-b23],[@pmed-0020232-b24]\]. The rate of horizontal transmission among humans is determined by the underlying biology of the pathogen, medical antibiotic use, and hospital infection control, but not by agricultural antibiotic use \[[@pmed-0020232-b22]\]. On the other hand, a farm where multiple antibiotics are used routinely, universally, and in low quantities for growth promotion is likely to be an excellent environment for the evolution of multiple resistance factors, including some variants that might never have evolved in humans. Thus, even very rare transmission resulting from agricultural antibiotics may have a medical impact by introducing new resistant variants to the human population. The epidemiology of spread in the human population dictates how the impact of agricultural antibiotic use should be assessed. Zoonotic pathogens, such as Campylobacter and Salmonella, are generally regarded as having low horizontal transmission rates in human populations. While resistance in zoonotic infections should be directly attributable to resistance in the zoonotic reservoir, the impact of agricultural antibiotic use remains controversial \[[@pmed-0020232-b18],[@pmed-0020232-b27]\]. Zoonotic species could acquire resistance genes from human commensal bacteria during the infection process, but this hypothesis is difficult to test. For pathogens with high horizontal transmission rates, resistant bacteria will spread rapidly once they have emerged, and prevalence will be maintained at a steady state by horizontal transmission. Thus, the impact of subsequent heterospecific transmission is limited ([Figure 1](#pmed-0020232-g001){ref-type="fig"}). Nevertheless, one or two heterospecific transmission events could be sufficient to cause the appearance of a highly successful ARB genotype in humans, affecting the timing, nature, and extent of spread within the human population \[[@pmed-0020232-b22]\]. Not only are such events difficult to trace, but their impact is impossible to measure, since there is no way to know what type of resistance would have appeared and with what temporal pattern, if transfers from animals had been prevented. ::: {#pmed-0020232-g001 .fig} Figure 1 ::: {.caption} ###### The Emergence and Spread of Antibiotic Resistance in Bacteria with High Horizontal Transmission Rates Emergence and spread begins with a honeymoon period following the approval of a new antibiotic; the honeymoon ends when resistance emerges. Prevalence increases exponentially at first, but it eventually approaches a steady state. The impact of agricultural antibiotic use must be assessed by comparing the observed situation with the counterfactual situation, an imaginary world in which antibiotics were never used in agriculture. The impact of agricultural antibiotic use is, then, the total number of cases of resistance that would not have happened without the use of antibiotics in agriculture. This is approximately the difference between the time of actual emergence and the counterfactual emergence, multiplied by the steady-state prevalence. While we don\'t know what would have happened in any particular case, we can estimate the likely magnitude of agricultural impacts. ::: ![](pmed.0020232.g001) ::: The case where horizontal human transmission rates are intermediate is particularly interesting. If each case in a population generates approximately one new case (a situation we call "quasi-epidemic" transmission), each instance of heterospecific transmission will initiate a long chain of horizontal transmission that eventually burns out. Quasi-epidemic transmission can amplify a relatively low amount of heterospecific transmission and substantially increase prevalence \[[@pmed-0020232-b23]\]. The effect is sustained as long as heterospecific transmission continues. A corollary is that banning agricultural antibiotic use would have maximal benefits if horizontal transmission is quasi-epidemic \[[@pmed-0020232-b24]\]. Moreover, the effects are most difficult to estimate because both heterospecific and horizontal transmission must be accounted for. These principles apply to bacteria associated with outpatient antibiotic use and community-acquired infections as well as those that are primarily hospital-acquired. Although quasi-epidemic transmission would seem to be a special case, it may in fact be the rule for many hospital-acquired bacteria because it is the natural endpoint of the interplay between economics and ecology \[[@pmed-0020232-b30]\]. By spending money on hospital infection control, hospital administrators can reduce nosocomial transmission rates for resistant bacteria. For example, hospitals may screen and isolate patients who are likely to be carriers (i.e., active surveillance) and implement infection-control measures, but this comes at the cost of isolating patients \[[@pmed-0020232-b31]\]. Total costs are minimized by spending just enough to eliminate (or nearly eliminate) the pathogen; thus, quasi-epidemic transmission is the economic optimum \[[@pmed-0020232-b30]\]. The Community as a Reservoir for Resistance {#s3} =========================================== Horizontal transmission is further complicated by population structure, such as the movement of humans through hospitals and long-term care facilities. Medical antibiotic use and horizontal transmission rates are high in hospitals, but this is counterbalanced by short hospital stays. An emerging view for hospital-acquired bacterial infections is that persistent asymptomatic carriage plays a key role in the epidemic of resistance. ARB can asymptomatically colonize a person for years: even if the number of other people infected during a single hospital visit is less than one, this number will exceed one when summed over several hospital visits \[[@pmed-0020232-b25],[@pmed-0020232-b32],[@pmed-0020232-b33]\]. Thus, the ecological reservoir of resistance in the community plays an important role in the increasing frequency of resistance in hospital-acquired infections. Short hospital visits and long persistence times of ARB in people guarantee that some of the costs associated with failed infection control are passed on to other hospitals---new carriers are frequently discharged from one hospital only to be admitted to another hospital later \[[@pmed-0020232-b30]\]. Thus, the harm done by these ARB is borne by the whole human population, particularly all of the health-care institutions that serve a single catchment population. In economic terms, the damage caused by the carriage of ARB is a kind of pollution. By comparing the total number of new carriers generated in the community, the impacts of agricultural antibiotic use on hospitals can be compared directly to the impact hospitals have on each other ([Figure 2](#pmed-0020232-g002){ref-type="fig"}). The rate of heterospecific transmission is intrinsically difficult to measure directly because the risk of exposure and colonization per meal is very small. Nevertheless, agricultural antibiotic use may generate as many carriers as hospitals for the simple reason that the population experiences many more meals than hospital discharges \[[@pmed-0020232-b34]\]. When agricultural and nosocomial transmission are equally rare in the population, the latter will be much easier to identify and quantify. ::: {#pmed-0020232-g002 .fig} Figure 2 ::: {.caption} ###### How Large Is the Impact of Antibiotic Use in Agriculture? Comparing the amount of antibiotics used in agriculture with the amount used in medicine means comparing fundamentally different things because they affect the emergence of medically important antibiotic resistance in different ways. For hospital-acquired infections, it is more appropriate to think about ARB carriage in the community as a kind of pollution that flows into hospitals. Thus, the appropriate way to measure impact is by counting how many new carriers are added to the community reservoir from hospital discharges versus from exposure to bacteria that originate on farms. Different formulas describe these processes. To count ARB carriers among hospital discharges, let x denote the proportion of patients from a hospital (or other institution) that are colonized on discharge. In some discharged patients, resistant bacteria clear quickly, but a fraction, p, become ARB carriers. Some proportion of patients were already carriers at the time of admission, denoted by k. Institutions vary by size, H, and average length of stay (1/s). Thus, the rate that new carriers are discharged from a hospital is given by the formula: sH(px − k). This formula measures the contribution of a hospital to the number of ARB carriers in the community. For example, a hospital with 400 filled beds (H = 400 people) serves a US population of about 250,000 people. With a five-day average length of stay (the discharge rate is s = 0.2 per patient per day), the hospital discharges about 80 patients each day. If we suppose that 20% of patients acquire resistant bacteria while hospitalized, and one in four of these patients become carriers (px − k = 0.05), a hospital would discharge about four persistently colonized people per day---about 1,460 carriers after one year, or approximately 0.58% of its catchment population. A different formula characterizes heterospecific transmission, following exposure to ARB on contaminated food. We let g denote the daily per-capita rate that ARB are ingested with a meal. Similarly, we let h denote the proportion of those ARB populations that survive the gastric barrier and persistently colonize. The number of new carriers generated in the community by agricultural antibiotic use in a population of size N is: ghN. For example, if the average person consumes some ARB in 1% of meals (g = 0.03 per person per day), followed by colonization with probability one in 2,000 (h = 0.0005), agricultural antibiotic use would generate about four new carriers per day in a population of 250,000 people, N, approximately the same number as a hospital. The formulas illustrate a general principle: "A large number of people exposed to a small risk may generate many more cases than a small number exposed to a high risk" \[[@pmed-0020232-b34]\]. ::: ![](pmed.0020232.g002) ::: A Natural Experiment: Glycopeptides and Vancomycin-Resistant Enterococci {#s4} ======================================================================== Is the impact of agricultural antibiotic use on the emergence and spread of ARB in humans large or small relative to medical antibiotic use? Put another way, are farms or hospitals bigger polluters? A large-scale natural experiment was conducted in the United States and several European countries when each adopted different policies on glycopeptide use in animals (avoparcin) and humans (vancomycin) \[[@pmed-0020232-b16],[@pmed-0020232-b17],[@pmed-0020232-b35]\]. Many European countries approved avoparcin for animal growth promotion in the 1970s, but the US did not. In the early 1980s, demand for vancomycin in US hospitals surged because of increasing aminoglycoside resistance among enterococci and methicillin resistance in Staphylococcus aureus. Physicians in US hospitals also used oral vancomycin for some Clostridium difficile infections \[[@pmed-0020232-b37]\]. In the late 1980s and early 1990s, vancomycin-resistant enterococci (VRE) emerged and spread through US health-care systems. In Europe, hospitals used less vancomycin because most enterococci were sensitive to aminoglycosides, and oral vancomycin was seldom used. VRE still emerged and spread through European hospitals, but the problem has been less severe than in the US \[[@pmed-0020232-b40]\]. A different pattern emerges for community prevalence of VRE. VRE are rarely found outside of hospitals in the US, except for patients who have a prior history of hospitalization. Community prevalence of VRE in the US is typically less than 1%. In contrast, community prevalence of VRE was estimated at 2%--12% in Europe during the late 1990s, including carriage by people with no history of hospitalization \[[@pmed-0020232-b17],[@pmed-0020232-b41]\]. In other words, the European community reservoir generated by vancomycin use in hospitals and avoparcin use in agriculture was apparently much larger than the US community reservoir generated only by vancomycin use in hospitals. The prevalence of VRE in the community declined after the EU banned avoparcin \[[@pmed-0020232-b15]\]. Thus, avoparcin is at least partly responsible for the reservoir of VRE in the European community, but how much of that reservoir came from avoparcin and how much came from hospitals? To weigh the impact, we subtract the community prevalence of VRE in the US (\<1%) from the community prevalence of VRE in Europe (\>2%). The remainder (\>1%) is attributed to avoparcin. This analysis probably underestimates the real impact because vancomycin was used less in European than in US hospitals. Thus, avoparcin use in Europe would appear to be responsible for generating a larger reservoir of VRE in the community than US hospitals. Put another way, the impact of avoparcin use on European hospitals was larger than the impact of US hospitals on one another. Conclusion {#s5} ========== Despite the evidence that avoparcin use has had a large impact on the emergence and spread of VRE by increasing the reservoir of VRE in the EU, some uncertainty continues to surround the clinical significance of VRE strains of animal origin and of the zoonotic origins of resistance in general. Bacterial strains circulating in hospitalized populations may be genetically distinct from those circulating in the general human population \[[@pmed-0020232-b13],[@pmed-0020232-b17],[@pmed-0020232-b49]\]. Thus, bacterial populations are some combination of zoonotic, quasi-epidemic, and epidemic strains. The complexity of bacterial population biology and genetics makes it practically impossible to trace bacteria (or resistance factors) from the farm to the hospital, or to directly attribute some fraction of new infections to agricultural antibiotic use. Asymptomatic carriage of resistance factors by nonfocal commensal bacteria adds to a general risk of resistance, but transfer of resistance among bacterial species is unpredictable and difficult to quantify. Until more evidence is available, it is prudent and reasonable to consider bacteria with resistance genes a general threat \[[@pmed-0020232-b50]\]. Some part of the controversy over agricultural antibiotic use has been a disagreement about how to weigh evidence and make decisions when the underlying biological processes are complex. In this case, the effects of agricultural antibiotic use on human health remain uncertain, despite extensive investigation, and the effects may be unknowable, unprovable, or immeasurable by the empirical standards of experimental biology. What should be done when complexity makes an important public-health effect intrinsically difficult to measure? What is an appropriate "null hypothesis" or its equivalent? Should the same standards of proof be used in science and science-based policy? Where should the burden of proof fall? Scientific assessments for policy should summarize the best state of the science, recognizing that the burdens and standards of proof are necessarily softer because of the uncertainty that is introduced by biological complexity. The best decisions weigh the evidence in light of the inherent uncertainty. The EU banned the use of antibiotics for growth promotion, based on the precautionary principle. The use of the precautionary principle was criticized by some as unscientific in this context. In fact, the intrinsic problem of knowability, posed by the biological complexity of the problem, makes the use of precautionary decision making particularly suitable in this arena. The assumption that plausible dangers are negligible, even when it is known that such dangers are constitutively very difficult to measure, may be more unscientific than the use of precaution. Summary Points -------------- The emergence and spread of ARB is complex and intrinsically difficult to study; mathematical models can help with understanding underlying mechanisms and guiding policy responses.Agricultural antibiotic use may generate novel types of ARB that spread to the human population; models can help estimate how much additional disease has been caused by agricultural antibiotic use.Transmission of ARB from animal to human populations is particularly difficult to measure, as it is the product of a very high exposure rate to potentially contaminated food, and a very low probability of transmission at any given meal.Depending on the assumptions used, the model suggests that transmission from agriculture can have a greater impact on human populations than hospital transmission.A comparison of patterns of colonization of VRE in Europe and the United States, which had different patterns of agricultural and hospital antibiotic use, suggests that agricultural antibiotic use can have important quantitative effects on the spread of resistance in the community. Citation: Smith DL, Dushoff J, Morris G Jr (2005) Agricultural antibiotics and human health. PLoS Med 2(8): e232. The views presented in this paper represent the personal views of the authors and do not construe or imply any official position or policy of the Fogarty International Center, National Institutes of Health, Department of Health and Human Services, or the US government. ARB : antibiotic-resistant bacteria VRE : vancomycin-resistant enterococci [^1]: David Smith is a mathematical epidemiologist and infectious disease ecologist at Fogarty International Center, National Institutes of Health, Bethesda, Maryland, United States of America. Jonathan Dushoff is on the research staff in the Department of Ecology and Evolutionary Biology at Princeton University, Princeton, New Jersey, United States of America, and at Fogarty International Center, National Institutes of Health. J. Glenn Morris, Jr., is a physician epidemiologist and specialist in infectious diseases, and chair of the Department of Epidemiology and Preventive Medicine in the School of Medicine, University of Maryland, Baltimore, Maryland, United States of America. [^2]: Competing Interests: DLS and JGM received funding four years ago from Pfizer to conduct a risk assessment for the emergence of streptogramin resistance in Enterococcus faecium. JD declares that he has no competing interests.	PubMed Central	2024-06-05T03:55:58.197092	2005-7-5	{ "license": "Creative Commons Zero - Public Domain - https://creativecommons.org/publicdomain/zero/1.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1167557/", "journal": "PLoS Med. 2005 Aug 5; 2(8):e232", "authors": [ { "first": "David L", "last": "Smith" }, { "first": "Jonathan", "last": "Dushoff" }, { "first": "J. Glenn", "last": "Morris" } ] }
PMC1168885	Introduction {#s1} ============ "Academic medicine" might be defined as the capacity of the health-care system to think, study, research, discover, evaluate, innovate, teach, learn, and improve. As such, little could be more important---particularly as new discoveries in science offer tremendous opportunities and emergent diseases pose huge threats. Yet something is not right with academic medicine. Worse, the diagnosis is not entirely clear, although many august bodies have reported on the issue ([Table 1](#pmed.0020207-t001){ref-type="table"}), and the treatment is unknown. Moreover, these previous consultations have been based in the industrialised world, particularly the United Kingdom and the United States, and few have taken a global view. ::: {#pmed.0020207-t001 .table-wrap} Table 1 ::: {.caption} ###### Reports of Major National Academic Medicine Organisations ::: ![](pmed.0020207.t001) ::: The International Campaign to Revitalise Academic Medicine (ICRAM) was founded to give young medical academics an opportunity to think about the future of academic medicine ([Box 1](#box1){ref-type="boxed-text"}). It started with only two premises: (1) at the start of the new millennium it was necessary to think globally, and (2) "more of the same" was not the answer. Reinvention was needed. Box 1. ICRAM ------------ In 2003 the BMJ, Lancet, and 40 other partners launched ICRAM, a global initiative that is committed to fostering a debate about the future of academic medicine ([bmj.com/academicmedicine](bmj.com/academicmedicine)). The campaign arose because of a persistent concern that academic medicine is in crisis around the world. At a time of increasing health burden, poverty, globalisation, and innovation, many have argued that academic medicine is nevertheless failing to realise its potential and global social responsibility. ICRAM is composed of the following: A core working party of 20 medical academics representing 14 countriesStakeholder groups representing the areas of academia, business and industry, government and policy makers, journal editors, patients, professional associations, and students and traineesRegional groups covering the worldA facilitating committee that helps plan and execute the ICRAM work Through a series of stakeholder and regional consultations, systematic review of the available evidence, and future scenarios building, ICRAM intends to produce a series of recommendations for reform in global academic medicine, including the following: Developing a vision and values for academic medicineRecommending strategies for building capacity in academic medicine, including better career pathsProposing how academic medicine could improve its relationships with "customers," including patients, policy makers, practitioners, and others ICRAM decided to undertake some scenario planning, a technique of thinking about both the future and the present. The technique works by gathering together a team who consider the instabilities in the present and the drivers of the future and who then imagine plausible but different futures. The aim is not to predict the future, which is impossible, but to enable richer conversations through stretching thinking on what the future might bring. Once the scenarios have been created, they can be used to think more deeply about the present and the near future. The group began by considering current instabilities in academic medicine around the world ([Box 2](#box2){ref-type="boxed-text"}). None of these ideas are new, and most people would probably agree with most of them. Box 2. Current Instabilities in Academic Medicine ------------------------------------------------- Widespread, even universal, agreement that things are not right but little agreement on the exact nature of the problemLack of capacity in "translational research"---that which brings innovations directly to patientsThe substantial gap between best, evidence-based practice and what actually happensThe canyon between academics and practitionersThe growing difficulty/impossibility of a single individual being competent in practice, research, and teachingUse of citation indices in research assessment, which overemphasises the value of basic research and underemphasises the importance of applied research that may bring more immediate benefit to patientsA lack of mutual respect among different categories of researchers---basic, clinical, public health, primary care, applied, etc.Problems with career progression for academicsShortages of doctors wanting to enter careers in researchIn many, probably most, countries those doctors who enter careers in research, who ideally would be the "best and the brightest," being likely to earn much less than those who can spend at least some time working in private practiceResearch often not being concerned with the biggest health problems (particularly true in a global context)Systems of accreditation of doctors making life difficult for doctors who want to pursue an academic careerAlthough no clinicians are openly against research, clinicians being often unimpressed with doctors who concentrate on researchToo much medical research being undertaken by doctors with limited training in research methods---making the research of poor qualityResearchers often setting little store by quality improvement projects, although such projects are one of the ways of ensuring that all patients receive the most up-to-date careMuch of the teaching in medicine being done by people with very little training in medicineThe ratio of teachers to students in many medical schools being so low that the quality of teaching is reducedIn many countries academic medicine lacking a well-resourced institution to speak for itAcademic medicine relating poorly to its stakeholders---patients, policy makers, practitioners, the public, and the mediaThe great pressures on health services, such that academic medicine is often squeezed and forgotten Much of what determines the future of academic medicine will lie outside the control of medical academics themselves. The world will change around them, and they will have to follow. But there will also be change that comes from within academic medicine. [Box 3](#box3){ref-type="boxed-text"} shows some of the drivers of the future considered by the group. Box 3. Drivers of Change in Academic Medicine --------------------------------------------- New science and technology, particularly genetics and information technologyThe rise of sophisticated consumersGlobalisationEmergent diseasesIncreasing gap between rich and poorDeath of distance"Big hungry buyers" demanding more from health careSpread of the internet and digitalisationManagerialismIncreasing anxieties about securityExpanding gap between what could be done and what can be afforded in health careLack of agreement on where "health" begins and endsAgeing of societyFeminisation of medicineIncreasing accountability of all institutionsLoss of respect for expertsRise of self careRise of ethical issues24/7 societyEconomic and political rise of India and China Five Scenarios {#s2} ============== In building scenarios, the group used a time span of 20 years, but some of the scenarios are more futuristic than others. We decided to write up the scenarios as if they were in the past in order to give some idea of how they might arise. The scenarios are summarised in [Table 2](#pmed.0020207-t002){ref-type="table"}, and described more fully in the rest of this article. ::: {#pmed.0020207-t002 .table-wrap} Table 2 ::: {.caption} ###### Summary of Scenarios ::: ![](pmed.0020207.t002) ::: Academic Inc.: "Academic medicine flourishes in the private sector". Slowly but surely the public sector around the world realised that it could not support the costs of academic medicine. Medical students had high earnings during a professional lifetime: why shouldn\'t they pay for their education? And if researchers were doing something valuable then shouldn\'t they be able to find a market for their product---accepting that sometimes payment would come from the public sector? The process of academic medicine moving almost entirely into the private sector began with an increasing number of medical schools becoming private. The most prestigious schools went first. In an increasingly global market these schools could charge high fees, pay their staff well, and improve their facilities. They also invested a great deal in information and communication technology, bringing state-of-the-art learning to their students. This meant too that the schools could run courses for students far away from their geographical base. As these schools developed they expanded internationally, sometimes forming alliances with other prestigious schools but also taking over weaker schools. Soon the best schools were operating on all five continents. In the branches in developing countries, medical student bodies tended to comprise both students from the developed countries along with a quota from the developing countries. Competition was intense and involved both cost and quality. Schools that managed to improve quality while reducing costs---usually through clever use of technology---flourished, but a great many medical schools disappeared. The numbers of students, however, increased, and the competition for talent was intense, with schools offering generous bursaries to poor but bright students and becoming ever more sophisticated at finding high-quality students in resource-poor populations. As happens in most intensely competitive markets, medical schools also competed by occupying niches. Schools offered very different kinds of courses, specialising in older students, basic science, rural medicine, surgical skills, training doctors for poor communities (in both the country where the medical school was based and lower and middle income countries), and many other subject areas. Sometimes the students\' fees were paid by governments, local communities, or the military in order to produce students who met their needs. Many students attended schools in countries other than their own. Health research happened almost entirely in the private sector, but in a wide range of organisations: pharmaceutical companies, medical schools, biotechnology companies, small companies offering a huge range of services, and charities. Companies were founded not only by researchers but also by patients, practitioners, and others. Many of the companies founded by academics offered complex and innovative heath services. As in all business, to be successful companies had to be highly responsive to the needs of customers, including patients and governments. Those that were innovative, flexible, responsive, and relentlessly cost-conscious flourished, but many companies "failed". Little stigma was, however, attached to "failure". Indeed, as in Silicon Valley at the end of the 20th century, experience of "failure" was seen by many as an important qualification in a leader. The injection of more competitive pressure and a competitive business model into academic medicine increased not only efficiency but also "effectiveness": research was much more relevant, and the time lag between the development of new ideas and their introduction into practice was dramatically shortened. Basic science was still well funded because both governments and investors recognised the potentially high returns. Research into the health needs of poor and marginal populations also improved because public sector bodies concentrated their resources into these problems, leaving the problems of the wealthier to the market. Applying a competitive business model to academic medicine meant that efficiency and effectiveness trumped equity. On the negative side, the scenario "Academic Inc." resulted in a two-tier system, with the rich finding it easy to create careers in academic medicine and the poor finding it hard to enter the profession---despite the generous bursaries available to some. In addition, much more attention continued to be paid to the health problems of wealthier people and countries, and the brain drain from poor to rich countries accelerated. Innovation also suffered. Private academic medicine enjoyed less lead time and had more direct and immediate accountability to its shareholders than when it was publicly funded. Reformation: "All teach, learn, research, and improve". Twenty years ago there was increasing concern about the gap between academic medicine and practice, with important research results not being implemented, too much irrelevant research, bored students, and practitioners who stopped learning. In some medical communities and among their medical leadership, the response was not to try and strengthen academic medicine and make it more responsive but rather to abolish it and instead to bring the processes of teaching, learning, researching, and improving into the main stream of health care, and tailor these to local needs. This innovative---though not initially welcomed---response proved to be highly successful and was copied everywhere. A century of separation of academic medicine from practice was ended. Professors disappeared. The entity "academic medicine" was dead. It was akin to the destruction of the monasteries and so became known as the reformation of academic medicine. Teaching, learning, researching, and quality improvement all began to take place in the practice setting and were everybody\'s business. But importantly the fiction that an individual could be competent in practice, teaching, research, and improvement finished. It was teams that had to have all these competencies not individuals, and substantial investment was made in getting teams to work well and to communicate to a degree rarely seen before in health care. The teams were supported by advanced technology that provided online learning, decision support, answers to questions that arose during practice, and access to research results. Much of the teaching and research was done in collaboration with patients, and all teams included patients as well as practitioners, students, professional researchers, and other health professionals. Research was built around the questions that arose when doctors (and other health professionals) and patients consulted together. The questions were collected by the National Question Answering Service, which provided evidence-based answers to questions when it could. The service also organised research to answer questions that arose commonly, were unanswered at that time, but could be answered. Teams of different sizes and skills were assembled to conduct research; some of the researchers were permanently in practice but others---particularly basic researchers---were resident in research institutions and joined teams as needed. Some research was driven by discoveries made in basic science rather than questions that arose during practice; the fact that practitioners and researchers were used to working together in teams facilitated "translational" research. The teams also switched back and forth from research to quality improvement, ensuring that research developments were fed through into practice. Studies reporting the results of quality improvement projects were published just as frequently as research studies, and highly efficient information systems ensured that relevant information reached practitioners quickly---unlike in the old days when practitioners had been deluged with research results, most of which were irrelevant to them. Intellectual leadership in health care was provided by specialist societies, which included patients, citizen juries, researchers, health professionals other than doctors, health managers, and policy makers as equal members. In most countries the specialist societies were gathered together into an academy or institute of health care, and there was an international academy that was much more than a talking shop: it had a strong influence on world leaders. Health students started their training with six months in institutions that taught them how to learn. The learning professionals who staffed these institutions were available to all the practice teams. Students then learnt through attachment to practice teams, starting with a spell in general practice. Some students specialised early, some becoming competent cardiologists within five years. Becoming an independent professional depended not on university degrees or exams but on a demonstration of competencies determined by a national body dominated by patients that used the most up-to-date methods for assessment. Learning was by doing, and the divisions of undergraduate, specialist, and continuing education disappeared---as did the divisions among teaching, learning, researching, and quality improvement. One problem with the "Reformation" scenario was that it lacked stability because it required shared values and beliefs, which not all teams held. Plus decision making could be slow, and it was hard for brilliant and charismatic individuals to shine as leaders and thinkers. One result was that such individuals eschewed careers in health care, research, and education. In developing countries, in particular, the lack of an academic medicine structure meant that there were fewer opportunities to influence medical research and training. In the public eye: "Success comes from delighting patients and the public". Academic medicine was slow to recognise the rise of global media, "celebrity culture," and the use of public relations (or spin) to drive the political process, but once it did recognise how the world had changed it responded dramatically. Whereas it had been suspicious of the media and public appeal and rather patronising to patients, academic medicine realised that to succeed it must delight patients and the public and learn to use the media. The most successful academics became those who were very responsive to patients and the public, capturing their imaginations, and appearing regularly on their television screens. Some medical academics became as well known as film and rock stars and were feted by politicians. All academic institutions became dominated by public citizens and patients, and the public and media relations department became the most important department in any institution. Money---from both public and private sources---followed "interest," which was often influenced through game and "reality" shows on television. Academic medicine learnt from sport, and large prizes were awarded to those who won academic competitions. Although some academics were horrified by these developments, others remembered how John Harrison had been stimulated to solve the problem of calculating longitude through the promise of a large prize. Not all decisions on research priorities and resource allocation were made in the glare of television cameras. Although all decisions put the public interest first and were made by the representatives of the public, some were made by more sedate and evidence-driven bodies like citizens\' juries, where randomly selected patients and members of the public were presented with detailed evidence by "experts." Medical training was also conducted in the public eye, with students receiving much of their training from expert patients. The agenda for training was set predominantly by the public and patients, and responsiveness to patients was the number one characteristic of successful doctors and students. There was much greater diversity than at the beginning of the millennium in the form and size of academic institutions, with both huge public and private universities and smaller institutions that were often built around one charismatic individual. Competition among the institutions was intense---particularly for "celebrity" teachers and researchers. Only those institutions that could attract and keep public attention could survive. In developing countries, the academic health community linked itself with strong consumer movements, such as those focused on HIV/AIDS, and the leading nongovernmental organisations established their own medical schools. These ensured a powerful public voice, that training was tailored to local need, and a committed group for field testing new research advances. On the negative side in the "In the Public Eye" scenario, medical academics felt more anxious about their job security and ability to succeed. Even celebrity academics worried their time in the spotlight was short lived. Advances in science, medicine, and technology were shaped by popular appeal, and thus subject to fads and fashion. Some patients struggled with their new-found status as governors, and there was little regulation of health information. Global academic partnership: "Academic medicine for global health equity". In 2005, the world began to find the growing global gap between the rich and poor unacceptable. The concern was driven partly by the media and global travel bringing the plight of the poor in front of the eyes of the rich, but it was also driven by anxieties over global security. Terrorism was recognised to be fuelled by the gap between rich and poor. Global policy makers also understood better---particularly after the report of the Commission on Macroeconomics and Health \[[@pmed-0020207-b2]\]---that investment in health produced some of the richest returns in not only social but also economic development. Health care was a "must have" not a "nice to have". Money flowed into health in the poor world, and governments required that the investment be accompanied by learning, research, planning, and evaluation. The primary concern of much of academic medicine became to improve global health, particularly through concentrating on the health problems of the 90% who had previously received only 10% of health care resources (the 90:10 gap). Academics became excited by this kind of work not only because it was intellectually exciting and highly rewarding in personal terms but also because it was where money and prestige were most likely to be found. The result was that it was impossible for an academic institution to be a world leader without a substantial investment in global health and extensive links around the world. The view and scope of academic medicine broadened. It was increasingly concerned with human rights, justice, economics, and the environment, recognising that these are the major drivers of health. This broader view meant that academic medicine (usually referred to as "global health innovation" by 2012) became the main institution concerned with the rights of those who will be alive 50 years from now, a group that previously had nobody to speak for them. But at the same time basic science remained important because of the contribution it could make to global problems like finding vaccines and new treatments for malaria, AIDS, and emergent diseases like rapidly spread respiratory virus, which appeared in 2010 and killed millions in a global pandemic. Academic medicine, in partnership with governments (and where corruption is prevalent, with nongovernmental organisations), became a major driver towards achieving the millennium development goals. The G8 governments had signed an accord that prohibited recruitment of academic health professionals from developing countries. Medical schools and research institutions formed themselves into networks linking with local nongovernmental organisations, joining developed and developing countries and forming links among developing countries. A network was formed whereby the universities in developed countries committed 10% of their faculty members\' time to addressing problems of the developing world. Some institutions formed developed country--developing country pairs, some merged, and researchers, teachers, and students moved regularly between both settings. The net flow was to the developing world, with the 90:10 divide beginning to correct itself surprisingly rapidly. Big investments in information and communication technology meant that those in developing countries had the same access to information and modern learning methods as those in developed countries. The networks of institutions developed a global governance structure with substantial input from politicians, practitioners, policy makers, the public, and patients. Academic medicine moved from being marginal to central in global affairs, and medical academics, particularly those with experience in both the developing and developed world, became global leaders. It was a development that happened naturally because of their broad interests in human rights, justice, and the environment. The "Global Academic Partnership" scenario for academic medicine, however, was idealistic and sometimes struggled with realising its full potential---despite the best intentions of its architects and practitioners---because it required enormous political will and global cooperation. Too often nations would revert to narrow self interest. Academics as well often longed for the comforts of the developed world and sometimes felt exhausted by extensive travelling and the enormous problems of the developing world. Fully engaged: "Academic medicine engages energetically with all stakeholders". Early in the new millennium academic medicine became concerned that its relationships with its stakeholders were mostly poor. The public had little or no understanding of what academic medicine was or why it mattered. Its very name implied irrelevance to many. Patients often felt patronised by academics, and many practitioners---including doctors---were unconvinced of the value of academic medicine. "I wouldn\'t want a professor of surgery touching me," was a commonly heard refrain. Although some leading academics did have good relationships with politicians, policy makers found that many academics were not interested in policy problems and that the studies they produced, even if relevant, came too late to be useful. Policy makers recognised that biotechnology might be very important in future wealth creation, but it was difficult to fund because the public profile of academic medicine was both low and clouded. Most medical academics recognised that they were doing a poor job of relating to stakeholders and that it was thus unsurprising that they were misunderstood, underappreciated, and seen as largely irrelevant. This, they thought, was particularly unfortunate as the ability of the system of health care to discover, think, study, learn, and evaluate had never been more important. The medical academic community thus decided that it had to do better, and across the globe medical academics organised themselves to engage fully with their stakeholders. In many countries this meant the creation of new organisations. In others it involved the transformation of existing organisations: the gongs and gowns were abandoned, and focus groups began. Fifty prestigious universities in developed countries with medical faculties partnered with universities in developing countries to help stop the "brain drain" and replace it with a "brain gain" through incentive programs that provided resources for training and research, academic recognition, travel funds, and family support. Everywhere medical academics had to learn how to communicate with the public, patients, and practitioners. They had to stop being elitist and patronising and recognise the messiness of public discourse. Crucially, they had to be much cleverer in handling the media, telling them not only about their successes but also sharing their uncertainties and problems. But communication on its own wasn\'t enough. Academic medicine had to bring its stakeholders inside its processes. The governance of academies included patients, the public, practitioners, health administrators, and policy makers. Sometimes the president of an academy was not a distinguished researcher but a prominent patient, journalist, or community organisation leader. The medical academics discovered that their arguments were taken much more seriously when advanced clearly by a patient rather than by themselves. Patients, health administrators, and community organisation representatives became involved not only in peer review of grants and studies but also in the prioritising, designing, and conducting of research. Medical students became the main drivers of medical education rather than simply its consumers. Slowly but surely medical academics became not a group apart but a highly diverse group of people with a broad set of skills and backgrounds. They were at the centre of a vibrant community of patients, members of the public, practitioners of all stripes, policy makers, members of the media, marketing experts, and politicians, all of whom were interested in learning, studying, researching, and thinking about health care. Some academics, it must be said, found the change uncomfortable and were unconvinced of its value. Critics talked of "dumbing down" and popularisation. They fretted that in abandoning its elitism academic medicine had lost its ability to be truly original and speak independently. For most, however, academic medicine was so much more fun than it used to be. Applications to medical schools increased. Health services invested more in evaluating what they did and paid more attention to the results. More funds flowed into basic research, and there were improvements in the connections between the many diverse groups involved in research---with the result that intellectual silos were breached. Lessons from the Scenarios {#s3} ========================== None of these scenarios will come to exist as we have described them, but the future is likely to contain some elements from each of them. We have tried to identify common features in the scenarios to learn lessons for now. These features are shown in [Box 4](#box4){ref-type="boxed-text"}. Our main hope for these scenarios is that other groups may find them useful in thinking about both the present and the future of academic medicine. The scenarios will need to be adapted to the particular social, economic, and political conditions of different regional and national settings. As such, they are tools that can be used globally, modified as required. We seek not agreement but broader thinking. Box 4. Common Features Shared By All Five Scenarios --------------------------------------------------- It is likely to be important in all scenarios for academic medicine to put more effort into relating to its stakeholders: the public, patients, practitioners, politicians, and policy makers. This may demand the development of new institutions that involve all these groups.Academic institutions will need to be more globally minded.Teaching, researching, improving, leading, and providing service will continue to be important, but expecting individuals to be competent in them all will be increasingly impractical.Teamwork will become ever more important, but it will also be necessary to allow individuals to shine and flourish.Competition among academic institutions is likely to increase, and the competition will increasingly be international.Academic institutions will need to become more "business-like" in all the scenarios. They will also need to be more adept at using the media.Teaching and learning will be increasingly important---not least because dissatisfied students may go elsewhere. Learning will be lifelong and will depend heavily on information technology.It will be increasingly important to combine research, both basic and applied, with implementation and improvement. The gap between knowledge and practice will become increasingly intolerable.The range of types of academic institutions is likely to become increasingly diverse, with medical schools or academic centres just one form.Academic medicine will need to be ever broader in its thinking and skill set, combining with and learning from other disciplines such as economics, law, ecology, and humanities.Thinking about the future will become increasingly important for academic institutions but also increasingly difficult. The members of the working party of the ICRAM are Tahmeed Ahmed (Scientist, Clinical Sciences Division, International Centre for Diarrhoeal Disease Research Bangladesh, Dhaka, Bangladesh), Shally Awasthi (Professor, Department of Paediatrics, King George\'s Medical University, Lucknow, India), A. Mark Clarfield (Professor, Department of Geriatrics, Soroka Hospital, Ben-Gurion University of the Negev, Beersheva, Israel), Lalit Dandona (Director, Centre for Public Health Research, Administrative Staff College of India, Hyderabad, India), Amanda Howe (Professor of Primary Care, School of Medicine, University of East Anglia, Norfolk, United Kingdom), John P. A. Ioannidis (Chairman, Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece), Edwin C. Jesudason (Academy of Medical Sciences National Clinician Scientist, Health Foundation Leadership Fellow, and Lecturer in Paediatric Surgery, School of Reproductive and Developmental Health, University of Liverpool, Liverpool, United Kingdom), Youping Li (Director, Chinese Cochrane Centre, West China Hospital, Sichuan University, Chengdu, China), Juan Manuel Lozano (Professor, Department of Pediatrics and Clinical Epidemiology, Javeriana University School of Medicine, Bogota, Colombia), Ana Marusic (Professor, Department of Anatomy, Zagreb University School of Medicine, and Editor, Croatian Medical Journal, Zagreb, Croatia), Idris Mohammed (Outgoing Provost, College of Medical Sciences, Department of Medicine and Clinical Immunology, University of Maiduguri, Maiduguri, Nigeria), Gretchen Purcell (Pediatric Surgery Fellow, Pittsburgh Children\'s Hospital, and Adjunct Assistant Professor of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania, United States), Karen Sliwa-Hähnle (Professor, Department of Cardiology, CH Baragwanath Hospital, University of the Witwatersrand, Johannesburg, South Africa), Sharon E. Straus (Associate Professor, Department of Medicine, Toronto General Hospital, University of Toronto, Toronto, Ontario, Canada), Tessa Tan-Torres Edejer (Scientist, Department of Health Systems Financing, World Health Organization, Geneva, Switzerland), Timothy Underwood (Medical Research Council/Royal College of Surgeons Clinical Research Training Fellow, University of Southampton, Southampton, United Kingdom), Robyn Ward (Associate Professor, Department of Medical Oncology, St Vincent\'s Hospital and School of Medicine, University of New South Wales, Darlinghurst, New South Wales, Australia), Michael S. Wilkes (Vice Dean and Professor of Medicine, School of Medicine, University of California, Davis, California, United States), and David Wilkinson (Deputy Head and Professor of Primary Care, School of Medicine, University of Queensland, Brisbane, Australia). Note Added in Proof Reference information for the summary of this article being published in the BMJ: Clark JP, for ICRAM (2005) Five futures for academic medicine: The ICRAM scenarios. BMJ. In press. Citation: Awasthi S, Beardmore J, Clark J, Hadridge P, Madani H, et al. (2005) Five futures for academic medicine. PLoS Med 2(7): e207. This article is a shortened version of a report entitled "The Future of Academic Medicine: Five Scenarios to 2025" \[[@pmed-0020207-b1]\] published by the Milbank Memorial Fund ([www.milbank.org](www.milbank.org)). A summary of this article is being published in the BMJ. ICRAM : International Campaign to Revitalise Academic Medicine [^1]: Shally Awasthi is Professor, Department of Paediatrics, King George\'s Medical University, Lucknow, India. Jil Beardmore is Research and Project Coordinator at the Centre for Global Health, University of Ottawa, Ottawa, Ontario, Canada. Jocalyn Clark is Associate Editor of BMJ, London, United Kingdom, and Project Manager of ICRAM, London, United Kingdom. Philip Hadridge is an organisation consultant in Cambridge, England. Hardi Madani is a fourth-year medical student at the Royal Free and UCL Medical Schools in London, United Kingdom, and a part-time pharmacist. Ana Marusic is Professor, Department of Anatomy, Zagreb University School of Medicine, Zagreb, Croatia, and Editor of Croatian Medical Journal, Zagreb, Croatia. Gretchen Purcell is Pediatric Surgery Fellow at Pittsburgh Children\'s Hospital and Adjunct Assistant Professor of Medicine at the University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America. Margaret Rhoads is a medical student at Imperial College London, London, United Kingdom. Karen Sliwa-Hähnle is Professor, Department of Cardiology, CH Baragwanath Hospital, University of the Witwatersrand, Johannesburg, South Africa. Richard Smith is former Editor of BMJ, co-founder of ICRAM, and Chief Executive of UnitedHealth Europe, London United Kingdom. Tessa Tan-Torres Edejer is Scientist, Department of Health Systems Financing, World Health Organization, Geneva, Switzerland. Peter Tugwell is Professor of Medicine and Head of the Centre for Global Health at the University of Ottawa, Ottawa, Ontario, Canada, and Leader of ICRAM. Tim Underwood is Medical Research Council/Royal College of Surgeons Clinical Research Training Fellow, University of Southampton, Southampton, United Kingdom. Robyn Ward is Professor, Department of Medical Oncology, St Vincent\'s Hospital, Darlinghurst, New South Wales, Australia, and School of Medicine, University of New South Wales, Darlinghurst, New South Wales, Australia. See Acknowledgments for the members of the working party of the International Campaign to Revitalise Academic Medicine. [^2]: Competing Interests: JC is an employee of the BMJ Publishing Group, which benefits from academic medicine. RS is an employee of the UnitedHealth Group, which benefits from the output of academic medicine. It\'s not easy to see how the company might be affected financially from any of the possible changes in academic medicine. He is also an unpaid governor of St George\'s, University of London, and an unpaid board member of the Public Library of Science, both of which will be affected by any changes.	PubMed Central	2024-06-05T03:55:58.199189	2005-7-5	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168885/", "journal": "PLoS Med. 2005 Jul 5; 2(7):e207", "authors": [ { "first": "Shally", "last": "Awasthi" }, { "first": "Jil", "last": "Beardmore" }, { "first": "Jocalyn", "last": "Clark" }, { "first": "Philip", "last": "Hadridge" }, { "first": "Hardi", "last": "Madani" }, { "first": "Ana", "last": "Marusic" }, { "first": "Gretchen", "last": "Purcell" }, { "first": "Margaret", "last": "Rhoads" }, { "first": "Karen", "last": "Sliwa-Hähnle" }, { "first": "Richard", "last": "Smith" }, { "first": "Tessa Tan-Torres", "last": "Edejer" }, { "first": "Peter", "last": "Tugwell" }, { "first": "Tim", "last": "Underwood" } ] }
PMC1168887	Background ========== From a public-health perspective, the major obstetric and perinatal problem in Australia is the poor health of Indigenous mothers and babies. Although Australia as a whole has one of the lowest perinatal mortality rates (PMR) of any of the established market economies \[[@B1]\], Indigenous Australians have PMRs that are at least twice as high as their non-Indigenous counterparts \[[@B2],[@B3]\]. Moreover, the relative excess of deaths among Indigenous babies has not improved over time. In Queensland in 1987--1989, the PMR was 2.2 times higher among Indigenous compared with non-Indigenous babies and in 2000--2002 the rate was still 2.1 times higher. A similar lack of progress is evident in other states \[[@B4]-[@B6]\]. The risk of perinatal death is strongly related to a baby\'s birthweight and gestational age. For example, the risk of death for a moderately preterm baby (33 to 36 weeks gestation) in Queensland is 8.2 times greater than that of a term baby and for a very preterm baby (27 to 32 weeks) the risk is 41.3 times greater. Similarly, a baby who was born at term but weighed less than 2500 g was 6.3 times more likely to die than a baby who weighed more than 2500 g. In Queensland, Indigenous babies are 1.6 (95%ci: 1.5,1.7) times more likely to be born preterm (\<37 weeks gestation) and 2.2 (2.0,2.5) times more likely to be low birthweight at term than non-Indigenous babies \[[@B3]\]. Similar disparities in birthweight and gestational age have been reported from the other states of Australia \[[@B2],[@B5],[@B7]-[@B10]\]. The aim of this paper is to estimate the number and proportion of excess deaths among Indigenous babies that can be explained by their higher risk of low birthweight and preterm birth. This analysis strategy assumes that the distribution of birthweight and gestational age among Indigenous babies reflects the prevalence of antenatal risk factors such as smoking, infection, maternal nutrition and psycho-social stress; while any excess of mortality that remains after the excess low-birthweight and preterm risk among Indigenous babies is removed might say something about the quality of medical care at the time of confinement. Although such reasoning has many proponents \[[@B11]-[@B13]\], risk factors such as smoking and infection are likely to also have at least a small effect on mortality that is independent of birthweight and gestational age. That is, adjustment of perinatal mortality rates by birthweight and gestational age is not a perfect way of assessing the quality of medical care at the time of confinement. It is similar to case-mix adjustments used in other settings to allow for differences in risk \[[@B14],[@B15]\]. Such adjustments are not expected to remove all confounding. Instead, the reasoning is that the adjusted rates, although not perfect, provide a useful way of identifying policy issues, setting agendas, and facilitating discussion. To be more specific, if a large proportion of excess deaths can be explained by the unfavourable birthweight and gestational age distribution of Indigenous babies, then that would suggest that priority should be given to implementing primary health care interventions to reduce the risk of low birthweight and preterm birth (eg, interventions to reduce maternal smoking or genitourinary infections). Conversely, if only a small proportion is explained by birthweight and gestational age, then other strategies might need to be considered such as improving access to high-quality hospital care around the time of confinement. Methods ======= Data were obtained from the population-based Queensland Perinatal Data Collection for the five years 1998 to 2002. This was the most recent five-year period for which complete data were available. The database includes information on all livebirths and stillbirths of at least 20 weeks gestation or 400 g birthweight. A perinatal death is defined as a stillbirth or the death of a liveborn baby within 28 days of birth. Our data set comprised 231,039 births and 2,255 deaths to non-Indigenous mothers and 13,920 births and 273 deaths to Indigenous mothers. Indigenous status is based on the self-reported Indigenous status of the mother and the gestational age is based on the best clinical estimate, which might be derived from the date of the last menstrual period, ultrasound in early pregnancy or maturity scoring of the neonate at birth. The method used is not recorded. In this paper we report the results of a Mantel-Haenszel procedure, which was used to determine the relationship between perinatal mortality and Indigenous status adjusted for the effect of low birthweight and preterm birth. The proportion and number of deaths that could be avoided were estimated by comparison of the crude and adjusted rates. We also used Poisson-regression models to adjust perinatal mortality rates for birthweight and gestational age. We variously fitted single week or four week categories of gestational age and 250 g and 500 g categories of birth weight. The results were the same as using the Mantel-Haenzel approach with broad categories of gestational age and birthweight, and we included these in preference to the Poisson-regression models for ease of interpretation. Results ======= The crude perinatal mortality rate among Indigenous babies was 19.6 per 1000 births, which was 2.01 (95%ci:1.77,2.28) times higher than the rate among non-Indigenous babies. When perinatal mortality rates were compared within each birthweight and gestational age strata, the point estimates suggested that Indigenous babies were only slightly more likely to die than non-Indigenous babies (Figure [1](#F1){ref-type="fig"}). The test for homogeneity of the rate ratios across strata was not significant (χ^2^(3) = 5.78, p = 0.1226) suggesting that the effect of Indigenous status is the same across the birthweight and gestational age strata (except for statistical noise) and that it is appropriate to use the adjusted combined estimate: M-H Adjusted RR = 1.13, 0.99--1.28. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Rate ratios^a^comparing Indigenous to non-Indigenous perinatal mortality stratified by preterm and birthweight status, 1998--2002. a) Rate ratios greater than 1.0 indicate higher mortality among Indigenous babies. b) Differences between stratum-specific rate ratios are not statistically significant (χ^2^(3) = 5.78, p = 0.1226). ::: ![](1743-8462-2-11-1) ::: These results suggest that if the population of Indigenous babies had the same birthweight and gestational age distribution as non-Indigenous babies, then the relative disparity would be reduced by 87% and there would be 20 fewer deaths of Indigenous babies per year. Discussion ========== Interpretation of results ------------------------- Perinatal mortality among Indigenous babies has remained twice that of their non-Indigenous counterparts for more than a decade. We found that most of this mortality excess is because Indigenous babies are at greater risk of being born too early and too small. In contrast, the case fatality rates of Indigenous babies who were born preterm or of low birth weight were similar to their non-Indigenous counterparts. Using a framework advocated by several perinatal epidemiologists \[[@B11]-[@B13]\], these results suggests that, broadly speaking, access to high quality care during confinement is adequate for Indigenous mothers and babies. That is, priority should initially be given to primary health care interventions to reduce the proportion of preterm and low birth weight babies. Policy implications ------------------- Risk factors for preterm birth and low birth weight include smoking, gentio-urinary tract infections, poor maternal nutrition and psycho-social stress \[[@B16]-[@B19]\]. Several studies have reported a higher prevalence of these risk factors among Indigenous compared with non-Indigenous mothers. More specifically, the prevalence of smoking among Indigenous women during pregnancy has been reported to be more than 60%, which is at least 3 times the prevalence for non-Indigenous women \[[@B9],[@B20],[@B21]\]. A recent Cochran review found that there are effective primary health care interventions to help and support women to stop smoking that lead to fewer preterm babies and better birthweights \[[@B22]\]. Further, we know that Indigenous women are more than two times as likely to have a urinary tract infection during pregnancy as non-Indigenous women \[[@B23]\]. In overseas studies, primary health care interventions to detect and treat asymptomatic bacteruria have been shown to decrease preterm birth by 40% \[[@B24]\]. In Australia, the best example we have of a primary health care initiative aimed at reducing risk factors among Indigenous mothers is the Strong Women Strong Babies Strong Culture program in the Northern Territory \[[@B25]\]. This program resulted in increased early attendance for antenatal care, reduced numbers of STDs and a reduced proportion of low birthweight babies \[[@B26]\]. Although such results are encouraging, if substantial progress is to be made across the whole of Australia, a properly funded national initiative is needed. Such an initiative would include funding to improve access to culturally appropriate primary health care during the antenatal period, which would deliver, inter alia, interventions for smoking cessation, screening and treatment of genito-urinary tract infections, screening for domestic violence, and programs aimed at reducing alcohol consumption and poor nutrition. It would not be a case of one strategy fits all. Instead local partnerships with possibly different types of service models would be needed to implement the national initiative. This approach will encourage creativity, innovation and risk taking, which will be essential ingredients to tackling a situation that has proved difficult to improve. Study limitations ----------------- Using vital statistics to set agendas has a long and continuing tradition in public health \[[@B27]\]. The advantages of such statistics are convenience, low cost and total enumeration. The disadvantages are insufficient and inaccurate data, which create uncertainty about the validity of the results \[[@B27]\]. This study used four variables: perinatal death, birthweight, gestational age, and Indigenous status. It is unlikely that an important number of perinatal deaths were missed because they are checked against notifications to the Registrar-General of Births, Deaths and Marriages. It is also unlikely that there are important errors in the measurement of birthweight. Consequently, the main areas of uncertainty are Indigenous status and gestational age. With regard to Indigenous status, some mothers may be reluctant to identify as Indigenous, others may be non-Indigenous with an Indigenous male partner, or midwives may not ask the mother or make an educated guess \[[@B28]\]. However, of all the types of mortality data, perinatal mortality provides the most accurate estimate of excess Indigenous mortality because the numerator (number of perinatal deaths) and denominator (number of births) for the rate can be obtained from the one data set. This is in contrast to adult death rates where identification of Indigenous people can be different in death registration data (the numerator for mortality rates) and population data (the denominator). This problem of the numerator not being appropriate for the denominator is not unique to comparisons of Indigenous and non-Indigenous Australians; it hinders interpretation of race-specific rates around the world \[[@B29],[@B30]\]. Thus, of all the routinely available mortality data, perinatal data provides the most valid estimate of the mortality excess for Indigenous people and provides robust support for policy discussions. For several reasons, gestational age is known to be less accurate among Indigenous than non-Indigenous babies \[[@B31]\]. Nevertheless, previous work in Queensland and elsewhere has shown that gestational age in combination with birthweight provides a better statistical adjustment of mortality rates than birthweight alone \[[@B32],[@B33]\]. We therefore considered it better to present birthweight and gestational age adjusted rates, rather than just birthweight-adjusted rates. Conclusion ========== Although perinatal mortality rates in Queensland have decreased over the last 16 years, the rates in Indigenous populations remain at least double those in the non-Indigenous population. Our analyses, stratified by birthweight and gestational age, suggest that the priority for reducing the excess mortality among Indigenous babies is primary health care to reduce the prevalence of risk factors during the antenatal period. A primary health care approach encompasses a much-needed component of an overall shift towards empowerment of Indigenous women and increased awareness and ownership of health which has the potential to play an important role in reducing the social inequality that has resulted in outcomes such as those found for perinatal mortality. Authors\' contributions ======================= TJ performed the statistical analysis and participated in drafting the manuscript. MC conceived of the study and its design and participated in drafting the manuscript. Both authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:58.203197	2005-5-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168887/", "journal": "Aust New Zealand Health Policy. 2005 May 27; 2:11", "authors": [ { "first": "Trisha", "last": "Johnston" }, { "first": "Michael", "last": "Coory" } ] }
PMC1168888	Background ========== Periodic phenomena are widespread in biology, including, among others, membrane potential oscillations, cardiac rhythms, smooth muscle contraction, calcium oscillations, protoplasmic streaming, glycolytic oscillations, cAMP oscillations, oscillations in neuronal signals, insulin secretion (pancreas), gonadotropic hormone secretion, cell cycle, circadian rhythms, and ovarian cycle (see e.g. \[[@B1]\]). Consequently, there are numerous biological applications where periodicities must be detected from experimental biological data. Because the data measured from biological applications are inherently noisy and usually sparsely sampled, efficient algorithms are being developed to extract as much information as possible. In the past few years there has been an explosion of available microarray gene expression data. Detecting periodicity in gene expression is of great importance because it indicates, e.g., cell-cycle regulation \[[@B2]\] as well as the effect of circadian rhythms \[[@B3]\]. The significance of the detection of cell-cycle regulated processes is further emphasised by the linkage between cell-cycle and cancer (see, e.g., \[[@B4]\] and \[[@B5]\]). To this end, microarrays have been used to study the circadian gene expression in Neurospora Crassa\[[@B3]\] as well as cell-cycle regulated genes, e.g., in budding yeast \[[@B6]\], in fission yeast \[[@B7]\], and in human cells \[[@B5]\]. The task of finding periodicity in time series measured from a biological system can be viewed as a decision problem based on spectral analysis together with multiple hypothesis testing. A formal statistical testing procedure for the detection of periodic expression profiles was recently introduced by Wichert et al.\[[@B8]\]. It relies on the use of a so-called Fisher\'s g-statistic for which the exact null-distribution can be derived under the Gaussian noise assumption. Recently, a number of other methods for detecting periodic transcripts have also been proposed \[[@B9]-[@B13]\]. A major difference between the method proposed by Wichert et al.\[[@B8]\] and other methods is that Wichert\'s method is capable of detecting unknown frequencies whereas other methods are designed for detecting fixed frequencies. From a computational point of view, the problem of finding unknown frequencies is even more demanding since no prior knowledge of the frequency to be detected is available. In many biological applications it is more important to search for periodicities having an unknown frequency. However, in some applications, such as large-scale cell cycle studies, the period length is usually known and thus provides additional information for testing. In this paper, our goal is to tackle both of the two problems. That is, we develop two methods, one for the detection of unknown frequencies and the other for testing fixed frequencies. In many applications, including those arising from bioinformatics, the exact noise characteristics are usually unknown and can be remarkably non-Gaussian. Furthermore, the observed gene expression time series can exhibit other non-idealities, such as outliers, short length and distortion from the original wave form. Therefore, it would be useful to have a robust method for detecting periodic components, i.e., a method that also works well when the original (Gaussian) noise assumption no longer holds. A robust, rank-based, non-parametric spectral estimator was recently introduced in \[[@B14]\]. In this paper, we extend the approach of \[[@B14]\] to the detection of periodic time series. This results in a robust testing procedure which is insensitive to a heavy contamination of outliers, missing-values, short time series, nonlinear distortions, and is completely insensitive to any monotone nonlinear distortions. We also consider a permutation-based alternative to the method proposed in \[[@B8]\] and show that, when the data is contaminated with the above mentioned non-idealities, this results in a more robust method. As discussed, e.g., in \[[@B15]\], the performance of a method can be proven using either extensive simulations, analytical proofs or multiple plasmode experiments (to use a term from \[[@B15]\]). Plasmode is a real data structure whose true structure is known. Although very useful, the kinds of benchmarkings proposed in the literature (related to periodically behaving gene expression time series) so far do not belong to any of the above categories. In particular, the proposed benchmarking frameworks cannot be considered as plasmodes since the true structure (periodic genes) is not known but, instead, is based on partial biological knowledge (or other measurements, such as protein-DNA binding). Hence, the performance cannot be assessed solely on real data. Therefore, as the analytical proofs are hard to obtain in this case, we perform extensive simulations to show the superior performance of the proposed methods. However, we also apply the proposed method to real experimental data to show that the methods perform well on real data and that the results are biologically meaningful. Results ======= Computational methods --------------------- In order to be consistent with the previously published methods, we use similar notation as in \[[@B8]\] and also consider the same model for the periodic time series y~n~= βcos(ωn+ φ) + ε~n~, (1) where β≥ 0, ω∈ (0, π), n= 1,\..., N, φ∈ (-π, π\], and ε~n~is an i.i.d. noise sequence. To test for the periodicity, define the null hypothesis as H~0~: β= 0, i.e., time series consists of the noise sequence alone, y~n~= ε~n~. In the following we develop a method for detecting unknown frequencies and later introduce a modification which can be applied to the detection of known frequencies. We first review the Fisher\'s test for the detection of periodic transcripts as introduced by Wichert et al.in \[[@B8]\]. ### Wichert\'s method The method proposed by Wichert et al.\[[@B8]\] is based on the periodogram spectral estimator, defined as ![](1471-2105-6-117-i1.gif) where Nis the time series length. The periodogram is further evaluated at (harmonic) normalised frequencies ![](1471-2105-6-117-i2.gif) where a= \[(N-- 1)/2\] and \[x\] denotes the integer part of x. To test for the periodicity formally, some kind of a test statistic must be chosen. The so-called g-statistic for one time series is given by ![](1471-2105-6-117-i3.gif) In plain words, the g-statistic is the maximum periodogram ordinate divided by the sum of all periodogram ordinates for l= 1, \..., a. Large value of gindicates a strong periodic component and leads to the rejection of the null hypothesis. Wichert et al.\[[@B8]\] resort to a result by Fisher that, under the Gaussian noise assumption, gives the exact distribution of the g-statistic under the null hypothesis (see, e.g., \[[@B16],[@B8]\]). The exact p-value for a realisation of the g-statistic is shown to be ![](1471-2105-6-117-i4.gif) where bis the largest integer less than 1/xand xis the observed value of the g-statistic. Because there are usually thousands of time series that are tested simultaneously, whether they exhibit periodicity or not, there is a possibility that a time series can have a small p-value by chance. To correct the p-values for multiple testing, Wichert et al.\[[@B8]\] use the method of Benjamini and Hochberg (see, e.g., \[[@B17]\]), which controls the False Discovery Rate (FDR). The FDR method controls the expected proportion of false positives (Type I errors) at a given rate q. The threshold of the FDR depends on the evaluated p-values. The FDR procedure for the ordered set of p-values p~(1)~, p~(2)~,\...,p~(M)~, where Mis the number of time series, is as follows 1\. Let i~q~be the largest ifor which ![](1471-2105-6-117-i5.gif) 2\. Reject the null hypothesis for the time series corresponding to the p-values p~(1)~, p~(2)~,\..., ![](1471-2105-6-117-i6.gif) The procedure based on the periodogram spectral estimator for periodicity detection has several well-known and important properties. For example, if the noise sequence ∈~n~is i.i.d. Gaussian and the true underlying frequencies are among the harmonic frequencies Ω = {ω~l~: l= 0,\..., \[(N- 1)/2\]}, then the (square root of the) periodogram is the minimum variance unbiased estimator of the frequency content at discrete frequencies Ω (see, e.g., \[[@B18]\]). However, it is also widely recognized that the standard periodogram is an inconsistent spectrum estimator (see, e.g. \[[@B19]\]). Despite this weakness, the periodogram is a standard method that is theoretically well-founded. Under the popular Gaussian working assumption, widely invoked in spectral estimation, the distributional characteristics of the periodogram are known and useful. Indeed, it is this distributional characterisation that forms the basis for Fisher\'s g-test for whiteness. In other words, Equation (5) provides the exact significance value for a realisation of the g-statistic. This provides a solid theoretical basis for the use of the method proposed in \[[@B8]\]. Concerning the problem of detecting hidden periodicities, some generalisations and improvements over the traditional methods have been proposed. A recent review of the proposed methods can be found in \[[@B20]\]. Many of the improved methods are some type of generalisations of the traditional methods, such as the correlogram or periodogram. Artis et al.\[[@B20]\] report that two particular methods are generally found to have a good performance: so-called mixed spectrum methods by Priestley and Bhansali (see, e.g. \[[@B19],[@B20]\]), and a modified method based on the maximum periodogram ordinate by Chiu \[[@B21]\]. The method by Priestley and Bhansali uses a certain type of windowing of the correlogram for the \"smoothing\" purposes (i.e., in order to reduce the variance). The method by Chiu in turn modifies the g-statistic by replacing the average spectrum in the denominator with a proper trimmed mean of the ranked periodogram ordinates. As we explain in the Discussion Section, the same modifications/improvements can be used in the proposed robust method (to be introduced shortly) as well. Indeed, the windowing can be utilized much in the same way as it is used in the method by Priestley and Bhansali. Similarly, the modification(s) of Chiu can be directly implemented in the case of the proposed robust estimator. Fisher\'s test and the proposed robust method can be considered to be analogous in the sense that neither of them incorporates any modifications for the traditional periodogram/correlogram approach. In Simulations Section we perform an extensive comparison of the unmodified, standard methods since this provides a better insight into the performance differences of the traditional methods and the proposed robust rank-based methodology. Furthermore, from an extensive simulation point of view, comparison of many different modifications together with the robust method would also result in a rather large number of simulations, especially because it is possible to use many different combinations of the modifications. Note, however, that the further modifications, such as the ones proposed by Priestly, Bhansali, and Chiu, can be implemented in both of the frameworks (the traditional and the robust rank-based) to gain further performance improvements. We discuss this issue further in Discussion Section. In many applications, particularly in bioinformatics, the noise distributions are usually unknown and can remarkably deviate from the Gaussian assumption. The methods based on Gaussian noise assumption may fail, or even produce invalid results, when the model assumptions do not hold. The goal of this paper is to introduce an alternative for the standard methods, aimed at providing a robust estimator. ### A robust alternative As a starting point we should like to remind that the periodogram I(ω) is equal equivalent to the correlogram spectral estimator (see, e.g., \[[@B25]\]) ![](1471-2105-6-117-i7.gif) where ![](1471-2105-6-117-i8.gif) is the biased estimator of the autocorrelation function ![](1471-2105-6-117-i9.gif) Note that the required values for ![](1471-2105-6-117-i8.gif) for m\< 0 are obtained by invoking the inherent symmetry of the autocorrelation function: r(-m) = r(m). Consequently, the g-statistic in Equation (4) as well as the corresponding significance value in Equation (5) can also be computed using S(ω) instead of I(ω). It is natural to try to obtain robustness by replacing ![](1471-2105-6-117-i8.gif) with a robust alternative. Before continuing to the robust method, it is important to note that, especially in the case of gene expression time series, the data is often contaminated with missing values. Therefore, the spectral estimation method must take missing samples into account. Let I~m~be the set of time indices kfor which both y~k~and y~k+m~are available and K~m~= \\|I~m~\\|. As long as K~m~≠ 0, a missing data-adapted version of the unbiased estimate of the autocorrelation can be obtained as ![](1471-2105-6-117-i10.gif) We cover only the versions adapted to missing data in the following text since they are equal to the standard estimators in case of complete data sets. Note that K~m~equals N- mfor the complete data sets. Next we consider a recently introduced rank-based autocorrelation estimator \[[@B14]\] for the problems of spectrum estimation. This estimator is a moving-window extension of the Spearman rank correlation coefficient, quantifying the association between the sequences {y~k~} and {y~k+m~}. The resulting quantity, ρ^S^(m) is actually an alternative estimator of the standard correlation coefficient ρ(m) between these sequences (see e.g. \[[@B19]\]) ![](1471-2105-6-117-i11.gif) where ![](1471-2105-6-117-i12.gif)\[·\] denotes the expectation operator and μ~yi~= ![](1471-2105-6-117-i12.gif) \[y~i~\] is the mean of the sequence. Recall that the sample correlation coefficient between two Nlength sequences {x~i~} and {y~i~} is defined as ![](1471-2105-6-117-i13.gif) where ![](1471-2105-6-117-i14.gif) denotes the sample mean of {x~i~}. Under the assumption of stationarity, it immediately follows from Equation (9) that the correlation coefficient ρ(m) is related to the autocorrelation function r(m) by ![](1471-2105-6-117-i15.gif), where ![](1471-2105-6-117-i16.gif) is the variance of the sequence. Since it is important to remove the mean of the sequence prior to spectrum estimation to avoid low frequency artifacts and since ![](1471-2105-6-117-i17.gif) is simply a scale factor, the problem of detecting periodic components in a data sequence may equally well be based on ρ(m) as r(m). Consequently, we consider spectral estimators of the form ![](1471-2105-6-117-i18.gif) where ![](1471-2105-6-117-i19.gif) estimates the correlation coefficient between {y~k~} and {y~k+m~} and Lis the maximum lag for which the correlation coefficient is computed. More specifically, we consider the correlation coefficient between the data ranks R~y~(i) and ![](1471-2105-6-117-i20.gif), defined by ![](1471-2105-6-117-i21.gif) where Cis a normalisation factor, R~y~(i) denotes the rank of y~i~in the set S= {y~j~: j∈ I~m~} and ![](1471-2105-6-117-i20.gif) denotes the rank of y~i+m~in the set S\'= {y~j+m~: j∈ I~m~}. By selecting either C= K~m~or C= Nin Equation (12) yields the unbiased or the biased estimate of the correlation coefficient between the rank sequences, respectively. Because both of these rank sequences assume every value from 1 to K~m~precisely once, their average is (K~m~+ 1)/2, independent of the data values {y~n~}, and their sample variance, when scaled by the sequence length K~m~, can be shown to be (![](1471-2105-6-117-i22.gif) - 1)/12. More generally, if C= K~m~, and since ρ^S^(m) is the correlation coefficients between ranks, it is bounded by -1 ≤ ρ^S^(m) ≥ 1 for all m. In the following we shall use L= N- 2 in Equation (11) because N- 2 is the largest lag for which Equation (12) can be computed. We shall use the biased estimate, i.e., C= N, here because of its connection to the equivalence between I(ω) and S(ω). Moreover, the use of the biased estimate in spectrum estimation (Equations (6) and (7)) can be interpreted as triangular weighting of the autocorrelation function estimate. Windowing is usually applied to reduce the scalloping loss effect which is the reason why some frequencies are inferior to others \[[@B25]\]. As in the case of the standard autocorrelation estimate, the values for ρ^S^(m) for m\< 0 are obtained by symmetry: ρ^S^(-m) = ρ^S^(m). This also helps in computing Equation (11) since it reduces to ![](1471-2105-6-117-i23.gif) where ![](1471-2105-6-117-i32.gif)(x) denotes the real part of x. As opposed to standard periodogram or the corresponding correlogram, the proposed robust spectral estimator is not guaranteed to be non-negative. We shall hence use the absolute value of ![](1471-2105-6-117-i24.gif) in the following. ### Significance values In the same way as Wichert et al.\[[@B8]\] do, we use the g-statistic and evaluate ![](1471-2105-6-117-i25.gif) for each time series spectral estimate. However, we do not have the luxury of resorting to an exact distribution of the g-statistic, e.g., under the Gaussian noise assumption. To obtain the significance values we consider two common ways of computing them: simulation and permutation-based methods. This also opens up a possibility for adjusting some parameters in the proposed robust method that were previously kept fixed. In particular, we vary the number of equidistant frequencies at which the spectral estimate is evaluated and change Equation (14) accordingly by incorporating more terms in the max-operator as well as in the sum in the denominator. Instead of having a fixed set of a+ 1 normalised frequencies as in Equation (3), we can evaluate the spectral estimate at \[(K- 1)/2\] + 1 equidistant frequencies ![](1471-2105-6-117-i26.gif) Although the method is rather insensitive to the selection of K, we found that K= 2Ngenerally provides a good performance. Evaluating ![](1471-2105-6-117-i24.gif) at more frequencies can be viewed as a smoothing or interpolation of the original discrete spectral estimate. From the implementation point of view it is worth mentioning that Equation (13) evaluated at frequencies shown in Equation (15) can be computed using the fast Fourier transform (FFT). As was already discussed in the Background Section, in some cases one might be interested in testing fixed instead of unknown frequencies. The proposed method can naturally be adapted to that case as well. If ω\' is the known frequency for which the spectral content is to be tested, then a modified g-statistic, g\', can be used ![](1471-2105-6-117-i27.gif) In the following, we mainly concentrate on the use of the standard g-statistic for detecting unknown frequencies. However, the same methods, such as simulation and permutation based significance values, can also be applied to the modified g-statistic. In Experimental results Section we apply both the standard and the modified g-statistics to real microarray data. When the noise satisfies the i.i.d. assumption, the decision between simulation and permutation test-based significance values is facilitated by the following observation. A statistic Tis said to be distribution-free over a collection of distributions ![](1471-2105-6-117-i28.gif) if the distribution of Tis the same for every joint distribution in ![](1471-2105-6-117-i28.gif). Consider the signal model shown in Equation (1) under the null hypothesis H~0~: β= 0, i.e., y~n~= ε~n~, and assume that ε~n~is again i.i.d. and has continuous distribution. It is easy to see from Equation (12) that, for any time series {y~n~}~n=\ 1,\...,N~, the g-statistic depends on {y~n~}~n=\ 1,\...,N~only through the rank sequence {R~y~(i)}~i=\ 1,\...,N~, where R~y~(i) denotes the rank of y~i~in the original sequence. This implies that the g-statistic is distribution-free over the class of all joint distributions of Ni.i.d. continuous univariate random variables (see, e.g., \[[@B22]\]). In other words, for each Nand independent of the type of the noise, the g-statistic has exactly the same null distribution as long as the noise term satisfies the continuous i.i.d. assumption. Therefore we can choose to use either the simulated distributions (must be simulated separately for each times series sequence length) or the permutation tests, depending on the circumstances. ### Simulation-based significance values The simulation-based method is simple. Given the model as in Equation (1) together with some distributional assumptions for ε~n~, generate a set of Prandom time series under the null hypothesis. Evaluate the test statistic shown in Equation (14) on each of the Ptime series. Use the obtained g-values to compute an estimate of the distribution of the g-statistic under the null hypothesis. The distribution can be estimated, e.g., using kernel density estimation methods. The testing can then be performed as explained above except that the significance values are computed/integrated relative to the estimated distribution. Note that the null distribution must be estimated for each time series length separately but, due to the distribution-free property, the null distribution is independent of the noise characteristics under the i.i.d. assumption. ### Permutation-based significance values A more flexible way of obtaining p-values is to use permutation tests \[[@B23]\]. Although they are a relatively old concept, permutation tests have only recently become interesting in practise because of the intensity of needed computing power. The idea is simple: 1\. Choose a test statistic. 2\. Evaluate the test statistic on the original data. 3\. Randomly permute the data and evaluate the test statistic on every permutation. 4\. Estimate the distribution of the test statistic with the help of the sample generated in point 3. 5\. Use the estimated distribution to get a p-value for the original test statistic computed in point 2. A sequence of random variables {X~n~}, n= 1, 2,\...,Nis exchangeable, if the joint distribution of ![](1471-2105-6-117-i29.gif) is the same as that of the original sequence X~1~, X~2~,\..., X~N~for all permutations π. Under the null hypothesis, the elements of the time series y~n~are i.i.d. and therefore exchangeable, and hence the permutation test can be applied. Alternatively, as the application of a random permutation destroys any periodic structure that is present in the original sequence, permutation tests can be used to assess how highly structured the given time point values are in the light of the chosen test statistic versus other permutations of the given sample. As the concept of permutation tests is non-parametric, they can be applied without knowing the exact distribution of the data at hand. Instead of performing all the N! permutations for each time series, we have chosen to permute each of the original time series for P= 5000 times. As our simulations show, this seems to be quite an adequate number of iterations. The selection of Pis always a compromise, because too high Pmakes computations too slow and too low Pweakens the accuracy and resolution of the calculated p-values. For example, time series having a very periodic structure can get a p-values of zero due to the low value of P. While we have mainly applied the permutation tests to the robust estimator, it must be noted that with the help of permutation tests the robustness of the periodogram can also be improved. As we show in Results section that if we add, e.g., some impulsive noise to the simulated data, the results when using the periodogram method as in Equation (5) are not as good as when we use permutation tests to find the p-values. ### Correction for multiple testing In order to facilitate the comparison between the proposed and previous methods, the obtained p-values are corrected exactly in the same way as in the method by \[[@B8]\]. Simulations ----------- We put the presented methods to a test by first going through simulated data, where the ground truth is known, and then by finding periodically behaving genes in real microarray data. In simulations, we use exactly the same test signal model as in \[[@B8]\] for comparison purposes, namely Equation (1) with β= ![](1471-2105-6-117-i30.gif) and φ= -π/4, i.e., ![](1471-2105-6-117-i31.gif) where n= 1,\...,N, ωis uniformly randomly chosen in the interval ω∈ \[0.05π, 0.45π\] (we wanted to avoid frequencies near zero and the Nyquist frequency) and ε~n~is an i.i.d. noise sequence. An essential parameter is the amplitude (β= ![](1471-2105-6-117-i30.gif)) which affects the signal-to-noise ratio and which we would like to have the same as in \[[@B8]\]. We chose to consider three types of non-idealities, namely (i) pure standard Gaussian noise (zero mean and unit variance, see Figure [1(a)](#F1){ref-type="fig"}), (ii) standard Gaussian and impulsive noise (number of impulses equals ten percent of the sequence length, amplitude ± 6 times the standard deviation of the Gaussian noise, see Figure [1(b)](#F1){ref-type="fig"}), and (iii) standard Gaussian noise and x^3^distortion, where all values were raised to the power of three after adding the noise (see Figure [1(c)](#F1){ref-type="fig"}). In each example sequence in Figure [1](#F1){ref-type="fig"}, the normalised frequency of the original sinusoidal is ω= 0.1. Figures [2(a)--(c)](#F2){ref-type="fig"} show the spectral estimates for the time series in Figures [1(a)--(c)](#F1){ref-type="fig"}, respectively, using both the standard periodogram and the proposed robust method. Note that the spectra have been scaled for viewing purposes. Figures [2(b)--(c)](#F2){ref-type="fig"} already illustrate a remarkable difference between the two methods. For more details about the performance of the proposed robust method as a spectrum estimator, see \[[@B14]\]. A detailed comparison of the periodicity detection capabilities is performed next. Let us first examine the power of the test, i.e., one minus the probability of the type II error (false negative). The power of the test is estimated for the three different test cases as well as for different time series lengths and for different noise parameters using 10000 Monte Carlo runs, see Figure [3](#F3){ref-type="fig"}. The significance level is set to α= 0.05. In all the three cases, the case-specific noise assumptions are used for both the null hypothesis (β= 0) and the alternative hypothesis (β\> 0). In this simulation, we use the signal model shown in Equation (17) to represent a periodic signal (i.e., the alternative hypothesis). In the right column of Figure [3](#F3){ref-type="fig"}, the length of the time series is set to 40 and the power is shown as the function of varying noise parameters. Figure [3](#F3){ref-type="fig"} clearly shows that the power of the proposed robust hypothesis testing method is remarkably better than that of the Fisher\'s test, especially in the case of outliers and non-linear distortion. More interestingly, however, the proposed method is also more powerful in the case of standard Gaussian noise. Next we consider another simulation. In the same way as in \[[@B8]\], two thousand time series of length N= 10, 20, 40, 45, 50 and 100 were generated to test the periodicity detection. One thousand and nine hundred of the time series were plain noise and one hundred time series were generated according to Equation (17). We again consider the three aforementioned noise models. As explained in the Computational Methods Subsection, we evaluated the g-statistic and p-value for each time series and then used the FDR rule to determine which of the time series were considered to be cyclic for a certain FDR level. The FDR level, at which the expected rate of false positives is controlled, was chosen similarly as in \[[@B8]\], i.e., q =0.15, 0.10, 0.05, 0.01 and 0.005. For each Nand qthe simulation was run for 99 times for the simulation-based cases and 9 times for the permutation-based cases. Median statistics are reported for the number of found periodic components, the number of correctly identified periodic components (shown in parenthesis) and the number of truly periodic time series among the top 100 ranked sequences (Z). If we take a look at the results in Tables [1](#T1){ref-type="table"} to 9 we can draw some immediate conclusions. First, when the noise is plain Gaussian, Tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"} show that both methods perform approximately equally well. There are no significant differences between the two methods in terms of the number of detected genes or in terms of the number of correctly detected genes. However, the numbers of truly periodic genes among the top 100 ranked sequences (Z-scores) show somewhat favorable performance for the robust method, especially for the short time series N= 20 and N= 40. Indeed, this observation agrees with previous findings \[[@B14]\] where the robust method was found to have a good performance as a spectrum estimator for short time series. By comparing Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"} and Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}, it is obvious that the permutation tests do not provide any significant performance gain over the traditional approach where the significance values are computed using the simulation-based method or Equation (5), respectively. In both cases, the Z-scores are about the same, as expected. The only notable difference is seen in the number of found periodic genes for short time series (e.g., N= 40, 45, 50) and small FDR levels (q =0.005, 0.01, 0.05) where the numbers are slightly higher when permutation tests are used. This suggests that the permutation-based method finds a bit smaller p-values than the simulation-based method. Tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}, [6](#T6){ref-type="table"}, [7](#T7){ref-type="table"}, [8](#T8){ref-type="table"} clearly show the superior robustness of the proposed method over the traditional Gaussian analysis. As can be seen from Tables [1](#T1){ref-type="table"} and [5](#T5){ref-type="table"} and Tables [2](#T2){ref-type="table"} and [6](#T6){ref-type="table"}, there is only a minor performance degradation between the Gaussian case and the combined Gaussian and impulsive case. On the other hand, Tables [3](#T3){ref-type="table"} and [7](#T7){ref-type="table"} and Tables [4](#T4){ref-type="table"} and [8](#T8){ref-type="table"} clearly indicate the sensitivity of the periodogram method to fluctuations from the original Gaussian noise assumption. As discussed above, Tables [5](#T5){ref-type="table"} and [6](#T6){ref-type="table"} show that, in the case of the robust method, permutation-based significance value computation performs approximately equally well as the simulation-based computation. The only notable difference is again seen in the number of found periodic genes for short time series (e.g., N= 40, 45, 50) and small FDR levels (q =0.005, 0.01, 0.05). Tables [7](#T7){ref-type="table"} and [8](#T8){ref-type="table"} in turn show that, apart from the Z-scores, the permutation-based method mitigates the sensitivity of periodogram method to the fluctuations from the model (Gaussian) distribution. The robustness of the proposed method is further demonstrated by its response to combined Gaussian noise and nonlinear cubic distortion. As explained in Computational Methods Section, the robust method depends on the observed time series only through the rank sequence. Any monotone distortion preserves the ordering of the samples. Therefore, the rank-based method is completely insensitive to any monotone distortions. Consequently, the results for the third test case are identical to those presented in Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}. The results for the periodogram method are shown in Table [9](#T9){ref-type="table"}. Experimental results -------------------- The data sets that are considered here are from the following papers: \[[@B6],[@B5],[@B7]\]. For each time series experiment (13 in total), we apply the proposed robust methods for detecting genes having both fixed and unknown frequency components. For the fixed frequency we use the one that corresponds to the length of the cell cycle. Following the idea presented in \[[@B8]\], a simple method for estimating the cell cycle length/frequency is to compute the average robust spectral estimate. For each time series, we present the number of statistically significant genes that are found to be periodically behaving at a specific level of the FDR (q =0.05). For the cdc15 experiment by Spellman et al.\[[@B6]\], the sampling time was not equidistant in the beginning and at the end of the data set. Considering the missing time points as missing values would result in a large number of missing values with a regular pattern of occurrence. Although the proposed robust methods can cope with missing values, such a regular pattern of missing values can artificially cause many small significance values and hence result in an unreliably large number of statistically significant periodic genes. This can be avoided e.g. by interpolating the expression values for the systematically missing time points, in which we used simple linear interpolation. Only non-missing expression values are considered in the interpolation. If the expression values of both the previous and the next time instants are missing, then the interpolated sample is defined to be missing as well. This results in a more conservative number of detected genes for the cdc15 experiment. We chose to consider only those genes that have less than 30% missing values and decided to rule out all except the Score3 experiment in the data by Whitfield et al.\[[@B5]\] because of high degree of irregular sampling and short time series length. The obtained results are shown in Table [10](#T10){ref-type="table"}. The total number of genes analysed in each data set is denoted as M. The number of found periodic genes having fixed and unknown frequency are denoted as P\'and P, respectively. The corresponding figures from \[[@B8]\] are shown in parentheses. For the detection of periodic components having unknown frequency, we used the permutation-based method. As was shown in Simulations Section, both the simulation and permutation based approaches performed approximately equally well. Hence, for the ease of implementation, we used the simulation-based method for the detection of periodic components having a fixed frequency. If we first take a look at the numbers of detected genes having a periodic component of an unknown frequency (P) shown in Table [10](#T10){ref-type="table"}, we can see that generally the numbers of periodically behaving genes are lower than those of the corresponding figures in \[[@B8]\]. In other words, the proposed robust method seems to provide more conservative estimates, although the cdc15 experiments shows an exception. In the case of real data further comparison between the two methods is much more subjective than in simulation experiments since the ground truth is not completely known. Based on the simulation results shown above, one could put more faith on the robust method, especially in the cases where the Gaussian noise assumption is violated. Hence one could argue that the robust method has ignored more non-periodic time series, particularly ones where outliers and other non-idealities have caused artificial variation in the periodogram. The number of detected truly periodic genes can be increased, at the cost of detecting more false positives, by using a higher level of q. Let us then focus on the numbers of found periodic genes when using a fixed frequency in the robust method (P\'). Concerning the numbers of detected periodic genes in the data sets by Spellman et al.\[[@B6]\], the results are in concordance with the previously published ones \[[@B24]\]. On the cdc28 data set, the proposed method finds a slightly higher number of periodic genes. Direct comparison between the numbers Pand P\'is not meaningful as the number of detected genes depends on the significance level, which may be dependent on the used method. The comparison is better done using the ordered gene lists which we will discuss shortly together with three benchmark gene sets (see below). We have not yet come across with another study, besides the original paper, that would have examined the periodicity in the data by Rustici et al.\[[@B7]\]. Rustici et al.investigate the global cell cycle control of gene expression in the fission yeast Schizosaccharomyces pombeusing DNA microarrays. Thus, the comparison of values in Table [10](#T10){ref-type="table"}, related to Rustici et al.data, is not feasible. Table [10](#T10){ref-type="table"} shows that the number of detected periodic genes ranges from 673 up to 3131. In the case of very large number of detected periodic genes, more insight into the data set can be gained by looking at the ordered list of genes. Table [10](#T10){ref-type="table"} only shows the number of detected genes. Further insight can be gained by looking at the enrichment of the genes assumed to be cell cycle regulated among the top ranked genes. In particular, we resort to the three different benchmark gene sets introduced in \[[@B24]\]. In order to provide a direct comparison with the results shown in \[[@B24]\], we show similar enrichment graphs for the both robust detection methods (fixed and unknown frequency) in Figure [4](#F4){ref-type="fig"}. Note that some of the benchmark genes are ignored during the analysis since they have more than 30% missing values. There are also a few benchmark genes for which no exact match was found among the genes in the data provided by \[[@B6]\]. Hence the graphs are drawn based on slightly smaller benchmark gene sets, namely, 101--112 (B1), 309--339 (B2), and 465--505 (B3), depending on the experiment. Figure [4](#F4){ref-type="fig"} reveals some interesting results. First, let us compare how well the proposed robust method (with the fixed frequency) finds the benchmark genes when compared to previously published methods (for comparison, see the corresponding graphs in \[[@B24]\]). The results for the benchmark gene set B1 are shown on the first row in Figure [4](#F4){ref-type="fig"}. For all the data sets, the performance of the robust method is either between the best methods and the amplitude-independent method by Zhao et al.\[[@B9]\], or close to the method by Zhao et al.\[[@B9]\]. This finding is not surprising. As discussed in \[[@B24]\], this benchmark gene set is biased towards periodic genes which are strongly regulated, i.e., have large amplitudes. In general, because high-amplitude periodic genes are more easily detected from noisy expression data, the gene sets identified from such studies are likely to be biased towards high amplitude genes. An advantage of amplitude-independent methods is, however, that they detect small-amplitude periodic transcripts better, and hence may identify genes which are not yet known to be periodic. For the benchmark gene set B2, the performance of the proposed robust method is approximately the same as that of the best methods reported in \[[@B24]\], except for the alpha experiments on which the robust method performs slightly worse. Noteworthy is that the benchmark gene set B2 is obtained from a separate Chromatin IP experiment and thus is independent of the previous gene expression studies. Concerning the benchmark gene set B3, the robust method performs better than the majority of the methods. Notably good performance is seen on the data from the Cdcl5 and the Cdc28 experiments. Interestingly, the benchmark set B3 is also likely to be biased, but towards small amplitude genes \[[@B24]\]. This strengthens the assumption that the potential of the proposed robust method is especially in detecting unknown, small-amplitude, periodic genes. Yet another interesting observation can be drawn by comparing the solid and dashed curves in Figure [4](#F4){ref-type="fig"}. As can be expected, the method which detects especially cell cycle frequencies ranks the benchmark genes higher than the method which detects unknown (all) frequencies, i.e., the solid line is above the dashed line. Another expected behavior is that the method which detects unknown frequencies also detected a great number of genes assumed to be related to the cell cycle. However, from another point of view, Figure [4](#F4){ref-type="fig"} also indicates that there are some statistically significant periodic patterns which are more significant than some of the cell cycle related ones. Possible sources of those significant periodic patterns may include, among others, systematic artifacts in the array/experiment preparation, unknown periodic biological processes, or simply the considerable amount of experimental noise (false positives). The top 300 ranked genes for all the data sets analysed, obtained using the proposed robust method, are provided on our companion website. Discussion ========== As discussed above, some extensions and improvements over the traditional periodogram/correlogram approach have been proposed in the literature. Two particular modifications, namely utilisation of windowing and a trimmed g-statistic, were reported to provide a good performance in a recent review by Artis et al.\[[@B20]\]. Although we provided an extensive comparison of the unmodified traditional and the (unmodified) proposed rank-based methodologies, the further modifications can be implemented in a straightforward fashion in both of the frameworks. These extensions for the robust rank-based method will be examined in future studies but let us give an overview of the possible modifications. As discussed previously, the biased version of the robust correlation estimator can be viewed as a type of weighting or windowing. More generally the windowing is typically incorporated into the computation of the spectral estimate (see, e.g., Equations (6) and (11)). Different windows provide different properties for spectral estimators. For example, the shape of the window can be used to control the smearing and leakage effects whereas the length of the window compromises with the spectral resolution and the variance \[[@B25]\]. In general, the used windows can be chosen from a general class of windows, including, among others, Bartlett, Daniel, and Parzen windows (see, e.g. \[[@B16],[@B18],[@B19]\]). Concerning the detection of hidden periodicities, windowing can be applied much in the same way as it is used in the method by Priestley. Similarly, the modification by Chiu \[[@B21]\], i.e., the use of a proper trimmed mean of the ranked periodogram ordinates in place of the average periodogram, can be applied to the robust rank-based estimator as well. A drawback associated with the use of the trimmed g-statistic in the traditional periodogram setting is that only asymptotic distribution of the test statistic is available. The discrepancy between the true distribution and the asymptotic one can be remarkable in the case of small sample size typical e.g. in gene expression studies. These difficulties can be circumvented by the computer intensive simulation and permutation-based methods explained above. The proposed method has other possible extensions as well. As a periodically behaving gene may be involved in several different biological processes, its expression pattern may contain several dominant frequencies. In that regard, the testing procedure can be extended to detect several frequency peaks from the spectral estimate. See, e.g., \[[@B19],[@B21]\] for extensions of Fisher\'s test to that direction. In cell cycle related studies, a cell population is usually forced to synchrony prior to taking the measurements using an external synchronisation method. The synchronisation is achieved by arresting the cells at a specific phase of the cell cycle after which they are released. However, as time evolves, the cell population gradually loses its synchrony. Such a phenomenon can be viewed as time-varying (low-pass) filtering of the expression values where the time-varying filter kernel corresponds to the distribution of the cell population over the cell cycle. Inverse methods have been developed to correct for the effect of the loss of synchrony \[[@B26],[@B27]\]. Several interesting questions remain to be studied. First, the inverse filtering problem as such is fairly sensitive to noise and is further complicated by the fact that the accuracy level at which the filter kernel (i.e., distribution of the cell population) can be measured is limited. Therefore, the corrected time series may contain even more obscure non-idealities than the uncorrected ones. Consequently, robust methods are potentially even more important when periodic components are sought from the time series which are corrected for the loss of synchrony. Future studies are also needed to compare the robust periodicity detection method, when applied to both uncorrected and corrected time series, to see whether the inversion of the loss of synchrony brings any additional gain in the case of robust periodicity detection. In addition to the simulation results presented in the Simulations Section, we also performed preliminary simulations where the amplitude of the periodic signal was attenuated to model the loss of synchronisation. We noticed that if the average amplitude of the sinusoidal signal remained the same, the results were similar to those in the tables of the Simulations Section. Further comparisons must also be made to assess the performance differences between the proposed method (possibly combined with a proper inversion method for the loss of synchrony) and alternative methods in which a model for the loss of synchrony is incorporated into the statistical testing framework \[[@B12],[@B11]\]. Although elegant, such combined approaches have potential difficulties in that they usually result in a computationally intensive optimisation problem \[[@B11]\] and/or include several distributional assumptions \[[@B12]\]. Furthermore, the inversion of the loss of synchrony is performed blindly, i.e., without any additional measurements, which the distribution of the cell population could be estimated with. Future experiments are needed to address these questions. Conclusion ========== The presented method yields a robust way of finding periodicity in short time series data. As illustrated in Simulations Section, the proposed robust detection method is remarkably insensitive to different kinds of non-idealities in the data, such as heavy contamination of outliers, missing values, short time series, nonlinear distortions, and is completely insensitive to any monotone nonlinear distortions. The results also show that the proposed method has clearly better performance than the Fisher\'s test, even in the case of the standard Gaussian noise. Furthermore, the results on real data demonstrate that the proposed method performs well on real data and that the results are biologically meaningful. As illustrated in Figures [2(a)--(c)](#F2){ref-type="fig"} and more extensively reported in \[[@B14]\], the robust method serves also as a good spectral estimator. As the time series measured from biological systems are usually short and prone to contain different kinds of non-idealities, we believe that the robust detection method presented in this paper will find many important applications in this field. Authors\' contributions ======================= MA carried out an implementation of the methods, performed the computations and co-drafted the manuscript. HL developed the statistical methods, helped in computations and mainly drafted the manuscript. RP helped in developing the statistical methods. OY-H conceived of the study and participated in its design and coordination. HH helped in the implementation of the computational methods. All authors read and approved the final manuscript. Acknowledgements ================ The support of Tampere Graduate School in Information Science and Engineering (TISE) and the Academy of Finland are gratefully acknowledged. We are indebted to Dr. Korbinian Strimmer for providing an implementation of their methods, stimulating discussions, and useful suggestions. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Examples of time series. An example of a time series composed of a sine and (a) additive standard Gaussian noise, (b) additive standard Gaussian and impulsive noise, and (c) additive standard Gaussian noise and cubic distortion. ::: ![](1471-2105-6-117-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Examples of spectral estimates. The spectral estimates for the time series in Figures 1 (a)-(c), respectively, using both the standard periodogram and the proposed robust method. ::: ![](1471-2105-6-117-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Power of the test. The power of the tests (y-axis) for the three different test cases as the function of the time series length and varying noise parameters (x-axis). The solid (resp. dashed) line corresponds to the proposed robust method (resp. Fisher\'s test). Three different types of non-idealities are considered, namely, pure standard Gaussian noise (the first row), standard Gaussian and impulsive noise (the second row), and standard Gaussian noise and x^3^distortion (the third row). The left (resp. right) column shows the results for different time series lengths (resp. different values of the noise parameters). ::: ![](1471-2105-6-117-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Benchmark results. The fraction of the benchmark set that is identified (y-axis) as the function of the highest ranked genes (x-axis). The solid (resp. dashed) line corresponds the robust detection having fixed (resp. unknown) frequency. The dotted line shows the performance of the random gene selection. The columns from left to right correspond to the Alpha, the Cdcl5 and the Cdc28 experiment by \[6\]. The rows, from top to bottom, correspond to the three different benchmark gene sets B1, B2 and B3. for more details about the benchmark gene sets, see \[24\]. ::: ![](1471-2105-6-117-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Number of inferred periodic time series: case 1. The number of inferred periodic time series using the robust method and standard Gaussian noise in the data. p-values were obtained by simulating the distribution of the g-statistic using 10000 time series composed of Gaussian noise. ::: q\\N 10 20 40 45 50 100 -------- ---- ------ --------- --------- --------- ---------- 0.15 0 2(1) 107(90) 109(96) 117(99) 115(100) 0.10 0 1(1) 96(87) 103(94) 110(98) 109(100) 0.05 0 1(0) 83(79) 95(89) 101(96) 105(100) 0.01 0 1(0) 59(59) 80(79) 90(89) 101(100) 0.005 0 0(0) 32(32) 62(61) 64(64) 100(100) Z 12 49 89 93 95 100 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Number of inferred periodic time series: case 2. The number of inferred periodic time series using the robust method and standard Gaussian noise in the data. p-values were obtained by using permutation tests. ::: q\\N 10 20 40 45 50 100 -------- ---- ------ --------- --------- --------- ---------- 0.15 0 4(3) 108(92) 113(96) 111(98) 119(100) 0.10 0 1(1) 99(90) 106(94) 106(97) 112(100) 0.05 0 1(0) 88(84) 97(89) 101(95) 106(100) 0.01 0 0 65(64) 80(78) 86(86) 101(100) 0.005 0 0 46(46) 61(61) 71(71) 100(100) Z 15 48 91 92 95 100 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Number of inferred periodic time series: case 3. The number of inferred periodic time series using the periodogram method and standard Gaussian noise in the data. p-values were obtained by using Equation 5. ::: q\\N 10 20 40 45 50 100 -------- ------ -------- --------- --------- --------- ---------- 0.15 1(0) 17(11) 111(89) 108(86) 117(97) 122(100) 0.10 0 9(6) 100(85) 98(80) 108(96) 115(100) 0.05 0 2(1) 83(78) 79(71) 99(93) 107(100) 0.01 0 0 56(55) 54(52) 84(83) 101(100) 0.005 0 0 23(23) 20(20) 54(54) 99(99) Z 13 39 85 83 93 99 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Number of inferred periodic time series: case 4. The number of inferred periodic time series using the periodogram method and standard Gaussian noise in the data. p-values were obtained by using permutation tests. ::: q\\N 10 20 40 45 50 100 -------- ---- ------ --------- --------- --------- ---------- 0.15 0 2(0) 109(94) 111(98) 114(98) 117(100) 0.10 0 2(0) 98(91) 103(96) 107(98) 112(100) 0.05 0 2(0) 87(84) 98(95) 103(98) 105(100) 0.01 0 2(0) 68(67) 88(86) 91(80) 102(100) 0.005 0 1(0) 55(55) 70(70) 80(79) 100(100) Z 14 40 92 94 97 100 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Number of inferred periodic time series: case 5. The number of inferred periodic time series using the robust method and standard Gaussian plus impulsive noise in the data. p-values were obtained by simulating the distribution of the g-statistic using 10000 time series composed of Gaussian noise. ::: q\\N 10 20 40 45 50 100 -------- ---- ------ -------- -------- --------- ---------- 0.15 0 1(0) 73(62) 84(73) 101(84) 114(100) 0.10 0 1(0) 60(55) 76(69) 91(81) 109(100) 0.05 0 1(0) 48(46) 58(56) 77(73) 105(100) 0.01 0 0 22(21) 21(21) 49(49) 100(99) 0.005 0 0 13(13) 17(17) 34(34) 95(95) Z 9 35 73 78 84 99 ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Number of inferred periodic time series: case 6. The number of inferred periodic time series using the robust method and standard Gaussian plus impulsive noise in the data. p-values were obtained by using permutation tests. ::: q\\N 10 20 40 45 50 100 -------- ---- ------ -------- -------- -------- ---------- 0.15 0 1(0) 70(57) 83(72) 93(82) 117(100) 0.10 0 0 59(51) 71(68) 89(79) 111(100) 0.05 0 0 48(45) 60(58) 77(72) 105(99) 0.01 0 0 24(23) 35(33) 54(53) 99(98) 0.005 0 0 22(21) 27(27) 41(41) 95(95) Z 9 38 72 78 82 99 ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Number of inferred periodic time series: case 7. The number of inferred periodic time series using the periodogram method and standard Gaussian plus impulsive noise in the data. p-values were obtained by using Equation 5. ::: q\\N 10 20 40 45 50 100 -------- ---- ---- -------- -------- -------- -------- 0.15 0 0 16(16) 24(22) 28(28) 77(71) 0.10 0 0 13(12) 20(12) 25(24) 73(69) 0.05 0 0 11(11) 13(13) 17(17) 65(63) 0.01 0 0 3(3) 6(6) 9(9) 52(52) 0.005 0 0 0(0) 1(1) 3(3) 40(40) Z 11 29 44 46 52 79 ::: ::: {#T8 .table-wrap} Table 8 ::: {.caption} ###### Number of inferred periodic time series: case 8. The number of inferred periodic time series using the periodogram method and standard Gaussian plus impulsive noise in the data. p-values were obtained by using permutation tests. ::: q\\N 10 20 40 45 50 100 -------- ---- ------ -------- -------- -------- -------- 0.15 0 6(5) 38(29) 40(33) 18(16) 82(76) 0.10 0 2(2) 34(26) 36(32) 14(13) 76(72) 0.05 0 1(1) 23(22) 32(29) 10(9) 69(69) 0.01 0 0 16(15) 18(18) 7(6) 61(61) 0.005 0 0 15(14) 17(17) 7(6) 53(53) Z 11 32 47 49 44 80 ::: ::: {#T9 .table-wrap} Table 9 ::: {.caption} ###### Number of inferred periodic time series: case 9. The number of inferred periodic time series using the periodogram method and standard Gaussian noise and cubic distortion in the data. p-values were obtained by using permutation tests. ::: q\\N 10 20 40 45 50 100 -------- ---- ---- -------- -------- -------- --------- 0.15 0 0 49(44) 79(64) 89(74) 107(93) 0.10 0 0 39(36) 71(62) 80(69) 98(90) 0.05 0 0 25(24) 52(49) 64(59) 90(88) 0.01 0 0 8(8) 28(28) 44(43) 82(82) 0.005 0 0 8(8) 19(19) 37(36) 67(67) Z 7 15 68 71 79 91 ::: ::: {#T10 .table-wrap} Table 10 ::: {.caption} ###### Number of inferred periodic genes from real microarray data. Results obtained by using real microarray data are presented here. Permutation tests were used to obtain significance values for the robust spectra. Symbols: Nis the length of the time series, Mis the number of genes analysed, P(resp. P\') is the number of found periodic genes having an unknown frequency (resp. frequency corresponding to the cell cycle length). Notes in the table: ^a^Since the elutriation time course did not show any significant periodic components, we did not perform the test with a fixed frequency. ^b^The average spectrum showed several major peaks. The one in the vicinity of the assumed cell cycle frequency was chosen manually. ::: Cell type Experiment N M P\' P\'/M P P/M Source -------------- ----------------- ----- ------- --------- ---------- ------------ ------- -------- S.cerevisiae cdc15 24 5287 981 18.6 946(766) 17.9 \[6\] S.cerevisiae cdc28 17 6103 363 6.0 32(105) 0.5 S.cerevisiae alpha 18 6056 346 5.7 139(468) 2.3 S.cerevisiae elutriation 14 6074 \_a \_a 4(193) 0.07 Human HeLa Score3 48 41508 1285^b^ 3.1^b^ 3580(6043) 9.0 \[5\] S.pombe cdc25-1 19 4373 1431 32.7 759 17.4 \[7\] S.pombe cdc25-2 36 4422 2605 58.9 2197 49.7 S.pombe cdc25-sep1 20 4700 2624 55.8 2295 48.8 S.pombe elutriation1 20 4229 1948 46.1 551 13.0 S.pombe elutriation2 20 3961 1453 36.7 384 9.7 S.pombe elutriation3 20 4236 673 15.9 355 8.4 S.pombe elutr.-cdc10-br 22 4647 3131 67.4 2431 52.3 S.pombe elutr.-cdc25-br 21 4272 2405 56.3 767 18.0 :::	PubMed Central	2024-06-05T03:55:58.204502	2005-5-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168888/", "journal": "BMC Bioinformatics. 2005 May 13; 6:117", "authors": [ { "first": "Miika", "last": "Ahdesmäki" }, { "first": "Harri", "last": "Lähdesmäki" }, { "first": "Ron", "last": "Pearson" }, { "first": "Heikki", "last": "Huttunen" }, { "first": "Olli", "last": "Yli-Harja" } ] }
PMC1168889	Background ========== The introduction of the Expectation value (E-value) in the Blast program in 1990 \[[@B1]\] was a major milestone in the development of sequence search algorithms. For any sequence match with a score S obtained in a given database, the E-value is the number of hits of same or higher score that can be expected by chance. E-values tell biologists how likely they are to encounter a specific sequence match in a database search, which is a more useful view of biological significance than a mere similarity score. Except in some special cases, low E-values are commonly accepted as an evidence of sequence homology. The recent years have seen a growing interest for RNA motif searches, driven by the discovery of important new classes of regulatory RNA genes and motifs in all organisms. Non-coding RNAs are characterized in a large part by long range base pair interactions, whereas linear sequence constraints are not as important as in protein coding genes. As a result, standard sequence alignment programs such as Blast are not suited to RNA motif search. Computational biologists have addressed this problem in several ways, notably through descriptor-based systems in which the topology of base paired regions is user-specified \[[@B3],[@B5],[@B4]\], Stochastic Context Free Grammars (SCFG) which use a complete statistical model of RNA elements \[[@B6]\], and Secondary Structure Profiles, which use position weight matrices describing stems and single strands in the RNA motif, as found in the ERPIN program \[[@B8]\]. Although the last two methods compute alignment scores, the complexity of the underlying models and search algorithms has hampered the estimation of an E-value to date. When the behavior of the score distribution is known, such as in sequence alignment scores, E-values can be computed either directly, or by empirically fitting a histogram of scores from a sample of random sequences to the assumed distribution function \[[@B1],[@B2]\]. Unfortunately, we will show here that a search algorithm such as the one used in ERPIN does not produce predictable score distributions. A possible workaround for this practical limitation is to run simulations on randomized sequences and use the observed hit count at score Sas the E-value for this score. However, since interesting high scoring solutions can be extremely rare (commonly less than one random occurrence per genome) simulations often require days of calculations. Can we then estimate score distributions without having to run such lengthy simulations? In this article, we show that ERPIN score distributions can be estimated a priori through a discrete convolution analysis of score profiles, based on a random model of nucleotide frequencies. This led us to develop a computational procedure that estimates the score distribution of ERPIN profiles in a very short time, before the actual search begins, so that each solution can be automatically assigned an E-value. ERPIN profiles ============== ERPIN is an RNA motif search software using as input a training set of aligned RNA sequences, and a target database in which the motif is to be identified. The training set contains both RNA sequence and their common secondary structure, specified as shown in Figure [1](#F1){ref-type="fig"} in the form of single strands and helices. Importantly, gap characters (insertions or deletions) are allowed in single strands but not in helices. Helices are composed of two distinct strands of equal length. A region is defined as a continuous stretch of complete single strands or helical elements. When only one strand of a helix is included in a region, this strand is considered as a single strand. A mask is a subset of a region constituted of single strands and/or complete helices. ERPIN converts each helix and single strand in the alignment into a lod-score profile. This involves two steps. First, columns in the alignment are converted into frequency profiles, recording the frequencies of bases or base-pairs in column cas: ![](1471-2105-6-118-i1.gif) Here n~ic~is the number of bases or base-pairs of type iin column c, and Nis the total base or base-pair count in a column. There is one frequency profile for each single strand or helix in the alignment. In the case of a single strand, i∈ {A, T, G, C}, whereas for a helix, i∈ {AA, AT, AG, AC, \..., CC} and each column in a helical profile actually refers to two positions in the initial alignment. Single strand profiles thus have 4 rows while helix profiles have 16 rows. The special case of gap-containing single strands, where a fifth character is added to the profile is discussed later on. Frequency profiles are then converted into lod-score profiles, where values for column care defined as: ![](1471-2105-6-118-i2.gif) where b~i~is the background frequency of base or base-pair iin the target database. Base-pair frequencies are considered as the product of individual base frequencies. The score of a helix or single-strand element is obtained by presenting a target sequence to this element\'s profile and summing the scores obtained for every profile column. For ungapped elements, the calculation is straightforward. For gap-containing single strands, a dynamic programming matrix of the profile and target sequence is constructed, which provides the best possible alignment score \[[@B8]\]. In a first stage, let us ignore this alignment procedure and focus on ungapped elements. Exclusions and pseudocounts --------------------------- We define as an exclusion a profile element for which no base or base pair is observed in the training set, and thus P~ic~= 0. Exclusions may be due either to an unsufficient size of the training set, or to a true avoidance of this particular base or base-pair at this position in the RNA molecule. In any case, the log-odd ratio formula would produce a value of -∞ for such cases, thus requiring a special treatment. Exclusions are dealt with either by using arbitrary low values (e.g. -30) or by introducing pseudocounts in the frequency matrix that simulate what could have been observed in a larger training set. Pseudocounts are based on some prior knowledge of \"typical\" substitution frequencies in RNA molecules, as observed in a model RNA sequence alignment. The pseudocount calculation procedure used in ERPIN is the same spirit as that of Henikoff and Henikoff \[[@B9]\], but we use a different definition of pseudocounts, as explained below. Let us first reformulate the values in any column cof a frequency profile: ![](1471-2105-6-118-i3.gif) Where n~ij~is the number of {i, j} couples in column c, P(i, j) is the joint probability of finding iand jin this column, and ![](1471-2105-6-118-i4.gif). This develops into: ![](1471-2105-6-118-i5.gif) Where P(i\\|j) is the conditional probability of observing i, knowing that jis observed in the column. This conditional probability amounts to the observed frequency of i→ jsubstitutions. To introduce pseudocounts in ERPIN frequency profiles, P(i\\|j) is replaced with the average substitution frequencies observed in a model RNA sequence alignment, expressed in the form of a substitution matrix M. Pseudocount-based frequencies can be expressed as: ![](1471-2105-6-118-i6.gif) where Mis a square matrix whose columns jare normalized, ∑~i~M~ij~= 1, so that ![](1471-2105-6-118-i7.gif). See Methods section for construction of M. The relative ratio of pseudocounts to true counts in the final frequency matrix is then controlled by a user-defined weight parameter α∈ \[0, 1\], such that: ![](1471-2105-6-118-i8.gif) Since M~ij~are generally ≠ 0 in the substitution matrices (either for single strand or helices), most exclusions in the frequency profile are replaced by nonzero values as soon as α≠ 0. Not only the resulting lod-score profiles are basically devoid of arbitrary low values, but they better reflect \"natural\" base and base-pair substitution frequencies observed in real RNA alignments. This is especially interesting in helical regions, since the substitution matrix for helices represent natural exchanges between frequent base-pairs such as Watson-Crick, G:U or even G:A, while incurring strong penalties for exchanges involving rare base pairs. This maintains a large fraction of strongly negative values in helix profiles, which is desirable for the sake of search specificity. Shapes of score distributions: finite and non-finite scores =========================================================== We define as a \"finite\" score the score obtained for a sequence that does not contain any match to a profile exclusion. When pseudocounts are used, almost all scores are finite, but when pseudocounts are not in use (α= 0), many scores, especially in helix profiles, are \"non-finite\", although in practice they are replaced by arbitrary low values. Let Sdenote a finite score obtained at a given site in a random sequence. For any x\> -∞, the conditional probability formula reads: P(S\> x) = P(S\> -∞).P(S\> x\\|S\> -∞) (7) In this decomposition, it is noteworthy that: • The first factor is independent of xas soon as xis finite, • The second factor can be computed based on profile elements that contain no exclusion. Let S~c~(c= 1, 2, .., w) denote a score for column cof a profile, and S= S~1~+ S~2~+ \... + S~w~the score obtained by presenting a given sequence to this profile. If wis large enough (w≳ 10) and the distributions of random variables S~c~are (i) independent and (ii) identically distributed, the sum Sfollows a normal distribution (central limit theorem \[[@B11]\]). Due to exclusions arising with different frequencies in different columns, condition (ii) is generally not fulfilled, but, using the decomposition given by formula (7) rewritten for column c, we can write: P(S~c~\> x) = P~fs,c~.P~f,c~(x) (8) where P~fs~is the probability that a score is finite, and P~f~(x) is the probability that a finite score is higher than x. This gives, for a complete profile: ![](1471-2105-6-118-i9.gif) In many cases, the probability distributions of finite scores P~f,c~are similar enough so that their sum Sis normally distributed. If this behavior was always observed, the score distribution could be fully determined by computing the mean value μand the standard deviation σthat characterize the normal law, which would enable a direct calculation of the E-value \[[@B2]\]: ![](1471-2105-6-118-i10.gif) Figure [2](#F2){ref-type="fig"} shows the score distributions obtained using the tRNA region spanning the anticodon and TΨCloop (two helices + three single-strands) at each position of a 100 Mb random sequence database, with pseudocounts switched off (α= 0). As expected, finite scores (Fig [2a](#F2){ref-type="fig"}) follow a normal distribution, while total scores (Fig [2b](#F2){ref-type="fig"}) are unevenly distributed. So called \"non-finite\" scores may be biologically relevant, since many valid substitutions are potentially absent from the training set. However, non-finite scores are detected only when high enough to fall into the extreme end of the distribution. This part of the distribution is composed mainly of finite scores and should thus behave like that of finite scores. Unfortunately finite scores may also deviate from a Gaussian distribution, for instance when score distributions in successive columns are too different from each other. Pseudocounts are used to inject missing substitutions into frequency profiles, resulting in most \"non-finite\" scores becoming \"finite\". But what is the behavior of finite scores when pseudocounts are in use? Fig [3a--d](#F3){ref-type="fig"} and [4a--b](#F4){ref-type="fig"} show finite score distributions for a variety of ungapped RNA motifs (shaded bars), obtained using a typical level of pseudocounts (α= 2.10^-4^) in frequency profiles. Although some training sets (tRNA, SECIS) have nearly gaussian distributions, others (let-7 miRNA, snoRNA, polyA sites) are more erratic. This is due to the larger number of exclusions in the later sets -- only partially compensated for by pseudocounts -- and/or their non-uniform distribution over profile columns. If we aim to address true biological problems with such imperfect or sparse training sets, we necessarily have to deal with this type of score distribution that cannot be approached with classical methods. Nonetheless, we will still be using the decomposition formula (7), as it provides an important reduction of the range of values of the random variables involved. Score distributions of helices and single-strands ================================================= Ungapped helices and single strands ----------------------------------- How can we estimate score distributions such as those in Fig [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"} ? Let us admit that profile columns are independent. After matching the profile to a purely random sequence, the resulting scores for each column would thus behave as independent random variables, say X~1~and X~2~for columns 1 and 2. Therefore, the final score for two columns would be: S= X~1~+ X~2~ Then, the probability of obtaining a score S= xis: ![](1471-2105-6-118-i11.gif) The last formula defines the discrete convolution product\[[@B11]\] of two distributions. The overall score distribution can be obtained by doing the calculation for every possible values of uand v. Using the separation formula (7) between \"exclusions\" and \"finite scores\", equation (10) can be written: ![](1471-2105-6-118-i12.gif) These operations can easily be extended to Ncolumns by iterating the products on successive columns in the same single-strand or helix profile. At each successive iteration, scores are discretized on a predefined grid so that the number of possible scores increases linearly with the number of columns (see Methods section for algorithm). We performed such an analysis on a variety of helix and single strand profiles, with grid intervals set at Δx= .05. In Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}, score distributions estimated from discrete convolution (solid lines) are compared to scores obtained through simulation on a random database of variable size (shaded bars). There is a very good agreement between the discrete convolution and simulation. Gap-containing single strands ----------------------------- The score of a gap-containing single strand in the ERPIN program is computed from the dynamic programming alignment matrix. Therefore, it is the maximum of several values, and could be expected to comply with an extreme value distribution. However, gapped single strands in ERPIN are very diverse entities that may include oddities such as single-nucleotide strands, or strands mostly filled with gaps. This results in very uneven distributions that we were not able to model satisfyingly. Therefore, the score distribution of a gapped single strand in ERPIN is currently estimated based on a short simulation performed on a random sequence (see Methods section for details). Score distributions of complete regions with gaps ================================================= Score of a configuration ------------------------ When presenting a sequence to a whole region, the presence of gaps in single-strands results in multiple allowed positions for flanking helical elements. A configuration is a specific arrangement of helix elements determined by the number of intervening gaps (Fig [1](#F1){ref-type="fig"}). There is one score for each allowed configuration, which is the sum of scores for all helices and single strands in this configuration. We therefore need to compose the different score distributions to obtain the distribution of the total score for one configuration. This is again done using a discrete convolution of these distributions, with the same procedure and grid parameter as above. This provides the score distribution of a single configuration. Note that, although a configuration may contain gapped single strands of which score distribution was not produced by a discrete convolution, such distributions can now be treated by this second convolution round applied to whole profiles. Score of a complete region -------------------------- The number of gaps in a single strand is bounded by the maximum number of gaps observed for this strand in the training set: mxgaps. For a simple hairpin-loop motif with mxgapspossible gaps in the loop, there are (mxgaps+ 1) possible configurations. For a whole region containing Nstrands with gaps (i= 1, 2, \..., N), the number of configurations is: ![](1471-2105-6-118-i13.gif) Erpin evaluates all possible configurations without any construction rule or strategy. A combinatorial explosion is avoided by implementing multi-stage searches, where search at each stage is limited to a defined mask, or subset of the region under study. The score of a region or mask at in given site is the maximum score obtained for all possible configurations. Since a motif is identified only after all possible configurations are evaluated at a given site, our estimation of motif scores requires taking into account this additional complexity. Let Kdenote the number of configurations for a given motif, and S~i~, the score obtained for the ith configuration. As the ERPIN program does not permit the of addition gaps relative to those present in the training set, Kis necessarily bounded but its value can be relatively large. We are now interested in the maximal score obtained for all configurations at each site. This is the extreme value distribution, or the distribution of a random variable Mdefined as: ![](1471-2105-6-118-i14.gif) If P~fs~= P(S~i~\> -∞) and p~i~(x) = P(S~i~\> x\\|S~i~\> -∞), and the random variables S~i~are statistically independent (s.i) and identically distributed (i.d), then: ![](1471-2105-6-118-i15.gif) ![](1471-2105-6-118-i16.gif) The most \"interesting\" scores are expected to be of the same order of magnitude as those obtained by training set sequences. For any realistic training set, these scores should be very high compared to scores obtained on random sequences and, therefore, their probability should be very low. In this case P(M\> x), given by formula (17), behaves at the first order approximation as K.P~fs~.p(x). For a database of size Ω, considering that individual sequences in the database are large enough compared to the search motif so that border effects can be ignored, the final E-value, is : E(x) = P(M\> x).Ω (18) Figure [5](#F5){ref-type="fig"} compares these computed E-values (solid lines) to simulations performed on a random database (circles), for complex RNA regions encompassing multiple helices and singles strands (gapped or ungapped). Overall there is a very good agreement between E-value and simulation, consistent with our hypothesis that configuration scores are independent and equally distributed. Importantly, E-values remain accurate for RNA regions containing large gapped single strands, such as snoRNA (Fig [5b](#F5){ref-type="fig"}), Let-7 miRNA (Fig [5c](#F5){ref-type="fig"}) and SECIS (Fig [5d](#F5){ref-type="fig"}), and over a wide range of scores. This last point is also important, since \"borderline\" solutions with an E-value around 0.1 or 1 are potentially more interesting biologically as low E-value solutions. Moreover, computing times for overall E-value calculations in all our tests motifs remained insignificant relative to database scan times. In the case where pseudo-counts are switched-off, profiles contain multiple \"non-finite scores\" which are excluded from the convolution process. This may imply a lack of accuracy in the left-hand side of the estimated score distribution, where scores have low values, but should have little effect in the region of biologically interesting scores. Therefore we do not expect E-values to deteriorate significantly in practise when pseudocounts are switched off. Conclusion ========== We have presented a method to estimate the score distributions of RNA helices or single strand profiles and of their combinations into larger motifs. This method is based on discrete convolutions. The computing time of the discrete convolution algorithm increases quadratically with profile size and remains in any case negligible relative to database scan durations. This procedure is implemented in the last release of the ERPIN software (V. 4.2) and provides accurate estimates of E-values for practical applications. Interestingly, the discrete convolution approach can be applied as well to others sequence scoring models -nucleic acids or proteins -- based on ungapped profiles with independent columns. Methods ======= ERPIN program and utilities --------------------------- The ERPIN program (sources and executables) is available at <http://tagc.univ-mrs.fr/erpin/>. Simulated score distributions of independent helix and single-strand profiles (Fig [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}) were obtained using the -hist(histogram) option of ERPIN and utility programs epnstat, convhstat, convsstatand mstatprovided in the distribution. For Fig [5](#F5){ref-type="fig"}, full motif searches were performed using ERPIN Version 4.2.5 with pseudocount weight α= 2.10^-4^. Graphical outputs for figures [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"} were produced using the Matlab \[[@B13]\] package. Training sets ------------- Training sets for profile statistics and ERPIN runs in Fig [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"} are available on the ERPIN web site and were obtained as follows: • tRNA: 903 type I tRNA sequences (all species) from the 1997 version of M. Sprinzl\'s nuclear tRNA alignment \[[@B17]\]. • SECIS (Selenocystein Insertion Sequence): 117 metazoan SECIS sequences, from our own compilation \[[@B7]\]. • snoRNA (Small nucleolar RNA): 217 archaean C/D box snoRNA sequences, compiled and aligned by Fabrice Leclerc at CNRS Nancy (Personal communication). • Let-7 miRNA: 27 animal miRNA precursor sequences from our previous compilation \[[@B15]\]. • PolyA sites: 2327 human polyadenylation sequences from our previous compilation \[[@B16]\]. RNA substitution matrices for pseudocounts ------------------------------------------ Pseudocount calculation requires substitution matrices obtained from a model RNA sequence alignment or \"training set\", annoted with secondary structure information (helix or single strand). Klein and Eddy have developed RNA substitution matrices previously \[[@B18]\], but we use a different type here. Training set columns are converted into profiles containing raw nucleotide or base-pair counts. Let Qdenote a nucleotide or base-pair count profile of width w, produced by a concatenation of all single strand or helix profiles from the pseudo-count training set. Qis either a 4-line matrix for single strands (h= 4) or a 16-line matrix for helices (h= 16). The substitution matrix Mintroduced in Section \"Exclusions and Pseudocounts\" is then a square matrix of size h× hdefined as: ![](1471-2105-6-118-i17.gif) ![](1471-2105-6-118-i18.gif) ![](1471-2105-6-118-i19.gif) The square matrix Nis actually a \"correlation matrix\" of the profile lines since element N~ij~is the scalar product correlation of lines iand j. Coefficients λ~j~make this matrix normalized, so that: ![](1471-2105-6-118-i20.gif) Probability conservation is verified for Pand therefore it is also verified for P\': ![](1471-2105-6-118-i21.gif) Finally, it is obvious from formula (6) that P\"also verifies verifies probability conservation, hence: ![](1471-2105-6-118-i22.gif) Substitution matrices can be generated from any RNA sequence alignment using the utility program mksumof the ERPIN distribution. Default matrices provided with distribution (SUM.datfile) were obtained using a 16S/18S rRNA training set from R.Gutell (\[[@B10]\]) containing 6310 sequences from all three phylogenetic domains. We used the secondary structure of E-coli 16S rRNA as the consensus structure, resulting in 481 columns of helix profile and 7512 columns of single strand profile. Option -pcwis used to set pseudo-count weight αin the ERPIN program. Default internal value is 2.10^-4^, but this has been rescaled for users by a factor of 2.10^-3^giving a default user value of 0.1 and a practical maximal value that should not exceed 1. Effects of pseudocounts and of the αparameter on profiles can be visualized using the utility program pview. Score distribution of gap-containing single-strands --------------------------------------------------- The score distribution of gap-containing single strands is evaluated by repeatedly computing profile scores with a random sequence of same length Las the profile and same composition as the target sequence database. The calculation is repeated C.L^2^times, with Ca constant, and L\<L~max~in order to limit CPU time for unusually large strands. Default values are C= 300 and L~max~= 12. Histograms and discrete convolution product ------------------------------------------- Although discrete convolutions can be computed using iterated Fast Fourier Transforms, this approach is subject to numerical approximations in practice. A direct calculation is more accurate and proved fast enough in all cases tested. Time complexity of the discrete convolution algorithm is O(N^2^) where Nis the total number of profile columns. This value remains tractable even for the largest RNA motifs. The convolution algorithm was adapted from those found in the Octave \[[@B12]\] and Matlab \[[@B13]\] packages. The linear sampling interval was set at Δx= .05. CPU time for the complete E-value calculation (including profile construction, convolution of independent profiles and convolution of configurations) for motifs in Fig [5](#F5){ref-type="fig"} ranged from 10^-4^s to 0.8s on a 2.6 GHz Intel Pentium workstation with 1 Gb of RAM. Extreme value distribution -------------------------- Formula (17) used for calculating the extreme value distribution is of the type (1 - (1 - x)^N^). If x.N\> 1 the result is obtained with the C library function x![](1471-2105-6-118-i23.gif)x^y^which lacks precision when x≪ 1. Otherwise we compute (17) using the binomial formula for (1 - x)^N^. Authors\' contributions ======================= AL conceived the discrete convolution approach, programmed the software and participated in drafting the manuscript, ML and JFF performed program runs and statistical analyzes of outputs, DG participated in the design and coordination of the study and wrote most of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We thank the ACI IMPBio program for their support to the development of the ERPIN software and Web server. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### An example of ERPIN training set containing two double helices (noted 2 and 4), and three single strands (noted 3, 5 and 7). Due to gaps in the alignment, helix 2 spans 9 to 11 nt, and helix 4 spans 6 to 7 nt. Combinations of these allowed ranges give 6 possible configurations for the whole RNA motif. ::: ![](1471-2105-6-118-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distributions of finite and total scores obtained from the motif encompassing the anticodon and TΨCloop of tRNA, at each position of a 100 Mb random sequence database. This region covers three single strand profiles and two helix profiles and spans a gap-containing single- strand profile that is not included in score calculation. ERPIN results were processed by the epnstatutility program. A: finite scores. B: total scores (both finite and non-finite). ::: ![](1471-2105-6-118-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Comparision of finite score distributions obtained from discrete convolution of helix profiles (solid lines) and simulation (shaded bars). The various helices in the region under study were combined into a larger 16 × Wprofile, where Wis the total number of base-pairs in the region. Lod-scores were computed based on a uniform nucleotide composition, by the convhstatutility program. ::: ![](1471-2105-6-118-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Comparision of finite score distributions obtained from discrete convolution of single-strand profiles (curve) and simulation (shaded bars). The various strands in the region under study were combined into a larger 4 × Wprofile, where Wis the total number of nucleotides in the region. Lod-scores were computed based on a uniform nucleotide composition, by the convsstatutility program. ::: ![](1471-2105-6-118-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparision of computed E-values (solid lines) and number of solutions obtained from simulation on a random database of uniform nucleotide composition (circles), for different RNA motifs. Numbers following \"region\" refer to secondary structure elements in the corresponding training set available from <http://tagc.univ-mrs.fr/erpin/>. E-values were computed using the mstatutility program. (a) tRNA region covering the anticodon and TΨCstem-loops; (b) C/D box snoRNA region covering the major stem and C+D boxes; (c) Let-7 miRNA region covering the complete precursor hairpin; (d) SECIS element covering the large 14 bp stem and apical stem+loops. ::: ![](1471-2105-6-118-5) :::	PubMed Central	2024-06-05T03:55:58.210379	2005-5-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168889/", "journal": "BMC Bioinformatics. 2005 May 13; 6:118", "authors": [ { "first": "André", "last": "Lambert" }, { "first": "Matthieu", "last": "Legendre" }, { "first": "Jean-Fred", "last": "Fontaine" }, { "first": "Daniel", "last": "Gautheret" } ] }
PMC1168890	Background ========== A diverse range of applications that impact on clinical diagnosis and prognosis rely on the real-time detection and quantification of Polymerase Chain Reaction products such as bacterial diagnostics \[[@B1],[@B2]\], viral load \[[@B3]-[@B6]\], cancerous cell burden \[[@B7],[@B8]\] and parasite quantification \[[@B9]\]. Each particular assay dictates the requirement for instrument capacity, data reproducibility and run length times. There is an expansive collection of different fluorescent chemistries and instruments available for real-time amplification reactions \[[@B10]-[@B12]\]. These platforms represent a technological advance from the traditional thermal cycler and typically offer faster, smaller (reduced volume) and more (high-throughput) reactions. Whilst the number of instruments with different software and fluorescent dyes is considerable, the obvious drawback is that the resultant data may not be repeatable or even directly comparable between instruments or even within instruments on different runs. A statistically appropriate number of samples and runs need to be conducted in order to have confidence in the results or to assign mathematical significance to any of the factors affecting real-time amplification. In addition, there is a shortage of reliable, universal DNA standards at present. Increasing diversity through technological advances in DNA quantitation instruments and software may further challenge the standardisation of results. It should be ensured that proper quality controls under-pin both existing and emerging technologies. The aim of this study was to evaluate the repeatability of three different real-time PCR platforms including the effect of PCR reaction mix. Previous work carried out has shown that fluorogenic 5\' exonuclease (\"TaqMan^®^\") probe chemistry was more repeatable than Scorpion probe or SYBR^®^Green I chemistries for genetically modified soya amplification on the LightCycler^®^(Roche Diagnostics, Lewes, UK) and the ABI PRISM^®^7700 (Applied Biosystems, Warrington, UK) \[[@B13]\]. Hence, for the amplification of a \'model\' plasmid dilution series we have selected \"TaqMan^®^\" probes to assess inter-platform repeatability of the LightCycler^®^, the ABI PRISM^®^7700 and the Rotor Gene 3000™ (Corbett Research, Cambridge, UK). These three real-time PCR platforms have different temperature control mechanisms, specifically a 96 well heating block (ABI PRISM^®^7700), an air fan and/or coil with 32-position carousel (LightCycler^®^), and ambient air with 72-position rotor (Rotor Gene 3000™) \[[@B10]\]. In addition the performance of two commercially available master-mixes were compared. Statistical analysis was performed on the data to assess the levels of inter-platform repeatability and dynamic range of the instruments with different master-mixes. Results ======= Repeatability Estimates ----------------------- Precision estimates were determined based on each platform with PCR reaction mix combination at every plasmid copy number across the three days using ISO guidelines \[[@B14]\]. Precision estimates were calculated as repeatability standard deviations of the Ct (threshold cycle number) values from the amplification of each plasmid dilution. In addition, the mean value and standard deviations of these repeatability standard deviations were also presented in Table [1](#T1){ref-type="table"} to provide a measure of the variance of these estimates. Percentage coefficients of variance (%CV) were not used to represent precision estimates as a low mean Ct value in this experiment is indicative of good precision and therefore percentage CV would generate misleading results. Although the assay was not designed to assess biological sensitivity of the system, and the dilution levels did not approach the limit of detection for the system, the 10^8^plasmid copy number data set had the best overall precision associated with it. The most probable reason for this is the availability of the target analyte in the sample. The ABI PRISM^®^7700 had the lowest mean repeatability estimate of all the platforms in this study when combined with the FastStart reaction mix, as shown in Table [1](#T1){ref-type="table"}. However, the means of the repeatability estimates for the two PCR reaction mixes when combined with this platform were different (0.202 for the FastStart mix and 0.362 for the Excite™ mix). With the exception of the 10^4^plasmid copy number, every repeatability estimate value for the reactions performed on the ABI PRISM^®^7700 using FastStart was lower than the corresponding Excite™ reaction mix. Indeed, consistently across the three platforms, the FastStart mix had the lowest average repeatability estimates in this trial. The value of the mean and standard deviations of the repeatability estimates were comparable when the FastStart PCR reaction mix was employed on the ABI PRISM^®^7700 and on the LightCycler^®^(0.202 and 0.219 respectively and 0.047 and 0.082 respectively). The lowest repeatability estimate calculated in this study was found to be 0.100. This value was generated on the LightCycler^®^using the FastStart reaction mix and a plasmid copy number of 10^4^. Conversely, the highest repeatability estimate (1.378) was also produced on the LightCycler^®^but the Excite™ PCR reaction mix was employed with a plasmid copy number of 10^6^. Unexpectedly, the highest repeatability estimates for each data set are not necessarily associated with the low plasmid copy numbers \[[@B15]\]. The lowest repeatability estimates and the highest repeatability estimates for each platform and PCR reaction mix data set show no discernible trend with respect to plasmid copy number. The least repeatable system in this trial based upon the data in Table [1](#T1){ref-type="table"} was the combination of the LightCycler^®^with the Excite™ PCR reaction mix. Comparison with the results of other published studies was not possible, as the statistical methods used were not clearly reported in the literature. Analysis of Variance -------------------- An ANOVA approach using the threshold cycle number was used to identify sources of variability within the data. Table [2](#T2){ref-type="table"} shows the significance levels of the four main factors of day, plasmid copy number, reaction mix, and platform, as well as the significance levels of the interaction terms. The factors and interaction terms were compared to the Error MS (mean square) in order to assess their significance. All four individual factors were highly significant. From the calculated F-variance ratio the factor of day had the least significant effect of the four factors, followed by PCR reaction mix (approximately 1.5 times the effect of day), then platform (approximately 7 times the effect of day), and then, as expected, plasmid copy number (approximately 45 times the effect of day). Even though the effect of day had a significant effect upon the results, this effect was very small when compared to the effect of platform and plasmid copy number. The interaction of the four main factors was also investigated, this demonstrated that the majority of interaction terms were significant. However, in general the magnitude of the effect of these interaction terms was very small compared to the effect of the four main factors. ### Day Effect Day had a statistically significant effect upon the Ct value of the raw data (P\<0.00001). The average Ct value in the full data set decreased from 19.76 to 18.54 across the 3 days (data not shown). ### Plasmid Dilution Effect As Figure [1](#F1){ref-type="fig"} illustrates, the number of plasmid copies was inversely proportional to the average Ct value \[[@B6],[@B12]\] within the instrument dynamic working range and above the LOD (limit of detection). Our results concur with Sanchez et al\[[@B5]\] who reported that cytomegalovirus quantification was linear between 10^1^and 10^6^copies, and with Moberg et alwho used 10^2^to 10^7^DNA copies in their real-time assay \[[@B16]\]. At extremely low or high plasmid copy numbers the linear relationship with Ct value may be affected. At high copy number it may be difficult to set an accurate baseline because threshold values will be reached after only a few amplification cycles. In addition, reactions with very low numbers of target molecules may be disproportionately affected by pipetting errors and by stoichiometric effects. This highlights the need to work within the linear dynamic range of the instrument and the potential inaccuracy introduced by extrapolation of calibration curves. Fresh plasmid dilutions were utilised to counteract the previously observed progressive loss of intact plasmid DNA, particularly at low plasmid copy number dilutions. This is probably due to physical adhesion of the DNA to the tube walls, although precautions, such as the use of siliconised tubes, were undertaken. Fresh plasmid dilutions were made each day. Figure [2](#F2){ref-type="fig"} illustrates how the average Ct value alters across the three days (which have three different dilution sets) for each plasmid copy number. ### PCR Reaction Mix Effect The results from reactions containing FastStart mix had significantly lower average Ct values than their Excite™ 2x mix reaction counterparts (data not presented) -- the difference is approximately 1 cycle per plasmid dilution. The smallest interaction term is that of PCR reaction mix with platform (P = 0.019), indicating good consistency of results when using the reaction mixes across different platforms. ### Platform Effect The platforms have different heating mechanisms and therefore it was expected that their performance characteristics such as repeatability might differ significantly. The largest significant interaction term (F = 150.8) was that of platform with day, indicating that each platform may not be behaving consistently between the three days. Closer inspection of the data shows that the majority of the variation associated with the platform by day interaction is attributed to the Rotor Gene day 2, as illustrated in Figure [3](#F3){ref-type="fig"}, which appears inconsistent compared to the other two platforms. This interaction effect of platform with day was not significantly different from the factor of day or PCR reaction mix. The interaction of PCR reaction mix and platform (shown in Table [2](#T2){ref-type="table"}) was found to be statistically significant at the 5% level. The full raw data set is available \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Discussion ========== Use of a \'model\' system that contained little or no contaminants, coupled with a balanced experimental design enabled accurate assessment of repeatability. Other factors such as extraction methods and complex matrices can potentially increase analytical variability. As demonstrated by the present study, a range of factors can affect performance of each of the instruments. In addition, results are influenced by the threshold value settings applied to the machine by the analyst. To underpin meaningful quantitative PCR measurement, internationally accepted guidelines should be established to standardise the analysis and interpretation of real-time PCR results. PCR reference standards could also be used as indicators of instrument reliability and amplification efficiency \[[@B17],[@B18]\], while internal reaction controls could additionally monitor effects of PCR inhibitors and normalise for any instrument non-uniformity. A recent publication has looked at the plateau phase and the effect of enzyme inhibition during this stage of amplification \[[@B19]\]. With regard to analysis and software compatibility, a study conducted by Kuhne and Oschmann \[[@B20]\] has demonstrated the feasibility of importing standard curves from different LightCycler runs. Historically, standard curves have been the method of choice for determining absolute DNA concentration. However, the quantity of the standards used should be accurately calculated by gravimetry or other primary methods, as absorbance at a wavelength of 260 nm is not sufficient alone. Our present study has shown the potential inaccuracies introduced by importing standard curves, considering the significance of the factor \'day\' on the data presented. Even under the controlled conditions used in the present study, significant variability was observed, suggesting that transferring standard curves between runs could be unreliable unless adequate controls are used. Recent research using more sophisticated software approaches has demonstrated the ability to use grand mean calibration curves between runs following compensation for differences in amplification efficiency \[[@B21]\]. Amplification efficiency changes throughout the course of a PCR reaction and is dependent on enzyme kinetics \[[@B22],[@B23]\]. The potential pitfall of assuming that the sample analyte has the exact same kinetics and amplification efficiency of the standards has been highlighted recently \[[@B15]\]. A Galton-Watson branching process has been proposed which involves considering the number of molecules present during PCR cycles and uses Michaelis-Menten enzyme kinetics together with a population growth approach \[[@B23]\]. One other source of uncertainty associated with using amplification curves is the issue of handling outlier data points. In a recent publication Bar et aldiscussed this subject and presented a new statistical method \[[@B24]\]. This may have applications for clinical samples that have been extracted from complex matrices or any analysis with high levels of variability between results. It is not surprising that difficulties in data comparison arise due to the many different mathematical and software approaches applied to real-time quantitative PCR, despite sharing the same mathematical fundamentals \[[@B25]-[@B28]\]. The data generated in this study is not intended to indicate any preference for a particular piece of equipment or PCR reaction mix. The instrument specification, reliability, operator training, servicing and comparability all contribute to overall assay precision. In choosing a real-time platform, we would recommend that users consider the statistical outcomes of our results and perform their own critical assessments to ensure that an instrument is fit for their purpose. Generally any such judgement would involve factors such as the volume of samples for processing, the speed at which results need to be obtained and the reproducibility or repeatability of the results. In addition to choice of appropriate real-time platform, accurate quantitative measurement is dependent upon rational experimental design, optimisation and careful data interpretation. Conclusion ========== We have demonstrated that in this trial the most repeatable system was the ABI PRISM^®^7700 with the FastStart PCR reaction mix. However the mean repeatability estimate of the LightCycler^®^combined with the FastStart mix was comparable. In addition, we have shown that the plasmid copy numbers used in this trial did not display a distinct relationship with instrument repeatability. An ANOVA established that plasmid copy number, platform, PCR reaction mix and day significantly affected the results in a descending order of magnitude. Inherent differences in the data handling between instruments still presents a challenge to the normalisation of real-time PCR assays between laboratories and users of different instruments. We have confirmed that standardised real-time PCR detection instruments and data analysis approaches are imperative in order to guarantee the integrity and the repeatability of DNA measurement experiments. Methods ======= Plasmid Composition and Dilution -------------------------------- The target DNA was a pPCR-Script Amp SK(+) plasmid (Stratagene, Amsterdam, The Netherlands) with an insert at the SrfI site in the multiple cloning site (supplied by N.I.S.T, Gaithersburg, MD, USA). A portion of the insert sequence follows with the primer binding sites indicated in bold face type and the probe binding site underlined: 5\'-*CAGACCGCTCGACGATAGGT[CAGCACTGTCTCGTTGACAGGCGTG]{.underline}GTCAATCAGCCTGAAATCCTCGATCAGAGTGTGCCG**ATCTCTGGTCCACGTCCT-3\'. From a stock solution of plasmid (3 × 10^13^copies) in molecular biology grade water (Sigma, Poole, UK), 1 in 10 serial dilutions were carried out to a copy number of 3 × 10^5^per mL using the same batch of water. Fresh dilutions were made every day from the same stock and using the same conditions and were kept at 4°C before and between use. At low copy numbers particularly, a reduction in target copies available for amplification was observed on prolonged storage, possibly due to DNA absorption to the walls of the tube (data not shown). Fresh dilutions were prepared each day in order to minimise this source of variation. Real-time PCR Conditions ------------------------ A 100-bp region of the insert sequence was amplified using HPLC purified oligonucleotide primers (Sigma-Genosys, Cambridge, UK) and a fluorogenic 5\' exonuclease probe (Sigma-Genosys, Cambridge, UK) diluted in molecular biology grade water (Sigma, Poole, UK). The primers and probe used were P1A (forward primer): 5\'-ACAGACCGCTCGACGATAGG-3\', P1B (reverse primer): 5\'-AGGACGTGGACCAGAGATCG-3\' and TAQ1 (probe): 5\'-{FAM}CAGCACTGTCTCGTTGACAGGCGTG{TAMRA}-3\' (FAM -- 6-carboxyfluoresceine; TAMRA -- 5\'-carboxytetramethylrhodamine). Primer Express^®^software (Applied Biosystems; Warrington, UK) was used to design the primers and probe. The thermal cycling conditions used were dependent on the PCR reaction mix employed and were optimised protocols based on the manufacturer\'s guidelines. Amplification conditions for those reactions that contained \'FastStart Master mix\' (Roche Diagnostics, Lewes, UK) were a 10 minute hold at 95°C followed by 55 cycles of 95°C for 15 seconds and 60°C for 30 seconds. For reactions with the Excite™ 2x kit (Biogene, Kimbolton, UK) the amplification conditions were a 2 minute hold at 50°C followed by a 10 minute hold at 95°C and then 55 cycles of 95°C for 15 seconds and 60°C for one minute. Instrumentation & PCR Reaction Mixes ------------------------------------ Three platforms compatible with \"TaqMan^®^\" probes were assessed -- LightCycler^®^(Roche Diagnostics, Lewes, UK), ABI PRISM^®^7700 (Applied Biosystems, Warrington, UK) and Rotor Gene 3000™ (Corbett Research, Cambridge, UK). All of these instruments were found to be compatible with FastStart (Roche Diagnostics, Lewes, UK) and Excite™ 2x (Biogene Kimbolton, UK) mixes following optimisation. The amplification performances of both PCR reaction mixes were investigated on each of the three machines using the plasmid dilution series as target. The probe was added at a final concentration of 200 nM per reaction for all experiments. The Primers P1A and P1B were used at a final concentration of 1500 nM each in the Excite™ mix reactions whereas the FastStart mix reactions contained the primers at a final concentration of 500 nM each. A supplement of magnesium chloride (final concentration of 3 mM) was added to those reactions containing Excite™ mix that were run on the LightCycler^®^. In addition, magnesium chloride at a final concentration of 3 mM was required for all those reactions containing the FastStart mix except for the equivalent amplification reactions run on the LightCycler^®^, where a final concentration of 2 mM of magnesium chloride was optimal. A solution of BSA (BSA 100x; New England Biolabs, Hitchin, UK) at a final concentration of 0.1 mg/mL was added to the Excite mix for reactions run on the LightCycler^®^. To every reaction except the negative controls, 1 μL of plasmid DNA was added. Additional molecular biology grade water was added to a final volume of 20 μL and 25 μL for the FastStart and Excite™ mixes respectively. Reactions were set up in 20 μL glass capillaries, 0.1 mL plastic tubes and 96 well optical plates for use on the LightCycler^®^, Rotor Gene 3000™ and ABI PRISM^®^7700 respectively. Experimental Design ------------------- Six experiments (two per platform) were run, and repeated for three consecutive days. Three replicates of each plasmid DNA dilution plus three negative controls (24 reactions in total) were amplified per experiment. One specific analyst prepared the complete master-mixes containing primers, probes, water and other supplements as already outlined, in a pre-PCR \'clean\' room using the same pipettes each day. A second specific analyst added the plasmid DNA in a dedicated template addition area and used the same instruments for all template additions. Those amplification reactions containing the Excite™ 2x mix were run at the same time each morning on the Rotor Gene 3000™, ABI PRISM^®^7700 and LightCycler^®^and the reactions containing FastStart mix were run at the same time each afternoon on the Rotor Gene 3000™, ABI PRISM^®^7700 and LightCycler^®^. For each run the same instrument, rotor and carousel (where applicable) were employed to minimise intra-run variability. Baseline Settings & Data ------------------------ The Ct (threshold cycle number) was used as the measurand in order to assess the level of product in proportion to the level of starting target. As recommended by the manufacturers\' guidelines, the Ct values were calculated from the mid-point of the exponential phase of amplification. However, each real-time platform required some manual intervention to set analysis parameters. In addition specific analysis software was associated with each instrument and inherent differences in data manipulation may introduce further inter-platform variability. As amplification efficiency has been shown to change over the course of the reaction \[[@B29]\], we have based our analysis on the mid-point of the exponential phase to measure threshold cycles irrespective of platform, run or PCR reaction mix. Statistical Analysis -------------------- The results were interpreted and analysed according to the instrument manufacturer protocols. For each platform, graphs of cycle number versus arbitrary fluorescence units were generated and from this raw data, Ct values were automatically calculated by the software provided with each instrument \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. The statistical package Statistica 6.0 (Statsoft; Tulsa, OK, USA) was used to analyse the data for all the instrument runs by ANOVA (Analysis of Variance) and then repeatability estimates were carried out. Abbreviations ============= FAM: 6-carboxyfluoresceine TAMRA: 5-carboxytetramethylrhodamine BSA: bovine serum albumin Ct: threshold cycle number \%CV: Percentage coefficient of variance MS: Mean square Authors\' contributions ======================= CED carried out the experimental work and drafted the manuscript; FQ carried out the experimental work; MJB participated in the design of the study and performed the statistical analysis; MJH conceived of the study; JRB produced the plasmid material; AJW conceived and designed the study. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Raw Data Table. Table 3 consists of the raw data results comparing the effect of day, plasmid copy number, PCR reaction mix and platform on the mean Ct value. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The work described in this study was supported under contract with the UK Department of Trade and Industry as part of the national measurement system valid analytical measurement (VAM) program. We would like to thank the Biotech Division at NIST, (USA) for providing us with the plasmid DNA. The authors wish to thank Jacquie Keer for editing and helpful discussion. In addition, David McDowell and Gavin Nixon are thanked for their help with primer and probe design. We would like to thank Corbett Research UK for the loan of the Rotor Gene 3000™ instrument. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Average threshold cycle number (Ct) plotted against plasmid copy number. The data have been pooled from each platform, PCR reaction mix and day to give an average Ct value. Linear regression performed on the ploted data values from this graph determined that the gradient was -2.92 and that the amplification efficiency \[21\] was greater than two. However the amplification efficiency cannot be greater than two since PCR is an exponential reaction where the amount of product doubles at each cycle. ::: ![](1472-6750-5-15-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Average threshold cycle number (Ct) plotted against day for each plasmid dilution illustrating the interaction of day and plasmid copies. The data have been pooled from each platform, PCR reaction mix and plasmid dilution. ::: ![](1472-6750-5-15-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Average threshold cycle number (Ct) plotted against day for each platform illustrating the interaction of day and platform. The data have been pooled from each PCR reaction mix and plasmid dilution. ::: ![](1472-6750-5-15-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Comparison of repeatability estimates of the PCR measurements using different reaction mixes and platforms over three days (using standard deviations of the Ct value). ::: Repeatability Estimates (standard deviations of the Ct value) based on Individual Plasmid Copy Number ------------------------------------------------------------------------------------------------------- ------------------------------- ------------- ----------- ----------- -------- ----------- Plasmid Copy Number Platform and PCR Reaction Mix ABI 7700 LightCycler RotorGene Excite FastStart Excite FastStart Excite FastStart 10^2^ 0.543 0.182 0.568 0.233 0.888 0.186 10^3^ 0.452 0.204 1.177 0.190 0.184 0.445 10^4^ 0.223 0.305 1.123 0.100 0.256 0.592 10^5^ 0.243 0.168 0.947 0.224 0.209 0.221 10^6^ 0.362 0.183 1.378 0.281 0.772 0.333 10^7^ 0.473 0.189 0.919 0.156 0.319 0.247 10^8^ 0.236 0.184 0.454 0.352 0.322 0.166 Degrees of Freedom 6--8 6 6--8 5--6 6--8 6 Mean 0.362 0.202 0.938 0.219 0.421 0.313 Standard Deviation 0.130 0.047 0.331 0.082 0.286 0.156 (Note that one replicate is missing for the LC platform coupled with the Fast Start reaction mix -- the capillary broke). ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Variation attributable to day, plasmid copy number, PCR reaction mix and platform following an ANOVA test using threshold cycle number (Ct) as the variable factor. ::: Univariate Tests of Significance for Ct value. Sigma-restricted parameterisation. Effective hypothesis decomposition.* ------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------- -------- -------- -------- ----------- Effect SS DF MS F P Day 96.0 2 48.0 221.4 \<0.00001 Plasmid Copies 12919.2 6 2153.2 9932.0 \<0.00001 PCR Reaction Mix 75.0 1 75.0 346.2 \<0.00001 Platform 700.5 2 350.2 1615.5 \<0.00001 Day with Plasmid Copies 53.2 12 4.4 20.5 \<0.00001 Day with PCR Reaction Mix 1.9 2 0.9 4.3 0.01442 Plasmid Copies with PCR Reaction Mix 8.5 6 1.4 6.5 \<0.00001 Day with Platform 130.8 4 32.7 150.8 \<0.00001 Plasmid Copies with Platform 51.0 12 4.2 19.6 \<0.00001 PCR Reaction Mix with Platform 1.7 2 0.9 4.0 0.01901 Day with Plasmid Copies with PCR Reaction Mix 14.6 12 1.2 5.6 \<0.00001 Day with Plasmid copies with Platform 18.3 24 0.8 3.5 \<0.00001 Day with PCR Reaction Mix with Platform 49.6 4 12.4 57.1 \<0.00001 Plasmid copies with PCR Reaction Mix with Platform 4.5 12 0.4 1.7 0.06337 Day with Plasmid Copies with PCR Reaction Mix with Platform 11.8 24 0.5 2.3 0.00091 Error 54.4 251 0.2 (Where SS = sum of squares; DF = degrees of freedom; MS = Mean Square; F = ratio of variances; p = probability of obtaining a specific result, given by the null hypothesis). :::	PubMed Central	2024-06-05T03:55:58.213023	2005-5-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168890/", "journal": "BMC Biotechnol. 2005 May 25; 5:15", "authors": [ { "first": "Carol E", "last": "Donald" }, { "first": "Fizza", "last": "Qureshi" }, { "first": "Malcolm J", "last": "Burns" }, { "first": "Marcia J", "last": "Holden" }, { "first": "Joseph R", "last": "Blasic" }, { "first": "Alison J", "last": "Woolford" } ] }
PMC1168891	Background ========== Gene therapy for treating cancers has been intensively investigated. Therapeutic gene transfer using viral or nonviral vectors has been shown to be clinically feasible and safe. However, the therapeutic outcome appears not to be high as expected, presumably because the in vivo gene delivery efficiency is low. In this sense, a more effective modality is needed to improve the therapeutic benefit. Combined transfer of genes with different mechanisms of action may potentate the therapeutic outcome. Indeed, combination cancer gene therapy using cytokine and suicide genes has been shown to be more effective than single gene transfer in animal models \[[@B1]-[@B9]\]. Interleukin (IL)-12 is a heterodimeric cytokine composed of 35- and 40-kDa subunits, which are covalently linked by a disulfide bond \[[@B10],[@B11]\]. IL-12 stimulates interferon-γ (IFN-γ) production by natural killer (NK) cells, enhances the cytolytic activities of NK cells and cytotoxic T lymphocytes (CTLs), stimulates the differentiation of Th1 cells and inhibits angiogenesis \[[@B12]-[@B14]\]. These biological activities have provided a rationale for the use of IL-12 in cancer immunotherapy. In fact, recombinant IL-12 (rIL-12) exhibits significant antitumor activity in various animal models \[[@B12]-[@B14]\]. Subsequent phase I and II studies on various cancer types using intravenous or subcutaneous administration of human rIL-12 \[[@B15],[@B16]\] showed some beneficial responses, but there were also adverse effects, which interrupted the clinical trials \[[@B17],[@B18]\]. Recently, it has been reported that intratumoral administration of rIL-12 shows dose-limiting toxicity and results in measurable locoregional immunological responses in head and neck squamous cell carcinoma patients \[[@B19]\]. Since intratumoral IL-12 gene transfer could minimize the adverse effects of systemic rIL-12 administration \[[@B17]\], the effectiveness against tumors of cell vaccines expressing IL-12, or viral or nonviral vectors encoding IL-12, has been evaluated in various animal models \[[@B18],[@B20]-[@B23]\]. The results indicate that, compared to IL-2, IL-4, interferons, and granulocyte-macrophage colony-stimulating factor, IL-12 is one of the most effective immunomodulators for cancer treatment \[[@B13],[@B18],[@B20],[@B21],[@B23]\]. Pilot clinical trials have been conducted in patients with advanced primary and metastatic liver cancer on the basis of these results \[[@B24]\]. Intratumorally administered adenoviral vectors expressing IL-12 had low efficacy but no cumulative toxicity \[[@B24]\]. Suicide gene therapy deploys genes such as cytosine deaminase (CD), herpes simplex or varicella zoster virus thymidine kinase (TK), or nitroreductase, all of which convert nontoxic prodrugs into cytotoxic agents \[[@B25]-[@B27]\]. Systemically administered prodrugs are converted to the active agents only in cells expressing these suicide genes, so the therapeutic effect is maximized while systemic toxicity is reduced. CD, which normally catalyzes the deamination of cytosine to uracil during RNA biosynthesis, is found in many bacteria and fungi but not in mammalian cells \[[@B28]-[@B30]\]. It also deaminates 5-fluorocytosine (5-FC) to 5-fluorouracil (5-FU), which is highly toxic to cells. 5-FU has been used clinically since the late 1950s to treat colorectal cancer \[[@B31]\]. A cellular metabolite of 5-FU, 5-fluorouridine triphosphate, is incorporated into RNA in place of UTP and consequently the multiple functions of RNA are impaired. Another metabolite of 5-FU, 5-fluorodeoxyuridine monophosphate, blocks thymidylate synthase activity and thus inhibits DNA synthesis. No currently available gene delivery system can deliver therapeutic genes including suicide genes to all the target cells in vivo. However, tumor cells expressing a suicide gene can kill neighboring tumor cells that do not express this gene by the by-stander effect, which could partly overcome the gene delivery problem \[[@B32]-[@B35]\]. A number of suicide gene therapies have been shown to be effective against various cancer cell lines in culture and in animal models. Subsequent phase I, II or III studies using CD or TK genes were performed in colorectal cancer patients with liver metastases and in brain tumor patients \[[@B36]-[@B47]\]. The antitumor activities appeared less than expected \[[@B46]\], but the results showed that suicide gene therapy is safe to apply in the oncology clinic \[[@B36]-[@B47]\]. A combination of IL-2 and TK genes exerted a more potent therapeutic outcome in a metastatic carcinoma model, and induced antitumor immunity \[[@B1]\]. Since then, combination gene therapies using various cytokine and suicide genes have been intensively investigated \[[@B2]-[@B9]\]. In the present study, we investigated the effectiveness of combined transfer of IL-12 and CD genes against renal cell carcinoma (Renca)-tumors in mice. Methods ======= Cells and mice -------------- HeLa, 293, Renca (a cell line spontaneously derived from BALB/c renal cell adenocarcinoma), and NK cell-sensitive YAC-1 cells were cultured in DMEM or RPMI-1640 medium (Gibco-BRL) supplemented with 10% fetal bovine serum (FBS) and penicillin/streptomycin. Cultures were maintained in a 5% CO~2~atmosphere at 37°C. Male or female wild-type BALB/c mice, 7--8 weeks of age, were purchased from the animal breeding laboratory at the Korea Research Institute of Bioscience and Biotechnology. They were kept (five mice per cage) in isolation under specific pathogen-free conditions, exposed to 12-h light/12-h dark cycles, and provided with standard feed and water ad libitum. Construction of replication-defective adenoviral vectors -------------------------------------------------------- cDNAs for Esherichia colicytosine deaminase, p35 and/or p40 murine IL-12 subunits \[[@B48]\] or green fluorescent protein (GFP) were inserted by a standard cloning procedure into pX.dCMV.pA, an adeno-transfer plasmid harboring the immediate early gene promoter from human cytomegalovirus \[[@B49]-[@B51]\]. The shuttle plasmids were cotransfected into 293 cells together with pJM17, which harbors the genomic DNA of adenovirus 5 (Microbix, Canada), by a standard calcium phosphate method. Primary plaques were isolated and screened by PCR to ensure the presence of the inserts. Positive clones were plaque-purified twice and a single plaque was chosen for further amplification in 293 cells. Recombinant adenoviruses were purified by cesium chloride gradient centrifugation as previously described \[[@B49]-[@B51]\], dialyzed to remove the cesium chloride and frozen at -70°C.The viral titer was determined by a 293 plaque assay and expressed as plaque-forming units (pfu). Cell viability assay -------------------- HeLa cells were plated on 60-mm dishes at 5 × 10^5^cells/plate and mock-transduced, or transduced with Ad.CD, for 2 h at a multiplicity of infection (MOI) of 50 in serum-free DMEM. Cultures were continued for a further 12 h. Transduced and non-transduced cells were then mixed in different ratios and incubated in 24-well dishes for 24 h. The culture media were replaced with fresh media containing various concentrations of 5-FC, and incubation was continued for 4 days. Cell viability was assessed by the trypan-blue exclusion assay. Western blotting ---------------- HeLa cells were plated on 60-mm dishes at 5 × 10^5^cells/dish and transduced with recombinant adenoviruses in serum-free DMEM for 2 h. The culture media from mock- or virus-transduced cells were retained for measurement of released IL-12 at 48 h postinfection, and the cells were harvested in a lysis buffer (10 mM Tris-HCl, pH 7.5, 10 mM NaCl, 1.5 mM MgCl~2~, 10 mM β-mercaptoethanol). Cell extracts were prepared by three cycles of freezing (-70°C) and thawing and precleared by centrifugation. Fifty micrograms of precleared cell lysates were subjected to SDS-10% polyacrylamide gel electrophoresis (PAGE). The separated proteins were transferred to PVDF membranes, which were blocked with 3% bovine serum albumin (BSA) in phosphate-buffered saline (PBS) and incubated with rat anti-mouse IL-12 antibody (PharMingen), then with horseradish peroxidase-conjugated rabbit anti-rat lgG antibody (Sigma). Proteins were detected according to Amersham\'s ECL protocol. The culture supernatants were incubated with anti-IL-12 antibody. A suspension of protein G-Sepharose beads (100 μl) was added and the incubation mixtures were further incubated at 4°C for 12 h. The immunoprecipitates were washed three times with 40 mM N-2-hydroxyethylpiperazine-N\'-2-ethanesulfonic acid, pH 7.4, 100 mM KCl, 0.1% NP-40, 0.1 mM DTT, and once with PBS, then immunoblotted with the rat anti-mouse IL-12 antibody described above. Animal model and treatment -------------------------- Tumors were generated on the flanks of BALB/c mice by subcutaneous (s.c.) injection of 1.5 × 10^5^of Renca cells in 0.1 ml of PBS. After visible tumors had developed for 7--9 days after inoculation, the mice were injected once intratumorally with recombinant adenoviruses at 1 × 10^9^pfu in 10 mM Tris-HCl, pH 7.4, 1 mM MgCl~2~, 10% glycerol. Animals undergoing 5-fluorocytosine therapy received intraperitoneal administration of 5-FC (400 mg/kg) daily for 10 days. Tumors were measured prior to virus injection and subsequently at intervals of 3 days. Linear calipers were used to measure the longest diameter (a), width (b), and depth (c). The tumor size was calculated by the formula: (a × b × c/2). ELISA for IFN-γ --------------- ELISAs were performed using essentially the reagents and procedures described in the Endogen (USA) product literature. Flat-bottomed 96-well microtiter plates (Nunc Maxisorp) were coated with rat anti-IFN-γ antibodies (1 μg/ml) at 4°C overnight and blocked with 4% BSA in PBS for 2 h at 37°C. Renca cells were treated with mitomycin C at 50 μg/ml for 20 min at 37°C. Splenocytes (2 × 10^6^cells/ml) obtained from the tumor-bearing mice 21 days after virus treatment were stimulated with the inactivated Renca cells in U-bottomed 96-well plates at an effector-to-stimulator ratio of 10:1. Supernatants (50 μl) from 24 h and 48 h cultures were added to the ELISA plate wells and incubated for 1 h. After several washing steps, rat anti-mouse IFN-γ antibody conjugated with biotin was added. Color was developed using avidin-conjugated horseradish peroxidase and OPD substrate (Sigma). The reaction was terminated by the addition of 2N H~2~SO~4~. The absorbance was measured at 490 nm with an ELISA plate reader. The amounts of IFN-γ were quantified by interpolation of a standard curve generated using known amounts of standard mouse rIFN-γ (Endogen). The biological activity of the IL-12 produced from the recombinant adenoviruses was assessed on the basis of its ability to induce IFN-γ secretion from murine splenocytes. Murine splenocytes (1.5 × 10^6^cells/well) were washed with cold medium and incubated with the culture supernatants. IFN-γ in the incubation media was then determined by an ELISA, as described above. Cytotoxicity assay of NK cells ------------------------------ The cytolytic activity of NK cells was determined by a standard 4-h ^51^Cr-release assay. The spleens from dead mice obtained 21 days after tumor inoculation were removed aseptically and single-cell suspensions, prepared by passing the spleens through a metal mesh, were suspended in 5 ml of 0.83% ammonium chloride to lyse the red blood cells. The splenocytes were then used as effector cells to assay NK cell activity. YAC-1 cells (2 × 10^6^) in 0.5 ml of RPMI-1640 with 20% FBS were labeled with 50 μCi of Na^51^CrO~4~(Amersham) for 90 min. The labeled cells were washed three times with serum-free DMEM and mixed (1 × 10^4^per well) with the effector cells in a U-bottomed 96-well plate for 4 h at 37°C at three different effector-to-target ratios (100:1, 33:1, and 11:1), each in triplicate. The mixtures of target and effector cells were incubated for 4 h at 37°C. The percentage of specific ^51^Cr release was calculated as \[(cpm experimental release - cpm spontaneous release/cpm maximum release - cpm spontaneous release)\] × 100. The maximum ^51^Cr release was determined from the supernatant when labeled cells were lysed by adding 0.1 ml of 3% SDS. Spontaneous release was determined from the supernatants after labeled cells were incubated in 0.1 ml of medium without effector cells. The standard deviations for each triplicate sample and for spontaneous release were less than 10% and 20%, respectively. Statistics ---------- Statistical analysis was carried out using the Student\'s ttest (unipolar, paired) and the chi-square test (two-tailed). The values were considered statistically significant when the Pvalue was less than 0.05. Results and discussion ====================== Recombinant adenoviruses encoding the p35 and/or p40 subunits of murine IL-12 express bioactive IL-12 ----------------------------------------------------------------------------------------------------- We constructed replication-defective adenoviruses containing the p35 and/or p40 subunits of murine IL-12 under the transcriptional control of the immediate early gene promoter of human cytomegalovirus (Figure [1](#F1){ref-type="fig"}). Ad.IL-12 contains murine cDNAs of both IL-12 subunits, which are linked by an internal ribosome entry site sequence from encephalomyocarditis virus. To test whether Ad.p40 and Ad.IL-12 express the p40 subunit and the p35-p40 heterodimer respectively, we transduced HeLa cells with the two constructs at the indicated MOI. Replication-defective adenovirus encoding GFP (Ad.GFP) was used as a control. Lysates of cells transduced with Ad.p40 or Ad.GFP were resolved by SDS-PAGE followed by immunoblotting with anti-IL-12 antibody, which reacts specifically with the p40 subunit of mouse IL-12 and with free p40 (Figure [2A](#F2){ref-type="fig"}, left panel). Ad.p40 was found to express p40. The culture media from cells transduced with Ad.IL-12 or Ad.GFP were immunoprecipitated with anti-IL-12 antibody. The immuno-complexes were resolved by SDS-PAGE under nonreducing conditions and immunoblotted with anti-IL-12 antibody. The two protein bands detected might be a p40 homodimer (p80) and a p35-p40 heterodimer (p70) (Figure [2A](#F2){ref-type="fig"}, right panel, indicated by arrows). Under nonreducing conditions, Ad.IL-12 produced a doublet of 79/75 kDa that was sensitive to a reducing agent \[[@B52]\]. Ad.p40 produces a band migrating with an apparent molecular mass of approximately 90 kDa, which on a reducing gel reveals a band consistent with the p40 monomer \[[@B53]\]. Thus, our result appeared to be consistent with previous findings. We could not assess p35 expression because no antibody specific for this subunit was available. We next examined whether cells transduced with adenoviruses encoding the p35 and/or p40 subunits release bioactive IL-12. HeLa cells were mock-transduced, or transduced with a 1:1 (pfu ratio) mixture of Ad.p35 and Ad.p40, Ad.IL-12, or Ad.GFP at the indicated MOI, and the culture supernatants were collected after 24 h incubation. Splenocytes from naïve mice were incubated with equivalent volumes of the culture supernatants, and the IFN-γ contents of the medium were measured using ELISA (Figure [2B](#F2){ref-type="fig"} and [2C](#F2){ref-type="fig"}). The culture supernatants from cells transduced with a mixture of Ad.p35 and Ad.p40 or Ad.IL-12 exhibited much higher IFN-γ inducing-activity than the controls. These results suggest that cells transduced with Ad.p35 plus Ad.p40 or Ad.IL-12 release bioactive IL-12. Recombinant adenovirus encoding cytosine deaminase (Ad.CD) expresses catalytically active cytosine deaminase ------------------------------------------------------------------------------------------------------------ We constructed a replication-defective adenovirus encoding cytosine deaminase (Figure [1](#F1){ref-type="fig"}). To examine whether the Ad.CD virus expresses cytosine deaminase capable of converting 5-FC to cytotoxic 5-FU, we transduced HeLa cells with Ad.CD at an MOI of 50. Naïve HeLa or HeLa/CD cells, or mixtures of the two at the indicated ratios (Figure [3](#F3){ref-type="fig"}), were incubated in culture media without or with the indicated concentrations of 5-FC for 4 days, and cell viability was assessed by the trypan-blue exclusion assay (Figure [3](#F3){ref-type="fig"}). In the presence of 5-FC, the viability of the HeLa/CD cells decreased markedly but the naïve HeLa cells were unaffected, suggesting that cytosine deaminase was expressed in the virus-transduced cells and metabolized 5-FC to 5-FU. The viability of the mixtures of naïve and HeLa/CD cells appeared to decrease more than the proportion of naïve cells in the mixtures, particularly with 500 μg/ml 5-FC. Thus, the Ad.CD virus appeared to exert a bystander effect, though not a drastic one. The results suggest that Ad.CD expresses catalytically active cytosine deaminase. Transfer of the IL-12 and cytosine deaminasegenes in combination, followed by 5-FC treatment, exerts potent antitumor activity in Renca-tumor bearing mice ------------------------------------------------------------------------------------------------------------------------------------------------------------ Renca-tumor bearing mice were established as described in the Methods section, randomly divided into 4 groups, and injected once intratumorally with Ad.GFP, Ad.CD/5-FC, Ad.p35 plus Ad.p40, or Ad.p35 plus Ad.p40 plus Ad.CD. Mice receiving Ad.CD were treated with 5-FC. Tumor growth was measured at the indicated times (Figure [4A](#F4){ref-type="fig"}). Mouse survival until day 42 was recorded (Figure [4B](#F4){ref-type="fig"}). The treatment outcomes are summarized in Table [1](#T1){ref-type="table"}. Inhibition of tumor growth was most marked in mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC (p\< 0.001 for Ad.p35 plus Ad.p40 plus Ad.CD/5-FC vs Ad.GFP and p\< 0.0053 for Ad.p35 plus Ad.p40 plus Ad.CD/5-FC vs Ad.CD/5-FC). The mean survival times were for 33.2 ± 10.4 days for Ad.p35 plus Ad.p40 plus Ad.CD/5-FC group and 27.2 ± 10.9 days for Ad.p35 plus Ad.p40 group; the difference is statistically significant (p\< 0.0018). Among the 15 mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC, four became tumor-free and four exhibited partial responses (Table [1](#T1){ref-type="table"}). Among the 15 mice treated with Ad.p35 plus Ad.p40, one became tumor-free and two showed partial responses (p\< 0.0464 for comparison of the two treatments). Ad.p35 plus Ad.p40 treatment appeared to be more effective than Ad.CD/5-FC, which had a modest antitumor effect compared to the control (Ad.GFP). IL-12 and cytosine deaminasegenes in combination, followed by 5-FC treatment, enhance the cytolytic activity of NK cells and increase IFN-γ production -------------------------------------------------------------------------------------------------------------------------------------------------------- We examined whether the increased antitumor activity of Ad.p35 plus Ad.p40 plus Ad.CD/5-FC (Figure [4](#F4){ref-type="fig"}) was associated with the induction of tumor-specific CTLs, activation of NK cells, and induction of Th1 cytokines such as IL-2 and IFN-γ. Effector cells were prepared from the splenocytes of tumor-bearing mice treated with the recombinant adenoviruses. Renca cells labeled with ^51^Cr were used as target cells. A standard 4-h ^51^chromium release assay was carried out. However, we did not detect specific lysis of the labeled Renca cells (data not shown). We performed an NK cell assay using the splenocytes as effector cells and YAC-1 cells labeled with ^51^Cr as target cells (Figure [5A](#F5){ref-type="fig"}). The effector cells from mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC lysed the target cells about 10- and 5-fold more effectively than Ad.CD/5-FC and Ad.GFP, respectively, at a 100:1 effector-to-target cell ratio. Although Ad.p35 plus Ad.p40 treatment also increased NK cell activity compared to Ad.CD/5-FC or Ad.GFP, the cytolytic activity of splenic NK cells from mice treated with this mixture was lower than that from mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC. The effector cells used in the NK cell assay were mainly splenocytes from mice in which the tumors were completely regressed or small. This may account for the increased NK cell activity in the Ad.p35 plus Ad.p40 plus Ad.CD/5-FC treatment group compared to the Ad.p35 plus Ad.p40 treatment group. We next examined whether the virus treatments induced Th1 cytokines, which may represent innate and/or adaptive immune responses. Splenocytes from tumor-bearing mice treated with the viruses were incubated with inactivated Renca cells. The IL-2 and IFN-γ contents of the culture supernatants were measured by ELISA 24 and 48 h after incubation, respectively. We detected no IL-2 induction in splenocytes from any of the treatment groups (data not shown). However, IFN-γ production was about twice as great in splenocytes from mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC as in those treated with Ad.CD/5-FC or Ad.GFP (Figure [5B](#F5){ref-type="fig"}, p\< 0.0179 and \< 0.0394 for Ad.p35 + Ad.p40 + Ad.CD/5-FC vs Ad.GFP and Ad.CD/5-FC, respectively). Ad.p35 plus Ad.p40 treatment also increased IFN-γ, but its effectiveness was significantly less than that of Ad.p35 plus Ad.p40 plus Ad.CD/5-FC (p\< 0.0076). The results suggest that Ad.p35 plus Ad.p40 treatment, but not Ad.CD/5-FC, may enhance NK cell activity and stimulate IFN-γ production. Thus, the enhancement of NK cell activity may be partially responsible for the inhibition of tumor growth by treatment with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC or with Ad.p35 plus Ad.p40. The rapid tumor cell killing effect induced by suicide gene products induces a strong antitumor immune response, such as induction of tumor-specific CTLs and increased NK cell activity \[[@B54]-[@B57]\]. However, we detected neither a tumor-specific T cell immune response nor an enhanced NK cell activity after Ad.CD/5-FC treatment alone. It is possible that the antitumor activity of the Ad.CD constructed here was so weak that the tumor cells were not killed rapidly (Figure [4](#F4){ref-type="fig"}). However, this is unlikely, because Ad.CD/5-FC treatment increased the antitumor activity of Ad.IL-12 (see Figure [6B](#F6){ref-type="fig"} and [6C](#F6){ref-type="fig"}). Alternatively, the bystander effect, which kills neighboring tumor cells not expressing suicide genes, may have exerted a negative effect on the generation of tumor-specific CTL and/or active NK cells. Immune cells or antigen-presenting cells located close to the tumor cells could have taken up, and been killed by, the toxic metabolites generated by the suicide gene products. This may explain why the NK cell activity in Ad.CD-treated mice was significantly lower than in Ad.GFP-treated mice (Figure [5A](#F5){ref-type="fig"}, p\< 0.001). Transfer of the cytosine deaminasegene followed by 5-FC treatment increases the antitumor effectiveness of IL-12 gene transfer in Renca-tumor bearing mice ------------------------------------------------------------------------------------------------------------------------------------------------------------ Since a mixture of Ad.p35 and Ad.p40 was used in previous experiments (Figure [4](#F4){ref-type="fig"}), we examined whether Ad.IL-12, containing p35 and p40 in a single adenoviral vector (Figure [1](#F1){ref-type="fig"}), was more effective against the tumors when combined with Ad.CD/5-FC. Renca-tumor bearing mice were divided randomly into four groups, treated intratumorally with (1) Ad.IL-12 plus Ad.CD/5-FC, (2) Ad.IL-12, (3) Ad.CD/5-FC, or (4) Ad.GFP (control). Tumor sizes were measured at the indicated times (Figure [6A](#F6){ref-type="fig"}). There was a significant inhibition of tumor growth in mice treated with Ad.IL-12 plus Ad.CD/5-FC compared to Ad.GFP (p\< 0.0456). The effectiveness of Ad.IL-12 (5 × 10^8^pfu) plus Ad.CD/5-FC (5 × 10^8^pfu) appeared to be similar to that of Ad.IL-12 (1 × 10^9^pfu) alone. This contrasts with the effectiveness of Ad.p35 plus Ad.p40 plus Ad.CD/5-FC as compared to Ad.p35 plus Ad.p40 (Figure [4](#F4){ref-type="fig"}). Presumably the in vivo gene delivery of the Ad.p35-Ad.p40 mixture was not so efficient as that of Ad.IL-12. We further examined whether CD gene transfer followed by 5-FC treatment increased the antitumor effectiveness of IL-12 gene transfer. Sixteen Renca tumor-bearing mice were divided into 2 groups. Since a higher dose than 1 × 10^9^pfu of adenoviral vector appeared to be toxic to the mice, the total dose given was 1 × 10^9^pfu. Mice in each group received Ad.IL-12 (5 × 10^8^pfu) with either Ad.CD (5 × 10^8^pfu) or Ad.GFP (5 × 10^8^pfu) as a control. Both groups were treated with 5-FC for 10 days. Tumor sizes were measured at the indicated times (Figure [6B](#F6){ref-type="fig"}). Combined transfer of the IL-12 and CD/5-FC genes resulted in significant inhibition of tumor growth compared to IL-12 gene transfer alone (p\< 0.0301). Six of the 8 mice treated with Ad.IL-12 plus Ad.CD/5-FC, but only 4 of the 8 mice treated with Ad.IL-12 plus Ad.GFP/5-FC, became tumor-free and survived at 88 days. The mean survival times were 80.5 ± 20.8 days for the Ad.IL-12 + Ad.CD/5-FC group and 67.3 ± 28.3 days for the Ad.IL-12 + Ad.GFP/5-FC group; the difference is statistically significant (p\< 0.0368). Conclusion ========== Among various cancer gene therapies, immunogene therapy is a promising but challenging modality. Recent clinical trials using Ad.IL-12 have shown that intratumoral IL-12 gene transfer is a feasible and well-tolerated procedure but exerts only mild antitumor effects \[[@B24]\]. An increased Ad.IL-12 dose may increase the antitumor efficacy but might entail a systemic, toxic effect that would limit the use of Ad.IL-12 alone. In this study we have attempted to improve the antitumor efficacy in Renca tumor-bearing mice by using adenoviral vectors to transfer the IL-12 gene in combination with cytosine deaminase, followed by 5-FC treatment. The results show a marked inhibition of tumor growth and significant prolongation of survival time, suggesting that IL-12 plus CD gene transfer followed by 5-FC treatment may be an alternative modality for the human cancer treatment. Abbreviations ============= Ad: adenovirus; IL-12: interleukin-12; CD: cytosine deaminase; 5-FC: 5-fluorocytosine; DMEM: Dulbecco\'s Modified Eagle\'s Medium. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Cho W-K and Yoo J constructed the adenoviral vectors and analyzed the functionality of the expressed gene products. Hwang K-S, Yun H-J, and Kim S performed the tumor growth studies in the animal model, and the immunological assays with murine splenocytes. Im DS drafted the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/51/prepub> Acknowledgements ================ We are grateful to Yong J. Lee at University of Pittsburgh School of Medicine, Pittsburgh, PA, USA, for providing us with cytosine deaminase. This study was supported by a grant from the KRIBB Initiative Program, a grant from the 21C FrontierFunctional Genome Project from the Ministry of Science and Technology, Republic of Korea, and a grant from Samyang Genex Biotech Research Institute. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic illustration of replication-defective adenoviral vectors constructed in this study. The E1 region of adenovirus type 5 (Ad 5) was replaced by the indicated genes under the transcriptional control of the immediate early gene promoter from human cytomegalovirus (pCMV). IRES; internal ribosomal entry sequence. ::: ![](1471-2407-5-51-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The recombinant adenoviruses encoding murine p35 and/or p40 subunits express biologically active IL-12. (A) HeLa cells were transduced with Ad.GFP, Ad.p40, or Ad.IL-12 at the indicated MOIs. Cells or culture media were harvested 48 h after transduction. Cleared lysates of cells transduced with Ad.GFP or Ad.p40 were immunoblotted with anti-IL-12 antibody (left panel). The culture media from HeLa cells transduced with Ad.IL-12 or Ad.GFP were immunoprecipitated with anti-IL-12 antibody. The immunoprecipitates were dissolved in SDS sample buffer without reducing agent and subjected to SDS-PAGE, followed by immunoblotting with anti-IL-12 antibody (right panel). Molecular size markers in kilodaltons are shown at the left. The positions of p40, p35--p40 heterodimer (p70) and p40 homodimer (p80) are indicated by arrows. (B and C) HeLa cells were mock-transduced or transduced with either a 1:1 (pfu ratio) mixture of Ad.p35 and Ad.p40, or Ad.IL-12, or Ad.GFP, at the indicated MOI. Splenocytes from naïve mice were incubated with the culture supernatants for 24 h. ELISA was performed to determine the IFN-γ levels after incubation. Data are means ± SD of two independent experiments, each performed in duplicate. ::: ![](1471-2407-5-51-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Ad.CD virus expresses functional cytosine deaminase. HeLa cells were transduced with Ad.CD at an MOI of 50 and incubated for 12 h. These cells (HeLa/CD) were then mixed with naïve HeLa cells at the indicated ratios. The HeLa/CD, HeLa, or mixed cells were incubated on 24-well dishes for 24 h, then treated with 5-FC at the indicated concentrations for 4 days. Cell viability was assessed by the trypan-blue exclusion assay. Data are means ± SD of two independent experiments, each performed in duplicate. ::: ![](1471-2407-5-51-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The antitumor activity effected by IL-12 gene transfer increases in vivowhen combined with cytosine deaminase/5-FC. (A) Renca tumors were established in BALB/c mice by s.c. inoculation of 1.5 × 10^5^cells. When the tumors reached 5--6 mm diameter, the animals were randomly divided into four treatment groups (n = 15 per group) and injected once intratumorally with (1) Ad.p35 (2.5 × 10^8^pfu) plus Ad.p40 (2.5 × 10^8^pfu) plus Ad.CD (5 × 10^8^pfu), or (2) Ad.p35 (5 × 10^8^pfu) and Ad.p40 (5 × 10^8^pfu), or (3) Ad.CD (1 × 10^9^pfu), or (4) Ad.GFP (1 × 10^9^pfu). Mice treated with Ad.CD were given 5-FC (400 mg/kg) intraperitoneally daily for 10 days. Graphic representations of mean tumor growth rates are shown. Bars indicate SD. (B) Mouse survival until day 42 was recorded. Mice were sacrificed and considered as death when tumor size exceeded 1.5 cm in long and short axes. The mean survival times (± SD) were 33.2 ± 10.4 days for (1) group, 27.2 ± 10.9 days for (2) group, 21.9 ± 9.1 days for (3) group, and 19.7 ± 7.1 days for (4) group. The difference between (1) and (2) groups was statistically significant (p\< 0.0018) by Student\'s ttest. The treatment outcome is summarized in Table 1. One animal in each of the groups treated with Ad.CD and Ad.GFP succumbed to toxicity, and two mice in the group treated with Ad.p35 plus Ad.p40 were excluded owing to peritoneal metastasis. ::: ![](1471-2407-5-51-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### IL-12 gene transfer combined with Ad.CD/5-FC increases splenic NK cell activity and IFN-γ production. (A) Cytotoxicity of NK cells in Renca tumor-bearing mice treated with various adenoviral vectors (see Figure 4 legend). Mice in which tumor growth was completely or partially inhibited were chosen from the treatment groups on day 21 after tumor cell inoculation. Splenocytes were isolated from the treated mice (n = 3 per group) and used as effector cells. A standard 4-h ^51^Cr release assay was performed using YAC-1 cells labeled with ^51^Cr as target cells. The percentage specific lysis was determined. Data are means ± SD. (B) Mouse splenocytes (n = 3 per group) were isolated from the treatment groups on day 21 as described above, and incubated with Renca cells inactivated by pretreatment with 50 μg/ml mitomycin C for 20 min at 37°C. Culture supernatants were obtained from the incubated cells at the indicated times. Amounts of IFN-γ in the supernatants were measured using ELISA. Bars represent mean ± SD. ::: ![](1471-2407-5-51-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### CD gene transfer followed by 5-FC treatment increases the antitumor effectiveness of IL-12 gene transfer. (A) Mice with tumors were established by subcutaneous inoculation of 1.5 × 10^5^cells of Renca. The experimental animals in which tumors reached a diameter of about 5 mm were randomly divided into 4 treatment groups (n = 7 per group) and injected once intratumorally (day 0) with Ad.IL-12 (5 × 10^8^pfu) plus Ad.CD (5 × 10^8^pfu), or 1 × 10^9^pfu of Ad.IL-12, or Ad.CD, or Ad.GFP. Mice treated with Ad.CD were given intraperitoneal 5-FC (400 mg/kg) daily for 10 days. Tumor growth was measured at intervals of three days until day 21. Graphic representations of mean tumor growth rates are shown. (B) and (C) Renca tumor-bearing mice were divided into two groups (n = 8 per group), which received single intratumorally injections (day 0) of Ad.IL-12 (5 × 10^8^pfu) plus Ad.CD (5 × 10^8^pfu) or Ad.IL-12 (5 × 10^8^pfu) plus Ad.GFP (5 × 10^8^pfu). Mice in both groups were given intraperitoneal 5-FC (400 mg/kg) daily for 10 days. Tumor growth was measured at the indicated times until day 25. Graphic representations of mean tumor growth rates are shown (B). Mouse survival until day 88 was recorded (C). The mean survival times (± SD) were 80.5 ± 20.8 days for the Ad.IL-12 plus Ad.CD/5-FC group and 67.3 ± 28.3 days for the Ad.IL-12 plus Ad.GFP/5-FC group. The difference between these mean survival times was statistically significant (p\< 0.0368) by Student\'s ttest. ::: ![](1471-2407-5-51-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Tumor regression following intratumoral injection of Ad.p35 plus Ad.p40, or Ad.CD/5-FC alone, or a combination of both ::: Treatment Response (%) ------------------------------ -------------- -------- -------- --------- --------- Ad.p35 + Ad.p40 + Ad.CD/5-FC 15 4 (27) 4 (27) 0 (0) 7 (46) Ad.p35 + Ad.p40 15 1 (7) 2 (14) ‡2 (13) 10 (66) Ad.CD/5-FC 15 1 (7) 0 (0) †1 (7) 13 (86) Ad.GFP 15 0 (0) 0 (0) †1 (7) 14 (93) N= number of Renca tumor-bearing mice injected once intratumorally with adenoviral vectors at the doses indicated in Figure 4 legend. \"Complete\" refers to animals undergoing total and permanent regression of the injected tumor. \"Partial\" refers to animals undergoing partial tumor regression followed by regrowth, or a significant delay in growth \"Death or peritoneal metastasis\" refers to animals that died or had peritoneal metastasis during the period of investigation. None refers to animals whose tumors have continued to grow at a rate comparable to controls. †Death probably by toxicity. ‡Excluded due to peritoneal metastasis during the experiment. :::	PubMed Central	2024-06-05T03:55:58.217180	2005-5-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168891/", "journal": "BMC Cancer. 2005 May 24; 5:51", "authors": [ { "first": "Kyung-Sun", "last": "Hwang" }, { "first": "Won-Kyung", "last": "Cho" }, { "first": "Jinsang", "last": "Yoo" }, { "first": "Hwan-Jung", "last": "Yun" }, { "first": "Samyong", "last": "Kim" }, { "first": "Dong-Soo", "last": "Im" } ] }
PMC1168892	Background ========== Comparison of intraspecies multi-strain bacterial genome sequence data has shown that, even over short evolutionary time scales, genome evolution is dominated by gene insertions/deletions and gene divergence \[[@B1]-[@B4]\]. Genome levels of intraspecies genetic diversity must be examined if we are to gain a better understanding of genome evolution \[[@B5]\] and if we are to maximize the practical use of bacterial genome sequence information, for instance for development of technical applications, e.g., vaccine or drug development. One of the aims of bacterial intraspecies comparative genomics is to determine the overall genetic similarity between strains. Where sequence information is available, this type of analysis relies heavily on sequence homology and centres on the determination of conserved genes, strain-specific (i.e. unique) genes and, where the sequence provides unambiguous evidence, determination of orthologous and paralogous genes \[[@B6]-[@B9]\]. Although it has become increasingly apparent that obtaining the sequence of multiple strains per species is highly desirable, currently these types of datasets are limited in number. In their absence, other methods for performing comparative genomics have been developed. Among them, microarray-based comparative genomic hybridization (M-CGH) based on genome-sequenced strains has shown enormous potential \[[@B10]-[@B12]\]. Two different microarray-based approaches have been used to study the genetic composition of unknown bacterial strains. In the first approach, a control genome-sequenced strain was used as a reference to generate the probes for a microarray \[[@B13]-[@B16]\]. In the second approach, microarray probes were derived from the tester strain, either from a tester-derived shotgun library or a library enriched for tester-specific DNAs \[[@B17]\]. With either approach, control- and tester-derived targets are co-hybridized to the microarray and control- and tester-derived signals are compared, often by computing the Log Ratio (LR) = log~2~(tester signal/control signal). Whereas genes with similar signal in either channel are expected to have LRs near zero, genes with LRs that deviate significantly from LR = 0 are likely to show copy number changes or sequence divergence between control and tester strains. The relatively small number of studies on bacterial M-CGH has demonstrated the power of the method in a comparative genomics context despite a lack of consensus in current methods for analyzing M-CGH data. Although potential methods for standardizing and improving analysis have been suggested \[[@B15],[@B18]\] in practice, M-CGH data has routinely been analyzed by categorizing genes into two groups: genes that are likely to be conserved and genes that are likely to be divergent. One notable problem with this approach is that no attempt is made to differentiate between gene divergence and gene absence, despite the significant biological and evolutionary differences implied by these two types of events. A framework for improved analysis would require empirical data on the relationship between Log Ratio (LR) from M-CGH experiments and sequence conservation levels, however, to our knowledge no studies exist that have directly examined this question. The availability of intraspecies genome data from two strains of Campylobacter jejuni\[[@B19],[@B20]\], has provided us with the opportunity to examine the quantitative relationship between the LR and probe/target identity (PTI) using our C. jejunimicroarray. This experimental design allows us to directly match microarray results to the a prioriinterpretation of gene divergence and gene absence patterns. The goal of this study is to define the analytical parameters for the accurate prediction of gene conservation levels, leading to improved interpretation of M-CGH data. We present here the results of a detailed analysis of M-CGH experiments using the two genome-sequenced strains of C. jejuni. Results and discussion ====================== Determination of technical variation in M-CGH experiments --------------------------------------------------------- In order to examine the Log Ratio (LR) distributions where no differential signals are expected, we performed control experiments in which dual-labelled NCTC 11168 (or RM1221) DNA was tested in a series of self-self M-CGH experiments. Although our microarray is based on strain NCTC 11168 the resulting LRs should remain close to zero because, regardless of the strain used in the self-self experiment, \"Control\" and \"Tester\" targets are identical. Thus, any observed deviation from this result is likely due to technical variability in the assay and can be used to determine a threshold for statistically significant differential signals (i.e., outliers). The LR distribution in six replicates follows a normal distribution with a mean LR for six replicates of \~0.01 ± 0.22 (Figure [1](#F1){ref-type="fig"}) and, as expected, the mean and standard deviation of the various replicates were uniform regardless of the strain used. The variances observed were due to stochastic differences in the competitive hybridization of targets to the probes on the microarray and a good estimate of the technical variation in our experimental platform. Based on this data a LR = ± 1.0, used by many experimenters to identify divergent or deleted genes in similar M-CGH studies, represents a conservative threshold for divergent gene detection, since genes in which tester and control sequences are identical have a probability of less than 3.0 × 10^-6^of having a LR greater than 1.0 ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Log Ratio distribution of self-self experiments. The LR distribution of self-self experiments was used to determine the level of experimental variability in our experimental platform. Standardized samples of genomic DNA from C. jejuniNCTC 11168 (or RM1221) labelled with Cy3 and Cy5 were co-hybridized to our microarray. Results from six replicates had mean LR = 0.01 with an average SD of 0.215. Because samples from the same genomic DNA were used in both channels, LRs were expected to remain close to 0 and any deviations could be attributable to experimental error. ::: ![](1471-2164-6-78-1) ::: Analysis of the Log Ratio distribution of highly conserved genes ---------------------------------------------------------------- We analyzed data from a set of M-CGH experiments comparing strain NCTC 11168 (Control) with strain RM1221 (Tester). Because the probes in our microarray were PCR-amplified from the Control strain, Control targets should have 100% probe/target identity (PTI) with the probes on the microarray, and the LR values observed should be a function of the PTI between Tester targets and the NCTC 11168-derived microarray probes. The LR distribution of genes with 100% identity between NCTC 11168 and RM1221 (n = 114) would be expected to behave much like that of self-self experiments because in both cases Control and Tester targets are identical and thus have 100% PTI. This was found to be the case although the distribution of genes with 100% PTI had larger standard deviation (σ = 0.28) than that of self-self experiments (σ = 0.21) (Figure [2](#F2){ref-type="fig"}). Genes for which the RM1221 sequence had less than 100% sequence identity with NCTC 11168 would be expected to yield LRs that deviate from 0 due to the decreased hybridization of targets that are imperfectly matched to probes on the microarray. We examined the behaviour of genes with high levels of PTI in order to determine the level of sequence divergence that would have an observable effect on LRs. We found that whereas genes with greater than 99% PTI had LR distributions that were nearly indistinguishable from those from self-self experiments, genes with as little as 2% sequence divergence (i.e., 98% PTI and below) deviated from the LR distribution of genes with 100% PTI. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Log Ratio distributions of highly conserved genes. LR distributions from a series of M-CGH experiments comparing two genome-sequenced strains of C. jejuni(NCTC 11168 vs. RM1221). Genes were binned according to PTI and the LR distributions of bins with greater than 95% PTI are presented here. Because the microarray was designed based on strain NCTC 11168, LR deviations from 0 would be the result of sequence divergence or gene absence in strain RM1221. The LR distributions of genes with greater than 99% PTI do not deviate significantly from the average distribution of a Self-self experiment whereas increasingly larger deviations are observed in the range from 98 to 95% PTI. ::: ![](1471-2164-6-78-2) ::: Analysis of the relationship between PTI and Log Ratio ------------------------------------------------------ In order to examine the relationship between % PTI and LR in greater detail, we plotted the mean LR of genes according to their % PTI (Figure [3](#F3){ref-type="fig"}). As shown previously, the LR distribution of genes with greater than 99% PTI were similar. However, in lower PTI ranges a small yet noticeable decrease in average LR was observed. Although the small number of genes with less than 93% identity makes it difficult to obtain meaningful LR trends because of high variance, decreasing PTI still led to increasingly negative LRs. One caveat of these observations is that the LR of individual data points within a given PTI range show sufficient variability to make PTI predictions based on LR values potentially inaccurate across most of the range of PTIs. For example, although the difference in mean LR of genes with 95% PTI and 96% PTI is 0.12, their standard deviations are 0.42 and 0.39, respectively. Thus, although the average LR decreased with decreasing PTI, there is considerable overlap between the distributions. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Relationship between Log Ratio and PTI. We plotted the average LR of genes with varying levels of % PTI from M-CGH experiments comparing two genome-sequenced strains of C. jejuni(NCTC 11168 vs. RM1221). Genes were binned according to PTI and the LR distributions of individual bins are presented here. The number of observations made within each PTI bin is shown in the lower axis. As seen in Figure 2, the average LR of genes with high levels of PTI are very similar, although a noticeable decrease in average LR is observed even at 98% PTI. Although the average LR becomes increasingly negative as PTI levels drop, given the SD observed within each group PTI category, there is considerable overlap between categories. ::: ![](1471-2164-6-78-3) ::: Analysis of the Log Ratio distribution of absent genes ------------------------------------------------------ Genes that are absent in RM1221 (ie. 0% PTI) should have highly negative LRs because they should yield detectable hybridization signal in the Control channel coupled with a lack of signal in the Tester channel. Although the LR distribution of genes with 0% PTI was shifted towards the left (Mean = -2.34 ± 1.35), it also appeared to be bimodal, with a number of genes with higher than expected LR. When these genes were examined more closely, a common feature was a short microarray probe size (\< 250 bp). We plotted the LR distribution of genes with probe sizes \<250 bp and \>250 bp separately, and found a significant difference in their respective LR distributions (Figure [4](#F4){ref-type="fig"}). Whereas the LR distribution of the former was -0.93 ± 0.82, the LR distribution for longer genes was -2.94 ± 1.04. This \"dampening\" in LR amplitude appears to be largely the effect of an overall diminished signal for short genes (results not shown), possibly due to a difference in hybridization kinetics or hybrid stability under the hybridization and washing conditions used. The effect of decreased signal is that of decreased dynamic range because a lower signal in the control channel restricts the amplitude of the LR that can be measured. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Log Ratio distribution of genes absent from strain RM1221. We plotted the LR distribution of genes predicted to be absent from strain RM1221 based on BLAST searches of NCTC 11168 against the RM1221 genome. Because of the lack of Tester signal predicted from these genes, LRs should be expected to be highly negative. The LR distribution (A) appeared to be bimodal with a significant number of genes bearing unusually high LRs. Further examination of these genes revealed a potential bias towards genes represented by short probes on the microarray (i.e. less than 250 bp). Separate re-plotting of the LR distributions of long (B) and short (C) probes confirms a higher average LR among genes with short probes. ::: ![](1471-2164-6-78-4) ::: Determination of thresholds for highly conserved and absent genes ----------------------------------------------------------------- One of our goals for this analysis was to determine whether the observed trends would enable us to predict the PTI in M-CGH experiments based on LR alone. Although the levels of technical variability mask the subtle effect that low levels of sequence divergence have on LR, the LR distributions at the two PTI extremes (\>98% and 0%), which correspond to highly conserved and absent genes, show very little overlap. This enabled us to establish thresholds that, with high confidence, can be used to predict absent and highly conserved genes in M-CGH data (Figure [5](#F5){ref-type="fig"}). After removal of genes with short amplicon-based probes from our analysis, we established that less than 1% of the observed LRs \> -0.8 originated from absent genes (31 of n = 7910). Similarly, less than 1% of observed LRs \< -3.0 originated from conserved genes (6 of n = 636) with all false-positive observations stemming from the pyrCgene, which has a PTI of only 81.2%. Of 808 observed LR measurements in the range between -3.0 and -0.8, only 221 (27.3%) originated from genes with greater than 90% PTI. Although, based on our empirical data, LR values that fall between these two thresholds are likely to be from either absent or significantly divergent genes and unlikely to be from highly conserved genes, there is significant overlap between LR distributions of absent and divergent genes. At LRs ≅ -1.4, an observation has a nearly equal likelihood of stemming from an absent gene as it does from a present gene and thus the two classes cannot be distinguished in this LR range. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Determination of thresholds for absent and conserved genes. We calculated the proportion of genes belonging to each of four PTI categories at 0.2 LR intervals in order to determine LR thresholds that could be used to predict absent and conserved genes with a high degree of certainty. Below a LR of -3.0, the false positive rate for conserved genes is less than 1%; similarly, the false positive rate for absent genes above LRs of -0.8 is also less than 1%. In the LR interval between -3.0 and -0.8, particularly approaching the -0.8 boundary, there are significant number of genes from more than one PTI category and thus there is significant risk of misclassification. ::: ![](1471-2164-6-78-5) ::: Conclusion ========== Microarray analysis, whether in the context of gene expression or M-CGH studies, is based on determining which genes have statistically significant differential hybridization signal between two samples. In M-CGH analysis, these differential signals are the result of sequence divergence or differences in copy number. Two critical issues rise to the forefront in M-CGH analysis: a) does a gene show genuine differential signal (i.e. outside the norms of variability due to experimental error); b) what is the nature of the event that gave rise to the differential signal (i.e. sequence divergence, copy number change)? Because M-CGH generates hybridization data as a proxy for sequence similarity data, it is important that it be analyzed as such. While some empirical work has been carried out on probe/target identity (PTI) and data analysis using the microarray platform, the focus has largely been on optimization of species detection and/or identification in complex samples \[[@B21]-[@B23]\]. In these applications, the primary goal is that of optimizing probe sets and hybridization conditions to maximize the specificity of species-specific probe/target interactions, possibly at the expense of decreased assay sensitivity and thus the majority of microarrays used for species identification are oligonucleotide-based. By contrast, in comparative genomics, the primary goal is that of gene detection for the purpose of characterizing gene content, and thus the focus must shift to detection sensitivity in order to minimize the likelihood of false positive calls on gene absence events. Because oligonucleotide-based arrays can lead to erroneous gene absence calls \[[@B24]\], the majority of M-CGH studies have used amplicon-based microarrays, which are more sensitive albeit at the expense of specificity \[[@B25]\]. A common thread among bacterial M-CGH studies has been the grouping of all outliers into a single category. Currently it is unclear whether divergent and absent genes can be distinguished based on LR data alone. Although the lack of distinction between these types of events does not negate the results from these studies, it can potentially restrict further analysis of the data. For example, in any pair of intraspecies genomes, sequence similarity can be used to define genes absent in one or the other strain as well as genes that are conserved in both strains. Although the \"biological interpretation\" in the case of gene absence is unambiguous, many possibilities arise when sequences share any level of similarity. For instance, single nucleotide substitutions can lead to truncated or inactive gene products. Additionally, the level of sequence similarity required for full functional homology varies from gene to gene, increasing the complexity of the analysis even when DNA sequences are directly available. The inexact nature of hybridization analysis further compounds the difficulty in interpreting signal from divergent genes by M-CGH, and thus focusing on conserved and divergent genes ignores the increased reliability of gene absence calls. In previous work, we presented data suggesting that highly negative LR values were consistent with gene deletion events, paving the way for making the distinction between divergent and absent genes based on LR data \[[@B16]\]. When M-CGH data is analyzed such that gene absence events are grouped together with all other gene divergence events (i.e. as a continuous character model), it represents a significant loss of information both from a technical and from a biological point of view. In addition to the greater ambiguity in data interpretation as LRs approach the threshold for gene conservation, the continuous character approach negates the functional distinction that can be made between gene absence and gene divergence events. Because the LR thresholds described here could be used to reliably predict gene absence and gene conservation, it would be advantageous to focus the analysis of M-CGH data on the accurate detection of conserved and absent genes. While the data between the two thresholds should not be altogether discarded, the two thresholds represent boundaries defining regions in which gene absence and gene presence can be predicted with high confidence and thus should be given greater weight in subsequent analytical steps. It is important to note that the exact value of the LR thresholds presented here is specific to our experimental platform. The prediction accuracy achieved was remarkably high because of the uniform levels of variance across the multiple replicates analysed and because of the high correlation coefficients between replicates (the average ρ ≅ 0.92). This dataset was highly idealized because the relatively small number of replicates was carried out in such a way as to minimize technical variation. Nevertheless, a previous study in which we applied the thresholds described here on a large dataset showed that LRs below our \"absence threshold\" correlated very highly with other potential indicators of gene deletion \[[@B16]\]. Given the many documented sources of technical variability that can influence microarray results (e.g. variation in handling between individual investigators, laboratory conditions, microarray print batches), thresholds for gene presence/absence detection should be calibrated to the differential levels of technical variance found in individual microarray experiments, especially in large datasets. Kim et al \[[@B18]\] have suggested a solution to array-specific variance and normalization bias by determining thresholds specific to each array based on the point at which the LR distribution deviates from its inferred normal distribution. In practice, we have found that this approach can be susceptible to \"narrow\" LR distributions, leading to relaxed thresholds that yield an increased number of false positives for gene divergence. An alternative approach to deal with unequal variances and normalization biases across a dataset is based on normalizing multiple microarrays using the Z-score transformation \[[@B26],[@B27]\], in which LR values are divided by the standard deviation of the LR data distribution. Z score-based metrics could be used to replace Log Ratio-based metrics, enabling direct comparisons that are more valid because data from each microarray is \"variance-calibrated\". Based on the higher than expected Log Ratio values obtained in the case of absent genes, the \"relative accuracy\" of Log Ratio measurements obtained from short probes is significantly compromised under the standard hybridization conditions we used. It is important to note however, that based on the average standard deviations observed (\< 250 bp = 0.68; \> 250 bp = 0.85), results obtained from short probes do not lack precision compared to those obtained from longer probes. Nevertheless, our results show that data obtained from short probes yield anomalously high Log Ratio values. It is only because our assay represents a closed system in which all components are known that we were able to determine that short probes can significantly underestimate Log Ratio measurements. These results would not have been readily apparent in a typical experiment since there would be no a prioriknowledge on expected Log Ratio values. Although these results were obtained in a series of CGH experiments, the anomalous Log Ratio data from short probes is likely to be encountered under any type of microarray hybridization experiment, including gene expression-profiling experiments. Although longer probes performed better in our assay, this is likely a result of the higher signal intensity obtained with long probes relative to short probes. Optimal hybridization and scanning conditions for long probes would likely be sub-optimal for short probes, leading to decreased signal and a concomitant drop in Log Ratio amplitudes. Thus the problem resides not in probe length per se, but rather in the mixed probe lengths encountered in our microarray. These results have important implications towards microarray probe design because the adverse probe-length effect could be mitigated through standardizing probe length. Failing that, it would be advantageous to incorporate probe length effects into any analytical framework. The results presented here have been used to examine the relationship between LR and sequence conservation. The variability inherent in hybridization-based approaches makes it unlikely that LR data from M-CGH experiments can be used to accurately predict the level of sequence identity among divergent genes. In view of the considerable ambiguity in interpreting the significance of gene divergence even when sequence information is available, the focus on gene divergence in M-CGH studies must be re-assessed. We have established thresholds for the use of LR values for the accurate detection of highly conserved and absent genes, which should increase the robustness of downstream data interpretation and should extend the range of biological interpretation of M-CGH data. An accurate determination of conserved and absent genes should increase the accuracy of strain genotyping, metabolic pathway prediction, and determination of conserved targets for vaccine or drug development from M-CGH data. Methods ======= Bacterial strains and genomic DNA isolation ------------------------------------------- Strain RM1221 was obtained from Food Safety and Health Research Unit, USDA. Strain NCTC 11168 was obtained from the American Type Culture Collection (Mannassas, VA). Genomic DNA isolation was carried out as previously described \[[@B16]\] Construction of a C. jejuniNCTC 11168 open reading frame DNA microarray ------------------------------------------------------------------------- The DNA microarrays used in this work were previously described in \[[@B16]\]. Additional information can be obtained at \[[@B28]\]. Microarray hybridizations ------------------------- Genomic DNA was sheared to a mean fragment size of 1.5 Kb by nebulization in 35% glycerol at 15 PSI for 45 seconds as described by \[[@B29]\]. For each sample, 5 μg of sheared DNA were fluorescently labelled using direct chemical coupling with the Label-IT (Mirus Corp., Madison, WI) cyanine dyes Cy3 and Cy5 as recommended by the manufacturer. Probes were purified by sequentially passing samples through SigmaSpin (Sigma, Oakville, ON) and Qiaquick (Qiagen, Mississauga, ON) columns. Labelled DNA sample yields and dye incorporation efficiencies were calculated using a Nanodrop ND-1000 spectrophotometer (Nanodrop, Rockland, DE). Microarray hybridizations were set-up by co-hybridizing 2 ug of differentially labelled genomic DNA samples and were carried out as previously described \[[@B16]\]. NCTC 11168 versus RM1221 hybridizations were carried out in triplicate. A set of dye-swap experiments was also carried out, giving a total of 6 replicate experiments. Self-self hybridization experiments were carried out in which separate NCTC 11168 (or RM1221) genomic DNA samples were labelled with each Cy-dye and co-hybridized to the array. Data acquisition and analysis ----------------------------- Microarrays were scanned and analyzed as previously described \[[@B16]\]. Briefly, microarrays were scanned using a Chipreader laser scanner (BioRad, Mississauga, ON) according to the manufacturer\'s recommendations. Spot quantification, signal normalization and data visualization were performed using the program ArrayPro Analyzer (version 4.5; Media Cybernetics, Silver Spring, MD). Net signal intensities were obtained by performing local-ring background subtraction and spots with a signal less than 5-times above background were excluded from the analysis. Signal intensities for replicate spots were averaged and data from each channel were adjusted by sub-array normalization using cross-channel Loess regression. The ratio of tester signal to control signal for each gene was transformed to its base 2 logarithm \[[@B30]\], log~2~\[Tester Signal / C. jejuniNCTC 11168 Signal\], hereafter referred to as \"Log Ratio\" (LR). LRs from the two \"within slide\" spot replicates were averaged. To increase the number of observations for statistical purposes, LR data from each microarray replicates were analyzed separately. Determining level of sequence identity between probes and targets (PTI) ----------------------------------------------------------------------- We used the BLAST software package \[[@B31]\] to determine the identity between microarray probes and predicted target sequences. Complete genome sequence information for C. jejuniNCTC 11168 and C. jejuniRM1221 was downloaded from the National Center for Biotechnology Information\'s Prokaryotic Genomes Database \[[@B32]\], GenBank records AL111168 and CP000025, respectively. We created BLAST databases from the nucleotide sequences of the open reading frames in each C. jejunigenome strain and queried them with the nucleotide sequences of each probe in our microarray using the BLASTN program. The percent identity of the best hit for each subject/query pair was determined. List of abbreviations ===================== M-CGH: microarray-based comparative genomic hybridization; LR: Log Ratio; PTI: Probe/Target Identity; SD: standard deviation (or σ) Authors\' contributions ======================= ENT designed M-CGH experiments, carried out downstream data analysis, and drafted the manuscript. RRA assisted with downstream data analysis. WAF carried out BLAST analysis of the two sequenced genomes. CCL carried out hybridizations and performed upstream data analysis. JHEN participated in the conception and supervised the design of the study and writing the manuscript. All authors submitted comments on drafts and read and approved the final manuscript. Acknowledgements ================ Funding for this work has been provided to ENT, RRA, WAF and JHEN through the National Research Council\'s Genomics and Health Initiative, and to CCL through Health Canada.	PubMed Central	2024-06-05T03:55:58.220342	2005-5-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168892/", "journal": "BMC Genomics. 2005 May 27; 6:78", "authors": [ { "first": "Eduardo N", "last": "Taboada" }, { "first": "Rey R", "last": "Acedillo" }, { "first": "Christian C", "last": "Luebbert" }, { "first": "Wendy A", "last": "Findlay" }, { "first": "John HE", "last": "Nash" } ] }
PMC1168893	Background ========== Due to technological developments, better communication facilities and further differentiation and specialisation of professionals, patient care is increasingly provided outside hospitals. Financiers, governments and patients advocate this development, as they consider outpatient care to be more efficient and patient centred \[[@B1]\]. This shift has repercussions for chronically ill and elderly patients with complex care needs. First, it is questionable whether optimal care for these patients can be delivered in an outpatient setting. They need sufficient time to recover and are often in frail health \[[@B2]\]. Early hospital discharge can present dangers for them. Moreover, alternatives are often scarce, as the capacity of long-term care facilities is either limited or too expensive. What is known as \'bed-blocking\' occurs when patients whose medical treatment has been completed cannot be discharged because of poor continuation of care outside the hospital \[[@B3]\]. Second, these patients often need care from health professionals in multiple settings. Consequently, patient care journeys encompass multiple transitions from one setting to another \[[@B4]\]. Systems of care often fail to organise these transitions, thus influencing the quality of care \[[@B5]-[@B7]\]. Bearing these issues in mind, a plethora of care models have been developed to substitute hospital inpatient care and to improve transitional care \[[@B8]\]. One model can be described by the term \'intermediate care\'. This model refers to a range of services intended to bridge acute, primary and social care. It is considered to serve goals like reducing the length of hospital stays, preventing hospital admissions and readmissions, improving transitions from hospital to consecutive settings, and retaining people\'s independence as long as possible \[[@B9]\]. Still, the benefits and transportability of intermediate care are ambiguous \[[@B9]-[@B12]\]. This can be attributed to the blurred definition of the concept and the wide range of services labelled as such. Models differ in focus, setting, case mix, staffing, professionals involved, commissioning and context. Hence, it is quite difficult to define the concept, to identify best practices and to compare different settings. Against this background, it is argued that the \'black-box\' of intermediate care (i.e. the processes of care) should be opened \[[@B13]\]. Process evaluations are therefore warranted and promoted \[[@B14],[@B15]\]. This paper provides such a process evaluation of an intermediate care model. The aim was to provide management with information to assess and improve the quality of care. Even so, it should also provide more generalisable insights into the accumulated impact of the model. In the Dutch health care system, a distinction is made between the financing and organisation of acute and long-term care. Acute care is financed either by social health insurance (covering about 60% of the population) or private health insurance (covering about 40% of the population). Compulsory national health insurance covers the whole population for long-term care. Health care providers organise their services in networks that are covered by either scheme. Cooperation and coordination of care takes place predominately within the acute care sector. But health care delivery is fragmented in the transition from acute to long-term care. To bridge the gap, various intermediate care models are being set up in hospitals, nursing homes and residential homes \[[@B16],[@B17]\]. However, due in part to the division of the insurance schemes, financing is often ad hoc and available only for the duration of a single project. The model evaluated in this study eventually became structurally embedded in the local health system of the South-eastern Amsterdam district. The Henriëtte Roland Holst House (HRHH), a residential home, and the Academic Medical Center of the University of Amsterdam (AMC) agreed to establish a low intensity early discharge model of intermediate care encompassing transitional care as well as a transfer unit. This model focuses exclusively on all AMC patients who no longer require hospital treatment but are not healthy enough to be discharged to their home situations. It supplements two more intensive rehabilitation models of intermediate care in nursing homes for stroke and orthopaedic AMC patients. Table [1](#T1){ref-type="table"} describes the model and its local context. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of the intermediate care model ::: Focus Starting a \'transfer unit\' in a residential home for AMC patients whose medical treatment has been completed, but are unfit to be discharged to their homes. The unit should serve as a substitute hospital ward that relieves the problem of \'bed-blocking\' in the AMC and improves transitional care to the home situation. -------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Admission criteria All AMC patients are eligible for admission to the transfer unit if they meet the following criteria: \- Patient is medically stable and curative treatment has been completed; \- Patient needs care that can be delivered by one nursing assistant; \- Patient is not eligible for other regular care services and cannot go home; \- Patient is insured; \- Patient does not need daily care and/or intensive physical therapy; \- Patient is not a drug addict, terminally ill or comatose, and does not have AIDS; \- Patient does not exhibit disturbing behaviour if he or she is a psychiatric or psychogeriatric patient. \- Patient has an official indication for discharge to a consecutive setting. Transitional care Three AMC liaison nurses control, plan and coordinate all transitions of AMC inpatients to the transfer unit systematised by agreed discharge procedures. The nursing home physician, occupational therapist, the liaison nurse and the head (an RN) assess whether an AMC patient will be admitted to the transfer ward. Setting 20 transfer beds located in 10 rooms. The unit was established outside the AMC in a residential home in the South-eastern Amsterdam district. This institution accounts for 110 residential home places, 7 places for day care, 4 community health beds and 218 apartments for assisted living. Staffing Head of the transfer unit 1.0 FTE; nursing home physician 0.33 FTE, registered nurses 0.89 FTE; liaison nurses 0.5 FTE; occupational therapist 0.5 FTE; licensed practice nurses 11.61 FTE. Two physiotherapists with a practice in the residential care home are directly available for patients of the transfer unit. An AMC geriatric nursing specialist attends multidisciplinary meetings once a week. Context The AMC and Henriëtte Roland Holst House are located in the South-eastern Amsterdam district. This region accounts for approximately 85,000 residents of whom 7,000 (8%) are older than 65, and 61% belong to an ethnic minority. A number of institutions in the region provide care for the elderly: 1 AMC, 1 nursing home, 4 residential homes, 1 public home-care agency, 1 public health agency, 1 social care agency, 5 primary care centres and 1 institution for psychiatric care. Commissioning The local public insurer structurally finances the transfer unit. The annual budget is 758,205 euros. Transitional care is financed by the AMC budgets. ::: Methods ======= Aim --- The intermediate care model was evaluated at the request of the collaborating institutions. During the course of 2000, both partners expressed concerns about how the model was functioning, and needed information to make informed decisions on how to improve the quality of care. This implied that the evaluation should be conducted from a \'managerial evaluation perspective\' \[[@B18]\]. The following questions were addressed: 1) Is the patient population admitted to the model in accordance with the ex anteexpectations of key players and staff members? 2) How does the model ensure quality of care? Design ------ A process evaluation based on quantitative and qualitative methods was considered the most appropriate and feasible way of answering both questions. Registration forms and patient questionnaires were used to quantify the patient population. Semi-structured interviews were conducted to explore the expectations and experiences of the key players and staff members, as well as to describe how the quality of care is ensured. The project proposal was reviewed by the medical ethical committee of the AMC in August 2000 and was considered not in need of formal approval according to the Dutch legislation on experiments with human beings. There were no ethical objections raised against the study. Patient questionnaires and registration forms (quantitative analyses) --------------------------------------------------------------------- ### Sample Initially, all candidates (n = 189) for admission to the transfer unit between 1 October 2000 and 31 October 2001 were in the study. RNs working in the AMC wards selected candidates after consulting the individual patients and their families. Liaison nurses at the AMC collected the submitted applications and sent them to the transfer unit for assessment. Using formalised admission criteria (Table [1](#T1){ref-type="table"}), the nursing home physician assessed whether a candidate could be admitted to the transfer unit. 162 out of 189 (85.7%) candidates met the admission criteria and 27 out of 189 (14.3%) did not. As researchers, we did not interfere in this process. During the assessments, we requested informed consent from patients participating in the study. There were two significant differences between consenting (n = 70) and non-consenting patients (n = 119). The age distribution differed significantly between both groups (chi^2^= 12.51; p = .03), and the youngest and oldest patients were less willing to participate. Moreover, consenting patients were less likely to be refused for admission than those who did not consent. Just 2 out of 70 consenting patients (3%) were refused, while this rate was 27 out of 119 (21%) for non-consenting patients. This selective non-response can be explained by health status. Patients who did not meet the admission criteria were more seriously ill and less inclined to participate in the study. Of the 68 positively assessed patients who consented, 54 were actually admitted. During the assessment period 2 patients died and 12 were discharged to another destination. The sample is shown in figure [1](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow chart ::: ![](1472-6963-5-38-1) ::: ### Data collection Two registration forms and two patient questionnaires measured process parameters. The first form, filled in before the assessment procedure was started, recorded \'medical diagnosis\', \'reason for hospital admission\', \'reason for application\' (rehabilitation, waiting for admission to a nursing home, waiting for admission to a residential home, oncology, wound therapy, other), and \'name of the submitting AMC ward\'. After the assessment procedure \'refusal or admittance to the transfer unit\', \'date of hospital admission\', \'date of hospital discharge\', \'discharge destination\' (transfer unit, home, nursing home, residential home, other) were registered. The second form was filled out at discharge from the transfer unit. In this form \'destination of discharge\' (home with or without home care, nursing home, residential home, hospital, other), \'experienced burden of care in relation to the expected burden of care based on the application\' (the head of the unit gave his or her assessment), \'whether the patient was in the appropriate place\' (the head of the unit gave his or her assessment), and \'care delivery problems experienced\' were registered. During the assessment procedure, the patients filled out two questionnaires, assisted by a researcher if necessary: the 36-item short-form health survey (SF-36) and the Groningen Activity Restriction Scale (GARS). The SF-36 includes one multi-item scale measuring eight health concepts: 1) limitations in physical activities because of health problems; 2) limitations in social activities because of physical or emotional problems; 3) limitations in usual role activities because of physical health problems; 4) bodily pain; 5) general mental health (psychological distress and well-being); 6) limitations in usual role activities because of emotional problems; 7) vitality (energy and fatigue) and 8) general health perceptions \[[@B19],[@B20]\]. The GARS aims at measuring both the ADL and instrumental ADL disability in community-based studies with respect to the aid and services provided by professional home help and district nursing agencies \[[@B21],[@B22]\]. ### Data analysis We analysed the quantitative data using SPSS 10.1. We used T-tests and chi-squared test statistics to assess significance. Semi-structured interviews (qualitative analyses) ------------------------------------------------- ### Sample We interviewed 21 key players and staff members selected \'purposively\' for their positions, disciplines and institutions. All disciplines involved were represented in the study. In addition, nursing assistants were selected on the basis of gender, age, ethnicity and work experience. It was not feasible to interview them all. This approach resulted in the sample presented in table [2](#T2){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Interviewed professionals ::: Respondent Position Institution ------------ ---------------------------------------------- ------------- Nr 1. General manager HRHH Nr 2. Director of integrated care AMC Nr 3. Head of care department HRHH Nr 4. Chair board of directors HRHH Nr 5. Chair medical specialist staff AMC Nr 6. Nursing home physician HRHH Nr 7. Liaison nurses / head discharge unit AMC Nr 8. Liaison nurse AMC Nr 9. Liaison nurse AMC Nr 10. Geriatric nurse specialist AMC Nr 11. Liaison nurse / occupational therapist HRHH Nr 12. Registered nurse internal medicine AMC Nr 13. Registered nurse / Head of the transfer unit HRHH Nr 14. Occupational therapist AMC Nr 15. Physical therapist HRHH Nr 16. Physical therapist HRHH Nr 17. Nursing assistant transfer unit HRHH Nr 18. Nursing assistant transfer unit HRHH Nr 19. Nursing assistant transfer unit HRHH Nr 20. Nursing assistant transfer unit HRHH Nr 21. Nursing assistant transfer unit HRHH ::: ### Data collection We interviewed the respondents at their places of work using an interview guide (see figure [2](#F2){ref-type="fig"}), which was developed around the research questions. During the interviews, the guide was used in an informal and flexible way in order to prevent the researchers from imposing their own preconceptions. The interviews took approximately one hour each; they were recorded and later transcribed. One researcher coded the transcripts and wrote memos to systematise the analysis. We completed the data collection after finishing 21 interviews -- all disciplines working with the model were then represented and no new findings were expected. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Interview guide ::: ![](1472-6963-5-38-2) ::: ### Data analysis We used open and framework approaches to analyse the interviews. Building on respondents\' perceptions, we conceptualised the dynamics underlying the admittance of patients to the transfer unit. To interpret quality assurance activities and their implementation, a theoretical framework was explicated. On the basis of a typology of quality systems consisting of five elements (structural assets, allocation of responsibilities, protocols, information transfer and monitoring/feedback cycles) we identified existing quality assurance activities in the model. We contrasted respondents\' notions on the implementation of the model with recent knowledge on effective implementation \[[@B23],[@B24]\]. Rigour ------ To monitor and consider the rigour of the interviewing process, we used several strategies to rule out validity threats \[[@B25]\]. First, due to our sampling strategy we were able to identify respondents who gave socially desirable answers. We noticed that three of them were much too positive about the functioning of the transfer unit and were defensive in their responses. Second, we conducted member checks by asking respondents to validate transcripts and interpretations of their interviews. Respondents had few actual corrections, which we adopted without discussion. Third, we aimed for \'triangulation\'. We verified respondents\' essential statements by contrasting them with our quantitative data, local documents and/or international and Dutch literature. Finally, we solicited feedback from a variety of senior and other researchers (peer review). The researchers in the team systematically monitored the data collection, analysis and emerging findings. Colleagues at the Department of Social Medicine and the Dutch National Institute of Health Services Research reviewed earlier drafts of the manuscript. Results ======= Foreseen versus actual/perceived patient population --------------------------------------------------- The target population was described in the initial intermediate care model \[[@B26]\]. AMC patients whose medical treatment had been completed but who were unfit to be discharged to their homes -- the bed-blockers- were eligible for admission to the unit. Admission criteria further specify this target population (table [1](#T1){ref-type="table"}). Quantitative as well as qualitative data were used to verify whether members of this target population were actually admitted to the unit. The quantitative profile of the patient population -------------------------------------------------- During the 13 months of the study, 189 candidates were assessed (table [3](#T3){ref-type="table"}). The majority were female, single and older than 65. Apart from this, the profile of these candidates was more heterogeneous. Although the most prevalent diseases were cardiovascular diseases and cancer, candidates suffered from a variety of different diseases. This was also shown by the diversity of the submitting clinical wards and the various reasons for application. A relatively small number of candidates were waiting for placement in a nursing home (n = 33) or residential home (n = 3) and could be identified as bed-blockers. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Characteristics of assessed candidates ::: Process parameter Outcome ------------------------- -------------------------------------------------- Gender Male n = 72 (38.3%) Female n = 117 (61.7%) Age distribution \< 65 years n = 34 (18.3%) 65--69 years n = 21 (11.3%) 70--74 years n = 19 (10.2%) 75--79 years n = 40 (21.5%) 80--84 years n = 33 (17.7%) \> 85 years n = 39 (21.0%) missing n = 3 Home situation Living alone n = 160 (84.7%) Living with a partner n = 29 (15.3%) Medical diagnoses Cardiovascular diseases n = 44 (23.3%) Cancer n = 37 (19.4%) Other n = 108 (57.3%) Reasons for application Recovery after surgery n = 30 (17.4%) Rehabilitation n = 65 (37.8%) Waiting for a nursing home n = 33 (19.2%) Waiting for a residential care home n = 3 (1.8%) Oncology therapy n = 10 (5.8%) Other n = 31 (18.0%) Missing n = 17 ::: The \'Health-related quality of life\' (SF-36) and the \'Activities of daily living\' (GARS DL-scale) were filled out by 59 of the 70 participating patients. Their scores were comparable with those of a population in a British geriatric day hospital and a population in a Dutch community GP ward for recuperating elderly people \[[@B27],[@B28]\]. The 70 patients participating in the study had an average length of stay (LoS) in the AMC of 31.8 days (range 3--109; stdv 22.2). The 54 out of the 70 patients participating in the study who were actually admitted to the transfer unit stayed there for an average of 46.5 days (range 0--340). This average lies within the maximum of three months LoS. Looking back on their stays, the head of the transfer unit considered 12 of the 54 patients (22.2%) to have needed more care than had been expected, and 5 patients (9.3%) to have needed less care than had been expected. For 6 of the 12 more serious patients (11.1%) and for 3 of the 5 less serious ones (5.6%), the head also considered the transfer unit to have been inappropriate. The final discharge destination was home (n = 18; 33.3%), home with home care (n = 18; 33.3%), hospital (n = 6; 11.1%) or elsewhere (n = 12; 22.2%). A more complex patient load than anticipated -------------------------------------------- Respondents felt that a patient population with a heavier burden of care than anticipated had been admitted to the transfer unit. Respondents experienced a limited inflow of bed-blockers: The focus of the model was to reduce the problem of bed-blocking in the AMC. Now we notice we admit hardly any bed-blockers to the transfer unit (...). In practice it seems there are no bed-blockers in the AMC. (respondent 10) This perception was difficult to verify in the quantitative data. Based on the discharge destinations, just 36 out of 189 candidates (19%) could be considered as bed-blockers. This seems to be opposed by the average LoS in the AMC of 31.8 days, which is three times the average LoS of patients in the AMC in 2001 (9.4) \[[@B29]\]. However, the high standard deviation (22.2) and the median of 24 indicate that outliers are increasing the average LoS. Moreover, average LoS is an indicator of bed-blocking, not a valid measure. To support their perception, respondents put forward two explanations. First, respondents felt the target population envisioned in the design of the intermediate care model either did not exist in the AMC or was smaller than expected. An initial assessment of the size of the target population did not take place. This implies that the relevance of the intermediate care model may have been overestimated from the outset: You instinctively know the size of this patient group (...). As far as I know, there were no data available \[during the planning of the transfer unit\]. (respondent 8) Second, respondents suggested that the assessment procedure took too long. The model is only appropriate during a short period of patients\' care episodes because their care needs vary. Therefore, the transition from the AMC to the unit must be flexible and fast, and this was not achieved in practice. The period between application and discharge took on average 8.7 days (range 1--39). Moreover, in the study group, 14 out of 68 patients meeting the admission criteria were not discharged to the transfer unit. In the meantime, patients died (n = 2), went home (n = 5) or were admitted to another institution (n = 7). This implies that 20% of the inflow was cancelled during transitional care processes. Respondents considered the main reason for delays to be the limited availability of the nursing home physician (only three afternoons a week), whose authorisation was needed for admission. Another reason mentioned for delays was discharge planning in the AMC. Time was sometimes lost waiting for the medical application/referral forms. Due to the limited inflow of targeted patients, filling transfer beds became problematic. To avoid empty beds, respondents felt that admission criteria had been applied subjectively. In their opinion, restrictive application of inclusion criteria was not in the financial interests of the collaborating institutions: There is a negative spiral nobody in health care can ignore. (\...) On the one hand you must deliver a certain volume of care. On the other hand you have your human resources. When you fail delivering this volume of care you will loose personnel, as you earn not enough money. (\...) What happens? You can admit even less patients. So, you must compromise. (respondent 6) The HRHH maximised production to prevent budget reductions. The AMC optimised the turnover of patients to be in a better position for the annual budget negotiations with insurers. These inverse incentives may have resulted in admitting patients with a heavier burden of care to the transfer unit: I think the patient population admitted to the transfer unit has a heavier burden of care than was originally expected. (respondent 8) The medical aspects become more serious. Although the patients\' medical treatment has been completed, their health is still frail. (respondent 6) This perceived pattern was partly supported by quantitative data. According to the official hospital statistics bed-blocking days increased from 2,818 days in 2000 towards 3,315 days in 2001 while bed-occupancy rates decreased from 61.5% in 2000 to 57.5% in 2001 \[[@B29]\]. This may indicate that hospital discharge of bed-blockers is postponed to maximise bed-occupancy rates, which supports the hypothesis of the respondents. However, the severity of the self-reported health status of the patients (SF-36) did not change during the study. Quality of care --------------- Almost all respondents felt there was insufficient quality of care assurance and questioned the functioning of the model: It works, but if a few things go wrong it doesn\'t work anymore -- then the quality of care goes down fast. (respondent 11) It doesn\'t function like it should, but in the past it was worse. (respondent 17) Although we cannot verify these quotes, we assume that the intermediate care model functions poorly because it was mentioned by the majority of the respondents. Two key players (respondents 6, 10) gave an overall explanation for the poor functioning of the model. They highlighted contextual differences between the HRHH and the AMC as the main stumbling blocks. Setting up and implementing intermediate care requires a certain level of know-how and expertise. In the AMC, these kinds of requirements could easily be met, while in the HRHH they could not. Initiators took this difference for granted, thereby overestimating the organisational capacity of the HRHH: I don\'t think all of the consequences \[of setting up a transfer unit\] were foreseen\...There was no experience available of caring for these patients in another setting. (respondent 6) This overall explanation is supported by respondents\' thoughts on the implementation and on the nursing staff mix. Because the interviews were planned to take place throughout the study period, we noticed during the course of 2001 that all planned working processes and quality assurance activities were put into place. The respondents interviewed early in the study reported fewer activities than those interviewed later. Table [4](#T4){ref-type="table"} presents the processes and activities categorised in the five dimensions of a quality system. Although these activities confirmed the existence of quality of care assurance practices, the rather late implementation suggested otherwise. Respondents said that people don\'t work enough according to the agreed working processes and quality assurance practices, and concluded the implementation process was flawed: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Quality assurance activities in the transfer unit ::: Structural assets \- Description of required staff ----------------------------------------- ----------------------------------------------------------------------------- Allocation of responsibilities \- Job descriptions \- Job assessment interviews Protocols \- Description of the target population \- Admission criteria \- Discharge criteria \- Routing of the patients using a flow chart \- Nursing care plans Information transfer and record-keeping \- Transfer procedures from the AMC to the HRHH \- Patient record \- Handover procedures during shifts Monitoring and feedback cycles \- Steering group meetings \- Weekly multidisciplinary meetings \- Supervision by an AMC geriatric nursing specialist \- Patient satisfaction questionnaire upon discharge from the transfer unit \- Training and education \- Management information system ::: On the transfer unit, the wheel is reinvented every day. Because people don\'t work according to the agreements, the implementation has been flawed. (respondent 6) In the implementation literature, good preparation -- involving the relevant people, developing a proposal for change and selecting a set of multifaceted strategies -- is emphasised \[[@B23],[@B24]\]. Even so, the necessity of an open culture for change and involved management is underscored. Contrasting these insights with our data revealed shortcomings in the implementation. First, the initial intermediate care model lacked a detailed implementation strategy. Various respondents who had worked in the unit from the start confirmed this. Second, it can be questioned whether all relevant people where involved soon enough. Chief executives confessed they involved the nursing home physician too late. Also, nursing assistants said they were insufficiently prepared. Third, there seemed to be a lack of communication, resulting in a \'closed organisational culture\' and resistance to change. Finally, involvement by management was considered insufficient. There was criticism that supervision and control was too lax to bring about the desired change. This lack of management was partly due to discontinuities in leadership. Although four persons headed up the unit from the start, reorganisation of the entire residential home distracted chief executives. All these shortcomings were reflected in the following quotes: At the start, there was very little idea of what might happen. (respondent 6) They told us almost nothing about what was going to happen. I had the idea they didn\'t really know, either. (respondent 18) A nursing home physician was represented in the project group, but it wasn\'t the physician who was going to do the job. I didn\'t think this was very smart.(respondent 2) Staff members act differently. Different perceptions, different realities can be observed, but they\'re not tried out on each other or debated. (respondent 11) Apart from the implementation, respondents were also concerned with the nursing staff mix. In their view, the staff was insufficiently qualified: At this moment, I do not have a positive image of their expertise. (respondent 8) Managers and staff at the HRHH attributed the lack of expertise to the shortage of skilled nurses. Local figures showed that in 2000, nursing homes and residential homes in Amsterdam had 56 vacancies per 1,000 nursing staff. This rate is the highest in the Netherlands \[[@B30]\]. The managers of the HRHH said they had enormous problems filling the vacancies. Respondents working in the AMC underscored this, but also questioned the chosen staff mix model of the transfer unit: I was and still am disappointed in the nursing staff mix. The number of staff is okay, but on a ward where discharged hospital patients are admitted, just one RN supplemented by nursing assistants isn\'t enough. (respondent 2) Limitations ----------- The process evaluation has its limitations. The large number of non-consenting patients (119 out of 189) has biased the quantitative results. Most of the non-consenting patients were unable or less inclined to participate in the study. We attribute this in part to a flawed informed consent procedure. Consenting patients had to fill out questionnaires immediately, which made various candidates less willing, especially the sicker ones. Consequently, healthier patients are over-represented in the sample. This explains why we could not verify the perception by respondents that the patient population was more seriously ill. The validity of the qualitative findings is rather high, although some weaknesses were revealed. Because of our managerial evaluation perspective \[[@B18]\], three respondents seemed to give socially desirable answers. Furthermore, we had the impression that the evaluation became a management intervention in itself (Hawthorne effect). During the study we noticed that our framework of analysis, which was implicitly communicated during the interviews, was translated into management actions. Another weakness is that just one researcher conducted the interviews and analysis. Still, we were able to overcome these sources of bias because of the combination of methods, the purposive sampling procedure undertaken and the comprehensive triangulation achieved. The interviews therefore provide a credible exploration of the functioning of an intermediate care model in its local context. Overall, the transportability of the findings to other settings is limited. As we conducted a process evaluation, findings only provide a detailed description of the intermediate care model in Amsterdam. Nevertheless, we assume that the accumulated impact of the model is not unique to the Amsterdam model. From this perspective, the findings provide a good starting point for developing and evaluating models elsewhere. Discussion ========== In the current climate of health care policy, many stakeholders advocate the development and implementation of intermediate care models. However, the widespread popularity of the concept is insufficiently supported by evidence. Intermediate care is still in need of evaluation, as the benefits and the deficits of the various models are ambiguous \[[@B1]-[@B4]\]. Nevertheless, available knowledge provides enough do\'s and don\'ts for initiators who want to set up and implement intermediate care. Basically, initiators should be cautious. As our study shows, the accumulated impact of setting up such models may result in a \'bad practice\'. To prevent this, initiators should adequately plan, organise and monitor intermediate care services. The following issues are important. First, the relevance of an intermediate care model must be clear at the outset and supported by sound information. This was not the case in Amsterdam. The low intensity early discharge model was set up alongside two more intensive intermediate care models for stroke and orthopaedic patients. Consequently, the transfer unit was targeted towards a remnant and heterogeneous patient population within which the number of eligible candidates was smaller than expected. The straightforward conclusion is that one should know the size and profile of the target population. Apart from that, patients may view intermediate care as unacceptable. Little is known about this topic \[[@B31]\], and this is reflected in the call for more user involvement in intermediate care development \[[@B32]\]. Even so, this study indicates that the dynamic health status of candidates requires fast and flexible transition processes. Because the health status of patients can change rapidly, intermediate care services are appropriate for short periods during patient journeys. Moreover, poorly organised transition processes may increase the average LoS in the acute care setting which undermines achieving the goal of early discharging patients. So, detailed knowledge of the care needs, desires and size of the target population is necessary to justify the relevance of an intermediate care model. Second, policy-makers and managers should pay attention to the dynamics resulting from organisational and financial incentives. These may influence a broadening of admission criteria to maximise bed occupancy rates, as shown in Amsterdam. In the literature, this phenomenon is also known as \'Roemer\'s law\': a bed built is a bed filled \[[@B33]-[@B35]\]. Perverse incentives induce improper use of intermediate care beds, which results in another patient profile than initially anticipated. The changed patient profiles reported in various studies support this hypothesis \[[@B13],[@B36],[@B37]\]. Continuous monitoring of the patient profile in relation to bed occupancy rates is necessary to detect and overcome this phenomenon. Third, intermediate care requires a minimum of nursing staff to guarantee quality of care delivery. Their skills should meet the needs of envisioned patients. This was inadequate in Amsterdam. Nursing assistants gave direct care to patients, while medical and registered nursing staff headed up the unit. This high percentage of direct care given by relatively unqualified staff is also reported in other studies. It is considered one of the main reasons for the absence of improved outcomes of intermediate care \[[@B13],[@B36]\]. Other studies as well indicate that nursing staff doesmatter: those hospitals and nursing homes with the most highly qualified staff provide better care \[[@B38]-[@B40]\]. This underscores the importance of thoroughly considering nursing staff mix in intermediate care. However, it is difficult to determine what nursing staff mix is appropriate. There is no minimum standard available. Fourth, the implementation of intermediate care needs to be given attention. It requires know-how and expertise. In Amsterdam, the initiators overestimated the ability of the staff of the residential home to develop and operate a transfer unit for early discharged patients. These homes provide minimal care services for elderly people in stable health. These homes are less prepared to deliver care to post-acute care patients. This is reflected in the less advanced stage of development of quality assurance activities in Dutch residential homes \[[@B41],[@B42]\]. Initiators of collaborative intermediate care models should be aware of this pitfall and plan a comprehensive implementation strategy. Such a strategy must contain multiple approaches at different levels, tailored to specific settings and target groups \[[@B23],[@B24]\]. Such a strategy should ensure that all requirements are met for delivering good care. From a more general perspective, these issues promote a more rational and evidence-based management of intermediate care. It is acknowledged that the uptake of evidence in managerial practice could be better \[[@B43],[@B44]\]. As intermediate care lacks a straightforward evidence base, managers run the risk of being persuaded by political willingness rather than by \'evidence\'. Uninformed decision-making is dangerous and may ultimately harm patients; future research should fill the existing knowledge gaps. Steiner \[[@B1]\] identifies three key questions for intermediate care research: 1) Which services are best for which patients at which point? 2) Which professionals should be involved, doing what at which point? 3) What is the bottom line financially? Our findings indicate that these key questions must be answered simultaneously. One cannot properly answer one of the key questions without knowing the answers to the other two. This implies that research designs should have a broad scope, which as our study illustrates, is at the expense of rigour. However, creative study designs are being promoted that try to balance rigour and validity by combining quantitative and qualitative approaches, purpose-collected and coincidental data, and multidisciplinary research perspectives \[[@B45],[@B46]\]. These evaluative approaches provide a good foundation for developing the evidence base for intermediate care. Conclusion ========== We conclude that setting up a low intensive early discharge model of intermediate care between a university hospital and a residential home is less straightforward than was originally perceived by management, and that quality of care needs careful monitoring to ensure the change is for the better. The AMC and HRHH management have taken on these lessons, and the findings of the process evaluation have been translated into management interventions: consistent use of discharge and admission criteria, increasing patient flows by working with a second general hospital in Amsterdam, increasing nursing and medical expertise on the ward, implementing specific nursing protocols and more systematic monitoring of care. A combined quantitative and qualitative evaluation approach executed in close collaboration with the actors involved was helpful in revealing the underlying mechanism leading to the shortcomings. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= TP designed the qualitative study, conducted the semi-structured interviews, analysed the qualitative data, analysed all data, and primarily wrote the paper. DMJD designed the whole study, drafted the main protocol, assisted with the quantitative data analysis, reviewed earlier drafts of the paper, and supervised the daily research activities TFK collected quantitative data, was responsible for the quantitative data entry, did the quantitative analysis, and critically reviewed earlier drafts of the paper TJ collected quantitative data, critically reviewed earlier drafts of the paper NSK was responsible for completing the study and the supervision, designed the whole study, and critically reviewed earlier drafts of the paper Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/38/prepub>	PubMed Central	2024-06-05T03:55:58.223083	2005-5-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168893/", "journal": "BMC Health Serv Res. 2005 May 24; 5:38", "authors": [ { "first": "Thomas", "last": "Plochg" }, { "first": "Diana MJ", "last": "Delnoij" }, { "first": "Tineke F", "last": "van der Kruk" }, { "first": "Tonnie ACM", "last": "Janmaat" }, { "first": "Niek S", "last": "Klazinga" } ] }
PMC1168894	Background ========== The rigorous evaluation of health care interventions requires potential users of these interventions to participate in clinical trials to assess their effectiveness. There are currently various concerns about both the small proportion of eligible people who participate in randomised controlled trials (RCTs) and the experiences of participants \[[@B1],[@B2]\]. Inadequate information provision is a contributing factor to several of these concerns. For example, some people who would like to enter RCTs are not invited to do so, and struggle to identify trials for which they would be eligible \[[@B3]\]. Problems have been identified with the quality of information given \[[@B4]\]. Thus some people who are invited to participate in RCTs do not develop an adequate understanding to make a well-informed decision to participate or not \[[@B5],[@B6]\]. People who agree to participate in trials sometimes struggle to make sense of their participation \[[@B7]\], may not know what to do if they identify questions during the course of the trial, and are often disappointed at the lack of information they are given about the progress or results of the trials \[[@B8]\]. An information resource about ongoing clinical trials designed for potential and current participants could help overcome some of these problems. It could serve as a source of additional, semi-independent information for people who have been invited to participate in trials, and it could help people to identify trials in which they might participate. If kept up to date, it could serve as a source of information about the progress and results of particular trials. A number of trial registers have been established in recent years (for example, metaRegister of Controlled Trials <http://www.controlled-trials.com>, UKCCCR National Register of Cancer Trials <http://www.ctu.mrc.ac.uk/ukcccr/home.asp>), and there has been increasing pressure for requirements that all clinical trials be registered in the public domain. \[[@B9]\]. Trial registers are undoubtedly helpful in terms of getting information about planned, ongoing and completed trials into the public domain and reducing the problems associated with underreporting of trials. However, the current registers are not comprehensive \[[@B10]-[@B12]\]. They were not designed to be easily accessible or useful to general public audiences and none were designed specifically to meet the information needs of members of the public as potential or current trial participants. It can be difficult to identify which registers are appropriate for particular purposes \[[@B12],[@B13]\], and it is hard to search across registers because they do not organise and present information in a standardised way \[[@B3],[@B13]\]. In this paper we report on a scoping exercise undertaken for Consumers in NHS Research (now known as INVOLVE) to assess the desirable features and feasibility of developing and providing an information resource about ongoing clinical trials for potential and current participants \[[@B14]\]. Methods ======= We recognised that people who had played different roles in relation to trials would have different perspectives on information needs relating to clinical trials. Reflecting this, we sought information and opinions from a range of people from across the UK that included: people seeking to participate in or considering participating in a trial; current and recent trial participants; representatives of national patient interest/support groups; and researchers involved in the design and conduct of trials. People who were seeking to participate and considering participating in a trial were identified through condition-specific consumer networks, in particular the National Association for the Relief of Paget\'s Disease (a trial investigating the treatment of Paget\'s disease was about to start at the time we conducted this scoping exercise). Current and recent trial participants were identified at three meetings of health care consumers and advocates. Representatives of consumer groups were selected from those patient interest/support groups with a known interest in trials (identified primarily by Consumers in NHS Research). Researchers involved in the design and conduct of trials were selected from the UK MRC trial managers network and individuals known to the research team. Several of the consumer advocates and researchers had current or recent personal experience of participating (or declining to participate) in clinical trials. We accepted their contributions from the perspective of trial participants, because the time frame of our scoping exercise did not permit us to obtain ethics committee approval to identify and approach current participants via trial offices. We used semi-structured interviews, focus groups and e-mail discussion lists to explore the kinds of information that people want and need when considering whether or not to participate in a particular trial, and the ways they might access and use this information. We asked people to discuss the feasibility and appropriateness of \'star rated\' information on the quality of particular aspects of ongoing trials, and the provision of generic guidance about how to interpret and appraise information about trials. We also asked researchers and consumer representatives what they thought made for a good quality trial and what issues might arise from an attempt to provide standardised information about ongoing randomised clinical trials in a publicly accessible participant-orientated information resource. We used topic guides to support semi-structured interviews and focus group discussions. An example of the topic guide used with potential trial participants is included \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Most interviews were audio-taped and transcribed in full. If audio-recording was not possible, members of the research team took detailed notes. Analysis of transcripts, discussion notes and e-mails was carried out using a modified \'charting\' approach \[[@B15]\]: to facilitate the identification of the range of views about particular issues, the points made by each participant were summarised under key headings that reflected the main areas of questioning. Over the course of the scoping exercise we formally interviewed 25 individuals on a one-to-one basis including: • nine potential trial participants; • six representatives from patient interest /support groups; • nine researchers (including trial principal investigators, trial managers and a research nurse); and • one pharmaceutical/medical devices industry representative. In addition, the views of 24 consumer advocates were elicited in focus groups held at national meetings of consumers and advocates, and nine consumer representatives contributed comments via e-mail. Several of the consumer advocates and researchers had personal experience of participating (or declining to participate) in clinical trials. Results ======= Information elements required by potential trial participants ------------------------------------------------------------- Our informants identified a range of information needs of potential trial participants. We grouped these into the following categories: (a) the interventions being compared; (b) the implications of trial participation for participants; (c) the scientific or methodological quality of a trial; (d) the ethical probity and governance of a trial; and (e) contact details (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The features identified as desirable for a centralised information resource about randomised controlled trials. ::: ![](1472-6963-5-39-1) ::: ### Interventions being compared There was a consensus that potential trial participants need to know about the \'new\' intervention being tested, its potential risks, and how it is known to compare with the current \"standard\" intervention and any other interventions with which it is compared in the trial. Information about the availability of relevant interventions outwith the trial was also mentioned as important. ### Implications for participants Information needs relating to implications for participants included practical issues such as the timescales for the trial interventions and follow-up, and the rationale for, and number of extra visits or investigations associated with the trial. Several informants highlighted the need for information about whether and how people could withdraw from or join the trial at a later date. They also thought that participants should be told whether or how they could gain access to data about their own outcomes and the trial findings, and about whether and how participants would be told if a trial was stopped early for any reason. ### Scientific or methodological quality, and ethical probity and governance There was a consensus that people should be able to be confident about the quality of any trial they might agree to participate in, but it was widely recognised that relatively few people would currently be able to judge the scientific quality or ethical probity of a trial for themselves on the basis of the information they are typically given for recruitment purposes. The provision of information about key features of trial quality (and/or about quality assessment) was thought desirable for a centralised resource. ### Contact details Several respondents from all categories mentioned that potential trial participants might benefit from the option to talk to someone before deciding whether or not to participate in a trial, and possibly also during the trial. The case was regularly made that two-way interaction was key to understanding information, as any misconceptions could be corrected and personal concerns could be addressed. For people who had been invited to participate in a particular trial, the options suggested were: • having a chance to speak with a member of the trial team (preferably someone who is not related directly to a person\'s normal care) -- someone who can explain things like randomisation; • having a chance to talk to an independent person (i.e. someone not associated with the trial) e.g. GP, consumer advocate, voluntary health organisations, advice services; and • having contact details, so know whom to contact if concerned about anything while participating in the trial. People who were themselves looking for trials in which they might participate would be likely to need: • contact details for initial enquiries about the specific trial and (possibly subsequently) local recruiting centres/clinicians; and possibly • an information helpline staffed by people who could offer general information, answer general questions about trials, and help explain information provided about specific trials. However, the capacity of trialists or trial centres to handle unsolicited enquiries was limited, they might be reluctant to have their contact details \'advertised\' in the context of an information resource that promised help to people identify trials in which they might participate. Aids to information appraisal ----------------------------- As indicated in the methods section, we explored two approaches that might help people evaluate information about trials and assess their quality: (a) a quality or \"star\" rating scheme in which key features of trials were allocated one- to three- star ratings; and (b) a guide to help people appraise trials for themselves. Informants registered some strongly felt concerns about the quality ratings option. These included doubts about the feasibility of producing valid and reliable ratings across a diverse range of trials, and fears that particular ratings were likely to be contested by trial sponsors and trialists, which could create a lot of \'hassle\' and work for the resource producers. Despite the limitations of a quality ratings scheme, there was general agreement that it would be useful to indicate that key quality assurance features were in place for particular trials. The most obvious features suggested were that a trial had been through an independent peer review process (a check on scientific quality) and that it had been assessed and approved by an independent ethics committee (a check on ethical probity). There was widespread support for the provision of a general information section within any centralised resource that could: explain the purpose of clinical trials; highlight key features and indicate points to look out for in clinical trials; provide a glossary of commonly used terms; and identify links to relevant organisations. The inclusion of a \'guide\' to help people appraise the quality of a trial for themselves was also thought appropriate. This guide could be based on existing checklists developed for peer reviewers, ethics committees and consumer representatives who appraise trials, but modified for more general public use. Key design features of a centralised resource --------------------------------------------- All informants supported the basic concept of a centralised information resource about clinical trials. This could be built using a database structure with a standard set of information elements for each trial reflecting the key areas outlined above. Respondents identified a variety of design features that could potentially enhance the usefulness of the information within the resource, including: • A sophisticated search facility that would allow people to enter their personal details (e.g. health condition, age, gender, geographical location etc), and retrieve information on trials that fulfilled their search criteria. • Options to view particular types of information in varying levels of detail. • The use of a variety of media within or as a complement to the resource (for example, diagrams or video-clips to illustrate and explain trial procedures or the interventions being compared in particular trials Potential models for providing a centralised information resource ----------------------------------------------------------------- There was widespread support for the provision of this centralised information resource on the Internet, but a clear recognition that some people would not be able to access the information effectively via this route. Opportunities for supported access via outlets in health care settings, public libraries or other community centres would also be useful. Health information services might usefully serve as valuable \'intermediaries\', adding value to the resource by helping people to appraise the information and interpret it for their own use. Discussion ========== One of the strengths of our study is that we canvassed views from a range of perspectives. We identified a broad-base of support for the concept of a centralised, publicly accessible, participant-oriented information resource about ongoing clinical trials. We also identified several concrete suggestions for the content and design of such a resource. However, time constraints prevented us from exploring the views of a larger sample of people, and from considering the issues we identified in more depth. We acknowledge in particular that further research is needed to explore the opinions of a wider spectrum of the \'general public\' as potential users of a centralised resource for information about clinical trials, and to consider the potential implications of presenting particular elements of information and in particular ways. We also acknowledge that the potential trial participants interviewed for this study were mostly people with a chronic condition. Their perspectives may differ from potential trial participants with acute health problems. However, individuals with acute conditions were canvassed for their views as current or recent trial participants. We do not claim to have identified all possible information requirements, but are confident that we have identified the main ones. In theory, a centralised information resource about clinical trials would have two main audiences in terms of trial participants: those who have been invited to participate in a trial and want to find out more about it, and those who have not been invited to participate in a trial, but would like to identify any ongoing trials for which they might be eligible. If the resource included updated reports on trial progress, it may also be useful to a third group of people who are currently participating in trials who want more information about the trials in which they are participating It is unclear whether and to what extent a centralised information resource would achieve the desired aims of improving recruitment rates, and people\'s experiences of information provision relating to trial recruitment and participation, nor whether it would have any unwanted effects. It is not clear to what extent a publicly accessible resource might increase demand for participation in trials, nor what the implications of such demands might be. At the moment, some health care providers are not in a position to offer trial participation to their service users. Increased public awareness of particular trials might place a variety of extra burdens on health care systems, for example if patients start demanding access to particular trials and/or changing health care providers in order to gain access to them. Other potential disadvantages of making information about trials more easily accessible should be explored. For example, breaches of privacy might become a problem if the knowledge that someone is eligible to participate in a particular trial (which is often readily obtained by third parties who observe markers on medical records or witness the receipt of mailings from a trial team) could quickly lead to the revelation of information about personal health status or health care experiences because trial eligibility criteria would be easily ascertained. Public access to trial progress reports might, depending on the nature of information provided, influence people\'s self-reporting of outcome status and willingness to continue to participate in a trial. Issues such as these require further consideration and investigation. While the development of the database framework, search engine and the features identified above as desirable are technologically feasible, there are several potential barriers to the development of a participant-oriented central information resource about clinical trials. Current trial registers do not contain all the information elements that potential participants might find useful. The process of gathering, checking and updating the requisite information would be logistically challenging and time consuming, and would require considerable and sustained investment. Any attempt to include trial progress reports to the database would significantly add to the logistical challenges. However, recent trends towards increasing disclosure of information about trials, including commitments to openness from some pharmaceutical companies, will facilitate the acquisition of relevant information. Conclusion ========== A centralised participant-oriented information resource about randomised trials could offer many benefits. The development of such a resource would require considerable investment, but the support for the basic concept that is evident among trial participants, consumer advocates and trialists suggest that it warrants further investigation and evaluation. Abbreviations ============= UKCCR United Kingdom Coordinating Committee on Cancer Research RCT Randomised Controlled Trial Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors contributed equally to the design and execution of the study. MKC was principal investigator. ALL wrote the first draft of the manuscript. All authors contributed to the development of subsequent drafts, and read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/39/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 2001\. Example Topic Guide. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ Consumers in NHS Research(now known as INVOLVE) funded the scoping study on which this paper is based. We would like to thank the secretarial staff of the Health Services Research Unit for transcription of the interviews. Most of all we would like to extend our many thanks to the people who agreed to be interviewed or submit opinions for this study and gave generously of their time and views. The Health Services Research Unit is funded by the Chief Scientist Office of the Scottish Executive Department of Health. The views expressed are those of the authors and not necessarily those of the funders.	PubMed Central	2024-06-05T03:55:58.227513	2005-5-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168894/", "journal": "BMC Health Serv Res. 2005 May 24; 5:39", "authors": [ { "first": "Anne L", "last": "Langston" }, { "first": "Marion K", "last": "Campbell" }, { "first": "Vikki A", "last": "Entwistle" }, { "first": "Zoë", "last": "Skea" } ] }
PMC1168895	Background ========== Adult outpatient parenteral antibiotic therapy (OPAT) programs have been reported in the literature for over 20 years \[[@B1]\]. These programs have now become an accepted alternative to inpatient therapy for select patients and infection types \[[@B2]-[@B4]\]. OPAT programs have been demonstrated to be a safe, effective and acceptable alternative to hospitalization \[[@B5]-[@B7]\]. Cost analysis of OPAT programs in the U.S., Canadian and other settings have been reported \[[@B5],[@B8]-[@B13]\]. An adult OPAT program was implemented at our hospital in 1995. A cost analysis performed at our institution from 1995 to 1998 showed significant cost avoidance for both the hospital and the Ministry of Health \[[@B14]\]. Informal patient satisfaction surveys have shown that the program has been well received by patients. However, there are no published reports quantifying preference for treatment location of OPAT program patients. Willingness to pay (WTP) is a method of quantifying preference and this methodology is gaining popularity in health care \[[@B15]-[@B19]\]. WTP provides a measure of how much an individual values a particular treatment preference. The objective of this study was to elicit hypothetical infection treatment location preferences, and the willingness to pay for this preferred treatment option from OPAT program referral patients. Methods ======= This was conducted as a multidisciplinary, single centre, prospective study at a 1000-bed Canadian adult tertiary care teaching hospital. Our OPAT program receives about 250 patient referrals and provides approximately 3,000 patient-days of outpatient parenteral therapy per year \[[@B20]\]. General characteristics of patients managed by this service are described elsewhere \[[@B14],[@B20]\]. Adult patients are accepted into the program if they had a proven or suspected infection requiring one or more parenteral antimicrobials for an expected minimum duration of 5 days, are medically stable, have an acceptable venous access, demonstrate a willingness and capability to perform the necessary self-management tasks and live in a suitable home environment with access to a telephone. Once enrolled, the OPAT pharmacist and nurses provide patient teaching, insert the appropriate vascular device, liaise with community nursing personnel, coordinate delivery of drugs and supplies and arranged appropriate patient follow up. These patients are treated in their home environment and return to the hospital periodically for follow-up purposes only. Patients requiring short courses of parenteral antibiotics are typically excluded from the program and are treated as inpatients or managed in a hospital medical daycare setting. This study was approved by the University Ethics Committee and the Hospital Research Committee. Patient enrollment ------------------ Patients referred to the OPAT program during the 9-month study (September 1999 -- June 2000) were considered eligible for the patient preference assessment. For inclusion into the study, the questionnaire had to be administered to the patient prior to assessment by the OPAT clinical staff for possible inclusion into the program. Only consenting patients were administered the WTP questionnaire. This precaution was taken to avoid the possibility of biased responses from the participants in an attempt to obtain the treatment location of their choice. Willingness to pay ------------------ Contingent valuation methodology (CVM) was used to quantitatively measure patient preference (i.e. WTP) for intravenous antibiotic treatment location \[[@B15]-[@B19]\]. CVM is a survey-based approach for eliciting a consumer\'s monetary valuations for program benefits for use in cost-benefit analysis. The specific methodology employed was similar to that adopted by Donaldson et al \[[@B21]\]. The consumer utility being measured was compensating variation and the survey measured WTP in the context of program availability. Since the questionnaires were given to individuals undertaking the valuation who were already consumers of the treatment in question (i.e. were receiving parenteral antimicrobials when enrolled) and thus the primary uncertainty at the time of the questionnaire was the probability of treatment course outcomes, an ex-post perspective was adopted \[[@B15],[@B16]\]. The WTP questionnaire consisted of two hypothetical scenarios that were created based upon OPAT program data collected during the period 1995--1998 (see [Additional File 1](#S1){ref-type="supplementary-material"}) \[[@B14]\]. Each hypothetical scenario involved an infection that required a 23-day course of intravenous antibiotics (the average duration of therapy for patients enrolled in the OPAT program). The first scenario described a hospital treatment course, while the second scenario described a similar treatment regimen that was administered in the home setting. Using these historic data, the risks associated with each treatment location were also provided to the patient \[[@B14]\]. Patients were asked to specify their preference for treatment location based upon these scenarios. Utilizing open-ended questions, patients were asked to quantify their preference by stating how much they would be willing to pay to obtain treatment in their preferred location. The survey was initially designed and tested on ten patients, four clinical nurse specialists and two infectious diseases specialists. We utilized comments from these individuals to modify our survey in order to improved readability and understandability. Data collection --------------- A single investigator coordinated all patient self-administered WTP questionnaires. Partial completion of the surveys was identified in four of the first eleven patients; thereafter, the investigator examined the questionnaires for completeness at bedside and encouraged patients to provide responses if necessary. If requested by the patient, the questionnaire was read aloud and the investigator recorded the patient\'s responses. In these cases, the completed questionnaire was subsequently reviewed with the patient to ensure accuracy. Demographic information, socioeconomic data, infection details and WTP for treatment location were collected for all patients. Data analysis ------------- Means, medians and ranges for WTP by treatment preference location were determined. Differences in means and medians were testing using t-tests and Mann-Whitney U tests, respectively. To estimate an overall monetary valuation, a total WTP was calculated for patients preferring hospital treatment and a total WTP was also calculated for those preferring home treatment. A positive association between WTP and income was assessed to determine the construct validity of our questionnaire \[[@B22]\]. WTP was examined for normality using histograms to determine if a natural log transformation was necessary to fit the assumptions of linear regression. Using WTP as the dependent variable, multiple linear regression was utilized. Univariate analyses were performed between each of the possible predictor variables (gender, marital status, level of education, employment, annual household income, infection type) and the dependent variable using ordinary least-squares linear regression. Variables associated with WTP with a p-value \< = 0.10 in the univariate analyses were considered in the multiple linear regression models. Adjusted r^2^was calculated for the multivariable models to determine the amount of variance in the outcome variable explained by the predictor variables in the final models. Among significant variables, two-way interactions were investigated. No adjustments were made to p-values to account for multiple comparisons. Studentized residuals and Cook\'s distance were examined to determine if assumptions of multiple linear regression were violated. Two-sided P values are reported for all analyses. A p value of less than 0.05 was considered to be statistically significant. All analyses were conducted by using SPSS, version 10. Results ======= During the 9-month study period, 131 patients were considered eligible for enrollment in the contingent valuation analysis. Of these patients, 40 were excluded, as the investigator was unavailable to conduct an interview prior to the assessment by the OPAT team, informed consent could not be obtained due to language barriers, decreased cognitive status was evident or patients simply declined to participate. The remaining 91 patients completed the WTP questionnaire. Patient demographics and socioeconomic status are reported in Table [1](#T1){ref-type="table"}. Participants were typically married males in their sixth decade of life with a secondary school education or greater. The majority of participants were retired or were employed with an annual household income of less than \$60,000. Osteomyelitis was the most common type of infection for which parenteral therapy was required. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient demographics ::: Parameter Value ------------------------------------ ------------- No. of Patients 91 Mean age, years(range) 56 (25--81) Gender Male (%) 63 (69) Marital Status (%) Married 60 (66) Divorced 11 (12) Widowed 9 (10) Single 10 (12) Highest Level of Education (%) Elementary School 7 (8) Secondary School 29 (32) Trades/Technical College 34 (36) University Degree 16 (18) Post-graduate 5 (6) Employment (%) Retired 45 (49) Employed 29 (32) Unemployed 16 (18) Unknown 1 (1) Annual Household Income (%) \< \$20,000 22 (24) \$20,000--39,999 20 (23) \$40,000--59,999 22 (24) \$60,000--79,999 9 (10) \$80,000--99,999 3 (3) \$100,000--149,000 7 (8) \> \$150,000 4 (4) Unknown 4 (4) Type of Infection (%) Osteomyelitis 39 (43) Infected pacemaker/wires 9 (10) Endocarditis 9 (10) Wound infection 7 (8) Abscess 6 (7) Bacteremia 4 (5) Other^1^ 17 (17) ^1^Meningitis (3), pneumonia (4), infected graft of lower limb (2), line sepsis (2), septic arthritis (2), cellulites (1), pyelonephritis (1), CMV (1), discitis (1) ::: Willingness to pay ------------------ Of the 91 patients who were enrolled in the study, 87 (96%) indicated a treatment location preference while the remaining four participants had no preference. Of those 87 patients who indicated a preference, 77 (89%) preferred treatment at home while 10 (11%) preferred treatment in hospital. Seventy-one (82%) patients provided an interpretable response regarding WTP for treatment in their preferred location. Of those 16 patients (13 patients with a preference for home therapy vs. 3 patient with a preference for hospital therapy) who did not provide an interpretable response, one registered an astronomically high \"protest\" WTP far exceeding their ability to pay, while 15 indicated a treatment preference but provided no monetary value. For those 71 patients who provided an interpretable response, 64 patients preferred treatment at home with a median WTP of \$490 CDN (mean \$949, range \$20 to \$6250), and 7 patients preferred treatment in the hospital with a median WTP of \$500 CDN (mean \$1123, range \$10 to \$3000). Tests for differences in means and medians revealed no statistically significant differences between WTP values between the treatment locations at the 5% level. The total WTP for the seven patients who preferred hospital treatment was \$7,859 versus \$60,712 for the 64 patients who preferred home treatment. The natural logarithm of WTP values approximated a normal distribution, thus satisfying this assumption of linear regression (Figure [1](#F1){ref-type="fig"}). Only seventy-five patients (71 patients with an interpretable response plus those 4 patients with no treatment location preference) were included in the regression analysis. Multiple linear regression analysis revealed that income and treatment location preference were independent predictors of WTP (Table [2](#T2){ref-type="table"}). There was a trend towards respondents with lower incomes being willing to pay slightly less for their preferred treatment location than those with the highest incomes (p = 0.067). In addition, people who stated preferences were willing to pay significantly more for than those who did not state a preference (p \< 0.001). In the multiple linear regression model that included interaction terms (adjusted r2 = 0.543), there was also a significant interaction between income and treatment location preference such that patients with the lowest income were willing to pay significantly more for hospital treatment than for home treatment (p \< \<0.0001). The fact that there was a significant association between WTP and ability to pay (i.e. higher income) validates the theoretical construct of our survey. ::: {#F1 .fig} Figure 1 ![](1472-6963-5-43-1) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### WTP regression analysis^1,2^ ::: Parameter β - coefficient p-value 95% Confidence Interval --------------------- ----------------- --------- ------------------------- -------- Intercept 0.75 0.35 -0.82 2.32 Income, \$ CAN 0.067 ≤ 20,000 -0.991 0.031 -1.89 -0.093 21,000--79,000 -1.011 0.037 -1.96 -0.060 ≥ 80,000 Reference Location preference \<0.001 Home 6.10 \<0.001 4.58 7.54 Hospital 6.13 \<0.001 4.32 7.94 None Reference ^1^Dependent variable is the natural logarithm of WTP ^2^Adjusted r2 = 0.478 ::: Discussion ========== To our knowledge, this is the first published report quantifying preference for treatment location in an adult OPAT program patients using WTP. According to our WTP analysis, candidates for the program expressed an overwhelming preference for treatment in the home setting. Our results also demonstrated that the WTP values were similar between those patients who preferred to be treated at home and those who wished to remain in hospital. Accordingly, the total WTP value was greater (in fact, almost 8-fold greater) for those patients preferring treatment at home. This reflects the overall magnitude of societal preference for the management of infectious diseases that require intravenous therapy, but does not require institutionalization. There were several limitations to this study. We conducted this trial in one adult acute care institution, thus caution must be exercised when attempting to generalize the results to other health care settings involving different patient populations, and other infectious diseases which will require different treatment regimens. We relied on a hypothetical treatment scenario in our attempt to solicit a preference location and willingness to pay for this patient population. As the scenario did not necessarily reflect the treatment that they were about to receive, we must be careful in our extrapolation of the results. Although we acknowledge the potential problems with using such scenarios in CVM, we believe that this effect was minimized by surveying patients who were currently experiencing an infection that initiated a consult from the OPAT team (i.e. the ex-post perspective). We believe that most of these individuals would be able to realistically comprehend the health outcomes described in our scenarios. We also relied on open-ended technique rather than a bidding-game technique to solicit a WTP value. While the bidding game technique forces an upper and lower limit to the patient response and can be criticized for introducing a starting point bias, the open-ended technique has also been questioned. As described by O\'Brien and Viramontes, patient naivety regarding health care costs due to the Canadian universal health insurance environment may lead to an inability to quantify the value of an expected health improvement \[[@B16]\]. The broad range of WTP values provided by our participants may be a reflection of this naivety. In addition, O\'Brien and Gafni discuss that open-ended questions often elicit large numbers of non-responses or protest zero responses \[[@B21]\]. Indeed, some patients in our study expressed difficulty in placing a dollar value on their choice of treatment location. In some cases, this appeared to be a protest against the interview question and reflected a concern that their response would be used to determine a future fee for their treatment preference. In other cases, this may have been related to the fact that patients are not typically aware of, nor directly pay for, the costs of health care services in the Canadian health care system. Finally, WTP surveys measure only what a patient claims they are willing to pay for a particular treatment. The magnitude of payment is not necessarily an accurate reflection of what they would actually be willing to pay if they were to encounter the actual scenario. As expected, ability to pay was associated with WTP and this functioned as a confirmation of construct validity of our questionnaire. Unfortunately, as mentioned by Drummond et al, there is not an actual market for most health programs and, thus, there is no \"gold standard\" against which one can compare WTP values \[[@B22]\]. It is, therefore, difficult to establish criterion validity in this context. Conclusion ========== This study reveals that the majority, but not all, of adult inpatients receiving parenteral antibiotic therapy who are referred to an outpatient parenteral antibiotic therapy program prefer to be treated at home. Income and treatment location appear to independently predict their willingness to pay. Competing interests =================== This was an unfunded study. The authors declare that they have no competing interests. Authors\' contributions ======================= CM made substantial contributions to the conception, design, analysis and interpretation of the data; he was involved in the drafting of the article and revising it critically for intellectual content and has given final approval for the current version to be published. LF made substantial contributions to the conception, design, analysis and interpretation of the data; she was involved in the drafting of the article and revising it critically for intellectual content and has given final approval for the current version to be published. AG contributed to the study design, made substantial contributions to data collection, and assisted in the interpretation of the data; he was involved in the drafting of the article, revising it, and has given final approval for the current version to be published. AW contributed to the study conception, design, daily supervision of and participation in data collection, analysis and interpretation of the data; she was involved in the drafting of the article, revising it, and has given final approval for the current version to be published. LC contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. RN contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. CL contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. ST contributed to the study design, data collection and interpretation of the data; she was involved in the drafting the article and has given final approval for the current version to be published. BF contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. PJ was the coordinating investigator, made substantial contributions to the conception, design, analysis and interpretation of the data; he was involved in the drafting of the article and revising it critically for intellectual content and has given final approval for the current version to be published. All authors take public responsibility for appropriate portions of the content of the manuscript and all authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/43/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 WTP BMC HSR Version 1 PJJ appendix 1.doc. This file contains the questionnaire used for this study ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The investigators are (or were) members of the Pharmaceutical Sciences Clinical Service Unit at VHHSC and this work was conducted as part of their professional and academic obligation to the institution, and to the patients we serve. We thank these patients, and our colleagues for their contributions to this effort.	PubMed Central	2024-06-05T03:55:58.230340	2005-6-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168895/", "journal": "BMC Health Serv Res. 2005 Jun 7; 5:43", "authors": [ { "first": "Carlo A", "last": "Marra" }, { "first": "Luciana", "last": "Frighetto" }, { "first": "Alan F", "last": "Goodfellow" }, { "first": "Amy O", "last": "Wai" }, { "first": "M Lynn", "last": "Chase" }, { "first": "Ruth E", "last": "Nicol" }, { "first": "Carole A", "last": "Leong" }, { "first": "Sally", "last": "Tomlinson" }, { "first": "Barbara M", "last": "Ferreira" }, { "first": "Peter J", "last": "Jewesson" } ] }
PMC1168896	Background ========== Transfer-messenger RNA (tmRNA), also known as 10Sa RNA or ssrA RNA, is a hybrid of a tRNA-like domain (TLD) and a mRNA-like domain (MLD) connected by a variable number of pseudoknots \[[@B1]\]. TmRNA is a stable and essential component of trans-translation, a quality-control process that rescues ribosomes stalled on mRNAs lacking stop codons. During trans-translation, ribosomes switch from a defective mRNA (lacking its translation-termination signal) to the MLD of tmRNA. Because a stop codon is provided by the tmRNA, the ribosomes can dissociate and recycle \[[@B2]\]. As an additional advantage, the tandem translation of the two templates generates a tagged polypeptide which is degraded by housekeeping proteases \[[@B3],[@B4]\]. For tagging, tmRNA has to be charged by aminoacyl-tRNA synthetases \[[@B5]\]. Assisted by protein SmpB, the charged tmRNA is delivered to stalled ribosomes as a quaternary complex with EF-Tu and GTP. Binding of tmRNA to ribosomes is facilitated by ribosomal protein S1, which interacts with the MLD and pseudoknots but not with the TLD \[[@B6]-[@B9]\]. Recently, cryo-EM revealed the shape of the tmRNA associated with SmpB and EF-Tu in its ribosome-bound form \[[@B10]\]. Despite this significant progress, high-resolution structures as obtained by NMR and X-ray crystallography are unavailable and expected to be difficult to obtain in the foreseeable future due to the relatively large size and flexibility of the tmRNA. In the present work we used a stepwise procedure for arriving at high-resolution models for the entire tmRNA molecule. First, 2-D structures were obtained by covariation analysis of a large number of tmRNA sequences. The basepairing information was submitted to the ERNA-3D modeling program \[[@B11]\] to build the helical sections. Structural motifs of the loops and turns were identified in SCOR \[[@B12]\], high-resolution data were extracted from known structures, and these data were incorporated into the models. Overall, significantly improved 3-D models were obtained which will be useful to understand the role of tmRNA in trans-translation. The described approach can be adapted to obtain high-resolution models of the members of other RNA families. Results ======= Identification of tmRNA sequences --------------------------------- The tmRNA sequences were identified previously and subjected to comparative sequence analysis (CSA) as described \[[@B1],[@B13]\]. New tmRNA sequences were obtained from the tmRNA website \[[@B14]\], through keyword searches of the literature and GenBank \[[@B15]\], or BLAST \[[@B16],[@B17]\], and various genome sequencing projects. The new sequences were examined iteratively as described in Materials and Methods to confirm tmRNA identity, remove sequence duplications, and create a meaningful alignment. New potential tmRNA sequences were maintained as a preliminary alignment in BioEdit \[[@B18]\], separate EMBL-formatted sequence files, and a HTML-formatted phylogenetic list. The sequences were ordered phylogenetically using the information in the Ribosomal Database Project (RDP) \[[@B19]\]. If the organism name was not listed in the RDP, the sequence was placed next to its closest relative using the NCBI Taxonomy resource \[[@B20]\]. Selection of tmRNA sequences ---------------------------- The new sequences were confirmed individually as tmRNAs by comparison with the closest relative using the pairwise alignment feature of BioEdit \[[@B18]\]. If there was a lack of obvious similarity, the sequence was inspected for evidence of biological features such as the ability to form a TLD and an open reading frame. Furthermore, the possibility of a two-part tmRNA was considered. A sequence suspected to be a new tmRNA was investigated further by CSA \[[@B13]\] as described in Materials and Methods. Potential new sequences of the alpha-Proteobacteria and some Cyanobacteria that were encoded in two separate sections of their genes \[[@B21]\], were compared to the two-part tmRNA sequence from a closest relative for effective comparison with the one-part tmRNAs. The 3\'- and 5\'-ends of each section were determined by pairwise alignment to generate a single sequence. Each of the 20 new two-part tmRNAs (14 sequences from alpha-Proteobacteria and six from Cyanobacteria) was subjected to this rearrangement. Comparative Sequence Analysis ----------------------------- Sequences were ordered phylogenetically using the RDP \[[@B19]\] as a guide or by alignment with the closest relative. Identical regions were aligned first. Subsequently, similar regions were aligned using invariant positions as signposts. Regions of biological significance, such as the resume and stop codons, were then considered. Finally, common secondary structure features were used to align regions that lacked primary structure similarity or biological features. Supported Watson-Crick basepairs and G-U interactions were indicated in the alignment by uppercase letters. Gaps were introduced to account for differences in sequence length and to avoid the alignment of dissimilar regions. Secondary structure was determined using covariation analysis as described \[[@B13]\] (see also Materials and Methods). The alignment was examined to identify compensatory base changes (CBCs) and other covariations. The numbers of CBCs and mismatches between the alignment columns were counted. CBCs provided positive evidence for the existence of a basepair; mismatches provided negative evidence. If the number of compensatory base changes was two times or greater than the number of mismatches, the basepair was considered supported. If a basepair was invariant, no evidence for or against its existence could be gained from CSA. A basepair was considered specific to a particular phylogenetic group if it was proven only in that group. Quality control --------------- To check for the proper assignment of basepairs, the alignment was sent through an automated pipeline of programs from RNAdbTools \[[@B22]\]. The output was inspected visually and corrections were made manually using the BioEdit program \[[@B18]\]. The revised alignment was resubmitted to RNAdbTools, and the review process was repeated until a satisfactory alignment was produced. TmRNA alignment --------------- The final alignment contained a total of 274 tmRNA sequences in 16 bacterial phylogenetic groups. A complete phylogenetic list is available at the tmRDB \[[@B23]\]. There was a substantial increase in the number of two-part tmRNAs for a total of 27 sequences: 20 from alpha-Proteobacteria (20 tmRNAs), one mitochondrial tmRNA, and six cyanobacterial tmRNAs. The nine organelle sequences included one from a cyanelle, six from chloroplasts, one from a plastid, and one from the Reclinomonas americanamitochondrion. The typical tmRNA was about 350 nucleotides long. The R. americanamitochondrion tmRNA contained only 189 nucleotides and, since it appeared to lack an ORF, may not be functional. Excluding this exception and any partial tmRNAs, the tmRNA of Synechococcusspecies PCC7009 was the shortest (250 nucleotides), and the longest was from Chlamydophila psittaci(425 nucleotides). The tmRNA alignment is provided as additional files [1](#S1){ref-type="supplementary-material"}: tmRNA-alignment.html, [2](#S2){ref-type="supplementary-material"}: tmRNA-alignment-wide.txt, [3](#S3){ref-type="supplementary-material"}:tmRNA-alignment-92col.txt, and [4](#S4){ref-type="supplementary-material"}: tmRNA-alignment.msf. Secondary structure of tmRNA ---------------------------- The tmRNA secondary structure features were extracted from the alignment and are listed in phylogenetic order in Table [1](#T1){ref-type="table"}. The representative secondary structure of Escherichia colitmRNA is shown in Figure [1](#F1){ref-type="fig"}. Secondary structures of Bacillus anthracisand Caulobacter crescentusare presented in as [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf and [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf, respectively. TLD (helices 1, 2a and 12) -------------------------- Although a prominent feature of each tmRNA, the TLD was relatively weakly supported by CSA due to a high degree of sequence conservation. However, the structure of this region is well established by experimental evidence \[[@B24]-[@B26]\]. Helix 1 contained seven basepairs and was usually continuous with the exception of the Anabaenaspecies tmRNA, which contained an insertion in the 3\'-portion of helix 1. The first pair (1G-C359 in E. colitmRNA) was conserved with one exception in Alcaligenes eutrophuswhere there was a 1U-C345 mismatch possibly due to a sequencing error. The second (2G-C358,E. colinumbering, Figure [1](#F1){ref-type="fig"}) and third pair (3G-U357) of helix 1 were invariant and therefore neither supported nor disproved by CSA. The identities of the bases involved in the fourth (4G-C356) and fifth pair varied. The closing pair of helix 1 (7G-C353) was conserved with the exception of a 7U-A388 pair the Trichodesmium erythraeumtmRNA. The single-stranded region between helices 1 and 2a ranged from ten in Dehalococcoides ethenogenesto 13 nucleotides in one Clostridium acetobutylicumsequence. A U-A basepair (U6 in chain A with A88 in chain B) was possible in the R. americanamitochondrion tmRNA. Helix 2a was equivalent to the anticodon stem of tRNA and contained eight supported basepairs as well as a short variable internal loop in the 5\'half of the helix that occurred in a few sequences (e.g. Caulobacter crescentus, see [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). The first position in the helix was a conserved cytosine (C21 in E. coli) which formed a weakly-supported basepair with the conserved G333. The partial tmRNA from the chloroplast of Pavlova luthericontained a uracil in the first position, but no information regarding the 3\'portion of helix 2a was available. The T-loop and helix 12 were highly conserved, although many sequences lacked information about helix 12 due to primer annealing during PCR amplification. Helix 12 contained four strongly supported basepairs and a fifth conserved G-C pair (340G-C348 in E. coli; Figure [1](#F1){ref-type="fig"}). The Dehalococcoides ethenogenestmRNA had the potential to form a sixth basepair in helix 12. Helix 12 was almost always continuous, except for the tmRNA of Carboxydothermus hydrogenoformanswhich possessed four basepairs and a mismatched U333 and C347. A 331-GG-332 preceded U333 in C. hydrogenoformansand followed the conserved 328-GAC-330. Therefore, U333 was unlikely to pair. In the T-loop, the U341 and U342 (E. colitmRNA) seen in most sequences were replaced by two guanines in the tmRNA from the R. americanamitochondrion (G79 and G80 in chainB) \[[@B21]\]. In the tmRNA from Caulobacter crescentus, the nucleotide corresponding to U342 in E. colitmRNA was changed to G62 in chainB (see [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). Helical sections 2b, 2c and 2d ------------------------------ Overall, sections 2b, 2c, and 2d were well supported. Sections 2a and 2b were separated by a variable loop ranging from one to seven nucleotides in the 5\'portion and from one to nine nucleotides in the 3\'portion. Sections 2b and 2c had the potential to form a continuously stacked helix (e.g. in Chlamydophila psittacitmRNA). Usually, a bulge of two to six nucleotides separated helical sections 2c and 2d (residues 309--311 in E. colitmRNA, Figure [1](#F1){ref-type="fig"}). An asymmetrical loop was present in some sequences (for example, residues 40--41 in chainA, and 27--31 in chainB of Caulobacter crescentustmRNA, see [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). Helix 2d was the most conserved of the three helical sections. The G43-U308 basepair (E. colinumbering) in helix 2d was only weakly supported, conserved in most phylogenetic groups, but altered in the Thermatogales, Cyanobacteria, alpha-Proteobacteria, and Gram-positive bacteria. A 46A-U334 pair was possible in the Synechocystisspecies PCC6803 tmRNA. Pseudoknot 1 (helices 3 and 4) ------------------------------ Pseudoknot 1 (pk1) was well supported. Of the three connecting regions, the two 5\'-regions were very short (no or only one residue) while the third was relatively long (one to 11 residues). All pseudoknots in tmRNA followed the same general design \[[@B27]\]. Most sequences contained helices 3 and 4, with the exception of the tmRNA from Oenococcus oeniand the partial sequence from the chloroplast of Pavlova lutheri, both of which lacked helix 4 and thus did not form a pseudoknot. Helix 3 usually contained five basepairs. However, a sixth pair was possible in some bacteria. Helix 4 could be split into helicalsections 4a and 4b by a bulge seen in 46 sequences (position57 in B. anthracistmRNA; see [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf) or an internal loop seen in 52 tmRNA sequences. The adenine-rich terminal loop between the downstream halves of helices 3 and 4 ranged in length from twoto13 nucleotides. The mRNA-like region (MLD) -------------------------- The MLD consisted of an open reading frame (ORF) preceding helix 5 and varied from 48 (Heliobacillus mobilis) to 126 nucleotides (Odontella sinensischloroplast). The resume codon usually coded for alanine, but for glycine in 30 sequences (e.g. Bacillus anthracis), aspartic acid in three sequences (e.g. Staphylococcus epidermidis), arginine in two uncultured species (FS1 and LEM2), serine in the uncultured species RCA1, and glutamic acid in Mycoplasma pulmonis. Although helix 5 was only weakly supported by CBCs, recent site-directed mutagenesis experiments combined with functional studies in vivoand in vitro\[[@B28]\] provide strong evidence for its existence. One to three stop codons were located within the helix 5 loop. A single UAA stop codon was present in 157 sequences. UAG (17 sequences) or UGA (10 sequences) were used less frequently. In 85 sequences there were two in-frame stop codons, where UAA was always the first codon, followed by another UAA (73 sequences), UAG (10 sequences) or UGA (2 sequences). Curiously, two sequences (Bacillus megateriumand Chloroflexus aurantiacu) were found to contain three tandem in-frame stop codons. Pseudoknot 2 (helices 6 and 7) ------------------------------ Pseudoknot 2 was well supported and similar in overall design to pk1. Helical sections 6b and 6c showed a potential to form a continuous helix in Thermotoga maritima. In beta-Proteobacteria, 6b was replaced by a short hairpin6d \[[@B1]\]. Helix 6d was observed also in three tmRNAs of the gamma-Proteobacteria Acidithiobacillus ferroxidansand Francisella tularensis. Pseudoknot 3 (helices 8 and 9) ------------------------------ Pseudoknot 3 was well supported but missing in Cyanobacteria and the organelles (Table [1](#T1){ref-type="table"}). Helical sections 8a and 8b were likely to be continuously stacked because a single helix was present in some species such as Aquifex aeolicus. The unusual purine-rich internal loop between helical sections 8a and 8b was present in most gamma-Proteobacteria suggesting a special function. Pseudoknot 4 (helices 10 and 11) -------------------------------- This feature was well supported and was similar in design to the other tmRNA pseudoknots. Helical sections 10a and 10b had the potential to stack because a single helix was present in Prevotella intermedia. In some Cyanobacteria sequences, however, pk4 was replaced with two smaller tandem pseudoknots. Secondary structure prediction of the MLD ----------------------------------------- Because CSA was unable to determine secondary structure for a large portion of the MLD, energy calculations were carried out aimed to predict structure for the single-stranded portion of the open reading frame. The region corresponding to residues 79--107 of E. colitmRNA (Figure [1](#F1){ref-type="fig"}) was extracted from the alignment. A representative alignment of 197 sequences was submitted to Mfold \[[@B29]\]. Each sequence had the potential to form at least one helix, designated \"m\" (see Figure [1](#F1){ref-type="fig"}, [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf, and [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). Two or more adjacent helices were predicted for 17 sequences. The number of basepairs varied from two in Chloroflexus aurantiacusto ten in Mycoplasma pulmonis. Secondary structures of three representative tmRNA molecules ------------------------------------------------------------ Secondary structures were determined for all sequences in the alignment but only three were extracted, diagrammed, and processed for 3-D modeling. Secondary structure of E. colitmRNA ------------------------------------- The 363-nucleotide tmRNA of the gamma-Proteobacterium Escherichia colirepresented the typical tmRNA containing the TLD, the MLD, and four pseudoknots (pk1 to pk4) encompassing the pseudoknot domain (PKD). The 90-GCA-92 resume triplet coded for alanine. Two in-frame UAA stop codons (positions 120--125) were located within the terminal loop of helix 5 (Figure [1](#F1){ref-type="fig"}). Three basepaired regions (shown boxed) were only weakly supported by CSA. Helixm (residues 87--98) was predicted only by energy calculations. A slightly different helix involving residues 88--100 has been suggested by footprinting of E. colitmRNA \[[@B30]\]. The evidence for the 112U-A133 basepair was weak, but was included due to the possibility of extending helix 5 (Materials and Methods). Helical section 5a (residues 108--113 and 134--137) was enlarged by the weakly supported 108G-C137, 110U-A135 and 111U-G134. The 109C-G136 pair was disproved. In helix 10ab, the basepair between 256G-C275 was only weakly supported. Helix 10ab (residues 248--256 and 274--283) could be extended by the boxed 257U-G274 pair. Secondary structure of Bacillus anthracistmRNA ------------------------------------------------ Overall, the secondary structure of Bacillus anthracistmRNA (see [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf) was similar to that of E. coli. One notabledifference was a bulged uridine (U57) between helical sections 4a and 4b in pk1. A three-basepair helixm was predicted. The resume triplet (residues 89--91) coded for glycine, and the UAA stop codon was located at residues 119--121. Two weakly-supported pairs (108C-G132 and 109U-A131) extended helical section 5a. Secondary structure of Caulobacter crescentustmRNA ---------------------------------------------------- Caulobacter crescentustmRNA ([additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf) consisted of two chains, A and B, of 213 and 83 residues, respectively. The resume codon (82-GCG-84 of chainA) coded for alanine and was followed by a UAA stop codon at residues 121--123. Helical sectionsm1 and m2 were predicted by energy calculations. There was weak support for 5a (G109-U135 and 111C-G133 in chainA), and the 106U-A138, 107U-A137, 108C-G136, and 110C-G134 in chainA were disproved. The pseudoknots were relatively small. Helix 11 corresponded to the absent pk4 (residues 1--18 in chain B). Tertiary structure modeling and visualization of tmRNA ------------------------------------------------------ ERNA-3D, a program developed to model RNA in three dimensions \[[@B11]\], was used on an SGI workstation as described in Materials and Methods. E. colitmRNA was selected because this tmRNA is the subject of extensive research. B. anthracistmRNA was chosen as an example of a tmRNA from a Gram-positive bacterium, and C. crescentustmRNA was selected it represents a two-part tmRNA. In order to create the initial models, the sequence and basepairing information were entered into an ERNA-3D input file to automatically generate A-form RNA for the helices sections and specify the single-stranded regions using ERNA-3D\'s algorithm \[[@B11]\]. Since ERNA-3D avoided an XYZ coordinate system as reference for the user, the manipulation of the model from the viewer\'s perspective was simple and intuitive. The coordinates of each model were saved in PDB format \[[@B31]\] for compatibility with other molecular modeling programs. Motifs (listed in Tables [2](#T2){ref-type="table"} to [4](#T4){ref-type="table"}) were selected to model the loops and turns of a particular tmRNA. ERNA-3D selection files were generated to define clusters and place the motif in 3-D without disturbance to the rest of the model. The 3-D cursor box was used to manipulate a cluster in three-dimensional space, similar to the manipulation of a section of a physical model. Numerous high-resolution structures determined by NMR or X-ray crystallography represented a rich source of detailed information for defining biologically meaningful motifs. The SCOR database \[[@B12]\] provided a way to find suitabletemplates. In rare cases when a SCOR search for a motif did not result in an acceptable match (e.g. motif 9, Table [2](#T2){ref-type="table"}), the nucleotides were positioned manually in ERNA-3D. Otherwise, the coordinates were obtained from the Protein Data Bank PDB \[[@B32]\], extracted using the program Swiss-PDBViewer \[[@B33]\], and imported into ERNA-3D. The source motif and the region to be modeled were selected as separate clusters and aligned in three dimensions using common features (usually a shared basepair). Once superimposed, the coordinates of the residues in the source motif were copied onto the corresponding residues in the model. The template was then deleted, leaving a biologically meaningful structure. The backbone connections between the motif and the rest of the model were inspected visually and, if needed, manual adjustments were made to correct bond lengths and tetrahedral angles involving the phosphorous atom at the joint between the extracted motif and the helical structures generated by ERNA-3D. As an example of the motif modeling process, the purine-rich loop in E. colipk3 (positions 204--206 and 223--225) was constructed using a similar loop in the 30S ribosomal subunit. First, the purine-rich loop was defined as motif 11a (Table [2](#T2){ref-type="table"}), and used to search the SCOR database. Positions 780--782 and 800--802 in the structure of the Thermus thermophilus30S ribosomal subunit \[[@B34]\] were found to conform to the motif. The 30S ribosomal subunit coordinates (1J5E.pdb in this case) were downloaded from the PDB and displayed using Swiss-PDBViewer. The coordinates of the loop and the closing basepairs were extracted and inspected to confirm that the structure was compatible. The clustered regions were aligned with the ends of helical sections 8a and 8b at the basepairs 203U-G226 and 207A-U222 of the E. colimodel and 779C-G803 and 783C-G799 of the template. Template positions 780-AAA-782 and 800-GUA-802 were then copied onto 204-GGA-206 and 223-GAA-225 of the model. The template was deleted and the bond lengths and angles involving the atoms of the phosphates of residuesU203, G222, A206, and U222 were adjusted. In some instances, the tmRNA sequence alignment was reinvestigated using ideas derived from the 3-D model. For example, the alignments of pk1 in Bacillus anthracistmRNA and relatives was changed from a two nucleotide bulge (56-AU-57) between helical sections 4a and 4b to a more feasible and equally well supported one-nucleotide bulge (U57, see [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf). The alignment of helix 10 in pk3 in B. anthracistmRNA and relatives was altered from a 237C-A269 mismatch and an asymmetrical loop (C239 and 266-GU-267) to a single looped-out C269. The alignment of pk3 of Caulobacter crescentusand relatives was changed from four basepairs and a weakly supported fifth pair in helix 8 (between 174G-C196 of chainA) to the four basepair structure seen in [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf. Information about spatial neighborhoods as obtained from cross-linking, site-directed mutagenesis, and functional testing of E. colitmRNA was introduced and is described in detail below. All models were inspected visually for correct bond angles and distances around the phosphorous atoms at the joints between the extracted motifs and the helical regions generated by ERNA-3D. The coordinates are provided as [additional file 7](#S7){ref-type="supplementary-material"}: Ecoli-closed.pdb, 8: Ecoli-open.pdb, 9: Banthracis-closed.pdb, and 10: Ccrescentus-closed.pdb. 3-D model of E. colitmRNA --------------------------- The model shown as a ribbon diagram in Figure [3](#F3){ref-type="fig"} consists of a compacted MLD and PKD with the TLD extending from the body of the molecule due to the near-coaxial stacking of the helix 2 sections. The coordinates for the TLD were taken from a previous model \[[@B35]\] which is based on two cross-linked sites, one formed between nucleotides U9/U10 near the 5\' end and nucleotides C346/U347 in the T loop, the other involving residues at positions 25--28 and 326--329 within helix 2a (motif 2 in Table [2](#T2){ref-type="table"}). Important features of the TLD include the non-Watson-Crick base pairs formed by 19-GA-20 and 333-GA-334 which have been confirmed by site-directed mutagenesis \[[@B36]\]. A very efficient UV-induced cross-link observed between the stop codon loop of helix 5 and pk2 of E. colitmRNA (Wower et al., unpublished) introduced a considerable constraint of helices 5, 6, and 7, and, as has been shown recently, is consistent with the cryo-EM structure of ribosome-bound tmRNA of the initial stage of trans-translation \[[@B10]\]. Also considered was the previously-discovered covariation \[[@B37]\] between C44 and C66 (E. colinumbering, Figure [1](#F1){ref-type="fig"}) which determines the orientation of helix 2 in relation to helix 3 and thus the approximate angle by which the TLD protrudes. The 44/66 covariation is strongly supported (26 covariations versusfour mismatches) in an alignment of 143 representative sequences (not shown). Since this is a non-Watson-Crick covariation, it is difficult to propose a precise structure in this region. More extensive studies will be required to better understand the nature of the 44/66 covariation. Overall, the distance between the 3\'end and A231 in pk3 was 180 Å, and 70 Å between the outside edges of pk1 and pk4. Helix 5 and pk2 were positioned in a parallel fashion. The nucleotides in the bulge between helical sections 6a and 6b (motif 9, Table [2](#T2){ref-type="table"}) were adjusted manually to allow for a close fit of helix 5 and pk2. This model is supported by the finding that mutations that destroy base pairing in helix 5 substantially decrease tmRNA tagging activity (Wower et al, in press) and abolish the long-distance interaction between helix 5 and pk2 as judged by the absence of a cross-link between the stop codon loop of helix 5 and pk2 (Wower, unpublished data). Evidence for the existence of helix 5 has been provided by the analysis of compensatory mutations which completely restore tagging \[[@B28]\]. Each of the four pseudoknots displayed the previously determined structural properties characterized by extensive helical stacking \[[@B27]\]. The MLD and pseudoknots were arranged in a central loop with the resume codon positioned near the junction between helices 2a and 2b (motif 3a, Table [2](#T2){ref-type="table"}). Our model reflects the finding that the pseudoknots are functionally interchangeable \[[@B38]\] and thus are likely to retain a considerable level of structural independence. Furthermore, data derived from cross-linking experiments showing that pk2 and pk3 are in close proximity whereas helix 5 and pk4 are further apart (Wower, unpublished data) agree with the presented model. 3-D model of B. anthracistmRNA -------------------------------- The 3-D model of the B. anthracistmRNA (Figure [4](#F4){ref-type="fig"}) was similar to the E. colimodel. A sharper angle was modeled between helix 2 axis and the PKD. The dimensions were 150 Å from the 3\'end to the distant edge of pk2, and approximately 80 Å between the outer edges of pk2 and pk4, respectively. 3-D model of C. crescentustwo-part tmRNA ------------------------------------------ Compared to the other two tmRNA models, the two-part tmRNA Caulobacter crescentusmodel (Figure [5](#F5){ref-type="fig"}) was slightly more elongated. It measured 195 Å from the 3\'-end of chainB to the single-stranded region between pk2 and pk3. The distance between helix 5 and the 3\'end of chainA was 55 Å. Discussion ========== We have compared a growing number of tmRNA sequences from all groups of bacteria to produce an alignment from which the secondary structure of any tmRNA could be easily extracted. Most basepairs were supported by phylogenetic evidence, whereas only a few helical sections required energy calculations. Uncertainties in assignment of basepairs, such as the pseudoknot region of chloroplasts and one-piece cyanobacterial tmRNAs, may be eliminated in the future when more sequences will become available. The common layout of the secondary structures indicated a similar function in all bacteria. The number and size of the pseudoknots varied, supporting the idea that the pseudoknots may only enhance the essential functions carried by the TLD and the MLD \[[@B28]\]. Differences in the secondary structure features were usually not random but occurred between groups of related organisms. For example, helix 6d was present only in the beta- and three close relatives of the gamma-Proteobacteria. Whether these group-specific features are responsible for differences in the trans-translation mechanism remains to be determined. However, strategies that exploit these differences, for example for developing new antibiotics targeted at a specific group of bacteria, can now be envisioned. In principle, tertiary structure models of any tmRNA in the alignment could be built using the described procedures. Here, we have shown how to generate a biologically more meaningful model of E. colitmRNA which represents a significant progress from a previous model \[[@B27]\]. We also constructed 3-D model of the tmRNAs of B. anthracisand C. crescentus. Overall, the three models were similar in shape and size confirming that all tmRNAs have the potential to function similarly in trans-translation. The TLD mimicked the L-shape of canonical tRNA \[[@B39]\] and may be necessary for proper association tmRNA with the EF-Tu, SmpB, and subsequent binding to the ribosomal A-site. The lack of a D-stem has been suggested to confer flexibility \[[@B40]\], but SmpB may be responsible for stabilizing this region \[[@B41],[@B42]\]. Differences in the shapes of the three tmRNA models (e.g. the angle between helix 2 and the main body of the molecule) may be due to the difficulty in determining the precise arrangement of the pseudoknots. Considering that the pseudoknots are likely to constitute relatively independent structural units, conformational changes might occur around the connecting single strands, as well as in the MLD and the weakly-supported helix m. TmRNA may become less flexible when bound to proteins such as SmpB and ribosomal protein S1. EF-Tu, however, likely binds to the coaxially-stacked helices 1 and 12 \[[@B43]\], and therefore appears to have little effect on the conformation of the TLD. Protein SmpB was found to bind near helix 2a \[[@B41],[@B44]\], has two RNA binding sites \[[@B45]\], and thus could make additional contacts with other regions. Protein S1 is the largest ribosomal protein, has been shown to be close to numerous sites, and to be required for the binding of tmRNA to the ribosome \[[@B7]\]. Since S1 is a flexible, beadlike protein \[[@B46]\] it may not restrict the conformational potential of the tmRNA molecule. Instead, the protein may instill some constraint to the large central loop formed by the PKD and the MLD. Because S1 is known to melt helices in mRNAs \[[@B47]\], it is also possible that it unwinds helix m and exposes the resume codon and the preceding nucleotides U85 and A86 for efficient trans-translation \[[@B48],[@B49]\]. The tmRNA models show the resume codon in close proximity to the internal loop formed between helical sections 2a and 2b. This arrangement would allow the ribosome to \"jump\" a relatively short distance from the end of the broken mRNA onto the ORF of tmRNA. In a recent cryo-EM study of the initial stage of trans-translation \[[@B10]\], the tRNA-like region, SmpB, EF-Tu, and part of pk4 were located in the A-site of the ribosome. We suggest that this more open arrangement is made possible due to the flexibility of tmRNA, the melting of helix m, and/or a change in conformation induced by the binding of tmRNA to the ribosome (Figure [6](#F6){ref-type="fig"}). The opening of the central loop seems to be accompanied by a rotation of the TLD around the helix 2 axis (compare Figure [6A](#F6){ref-type="fig"} and [6B](#F6){ref-type="fig"}) and thus might properly align the resume codon with the 3\'-end of broken mRNA in the ribosomal decoding centre. At the later stages of the transit of tmRNA across the ribosome even more dramatic conformational changes were shown to disrupt helix 5 and the pseudoknotted regions (Wower et al, in press). These downstream alterations are likely mediated not by protein S1 but by the intrinsic helicase activity of the ribosome \[[@B50]\] and are required to maintain the ribosomal subunits in close proximity to the unfolded tmRNA in order to monitor trans-translation. Conclusion ========== This study significantly advances our understanding of trans-translation by providing biologically feasible 3-D models for the entire tmRNA molecule. Although the modeling of only three tmRNAs has been described here, 3-D models of every tmRNA can be extracted from the alignment. The models are characterized by a functionally significant features, including biologically relevant structures for the single-stranded regions and the close proximity between the TLD and the resume codon. Conformational changes induced by binding of tmRNA to SmpB, ribosomal protein S1, and the ribosome suggest a transformation of a free compact tmRNA to a more open ribosome-bound structure. The comparative modeling approach described here for tmRNA is easily adapted for other RNA classes. Methods ======= Comparative sequence analysis ----------------------------- The tmRNA sequences were arranged in phylogenetic order using information available in the RDP \[[@B19]\]. When the phylogenetic order could not be determined, the sequence was placed next to the closest relative as determined by the ClustalW plug-in of BioEdit \[[@B18],[@B51]\]. The sequences were made available at the tmRDB \[[@B52]\]. Aligning was done manually using BioEdit \[[@B18]\] with details described previously \[[@B13]\]. Briefly, closely related sequences were aligned first. Then, invariant positions were used as guides to align the dissimilar regions. Next, common secondary structure elements were identified by observing covariations and find support for basepairs, tertiary interactions, or other structural features. Compensatory base changes (CBCs) were observed if a change in one residue of a Watson-Crick or G-U pair was compensated by a second change to maintain basepairing. Two residues were mismatched if they did not form a Watson-Crick or G-U pair. CBCs and mismatches were counted to determine positive and negative evidence in order to prove or disprove the existence of a particular pair. A basepair was considered proven if there was at least twice as much positive than negative evidence. Invariant pairs provided neither positive nor negative evidence. If a basepair was proven in one phylogenetic group and disproved in another group, the basepair was considered to be specific to that group. The alignment and suggested CBCs were checked using RNAdbTools \[[@B22]\] to eliminate incorrectly-paired nucleotides, suggest extensions of helices, and determine the phylogenetic support for each basepair. Weakly supported basepairs adjacent to supported basepairs were considered an extension of the helix and usually included in the secondary structures (Figure [1](#F1){ref-type="fig"}, [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf and [6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). 3-D model building ------------------ The secondary structure information was used as input for ERNA-3D \[[@B11]\] installed on an SGI workstation running IRIX 6.5. ERNA-3D generated A-form RNA for each helix and calculated the conformations of single-stranded regions. The models were examined using CrystalEyes stereovision goggles and an StereoGraphics infrared emitter. Structural motifs were identified using SCOR \[[@B12]\], the coordinates were obtained from the Protein Data Bank (PDB) \[[@B32]\], extracted using Swiss-PDBViewer \[[@B33]\], and superimposed onto the model. Data obtained from site-directed mutagenesis, cross-linking experiments, or the literature were incorporated, and bond lengths and angles were adjusted manually to produce biologically feasible models. The final models were saved in PDB format (additional files [7](#S7){ref-type="supplementary-material"}: Ecoli-closed.pdb, [8](#S8){ref-type="supplementary-material"}: Ecoli-open.pdb, [9](#S9){ref-type="supplementary-material"}: Banthracis-closed.pdb, and [10](#S10){ref-type="supplementary-material"}: Ccrescentus-closed.pdb and viewed in iMol \[[@B53]\] to create the ribbon diagrams shown in Figures [3](#F3){ref-type="fig"} to [5](#F5){ref-type="fig"}. Authors\' contributions ======================= JB identified and aligned a large number of the tmRNA sequences, carried out the 3-D modeling experiments, and drafted portions of the text. CZ conceived the study, participated in all of its aspects, and wrote the final manuscript. FM developed the ERNA-3D program and incorporated new functions to allow comparative modeling of RNA. IW developed assays which allowed to test the biological significance of the Escherichia colitmRNA model. JW participated in the coordination of the study and provided RNA-RNA cross-link data which were crucial for constraining the models in 3-D. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 tmRNA alignment. Species names are shown on the left with their tmRDB ID (see <http://psyche.uthct.edu/dbs/tmRDB/tmRDB.html>). Supported base pairing are shown with upper case letters and are indicated on the bottom. Secondary structure features are indicated on the top. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 tmRNA alignment. Species names are shown on the left with their tmRDB ID (see <http://psyche.uthct.edu/dbs/tmRDB/tmRDB.html>). Supported base pairing are shown with upper case letters and are indicated on the bottom. Secondary structure features are indicated on the top. Each sequence is shown on a single long line suitable for import into other programs. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 tmRNA alignment. Species names are shown on the left with their tmRDB ID (see <http://psyche.uthct.edu/dbs/tmRDB/tmRDB.html>). Supported base pairing are shown with upper case letters and are indicated on the bottom. Secondary structure features are indicated on the top. The alignment is shown in sections for easier viewing and printing. This arrangement is not suited for import into other programs. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 tmRNA alignment. tmRNA alignment in msf format suitable for import into editors which support the msf format. Can be used to convert into other sequence formats using, for example the Readseq server at <http://bioweb.pasteur.fr/seqanal/interfaces/readseq.html> ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Secondary structure of Bacillus anthracistmRNA. Phylogenetically-supported helices are highlighted in gray and numbered from 1 to 12. The 5\' and 3\' ends are indicated. Arrows represent connections from 5\' to 3\'. Residues are numbered in increments of ten. Weakly supported regions and basepairs are show in boxes. The star labels the first nucleotide of the resume codon. The tag peptide sequence is shown below the mRNA-like region. The stop codon is indicated with a solid arrowheads. Three domains are distinguished: The tRNA-like domain (TLD), the mRNA-like domain (MLD), and the pseudoknot domain (PKD). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 Secondary structure of Caulobacter crescentustmRNA. Phylogenetically-supported helices are highlighted in gray and numbered from 1 to 12. The 5\' and 3\' ends of both chains are indicated. Arrows represent connections from 5\' to 3\'. Residues are numbered in increments of ten. Weakly supported regions and basepairs are show in boxes. The star labels the first nucleotide of the resume codon. The tag peptide sequence is shown below the mRNA-like region. The stop codons are indicated with solid arrowheads. Three domains are distinguished: The tRNA-like domain (TLD), the mRNA-like domain (MLD), and the pseudoknot domain (PKD). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 PDB coordinates of the closed form of Escherichia colitmRNA. 3-D model of the closed form of Escherichia colitmRNA. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 PDB coordinates of the open form of Escherichia colitmRNA. 3-D model of the open form of Escherichia colitmRNA. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 9 PDB coordinates of the closed form of Bacillus anthracistmRNA. 3-D model of the open form of Bacillus anthracistmRNA. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 10 PDB coordinates of the closed form of Caulobacter crescentustmRNA. 3-D model of the closed form of Caulobacter crescentustmRNA. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was supported by Grant GM58267 to JW from the National Institutes of Health. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *Secondary structure of E. colitmRNA. Phylogenetically-supported helices are highlighted in gray and numbered from 1 to 12. The 5\' and 3\' ends are indicated. Arrows represent connections from 5\' to 3\'. Residues are numbered in increments of ten. Weakly supported regions and basepairs are shown in boxes. The disproved potential pairing of C109 with G136 is labeled with an open arrowhead. The star labels the first nucleotide of the resume codon. The tag peptide sequence is shown below the mRNA-like region. The stop codons are indicated with solid arrowheads. Three domains are distinguished: the tRNA-like domain (TLD), the mRNA-like domain (MLD), and the pseudoknot domain (PKD). ::: ![](1471-2199-6-14-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Motif modeling procedure. Motifs, for example the nonamer-loop shown in the top-left panel, were identified in the known high-resolution structures (top-right) with the help of SCOR \[12\]. The PDB coordinates were extracted (bottom-right) and compared with the 3-D model generated by ERNA-3D (bottom-left) to deduce relevant models. ::: ![](1471-2199-6-14-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### 3-D model of Escherichia colitmRNA. The 3-D model of Escherichia colitmRNA is viewed as a ribbon diagram from the side in panel A, the top in panel B, and in panel C turned by approximately 90° around the y-axis in relation to A. Panel D shows a representation of the corresponding 2-D structure using the identical coloring scheme. Labeled are the 5\' and 3\' ends, the resume (R) and stop codons (S), and the three domains (TLD, MLD, PKD). The figure was produced with iMol \[53\] and the PDB coordinates of [additional file 7](#S7){ref-type="supplementary-material"}: Ecoli-closed.pdb. ::: ![](1471-2199-6-14-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### 3-D model of Bacillus anthracistmRNA. The 3-D model of Bacillus anthracistmRNA is viewed as a ribbon diagram from the side in panel A, the top in panel B, and in panel C turned by approximately 90° around the y-axis in relation to A. Panel D shows a representation of the corresponding 2-D structure using the identical coloring scheme. Labeled are the 5\' and 3\' ends, the resume (R) and stop codons (S), and the three domains (TLD, MLD, PKD). The figure was produced with iMol \[53\] and the PDB coordinates of [additional file 9](#S9){ref-type="supplementary-material"}: Banthracis-closed.pdb. ::: ![](1471-2199-6-14-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### 3-D model of Caulobacter crescentustmRNA. The 3-D model of Caulobacter crescentus stmRNA is viewed as a ribbon diagram from the side in panel A, the top in panel B, and in panel C turned by approximately 90° around the y-axis in relation to A. Panel D shows a representation of the corresponding 2-D structure using the identical coloring scheme. Labeled are the 5\' and 3\' ends, the resume (R) and stop codons (S), and the three domains (TLD, MLD, PKD). The figure was produced with iMol \[53\] and the PDB coordinates of [additional file 10](#S10){ref-type="supplementary-material"}: Ccrescentus-closed.pdb. ::: ![](1471-2199-6-14-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Conformational changes in Escherichia colitmRNA. Panel A: closed form of the E. colimodel as shown in Figure 3. Panel B: open conformation adjusted to more closely resemble the ribosome-bound form as determined by cryo-EM \[10\] using [additional file 8](#S8){ref-type="supplementary-material"}: Ecoli-open.pdb. Panel C: coordinates extracted from the cryo-EM model \[10\]. The TLD is shown in dark purple, helix 2 in green, pk1 in yellow, helix 5 in pink, pk2 in turquoise, pk3 in red, and pk4 in dark blue. Panel D: Electron density map of the 50S subunit in light blue, the 30S subunit in yellow, and the bound tmRNA (in the absence of ribosomal protein S1) in dark blue (from \[10\]). ::: ![](1471-2199-6-14-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Phylogenetic distribution of tmRNA features ::: Phylogenetic Group* TLD MLD pk1 pk2 pk3 pk4 Other -------------------------------------- --------- ------- --------- --------- --------- --------- ----------- Thermophilic Oxygen Reducers Thermatogales Green Non-sulfur & Bacteria elatives Flexibacter Cytophaga Bacteroides Green Sulfur Bacteria Planctomyces & elatives Cyanobacteria 1 1,2 3 Plastids - - Mitochondria - - - - - Fibrobacter/Acidobacter & elatives Spirochetes & elatives Proteobacteria, alpha 4 5 3 Proteobacteria, beta 6 3 Proteobacteria, gamma 6 Proteobacteria, delta Proteobacteria, epsilon Fusobacteria Gram Positive Bacteria The tRNA-like domain (TLD), mRNA-like domain (MLD), and the four pseudoknots pk1 to pk4 are shown on the top. Other features peculiar to a phylogenetic group are in the right column. White fields indicate the presence, dashes the absence of a feature. Numbers suggest the following structural features: (1) certain Cyanobacteria lack these pseudoknots. (2) One-chain cyanobacterial tmRNAs contain two smaller tandem pseudoknots named pk4a and 4b. (3) The tmRNAs of some species in this group consist of two basepaired molecules \[14, 21\]. (4) The genus Ricketsiaand its relatives lack pk2. (5) pk4 of the alpha-Proteobacteria has been reduced to a single helix (named helix 11). (6) Some species in this group contain an additional helix (helix 6d). ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Structural motifs used for the Escherichia colitmRNA model ::: Motif SCOR class tmRNA Res. Source Res. Coordinates Comments ----------- --------------------------------------------- -------------------- ------------------------ ----------------------- --------------------------------- 1 1--7, 353--363 1--7, 12--22 1IKD.pdb (chain W) ACCA end and G3-U357 pair 2 8--28, 325--352 8--28, 325--352 tmx-34.pdb from tmRDB 3a internal loop with unpaired stacked bases 29--33, 321--324 1775--1779, 1765--1768 1JJ2.pdb 3b stacked duplex with one non-WC pair C35, A319 ERNA-3D 4 stacked duplex with two non-WC pairs 38--39, 315--316 2874--2875, 2882--2883 1JJ2.pdb 5 309--311 ERNA-3D 6a pseudoknot 49--78 1--33 1RNK.pdb pk1 6b tetraloop 87--98 5--8 1AFX.pdb the only YRRR tetraloop in SCOR 7 nonaloop 118--126 1834--1842 1JJ2.pdb 8 one unpaired and stacked U U131 U30 1B36.pdb 9 171--174 ERNA-3D 10a stacked duplex with two non-WC pairs 149--150, 165--166 288--289, 363--364 1JJ2.pdb 10b pseudoknot 138--196 6a pk2 11a internal loop 204--206, 223--225 780--782, 800--802 1J5E.pdb 11b pseudoknot 197--247 6a pk3 12 stacked duplex with one non-WC pair G258, A273 A-G6, B-A27 420D.pdb 13a stacked duplex with one non-WC pair C266, U296 ERNA-3D 13b pseudoknot 248--299 6a pk4 Shown in columns one to four are the motif numbers in bold, their SCOR classification \[12\], the residue positions in the tmRNA model and the source structure. column five lists the filenames containing the atomic coordinates that were derived from the PDB \[31\], the tmRDB \[52\], or were generated by ERNA-3D \[11\]. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Structural motifs used for the Bacillus anthracis tmRNA mode ::: Motif SCOR class tmRNA Res. Source Res. Coordinates Comments ----------- ------------------------------------------- --------------------------------- -------------------------------- --------------------------- ------------------------------------- 1 1--11, 12--27, 316--355 1--11, 13--28, 324--363 E. coli1 ACCA end and G3-U349 pair 2 loop with stacked interdigitated bases 28--33, 312--316 477--482, 451--455 1J5E.pdb 3 300--301 ERNA-3D 4a 43--53, 61--65, 294--299 43--53, 63--67, 302--308 E. colimodel helix 2d, helix 3 4b pseudoknot 58--60, 66, 67--73 12--14, 20, 23--29, 1RNK.pdb pk1, helix 4b 4c 54--56, 74--75 54--56, 77--78 E. coli6a pk1, helix 4a 5 heptaloop 82--88 335--341 1JJ2.pdb 6a one looped out A A112 210 1GID.pdb 6b C114, C127 ERNA-3D 6c octaloop 117--124 1499--1506 1JJ2.pdb 7a 134--138, U163, A164, 165--169, 140--144, A171, A174, 175--179 E. coli10b helix 6a, looped out U163 and A164. 7b one looped out U U158 U87 1JJ2.pdb 7c 144--147, 148--156, 180--185 259--262, 264--272, 293--298 E. coli13b helix 6c, helix 7, pk2 8 G133, C170, 171--177, 178--179 C138, G181, 182--188, 190--191 E. coli10b 9a C190, 191--198, 208--215, 216 G200, 203--210, 219--226, U229 E. coli11a, 11b purine-rich loop, pk3 9b pseudoknot 202--206, 228--232 9--13, 28--32 1RNK.pdb pk3, helix 9 10a 199--201 154--156 E. coli6a pk3 10b 217--227 229--239 E. coli6a pk3 11a one looped out C C269 C6 1BVJ.pdb 11b stacked duplexes with four non-WC pairs 241--245, 260--263 795--798, 815--818 1JJ2.pdb 12 246--259, 275--290 259--277, 284--299 E. coli6a helix 10c, helix 11ab, pk4 Motif numbers in bold, their SCOR classification \[12\], residue positions in the tmRNA model and the source structure are shown in columns one to four. Column five lists the filenames containing the atomic coordinates that were derived from the PDB \[31\], the E. colimodel (see Table 2), or were generated by ERNA-3D \[11\]. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Structural motifs used for the Caulobacter crescentustmRNA model ::: Motif SCOR class tmRNA Res. Source Res. Coordinates Comments ----------- -------------------------------------------------- ------------------------------------------------------------------ ------------------------------------------------------------ ----------------------------------- --------------------------------------------------------------- 1 A-1-13, A-14-21, B-50-83 1--13, 15--22, 322--355. B. anthracis1 \- 2a one non-Watson Crick pair and one looped-out A A-22, B-48-49 14, 4--5 5MSF.pdb disruption in anticodon stem 2b loop with base triple A-26-31; B-40-44 8--13, 24--28 1FMN.pdb 2c A-C36, B-C35 ERNA-3D 3 A-40-41, B-27-31 ERNA-3D 4 A-47-61, A-62-70 196--210, 224--232 B. anthracis8a, 8b, 9a, 9c pk1 5 A-87-93 82--88 B. anthracis5 terminal heptaloop of helix m2 6 one looped out A and one or more non-WC pairs A-78-81, A-99-101 113--116, 205--207 1GID.pdb internal loop between helices m1 and m2 7 B-8-11 284--285, 291--292 1D6K.pdb 8a A-112-114, A-131-132 ERNA-3D 8b A-119-125 82--88 B. anthracis5 9a U148, C163 ERNA-3D 9b A-156-165, A-146-153, A-167-171 157--166, 149--156, 192--196 E. coli10b pk2. Single-stranded regions were adjusted to connect helices 10 A-154-156, A-170-171 54--55, 76--77 E. coli6a Correction of two pairs in pk2 based on E. colipk1 11 A-211-214, A-177-182, A-184-193, A-177-178, A-192-198, A-194-209 360--363, 207--212, 213--222, 200--201, 228--234, 230--245 E. coli1(ACCU tail) and 11b pk3 and ACCU tail of pk3 Motif numbers in bold, their SCOR classification \[12\], residue positions in the tmRNA model and the source structure are shown in columns one to four. Column five lists the filenames containing the atomic coordinates that were derived from the PDB \[31\], the E. colior B. anthracismodels (see Tables 2 and 3), or were generated by ERNA-3D \[11\]. :::	PubMed Central	2024-06-05T03:55:58.233010	2005-6-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168896/", "journal": "BMC Mol Biol. 2005 Jun 15; 6:14", "authors": [ { "first": "Jody", "last": "Burks" }, { "first": "Christian", "last": "Zwieb" }, { "first": "Florian", "last": "Müller" }, { "first": "Iwona", "last": "Wower" }, { "first": "Jacek", "last": "Wower" } ] }
PMC1168897	Background ========== The human small intestine is the longest segment of the gastrointestinal tract, approximately five to six meters long in adults. Although it is three to four times longer than the colon, small intestinal tumors are rare, forming about 1% of all gastrointestinal malignancies \[[@B1]\]. Like the colon, the small intestine is composed of multiple crypts (of Lieberkühn), which also contribute cells to finger-like projections called villi (Figure [1](#F1){ref-type="fig"}). Stem cells appear to reside near the crypt bases, producing numerous differentiated progeny that die within a week as they migrate upwards \[[@B2]-[@B4]\]. Stem cells also produce Paneth cells that migrate downwards, and subsequently survive a few weeks. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic of a crypt of Lieberkühn and villus.Stem cells probably reside within a niche just above the Paneth cells. Differentiated cells migrate out of the niche and up a villus, and die within a week. Illustrated are an isolated crypt and villus fragments. A cross-section illustrates that the villus fragments were essentially hollow tubes of epithelium. The schematic of stem cell lineages illustrates that cell division is always asymmetric (one stem cell and one differentiated daughter) in immortal stem cell lineages, and predominately asymmetric in a stem cell niche. However, niche symmetric divisions occasionally occurs (yellow region) leading to loss of one lineage and the expansion of another, such that total niche stem cell numbers remain unchanged. In this way, niches allow heterogeneous crypts to become homogenous. ::: ![](1741-7015-3-10-1) ::: Although there are likely to be many stem cells per crypt \[[@B2]-[@B4]\], intestinal stem cells have not been directly isolated or identified, and their properties and numbers are uncertain. Stem cell lineages may be immortal if at least one daughter retains stem cell properties after each division. However, mouse studies are more consistent with periodic losses of crypt stem cell lineages. The primary observation is that heterogeneous crypts created through mutagenesis or with chimeras eventually become homogenous \[[@B5]\]. If stem cell lineages were immortal, heterogeneous crypts would not become homogeneous. Stem cell turnover by a niche mechanism \[[@B5],[@B6]\] can explain how heterogeneous crypts become homogeneous (Figure [1](#F1){ref-type="fig"}). Niche stem cells divide but total niche stem cell numbers remain unchanged as daughter cells constantly leave the niche and differentiate. Usually a stem cell produces one stem cell and one differentiated daughter (\~95% of the time in mice \[[@B7]\]), but sometimes a stem cell lineage may become extinct when both daughter cells leave the niche, balanced by another stem cell lineage that expands when both daughters remain within the niche. Such a niche population-type mechanism (random stem cell loss with replacement) eventually results in the loss of all stem cell lineages except one, allowing heterogeneous crypts to become homogeneous (Figure [1](#F1){ref-type="fig"}). Human stem cell studies are difficult because the powerful visible fate-marking strategies of animal models are impractical. Most human crypts are visually homogeneous and therefore consistent with either immortal or niche stem cell mechanisms. A study of human colon crypts after therapeutic radiation was consistent with stem cell niches because the frequencies of mutant heterogeneous crypts progressively declined after radiation \[[@B8]\]. Similar studies have not been performed on human small intestines. One method that can reconstruct the past without direct serial observations infers ancestry from present day sequences. The past is encoded in sequences because errors may accumulate in a clock-like fashion (the molecular clock hypothesis \[[@B9]\]) and record relationships between cells and times since the most recent common ancestor or bottleneck. Just as the age of the human race may be inferred by comparing contemporary genomes, so the ages of crypt stem cell populations may in theory be inferred from their sequences. With immortal stem cell lineages, adult crypts will have greater sequence diversity compared to crypts maintained by stem cell niches that periodically \"bottleneck\" whenever all but one current stem cell lineage are lost. Therefore, crypt diversity measurements potentially reveal whether stem cell lineages are immortal (last bottleneck around birth) or turn over by a niche mechanism (more recent bottlenecks). Mutations are rare in normal cells, but recent studies illustrate that stem cell ancestries may be automatically recorded by random somatic epigenetic errors that frequently accumulate with aging \[[@B10]\]. Methylation increases with age at certain CpG-rich regions in the colon \[[@B11],[@B12]\]. This age-related methylation does not resemble a stereotypic developmental process because different cells within the same intestine may have different methylation patterns (5\' to 3\' order of methylated sites). Such random somatic errors resemble the drift of genomes during species evolution, and by analogy, such drift can be used to reconstruct somatic cell ancestry because methylation exhibits somatic inheritance \[[@B13]\] and all cells are related. Therefore, somatic methylation patterns potentially encode numbers of divisions since birth, and ancestral relationships among cells within a single crypt or villus. Here we infer from methylation patterns that crypt niches containing multiple stem cells maintain the human small intestine. Methods ======= Specimens --------- Crypts or villi were isolated from 1--2 cm^2^patches of fresh small intestines obtained from 13 patients undergoing surgery for non-small intestinal diseases (Table [1](#T1){ref-type="table"}), using an EDTA-containing solution as previously performed for colon crypts \[[@B10]\]. The research was approved by our Institutional Review Board and is in compliance with the Helsinki Declaration. Individual crypts or villus fragments were identified under a dissecting microscope (Figure [1](#F1){ref-type="fig"}) and placed in 0.5 ml microfuge tubes. Villus fragments were essentially hollow tubes of epithelium without stroma. Nine crypts were analyzed from 13 intestines for CSX, eight crypts from six intestines for MYOD, and 4--10 villi (average of seven) were analyzed for CSX from six intestines. The DNA was extracted and bisulfite converted as previously described \[[@B10]\]. DNA was also extracted from microdissected paraffin-embedded small intestinal tissues of an eight month old and a four year old. The bisulfite treated DNA from about half a crypt was amplified by PCR at CpG-rich segments of CSX or MYOD \[[@B10]\]. PCR products were cloned into bacteria and eight clones were sequenced from each crypt or villus. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Specimens ::: Patient Age/Sex Site Surgery Specimen Tags --------- --------- ---------- -------------------- -------------- ----------- A 17/F Ileum Constipation Crypt CSX, MYOD B 27/F Ileum Inflammatory Bowel Crypt, Villi CSX C 46/F Ileum Ovarian Cancer Crypt, Villi CSX, MYOD D 51/M Duodenum Whipple Crypt CSX E 52/M Ileum Colorectal Cancer Crypt, Villi CSX F 58/F Duodenum Whipple Crypt CSX, MYOD G 58/F Ileum Colorectal Cancer Crypt,Villi CSX, MYOD H 63/F Duodenum Whipple Crypt CSX I 67/F Duodenum Whipple Crypt CSX, MYOD J 72/M Ileum Colorectal Cancer Crypt CSX K 75/F Ileum Colorectal Cancer Crypt,Villi CSX L 76/M Ileum Colorectal Cancer Crypt CSX, MYOD M 78/M Duodenum Whipple Crypt,Villi CSX N 0.07/F Ileum Obstruction Paraffin CSX, MYOD O 4/M Ileum Trauma Paraffin CSX, MYOD Ileum specimens were proximal portions of total colectomies. ::: Stem cell dynamics from methylation patterns -------------------------------------------- Small intestinal methylation patterns were analyzed by the same quantitative approach we applied to colon crypt methylation patterns (\[[@B10]\], and its Supplementary Materials). Human crypts of Lieberkühn containing a total of \~500 cells \[[@B14]\] were simulated on a computer, modeling the processes of stem cell division (symmetrical or asymmetrical), differentiation, death, and random methylation errors. The parameters of the model (Table [2](#T2){ref-type="table"}) are numbers of stem cells per crypt, probabilities of symmetric or asymmetric stem cell division, a methylation error rate, and numbers of divisions since birth. These parameters are uncertain for the human small intestines, and the values are just guesses chosen to be consistent with the biology of the intestines, and are similar to the values we estimated for the colon \[[@B10]\]. We emphasize that they are just estimates, and further experiments are needed to define these values better. The model starts with unmethylated sites in the CpG-rich regions of CSX or MYOD, and errors are assumed to accumulate independently between CpG sites with the same rates as simulated in the colon \[[@B10]\]. The output is the methylation patterns of these regions after a given number of divisions. A total of 1000 simulations were performed for each crypt scenario. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Model Parameters ::: Parameter Stem Cell Niche Immortal Lineages ------------------------------------------------------------------------------------------ ----------------------------------------- ----------------------------------------- Stem Cells per Crypt 4--256 2 Probability of Asymmetric Stem Cell Division (P1)\* 0.98 to 0.89 1.0 Probability of Symmetric Stem Cell Division: zero (P0) or two (P2) stem cell offspring\* 0.02 to 0.11 0 Methylation Error Rate 2 × 10^-5^per CpG site per division\\ 2 × 10^-5^per CpG site per division\\ Stem Cell Division Rate 0.75 per day 0.75 per day \P0+P1+P2 = 1, with P0 = P2 \\* Equal probabilities for methylation and demethylation errors ::: A stem cell hierarchy is simulated with small numbers of stem cells that produce differentiated mitotic cells, which subsequently become differentiated non-mitotic cells. For example, if there are two stem cells per crypt, differentiated mitotic cells divide six more times before becoming non-mitotic. Differentiated non-mitotic cells remain through two more divisions with the oldest cells dying after each division cycle to maintain a constant number of crypt cells. In this scenario, there are 512 cells per crypt -- two stem cells, 126 differentiated mitotic cells, and 384 differentiated non-mitotic cells. With greater numbers of stem cells, numbers of differentiated mitotic cells and their divisions are correspondingly reduced to maintain a constant size of the mitotic compartment and 512 cells per crypt. Immortal stem cell lineages were simulated with strictly asymmetric division producing one stem cell and one differentiated daughter (P1 = 1.0). After a given number of stem cell divisions, eight alleles are randomly sampled from each simulated crypt, as in the experimental approach. Niche and immortal scenarios were identical except niche stem cells also exhibit symmetric divisions (P1 \< 1.0). Total niche stem cell numbers remain constant because divisions that produce two differentiated daughters are balanced by divisions that produce two stem cell daughters (P0 = P2). Symmetric divisions tend to reduce crypt diversity because methylation patterns can be lost through lineage extinction. The Paneth cell compartment was not specifically modeled because its dynamics are uncertain and these cells would be rarely sampled because their relative numbers are small (\~2--4 Paneth cells per crypt section \[[@B15]\]). Paneth cell methylation patterns should be similar to those of other differentiated cells because they survive only a few weeks in mice \[[@B16]\], although human Paneth cells may potentially have different lifetimes. Methylation of CSX or MYOD \"tags\" can be converted into a 5\' to 3\' binary code, with \"0\" representing an unmethylated site and \"1\" representing a methylated site. For example, an unmethylated CSX tag (eight CpG sites or 256 different possible tags) is \"00000000\" and a fully methylated MYOD tag (5 CpG sites or 64 different possible tags) is \"11111\". Tags can be summarized with three statistics that reflect numbers and types of stem cell divisions: 1\) Percent Methylation. Numbers of methylated tag sites can be summarized by percent methylation. For example the CSX tag \"01010101\" is 50% methylated. In general, percent methylation reflects numbers of divisions since birth. 2\) Unique Tags per Crypt. Diversity, or numbers of unique tags among the eight tags sampled from a crypt, reflects numbers of stem cells and stem cell lineage survival. Greater numbers of stem cells or longer-lived stem cell lineages would lead to greater crypt tag diversity. 3\) Intracrypt Distance. This is the average number of site differences between crypt tags (Hamming distance). For example, the distance between CSX tags 00000011 and 11000000 is four. Intracrypt distance is another measure of crypt diversity. Results ======= Crypts of Lieberkühn and villus fragments were isolated from fresh human small intestine (Figure [1](#F1){ref-type="fig"}) and individual allele methylation patterns were analyzed by bisulfite sequencing. Examples of crypt CSX methylation patterns (5\' to 3\' order of methylated sites) or \"tags\" are illustrated in Figure [2](#F2){ref-type="fig"}. Methylation represents somatic errors because CSX tags, like most CpG-rich sites \[[@B13]\], are initially unmethylated at birth. Although the tag patterns were complex, with differences within crypts and between crypts in the same intestine, the average percent methylation generally increased with age (Figure [3A](#F3){ref-type="fig"}). Crypt diversity (unique tags per crypt and intracrypt tag distances) also exhibited age-related changes (Figure [3A](#F3){ref-type="fig"}). Duodenum and ileum tag patterns appeared similar. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Examples of CSX crypt methylation patterns.Each group represents eight CSX tags (5\' to 3\' horizontal orientation, each of the eight CpG sites is represented by a circle) sampled from a single crypt, with nine crypts per patient. Open circles are unmethylated sites and filled circles are methylated sites. Numbers of unique tags per crypt and crypt percent methylation are summarized below each crypt. The seemingly random methylation patterns collectively encode both numbers of divisions since birth and ancestry. ::: ![](1741-7015-3-10-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Methylation tag analysis. A:Summary of CSX methylation tags sampled from crypts of Lieberkühn. Individual crypt values (\"X\" for duodenal crypts and \"+\" for ileal crypts) and average values (open circles for the duodenum and filled circles for the ileum) were plotted for 12 small intestines. Black filled circles represent the methylation measured in fixed small intestine from children. Solid green lines represent average values for a stem cell niche simulation best matching the experimental data. Consistent with the scatter of experimental crypt values, niche simulations also produced a variety of different outcomes summarized by dotted green lines representing 95% of simulated outcomes. Gray solid lines represent average simulated values consistent with the colon \[10\]. Small intestine percent methylation appeared to increase more slowly than the colon with age, consistent with fewer stem cell divisions relative to the colon. B:Summary of MYOD crypt of Lieberkühn methylation tags. C:Summary of CSX villus methylation tags. Average crypt values (black circles) for the same intestine are also plotted for comparison with average villus values (red circles). Unique tags and distances were significantly greater in villi compared to crypts whereas percent methylation was not significantly different. ::: ![](1741-7015-3-10-3) ::: Different tags within the same intestine are consistent with stochastic errors that accumulate independently in different cells. To extract ancestral information encoded by seemingly random tags within a single intestine, crypts were modeled assuming stochastic methylation errors and either immortal or niche stem cells (see Methods). The primary difference between the models is that the probability of symmetric division (yielding two stem cell (P2) or two differentiated daughters (P0)) is zero with immortal stem cells, whereas the probability of symmetric division is greater than zero with niche stem cells (Table [2](#T2){ref-type="table"}). Greater inter-crypt variability is expected with stem cell niches because both methylation errors and stem cell survival are stochastic. As in the experimental data, the simulated crypts contained tags with different patterns, summarized by averages and intervals including 95% of simulated crypt outcomes (Figure [4A](#F4){ref-type="fig"}). Average simulated crypt values were consistent with average experimental values for either two immortal stem cells per crypt, or 4 to 256 niche stem cells. However, niche stem cells produced more inter-crypt tag variation, which better matched the variability observed in the small intestines. Inconsistent with immortal stem cell lineages were the observations that unique crypt tag variances were greater than simulated average values for all 12 intestines, and only 4 of the 12 intestines were within 95% simulation intervals for immortal stem cells (Figure [4B](#F4){ref-type="fig"}). In contrast, experimental variances for 10 of 12 intestines fell within 95% simulation intervals for the niche scenarios. In addition, more than four unique sequences per crypt were often observed, which should occur only rarely with two immortal crypt stem cells (four possible alleles), but may occur more frequently with 4 to 256 niche stem cells (Figure [4A](#F4){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Niche versus Immortal Stem Cells. A:Comparisons between niche (green) and immortal (red) stem cell simulations with CSX tags. Niche simulations assume a constant number of stem cell (from 4 to 256 stem cells per niche) and probabilities of asymmetric division (P1) from 0.98 to 0.89 (see Table 3). Outcomes of all niche simulations were similar and therefore one cannot be certain of the exact stem cell niche size. Immortal lineage simulations assume two stem cells per niche. Parameters (see Table 2) were otherwise identical between the simulations. Outcomes were more variable with niche simulations with wider 95% simulation intervals (dotted lines). The differences are most obvious with unique tags per crypt. B:Variances of unique CSX crypt tags per intestine (for example from Figure 2, Patient A had 2, 2, 2, 2, 3, 3, 3, 4, and 5 unique tags per crypt for a variance of 1.1, and Patient M had 2, 2, 2, 3, 3, 3, 4, 4, and 6 unique tags per crypt for a variance of 1.7). Simulations with the immortal lineage scenario (average is a solid red line, 95% intervals are dotted lines) were less consistent with the experimental data (circles) than the niche simulations (green lines). ::: ![](1741-7015-3-10-4) ::: Tag data from human large \[[@B10]\] and small intestinal crypts were similar, except the percent methylation appeared to increase more slowly with age in the small intestine (Figure [3A](#F3){ref-type="fig"}). Small intestine percent methylation was generally lower than published \[[@B10],[@B17]\] measurements of human colon crypts, with a trend towards a slower increase with age (Figure [5](#F5){ref-type="fig"}), but this difference was not significant (p = 0.24, F-test). Potentially methylation error rates are lower in the small intestine, but assuming equivalent error rates (a molecular clock hypothesis \[[@B9]\]), stem cell division rates are either equivalent or lower in the small intestine compared to the colon. For example, simulations with a mitotic rate 75% of the colon (0.75 divisions per day) and a methylation error rate identical to the colon (2 × 10^-5^per site per division) better fit the small intestine data (Figure [3A](#F3){ref-type="fig"} and Figure [5](#F5){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparisons of large and small intestine age-related CSX methylation rates. Average percent methylation increased more slowly with age in the small intestine (black circles) relative to previously published \[10, 17\] colon measurements (red triangles). However this trend (black and red dotted lines) was not significant. A lower stem cell division rate was more consistent with the small intestinal data (green line, 0.75 divisions per day), relative to a model \[10\] postulating one division per day for the colon (grey line). ::: ![](1741-7015-3-10-5) ::: Crypt methylation patterns encode niche dynamics, but exact niche stem cell numbers and probabilities of symmetric divisions remain uncertain because different combinations yield similar outcomes (Table [3](#T3){ref-type="table"}). These niche combinations (4--256 stem cells and P1 from 0.98 to 0.89) are the same in large \[[@B10]\] and small intestines. For example, assuming that 95% of stem cell divisions yield one stem cell daughter, the crypt of Lieberkühn niche size is 64 stem cells. As in the colon \[[@B10]\], loss of all niche stem cell lineages except one or a \"bottleneck\" would recur on average every 3,000 divisions (with a 95% interquantile range of 1,000--7,000 divisions). Small intestinal crypt diversity (unique tags per crypt) did not change with age (Figure [3A](#F3){ref-type="fig"}), suggesting that niche dynamics or numbers of niche stem cells do not change with aging. MYOD tags sampled from six small intestines were also consistent with the same niche parameters inferred with the CSX tags (Figure [3B](#F3){ref-type="fig"}). For a further test of our model, CSX tags were sampled from villi of six intestines. Villi should exhibit more tag diversity than crypts because they receive contributions from four to ten surrounding crypts \[[@B18]\]. Furthermore, if methylation reflects the lifetime accumulation of stem cell mitotic errors, percent methylation should be similar between villi and crypts because crypt cells are only a few divisions older than villus cells. Consistent with these expectations (Figure [3C](#F3){ref-type="fig"}), there were significantly more unique tags per villus than per crypt (average 4.8 versus 3.0, p \< 0.001, two-tail t-test) whereas methylation differences were not significant (average 21.8% versus 22.9%, p = 0.76). Distances between tags within a villus were also significantly greater than tag intracrypt distances (2.5 versus 1.7, p \<0.001). Villus and crypt comparisons illustrate the ability of methylation tags to encode simultaneously both mitotic age (numbers of divisions since birth) and distinctly different ancestries. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Stem Cell Niche Parameters ::: Niche Size P asymmetric (P1)\* P symmetric (P0+P2) ------------ --------------------- --------------------- 4 0.98 0.02 16 0.97 0.03 64 0.95 0.05 256 0.89 0.11 \P0+P1+P2 = 1, with P0 = P2 ::: Discussion ========== The adult small intestine is physically composed of millions of crypts, and cells that constantly divide and die. Underlying the homeostatic visual appearance of a normal small intestine is an ancestral somatic cell tree that grows with age (Figure [6](#F6){ref-type="fig"}). Starting with the zygote, millions of related cells create an intestine. Past and present cells are all connected through time by ancestral lineages or branches, which ultimately root in the zygote or first common ancestor. Although stem cell definitions may differ \[[@B2]-[@B4]\], a crypt common ancestor is equivalent to a stem cell because all cells in a clonal crypt unit are progeny of this single common stem cell ancestor. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Somatic cell trees. A:A niche somatic cell tree. Starting from the zygote, millions of related cells create an intestine. Within each crypt are multiple stem cells (a niche with four stem cells is illustrated). With age, branches become longer reflecting more divisions since birth. Most niche stem cell lineages are lost (dotted lines) from random stem cell loss with replacement, which continuously creates newer more recent common crypt ancestors (circles). Such common crypt ancestors are stem cells because all current crypt cells are progeny of such cells. Any one of the four current niche stem cells may become a common ancestor, but only one will. In this way, crypts can appear to have both a single stem cell (most recent common crypt ancestor) and multiple potential stem cells. A villus tree (red lines in bottom tree) has essentially the same age as the crypt tree (same number of divisions since birth) but villi should exhibit more diversity than crypts because villus cells have older most recent common ancestors. B:Somatic cell tree with immortal stem cell lineages. Stem cells always divide asymmetrically and their lineages never become extinct. Unlike a niche somatic cell tree, branches are never lost or \"pruned\". ::: ![](1741-7015-3-10-6) ::: Both immortal and niche stem cell mechanisms can theoretically produce appropriate numbers of differentiated cells consistent with the morphology of normal small intestine. However, the shape of an intestinal somatic cell tree differs whether stem cell lineages are immortal or maintained by niches (Figure [6](#F6){ref-type="fig"}). Immortal stem cell lineages result in multiple long crypt branches, with the most recent common crypt ancestor present around birth. In contrast, a niche mechanism constantly creates newer more recent common crypt ancestors as most niche stem cell lineages become extinct. Methylation tags can distinguish between immortal or niche stem cell lineages because random somatic epigenetic errors will accumulate differently. Specifically, a niche mechanism produces more variability between crypts (wider variation in unique tags per crypt within a single intestine) and less intracrypt diversity (smaller average intracrypt distances) than immortal stem cells (Figure [4](#F4){ref-type="fig"}). Small intestine and colon crypt methylation patterns \[[@B10]\] were similar and more consistent with stem cell niches. The same stem cell dynamics (Table [3](#T3){ref-type="table"}) fit both colon and small intestinal niches. There were no apparent differences between ileal and duodenal crypt niches, and niche stem cell numbers (like the colon \[[@B10]\]) did not appear to change or decline with age. Niche stem cells defined by ancestry are physically intangible because common ancestors are defined by past events and no longer exist. Stem cells with immortal lineages are more readily identified because their pasts and futures are predictable. However, niche stem cell fates are unpredictable because all may potentially become common ancestors but only one will. The inability to predict niche survival may help explain why some adult stem cells have been so difficult to isolate or characterize. A niche somatic cell tree has few branches because random stem cell turnover eventually \"prunes\" all niche lineages except one (Figure [6](#F6){ref-type="fig"}). Such bottlenecks are predicted by our analysis to recur on average after 3,000 divisions. Other mechanisms may also be consistent with crypt methylation patterns and the assumptions of our model remain unverified \[[@B19]\]. For example, the observed variability in methylation patterns could also be generated if stem cell numbers were different between crypts within an intestine \[[@B19]\]. Although our bisulfite sequencing appears technically adequate (no evidence of nonconversion of C to T and only rare mutations (\<1 per 1,000 bases) at non-CpG sites), it is difficult to check the accuracy and reproducibility of the data. Repeat sampling of a crypt typically yields similar but not identical methylation patterns (data not shown), which may reflect crypt heterogeneity or experimental artifact. One strategy to test empirically a stochastic model that inherently yields scattered results is to examine similar but biologically distinct entities. Both small intestine and colon crypts are thought to be maintained by stem cell niches \[[@B5],[@B6]\] and the successful application of our model to the small and large intestines with methylation patterns from CpG-rich sequences on different chromosomes (CSX and MYOD) is empirical support for its ability to infer stem cell dynamics. In addition, small intestinal villi are physically different from crypts and represent mixtures from multiple adjacent crypts. Villi should be more diverse than crypts, yet villi and crypts should have similar numbers of divisions since birth. As one would expect if methylation records somatic cell histories, villus and crypt tag percent methylation were not significantly different, and villi contained significantly more tag diversity than crypts. Therefore, methylation patterns and our model are consistent with colon and small intestine crypt niches, and small intestine villi. Genetic alterations also appear to modulate stem cell dynamics because certain germline APC mutations are associated with significantly more diverse tags in normal-appearing familial adenomatous polyposis colon crypts, consistent with increased niche stem cell survival \[[@B17]\]. Mouse studies suggest that stem cell division rates are higher in the small intestines than the colon \[[@B2],[@B3],[@B20]\]. However, it is difficult with most assays to distinguish between stem and non-stem cell proliferation, and there are few human small intestine stem cell studies. Of interest, neoplasia, as with Min mice \[[@B21]\], is more frequently observed in the murine small intestines relative to the colon, whereas human small intestinal tumors are a small fraction of colon tumors \[[@B1]\]. In contrast to murine studies, human small intestinal stem cell division rates appeared to be slower or similar to those in the colon because age-related methylation appeared to increase more slowly in the small intestine (Figure [5](#F5){ref-type="fig"}), although the difference was not significant. Assuming a molecular clock hypothesis \[[@B9]\] or equivalent CSX methylation error rates throughout the lower gastrointestinal tract, the data are more consistent with equivalent or slightly lower small intestinal stem cell division rates relative to the colon. There is no definitive evidence supporting a somatic cell molecular clock hypothesis, but a priori a slower small intestinal stem cell division rate is more consistent with human intestinal cancer epidemiology \[[@B1]\], because fewer small intestinal errors (genetic or epigenetic) would be expected to accumulate during aging. A slower small intestinal stem cell division rate could also help explain why crypt purification is faster in the murine colon relative to the small intestine \[[@B5]\]. Moreover, an ability to count relative divisions since birth in different human tissues allows distinctions (Figure [7](#F7){ref-type="fig"}) between mitotic tissue ages (divisions since birth) and chronological ages (years since birth). The colon and small intestines within an individual have equivalent chronological ages but may have different mitotic ages. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### An intestinal tree based on mitotic ages or numbers of divisions since birth*. Each horizontal branch reflects a single crypt. Although all tissues have the same chronological ages, differences in stem cell division rates allow tissues within the same individual to have different mitotic ages. A small intestine tree has a more crypt branches (greater numbers of crypts) than a colon, but its branches may be shorter, which can possibly help explain why despite the greater physical size of the organ, small intestinal tumors are infrequent relative to colonic ones. ::: ![](1741-7015-3-10-7) ::: Conclusion ========== Methylation patterns suggest that niches containing multiple stem cells maintain crypts throughout the lower gastrointestinal tract. Small intestine stem cells appear to divide at equivalent or slower rates relative to the colon, and niche dynamics remain stable during aging. Although methylation tags only indirectly track stem cell dynamics and their exact interpretations are uncertain, they potentially allow for the systematic investigation of any intestine without prior experimental intervention. A somatic cell tree must underlie all human tissues and studies based on the hypothesis that certain somatic methylation errors record somatic cell histories may better define how cells divide and die during normal and abnormal aging. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= JYK performed most of the experiments and helped write the manuscript. KDS and ST helped with the statistical and quantitative analysis. DS analyzed the data and wrote the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1741-7015/3/10/prepub> Acknowledgements ================ We thank Mario Campuzano for technical assistance and Dr. Robert W. Beart for providing the specimens and clinical information. This work was supported by a grant from the National Institutes of Health (DK61140). ST is a Royal Society-Wolfson Research Merit Award holder and is supported in part by a grant from Cancer Research UK.	PubMed Central	2024-06-05T03:55:58.238746	2005-6-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168897/", "journal": "BMC Med. 2005 Jun 23; 3:10", "authors": [ { "first": "Jung Yeon", "last": "Kim" }, { "first": "Kimberly D", "last": "Siegmund" }, { "first": "Simon", "last": "Tavaré" }, { "first": "Darryl", "last": "Shibata" } ] }
PMC1168898	Background ========== Despite the advances in hospital care and the introduction of a wide variety of antimicrobial agents, Pseudomonas aeruginosacontinues to be a major nosocomial pathogen particularly in patients who suffer from immunosuppression \[[@B1]\]. P. aeruginosais a ubiquitous pathogen prevalent in the hospital environments, and can cause severe nosocomial infections \[[@B2]\]. The latter involve a broad spectrum of infections including the respiratory, gastrointestinal, and urinary tracts as well as wound infections, sepsis and others \[[@B3]\]. Various possible sources of P. aeruginosainfection in hospitals have been identified; such as tap water, medical equipment, hospital personnel and other patients \[[@B2],[@B4]\]. P. aeruginosaaccounts for 10% of all hospital acquired infections, a site specific prevalence which may vary from one unit to another and from study to study \[[@B5]\]. Among data on site-specific infections, P. aeruginosaappears to be the major cause of ventilator-associated pneumonia with a high rate of attributable mortality \[[@B6]\]. Moreover this organism can contaminate a number of other medical equipment such as respirators, endoscopes, bronchoscopes, transvenous pacemakers, urinary catheters, and dialysis equipment, leading to site-related infections \[[@B7],[@B8]\]. During the last year, the average prevalence of P. aeruginosanosocomial infections in our medical center was 18%. Such a high rate prompted us to study the P. aeruginosagenotypes circulating in the various units to reveal the clonal relationship between clinical and environmental isolates and to allow elucidating the source and mode of transmission of this important bacterium at this medical center. Results ======= Our data have shown that several mechanical devices were associated with P. aeruginosainfection in our patients with 55% due to mechanical ventilation and the remaining 45% due to polysite catheters, Foley catheters and different types of surgery. RAPD analysis have shown 31 genotypes present among the clinical and the environmental isolates (figure [1](#F1){ref-type="fig"}). Thirty eight of ninety (42%) of the clinical isolates and 10/23 (43%) of the environmental isolates showed genotype 1 to be distributed among the medical center units. Each of the genotypes 2--30 represented from 1--8% of the strains. Antimicrobial susceptibility testing has shown that susceptibility patterns between clinical and environmental isolates were similar with most drugs showing \> 90% susceptibility, with the exception of tobramycin and gentamycin which showed lower susceptibility (83%) in the environmental strains. Data is in concordance with previous reports in this institution \[[@B13]\]. Enzymatic assays determined that genotype 1 in addition to other 10 genotypes, in clinical and environmental isolates, were positive for all virulence factors tested for. All other genotypes were all positive for the production of protease, lecithinase and coagulase and positive in 83--89% of the cases for the production of the other 4 virulence factors (Figure [2](#F2){ref-type="fig"}). Table [1](#T1){ref-type="table"} shows the relationship between the various genotypes, to source, antibiograms and virulence factors. Discussion ========== The strain distribution based on RAPD analysis showed that clinical and environmental isolates distributed in one or more than one unit, included genotypes 1--8 and 9, with predominance of genotype 1 in all investigated units. This may indicate that cross contamination among patients lead to the spread of this genotype among the various units, possibly through transient hand carriage by health care personnel due to contact with contaminated surfaces or by patient contact with contaminated surfaces or medical equipment \[[@B4]\]. Our findings suggest that cross colonization may be an important means of P. aeruginosaspread and infection especially after identification of a potentially virulent clone (genotype 1) of this organism that had been propagated in various units over a period of 9 months. This could indicate that the patients were continuously infected with a strain originating from an exogenous source. The importance of cross colonization of P. aeruginosain nosocomial infections was previously reported, by Bergmans et alwho studied 100 patients admitted to an ICU ward; cross colonization accounted for 50% of all cases of acquired P. aeruginosacolonization, the other 50% of patients were probably colonized from endogenous sources \[[@B14]\]. The remaining genotypes were isolated exclusively from the clinical specimens of patients and were not detected among environmental isolates. Most of these genotypes did not harbor all the virulence factors tested for. Not being encountered in environmental sources may indicate that these strains may have been endogenously acquired. High rate of colonization with P. aeruginosafrom endogenous sources occurs mainly in the respiratory and gastrointestinal tracts \[[@B2]\]. Hospitalization may lead to increased rates of carriage, particularly in the lower respiratory tract in patients undergoing mechanical ventilation, upper respiratory tract due to broncho- pulmonary colonization and infection, in the gastrointestinal tract of patients receiving chemotherapy for neoplastic diseases, or at virtually any site in patients treated with antibiotics \[[@B3],[@B15],[@B16]\]. Prevalence of strains with resistance to all antimicrobial agents will constitute a major risk for hospitalized patients. In our study, though, the most prevalent strain of genotype 1 was susceptible to all antimicrobial agents and does not constitute a problem in treatment. However, its potential of producing all virulence factors and being spread by various means in the hospital, unlike other genotypes that harbor all virulence factors, may render genotype 1 a high risk pathogen specially in the immunosuppressed and debilitated patients. The fact that genotype 1 remained susceptible to all antibiotics lead to the conclusion that patients could be continuously infected with a strain originating from an exogenous source. In addition to mechanical spread by personnel, this genotype may carry a number of adhesins that enhance its colonization in the hospital environment and render it more accessible to patients. Studies are underway in our laboratory to detect by PCR, genes encoding a number of adhesins and determine their transcription levels in genotype 1, in comparison to other encountered genotypes. This will shed light on the possible role of adhesins on the prevalence of this genotype in the hospital environment. The extracellular enzymes or toxins produced, on the other hand, will contribute to the breaking down physical barriers and help the organism to penetrate, impair host defenses, and render its new milieu more conducive to its physical, nutritional and reproductive requirements \[[@B17]\]. In summary, our data have shown the predominant prevalence of a potentially virulent P. aeruginosagenotype 1 in clinical and environmental specimens, circulating in the various hospital units. From a practical point of view, the results of our study emphasize the need to reinforce implementation of infection control measures, and to limit the transmission of P. aeruginosaamong patients and from environmental sources to patients. Screening for P. aeruginosacarriage in all patients with nosocomial colonization or infection should also be done in research settings using both rectal and respiratory tract specimens to determine the source of colonization/infection and hence get informed on whether it was acquired endogenously or exogenously. Conclusion ========== In conclusion, our data have shown the predominant prevalence of a potentially virulent P. aeruginosagenotype 1 in clinical and environmental specimens, circulating in the various medical center units. Methods ======= Consecutive P. aeruginosa(90) isolates were recovered from different patients specimens (one per patient) submitted for bacteriological investigations at the Clinical Microbiology Laboratory at the American University of Beirut Medical Center (AUBMC), between September 2003 and May 2004. Patients acquiring a nosocomial infection due to P. aeruginosaas determined by clinical and laboratory testing and indicated in their medical records, were only considered in this study. Fever, and recovery of P. aeruginosafrom the site of infection during stay at the medical center, constituted the most important criteria that defined patient\'s infection with this organism. Patients\' data collected from the medical records, included age (2 to 91 years), sex (Males: 43 and Females: 47), admission date, admission diagnosis, invasive procedures used on the patients, symptoms, and the date of the first positive culture for P. aeruginosa. The drugs used in the treatment of these patients, mainly included: amikacin, azactam, gentamicin, tobramycin, and tazocin. Patients were distributed within ten different units in the medical center, mainly the Respiratory Care Unit (RCU), Intensive Care Unit (ICU), Coronary Care Unit (CCU), the Surgery Unit, as well as other units. Twenty three isolates were also collected from different environmental sources, such as, respirators, respirators\' filters, water irrigation, tap water, basins, trays, bed side tables, side rails, and sink sides. Statistical analysis determined the sample size required to estimate the true proportion (percentage of P. aeruginosainfections during a given period of time) to within 0.10, with 95% confidence was calculated. Calculations estimated the minimum number of samples required to be 54 \[[@B9]\]. The presumptive identification of P. aeruginosaon culture based on colonial morphology, Gram stain microscopy, and oxidase test was further confirmed by the API NE Kits and growth at 42°C. Susceptibility of all isolates to a panel of antimicrobial agents (amikacin, aztreonam, ceftazidime, ciprofloxacin, gentamicin, imipenem, piperacillin/tazobactam, tobramycin) was determined according to the guidelines of the National Committee for Clinical Laboratory Standards (NCCLS) \[[@B10]\]. P. aeruginosaATCC 25321 strain was used as positive control in all tests. DNA was extracted from P. aeruginosaATCC strain and from all isolates of P. aeruginosaby the GFX™ Genomic Blood DNA Purification Kit (Amersham PharmaciaBiotech, Uppsala, Sweden) according to the manufacturers\' specifications. Random amplified polymorphic DNA (RAPD) analysis of the clinical and environmental isolates using two in-house oligonucleotide primers, Pa1 (5\'AGGGGTCTTG 3\') and Pa2 (5\' CTTCTTCAGCTCGACGCGACG 3\') was done. RAPD was carried out according to Matar et al \[[@B11]\] using the PTC-100™ Programmable Thermal Controller (MJ Research, Inc., Watertown, Mass., USA). Briefly, RAPD was carried out on all isolates in 100 μl reaction mixtures containing each: 10 μl of template DNA, 16 μl of dNTPs (0.2 mM), 10 μl of 10X PCR buffer (100 mM TrisHCl \[pH 8.3\], 500 mM KCl, 4 mM MgCl~2~), 1 μl of primer 1 (0.5 μM), 1 μl of primer 2 (0.3 μg/ μl), 2.5 U of TaqDNA polymerase and 61.5 μl of nanopure sterile water. The amplification program, included the following steps: denaturation at 94°C for 3s, annealing at 53°C for 1 min and extension at 72°C for 1 min, for 44 cycles. The cycles were followed by a final extension step at 72°C for 10 min. Amplicons were subjected to electrophoresis on 2% agarose gels at 107 volts for 2 hours. Patterns that had the same number of bands and similar fragments size were considered identical. The enzymatic activities of the isolates of P. aeruginosawere evaluated by spot inoculation containing 10^6^CFU/ml of the organisms in various media \[[@B12]\]. Media used: Brain heart infusion for the protease activity, 1% elastin (Sigma Chemical Co. St Louis, Mo. USA) for the elastase activity, human fibrinogen type 1 (Sigma Chemical Co. St Louis, Mo. USA) for the fibrinolytic activity, trypticase soy agar (TSA) supplemented with egg yolk (Difco Laboratories, Detroit, Mi., USA) for the lecithinase production, TSA with tween 80 (Sigma Chemical Co. St Louis, Mo. USA) for the lipase activity, DNase test agar with toluidine blue 0, (Difco Laboratories, Detroit Mi., USA) for the DNAase production, and rabbit plasma (Difco Laboratories, Detroit Mi., USA) for the coagulase activity. Positivity of tests was assessed as follows: clearing of opacity around the inoculum spots for the protease, elastase and fibrinolytic activities, white precipitate around or beneath the inoculum spots for the lecithinase activity, a turbid halo around the inoculum spots for the lipase activity, formation of a pink halo around the inoculum spots for the DNase activity, and gelling of rabbit plasma after 48 hours for the coagulase activity. Authors\' contributions ======================= GM supervised the study and wrote the manuscript. MC did the bench work and helped in writing the manuscript. GA provided bacterial isolates. ZS provided help on wards. GJ provided clinical support. UH provided clinical support. All authors read and approved the final manuscript. Acknowledgements ================ Thanks are due Miss Marie Risk and Mr. Elias Rahal for technical assistance. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### \(A) Genotypes of P. aeruginosaisolates found in 10 units. Lane 1: 50-bp ladder, lane 2: negative control, lane 3:ATCC25321 strain, lanes 4--18: Genotypes 1,2,3,4,5,6,7,8,9,10,11,12,13,14, and 15. (B) Lane 16: 50-bp ladder, lane 17: negative control, lane 18: ATCC 25321 strain, lanes 19--33: Genotypes 16,17,18,19,20,21,22,23,24,25,26,27,28,29, and 30. (C) Genotypes of P. aeruginosafrom environmental sources found in the RCU. Lane 1: 50-bp ladder, lane 2: negative control, lane 3:ATCC 25321 strain, lane 4--11: Genotypes (1,4,17,25,27,28,30,31). ::: ![](1471-2180-5-29-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Percentage of clinical and environmental P. aeruginosaisolates producing the virulence factors. ::: ![](1471-2180-5-29-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Correlation between P. aeruginosagenotypes, source and site of infection, antimicrobial susceptibility and enzymatic activities of virulence factors ::: Genotypes types of unit Number of clinical isolates Isolation site/ Number of isolates Antibiograms Virulence factors --------------- ------------------- --------------------------------- ----------------------------------------------------------- ------------------ ----------------------- ------- ------- ------- ------- ------- ------- ---- ---- ---- ---- ---- ---- 1 RCU 15 DTA^1^/ 13 blood/ 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ ICU 3 DTA/ 2 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ CCU 1 Urine/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ PICU 4 DTA/2 Tongue/ 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ St Jude 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ Surgery 3 Wound/ 3 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 South 6 Abscess/ 1 Sputum/ 2 DTA/ 1 Bronchial washing/ 1 Urine/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 4 South 2 Sputum/ 1 Swab/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 9 South 3 DTA/ 1 Sputum/ 1 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 2 RCU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ ICU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ Surgery 2 Wound/ 2 S S S S S S S S \+ \+ \+ \+ \+ \+ Oncology 1 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 south 1 Pleural fuild/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 3 RCU 2 DTA/ 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 South 2 Sputum/ 2 S S S S S S S S \+ \+ \+ \+ \+ \+ Oncology 3 DTA/ 2 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 4 RCU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ St Jude 2 Blood/ 1 Catheter/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 South 1 Bile/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 PICU 1 DTA/ 1 S S S S S S S S \+ \+ \- \+ \+ \+ 4 South 1 DTA/ 1 S S S S S S S S \+ \+ \- \+ \+ \+ 6 RCU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ ICU 1 Sinus/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 7 ICU 2 DTA/ 2 S S S S S S S S \+ \+ \+ \- \+ \+ Oncology 1 Blood/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 8 ICU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \- \+ Oncology 1 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \- \+ 9 RCU 1 Urine/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ Surgery 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 10 4 South 2 Wound/ 2 S S S R R R R R \+ \+ \- \+ \- \+ 11 Oncology 2 leg ulcer/ 1 Bronchial washing/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 12 5 South 2 DTA/ 2 S S S S S S S S \+ \+ \+ \- \- \+ 13 9 South 1 Sinus discharge/ 1 S S S R R S S R \+ \+ \- \+ \- \+ 14 9 South 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 15 ICU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 16 CCU 1 Sternal tissue / 1 S S S S S S S S \+ \+ \+ \+ \+ \- 17 CCU 1 Urine/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 18 CCU 1 DTA/ 1 S I S S S S S S \+ \+ \- \- \- \+ 19 RCU 1 DTA/ 1 S I S S S S S S \+ \+ \+ \+ \+ \+ 20 RCU 1 DTA/ 1 S S S S S R S S \+ \+ \- \+ \- \+ 21 RCU 1 DTA/ 1 S S S S R S S R \+ \+ \- \+ \- \+ 22 RCU 1 DTA/ 1 S I S S S S S S \+ \+ \+ \+ \+ \+ 23 RCU 1 DTA/ 1 S I S S S S R S \+ \+ \- \+ \- \+ 24 RCU 1 Wound/ 1 R S R R R R R R \+ \+ \- \+ \- \+ 25 RCU 1 DTA/ 1 S S S S R S S I \+ \+ \+ \+ \+ \+ 26 RCU 1 Sputum/ 1 S R R S S R R S \+ \+ \- \+ \- \+ 27 PICU 1 DTA/ 1 S R R S S S R S \+ \+ \+ \+ \+ \+ 28 ICU 1 DTA/ 1 S R R S S I R S \+ \+ \+ \+ \+ \+ 29 4 South 1 DTA/ 1 S I S S R R S S \+ \+ \+ \+ \+ \- 30 RCU 1 DTA/ 1 S S S S R S S I \+ \+ \- \+ \- \+ NB: Genotype 31 was only found among the environmental isolates ^1^DTA: Deep Tracheal Aspirate AN: Amikacin CIP: Ciprofloxacin TZ: Tazocin ATM: Aztreonam GM: Gentamicin NN: Tobramycin CAZ : Ceftazidime IPM: Imipenem Pro: Protease Lip: Lipase Lec: Lecithinase Fib: Fibrinolysin Elas: Elastase Coag: Coagulase :::	PubMed Central	2024-06-05T03:55:58.242123	2005-5-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168898/", "journal": "BMC Microbiol. 2005 May 20; 5:29", "authors": [ { "first": "Ghassan M", "last": "Matar" }, { "first": "Mira H", "last": "Chaar" }, { "first": "George F", "last": "Araj" }, { "first": "Zaher", "last": "Srour" }, { "first": "Ghassan", "last": "Jamaleddine" }, { "first": "Usamah", "last": "Hadi" } ] }
PMC1168899	Background ========== Yersinia pestis, the causative agent of plague, causes rapidly progressing disease in humans with a high mortality rate, especially in the pneumonic form of the disease. Due to the severe nature of plague, its ability for aerosol transmission, and the potential for human to human transmission plague is considered to be a disease of high concern as an agent of biological warfare or biological terrorism \[[@B1]\]. For this reason, an improved vaccine for plague is desirable. Current efforts for vaccine development have focused on two proteins: LcrV (also known as the V antigen) and the capsular F1 antigen \[[@B2]\]. The best results to date have been obtained by using a combination of recombinant LcrV and F1 subunits \[[@B3]\] separately or as a fusion protein \[[@B4],[@B5]\]. These subunit vaccines demonstrate very good protection against both pneumonic and systemic forms of plague \[[@B2]\] in mouse models. One of the potential limitations of these subunit vaccines is that F1 is not required for full virulence of Y. pestis, as F1-negative strains have the same LD~50~value as F1-positive strains \[[@B6]-[@B9]\]. A second limitation that could result in undesired side-effects in immunized individuals is the demonstrated immunosuppressive effect of LcrV \[[@B10]-[@B13]\]. Additionally, serologic diversity of LcrV has been reported, in Yersiniaspecies other than Y. pestis, that could theoretically limit the usefulness of an LcrV based vaccine. While the recombinant subunit vaccines are very effective in experimental animals and offer protection against F1 minus strains of Y. pestis\[[@B2]\], the inclusion of other antigens with the LcrV/F1 subunit vaccine candidates could improve the ability of the resulting vaccine to offer protection against multiple Y. pestisstrains, or the new antigens could be developed as separate vaccine candidates. The type III secretion apparatus encoded on the low-calcium response (LCR) virulence plasmid, pCD1 in strain KIM \[[@B14]\], of Y. pestisis a conserved virulence mechanism that is absolutely required for virulence of Y. pestis\[[@B15]\]. YscF is a surface localized protein that is required both to secrete Yops and to translocate toxins into eukaryotic cells \[[@B16]-[@B19]\]. One report speculates that YscF polymerization is required for a YscF needle to puncture eukaryotic cell membranes \[[@B18]\]. Other researchers suggest that YscF and its homologs function to provide a base that a translocon complex is built upon, or that YscF builds a conduit from the bacterium to the eukaryotic membrane \[[@B20]\]. This suggestion seems more likely given that other proteins such as YopB, YopD, and LcrV are also required for translocation into eukaryotic cells \[[@B21]-[@B28]\]. Additionally, YscF needle producing Y. enterocoliticadeficient in production of the translocators (LcrV, YopB, and YopD) do not translocate Yops into macrophages, demonstrating that the YscF-needle is not sufficient for translocation \[[@B19]\]. Most currently described pathogenesis-related type III secretion systems possess homologs to YscF. In pathogenic Salmonellaand Shigella, the YscF homologs (PrgI and MxiH, respectively) have been demonstrated to form a needle structure that protrudes from the surface of bacterial cells \[[@B29]-[@B31]\]. The best-characterized homolog of YscF is EscF of enteropathogenic E. coli(EPEC). EscF is required for \"attaching and effacing\" (A/E) lesion formation on the intestinal mucosa and for type III secretion of effector proteins \[[@B32]-[@B34]\]. EscF is thought to be a structural component of the needle complex on the bacterial surface as it binds EspA, the major component of a filamentous surface organelle, and is required for formation of the EspA filaments. \[[@B32]-[@B34]\] However, this surface localization has never been visualized directly, as the only EscF antiserum generated was unable to recognize the native protein \[[@B33]\]. Based on the fact that YscF is thought to be a surface-expressed protein in Y. pestisand is required for virulence, we sought to determine if YscF could serve as a protective antigen against experimental plague infection. Immunization of mice with His-tagged YscF resulted in a high anti-YscF titer and significant protection against i.v. challenge with Y. pestis. The findings of this study suggest that YscF may be a potential plague vaccine candidate that could be used in conjunction with other plague antigens, or possibly alone if its efficacy can be improved by alternative delivery methods. Results and discussion ====================== Expression and purification of HT-YscF -------------------------------------- To facilitate the purification of YscF, yscFwas cloned into the overexpression plasmid pET24b (Novagen) to yield a hexahistidine-tag on the C-terminus of YscF (HT-YscF). E. coliBL21(DE3) harboring the HT-YscF expression plasmid, pJM119, was grown in one liter of LB broth \[[@B35]\] containing kanamycin at 37°C. Expression of HT-YscF was induced after 2 h of growth with 0.3 mM IPTG and then incubated until the A~550~reached \~1.0. Cells were harvested by centrifugation and disintegrated by passage through a French pressure cell at 20,000 lb/in^2^. Subsequent to disintegration, the extracts were clarified by centrifugation at 3,200 × gfor 20 min at 4°C. Affinity purification of HT-YscF was performed using Talon resin (BD Clontech) using standard methods. Purity of the recovered protein (\> 95 %) was estimated by SDS-PAGE on a 15% (wt/vol) gel followed by staining with a Coomassie blue stain. The purified HT-YscF ran as multiple bands on the gel, regardless of the presence or absence of a reducing agent, such as dithiothreitol (data not shown). A band that corresponded to the predicted size of HT-YscF was the dominant species and other larger bands could also be visualized (Figure [1A](#F1){ref-type="fig"}). Based on the sizes of these bands, it is likely that they represent dimers and other multimers of YscF. The presence of YscF multimers is not surprising as YscF and its homologs are known to form multimeric structures \[[@B29],[@B36]\]. Support for the contention that the larger bands are multimers of YscF is seen in Fig. [1B](#F1){ref-type="fig"}. Purified recombinant HT-YscF was immunoblotted with a Penta-His specific antibody (Qiagen). The results seen in Figure [1](#F1){ref-type="fig"} show that the larger bands seen in panel A are also recognized by the penta-His antibody (Fig. [1B](#F1){ref-type="fig"}) demonstrating that the larger bands are likely recombinant proteins containing a poly-histidine tag and, therefore, contain YscF. Additionally, the purified HT-YscF was immunoblotted and probed with serum from immunized mice (Fig. [1C](#F1){ref-type="fig"}) to demonstrate that the immunized mice mounted an immune response against the purified HT-YscF. The larger bands seen in the Coomassie stain and visualized on the immunoblots (with both anti-Penta-His antibody and post-immune sera) could also be contaminating E. coliproteins, that cross-react to the penta-His antibody, this possibility cannot be currently discounted. The presence of the higher molecular weight proteins, if they are not YscF multilmers and they are from E. coli, likely had little influence on this study as immunization with HT-YscF induced a specific antibody response to native YscF produced by Y. pestis(discussed below). Specificity of the antibody response to YscF -------------------------------------------- To confirm that the recombinant HT-YscF antibody response generated in immunized mice was directed against YscF from Y. pestis, Y. pestisprotein extracts from the parental strain KIM8.3002 (pgm^-^, pla^-^) and extracts from a isogenic yscFdeletion strain (kindly provided by G. Plano, University of Miami, Miami, Fl.) were immunoblotted and probed with pooled antiserum from HT-YscF immunized mice (Figure [2](#F2){ref-type="fig"}). The results in Figure [2](#F2){ref-type="fig"} demonstrate that the pooled mouse antiserum specifically recognizes YscF produced by Y. pestis. As seen in Figure [2](#F2){ref-type="fig"}, YscF is visualized on the immunoblot as a highly reactive band of the correct predicted size and the YscF band is only seen only in strains containing the yscFgene (Figue [2](#F2){ref-type="fig"}, lanes 1--4). Importantly, no bands are seen in lanes 5 and 6 that contain proteins derived from the yscFdeletion strain. In lanes 1 and 2, calcium regulation of the YscF band is seen as expected for the LCR-regulated yscFgene. Transcomplementaion of the ΔyscFstrain with pBAD-YscF (G. Plano, University of Miami) restored YscF reactivity to the HT-YscF serum in the ΔyscFstrain (Figure [2](#F2){ref-type="fig"}, lanes 3 and 4). Demonstrating that the lack of YscF reactivity in the ΔyscFstrain was due to the deletion of yscF. The higher molecular weight bands seen in the whole cell fraction (Figure [2](#F2){ref-type="fig"}, lanes 1--6) represent cross-reactive Y. pestisbands not specific to YscF, as they are present in the ΔyscFsamples. The same cross-reactive bands are also present in whole cell samples probed with pre-immune serum (data not shown) suggesting that the reactivity seen is not induced by immunization with HT-YscF. The higher molecular weight band seen in the culture supernatant fractions could represent a multimer of YscF (possibly a trimer) or YscF in complex with another secreted protein. This higher molecular weight species is likely related to YscF as a similar band is not seen with pre-immune sera (data not shown) or in the ΔyscFstrain (Figure [2](#F2){ref-type="fig"}). These results demonstrate that mice immunized with recombinant HT-YscF produce antibodies that specifically recognize YscF produced by Y. pestis. Active immunization of outbred mice followed by challenge with Y. pestisKIM5 ------------------------------------------------------------------------------ To examine the ability of HT-YscF to protect mice from Y. pestisinfection outbred mice were immunized intra peritoneally with purified HT-YscF using complete Freund\'s adjuvant (CFA) for primary immunization and incomplete Freund\'s adjuvant (IFA) for booster immunizations or with a phosphate-buffered saline (PBS, \[[@B37]\]) control in CFA or IFA to control for adjuvant effects. For these studies, 6-to 8-week-old female Swiss-Webster mice were immunized i.p. with 40 μg/mouse HT-YscF in PBS or PBS (control mice) alone emulsified 1:1 with CFA. Experimental mice were boosted with 40 μg/mouse HT-YscF in IFA after two weeks and with 20 μg/mouse HT-YscF in IFA at 4 weeks post-immunization. Negative control mice were boosted with PBS emulsified with IFA according to the same schedule. Two weeks following the final booster immunization, sera were collected from the HT-YscF-immunized and the PBS-immunized mice to assay for HT-YscF reactivity. Sera from 22 mice from the HT-YscF-immunized and the PBS-immunized groups were tested for total IgG reactivity. HT-YscF immunized mice had a geometric mean titer of 1:40,000 for IgG specific to HT-YscF while PBS-immunized mice had HT-YscF titers less than the lowest dilution 1:800 (Table [1](#T1){ref-type="table"}). After establishing that the HT-YscF immunized mice had developed a strong antibody response to HT-YscF, the mice were challenged with Y. pestis. Two weeks after the final immunization, groups of 10 mice were challenged intra venously via the retro-orbital sinus with 10^1^to 10^6^CFU Y. pestisKIM5 (pgm^-^) in PBS (grown at 26°C). The mice were observed for 19 days after challenge, and the average doses required to kill 50% of the mice (LD~50~) for the treatment groups were calculated using the extrapolation method of Reed and Muench \[[@B38]\]. Mice that were immunized with HT-YscF demonstrated a 134-fold increase in the calculated LD~50~value as compared to PBS-immunized mice (Table [1](#T1){ref-type="table"}). These results demonstrate that immunization of mice with HT-YscF was able to provide a degree of protection to the immunized mice from a subsequent challenge with Y. pestis(Table [1](#T1){ref-type="table"}). While the protection provided by HT-YscF is not of the same magnitude as that seen with the protective antigens F1 or LcrV, the increased LD~50~value clearly shows that immunization with YscF affords a significant level of protection. Thus, HT-YscF becomes the only other reported antigen apart from LcrV or F1 to induce a significant protection in F1-positive Y. pestis. Immunization with YopD has been demonstrated to provide significant protection against a F1-minus mutant of Y. pestis\[[@B39]\]. The protection generated by HT-YscF suggests that YscF could be potentially developed as a novel subunit vaccine for Y. pestisor could serve as an additional antigen in a multivalent Y. pestisvaccine comprised of YscF, F1, and LcrV for example. Characterization of the antibody response to HT-YscF ---------------------------------------------------- Mice immunized with HT-YscF demonstrated a strong antibody response to YscF and provided protection to the vaccinated mice from lethal Y. pestischallenge. Due to the strong antibody response, isotyping analysis was performed to determine the predominant isotypes of antibodies produced by mice in response to vaccination with HT-YscF in CFA/IFA. Anti-YscF antibody titers were determined two weeks following the last immunization, prior to challenge. The YscF-specific antibody titers of PBS-immunized mice were below the ELISA assay baseline of 1:800 (Table [1](#T1){ref-type="table"}), as was the pre-immune serum (data not shown). However, the HT-YscF immunized mice reached a GMT (geometric mean titer) of 1:40,000 (Table [1](#T1){ref-type="table"}) for total IgG. Isotyping analysis was performed on the pooled sera obtained from the 22 mice selected for total the IgG analysis reported in Table [1](#T1){ref-type="table"}. Pooled sera were used to minimize the animal to animal variation expected from using outbred mice. The isotyping analysis demonstrated no significant IgA or IgM production (Table [2](#T2){ref-type="table"}) and a very high IgG titer (Table [2](#T2){ref-type="table"}) as expected from the data in Table [1](#T1){ref-type="table"}. Among the IgG subclasses IgG2b appeared to have the highest levels (Table [2](#T2){ref-type="table"}), although IgG1 and IgG2a levels were also very high (Table [2](#T2){ref-type="table"}). IgG3 levels were the lowest (Table [2](#T2){ref-type="table"}). Generally, immunization with CFA tends to drive a strong Th1 response. Immunization with HT-YscF in CFA induced a strong Th1 response, evident by the high IgG2a response. However, a strong Th2 response is also present as seen by the high IgG1 and IgG2b levels. Immunization with HT-YscF induced a strong IgG1 response in mice and interestingly, Titball et al. have shown that high IgG1 titers to a F1/LcrV recombinant subunit vaccine correlated very well with protection against pneumonic plague in mice \[[@B40]\]. This suggests that YscF, which also induces a strong IgG1 response, could possibly afford some protection against pneumonic challenge as well as against systemic challenge. Ability of α-YscF to effect Yop translocation --------------------------------------------- To examine one possible method that anti-YscF could be functioning to provide protection in immunized animals the ability to translocate Yops in the presence of anti-YscF antiserum was examined. Antibody to the surface-localized LcrV has been shown to block the ability of the TTSS in Y. pestisto translocate Yops into cultured macrophages \[[@B41],[@B42]\] but anti-LcrV was unable to block translocation into HeLa cells \[[@B21]\]. However, anti-LcrV was able to block Yops translocation by Y. pseduotuberculosisinto HeLa cells \[[@B26]\]. Since YscF is also surface-localized the ability of anti-HT-YscF to block Yop translocation into HeLa cells was tested. Day et al have described an elegant methodology to follow the translocation of Yop effector by fusing them to a Elk reporter \[[@B43]\]. Elk, a eukaryotic transcriptional activator, becomes phosphorylated only after entering the nucleus, providing a reporter for translocation into eukaryotic cells \[[@B43],[@B44]\]. This methodology has the advantage of not requiring cell fractionation and protease protection assays to establish the intracellular localization of translocated proteins. To test the ability of anti-YscF to block translocation Y. pestisKIM8-3002 was transformed with plasmid pYopE~129~-Elk \[[@B43]\]. HeLa cells were infected at an MOI of 10 and infection was allowed to progress for 4 h. After the 4 h incubation infected HeLa cells were harvested and immunoblotted to analyze YopE, Elk and PO~4~-Elk production. Y. pestisKIM8-3002 (wt; \[[@B24]\]) and an isogenic translocation defective strain containing a yopDdeletion (KIM8-3002.2, ΔyopD; \[[@B45]\]) both containing pYopE~129~-Elk were used as positive and negative translocation controls, respectively. Immunoblots of HeLa cells infected with the wt and the ΔyopDstrains showed that only the wt strain elicited the production of PO~4~-Elk while the ΔyopDstrain had no production of PO~4~-Elk. The wt and the ΔyopDstrains were used to infect HeLa cells in the presence of a 1:10 or a 1:25 dilution of anti-YscF (titer for HT-YscF, 1:100,000) or in the presence of anti-PcrG (a Yersinianon-reactive antibody control, titer for PcrG, 1:20,000). The wt strain was capable of translocating YopE-Elk in presence of both anti-sera, demonstrating that anti-YscF was not capable of blocking Yops translocation and expectedly the ΔyopDstrain was still defective for translocation. The experiment likely contained sufficient antibody against YscF to block translocation. In a previous report Pettersson et al used as low as a 1:100 dilution of an anti-LcrV anti-sera with a titer of 1:20,000 for LcrV and in that experiment translocation of Yops into HeLa cells was blocked \[[@B26]\]. These results suggest that antibody to YscF may not exert its protective effect by blocking Yops translocation. The results also suggest that while YscF is surface-exposed in the yersinae, antibody directed against YscF, unlike, anti-LcrV cannot block translocation. This may imply that YscF activity is shielded from neutralization by antibody, unlike LcrV activity that is blocked by antibody in some cases. However, since anti-LcrV was unable to block Yops translocation into by Y. pestisinto HeLa cells \[[@B21]\] but could block translocation into cultured macrophages \[[@B41],[@B42]\], the possibility remains that anti-YscF also display this type of differential blockage. Conclusion ========== In this study we have determined that immunization of mice with recombinant YscF can protect mice from an i.v. challenge with Y. pestis. This is the first report to our knowledge that has utilized a conserved protein from the type III secretion complex of a gram-negative pathogen as a candidate for vaccine development. This result suggests that type III secretion complexes of other gram negative pathogens could also serve as vaccine targets. YscF and its homologs are obvious targets for use as vaccine candidates as they are surface exposed and are required for virulence in all the systems examined. The protective antibody response elicited by HT-YscF is evidence that YscF is not only expressed during the course of a plague infection, but is also in a location accessible to antibodies at some point in the infectious process. The mechanism of protection by the YscF antibody response is currently under investigation. Essentially three possibilities exist to explain the antibody\'s protective activity: increased opsonization of the bacteria, enhanced complement binding to the bacterial surface, or direct blocking of Yops translocation into the host cells. Cytotoxicity assays (data not shown) and the ability to translocate Elk-tagged YopE into HeLa cells (Figure [3](#F3){ref-type="fig"}) have shown that it is unlikely that anti-YscF directly blocks Yops translocation into HeLa cells. Suggesting that a blockage of Yops translocation may not be the mechanism whereby anti-YscF antibodies are protective. The degree of protection observed after immunization with YscF is not as great as that seen for the two known protective antigens, F1 and LcrV. This result suggests that YscF could be best used in combination with the other known antigens to formulate a tri-valent vaccine for Y. pestis. However, further work could lead to the development of YscF as a monovalent vaccine or combined with other antigens that could be efficacious not only against Y. pestis, but also against Y. enterocoliticaand Y. pseudotuberculosis. Methods ======= Cloning of yscFfor overexpression and HT-YscF purification ------------------------------------------------------------ Plasmid pJM119 was constructed by cloning a BamHI-and XhoI-cleaved PCR product into pET24b (Novagen, Madison, WI). The primers used to amplify yscFwere HT-YscF Start (5\' CGG GAT CCG ATG AGT AAC TTC TCT GGA TTT 3\') and HT-YscF Stop (5\' CCG CTC GAG TGG GAA CTT CTG TAG GAT GCC 3\'). E. coliBL21(DE3) (Novagen) harboring pJM119 was used for HT-YscF overpexpression according to the manufacturer\'s suggestions. HT-YscF was purified using Talon resin (BD Clontech, Palo Alto, CA) according the manufacturer\'s directions. Immunization of mice and infection with Y. pestisKIM5 ------------------------------------------------------- For primary immunization 6-to 8-week-old female Swiss-Webster mice were immunized i.p. with 40 μg/mouse His-tagged YscF or phosphate-buffered saline \[[@B37]\] PBS (control mice) emulsified 1:1 with complete Freund\'s adjuvant (CFA). Booster immunizations were performed the same as the primary immunization with the substitution of Incomplete Freund\'s Adjuvant for CFA. Mice were challenged with Y. pestisvia the retro-orbital sinus using blunt-feeding needles. Y. pestisused to infect mice was grown overnight in HIB broth, sub-cultured to an A~620~of 0.2 absorbance units and grown with shaking to an A~620~of 1.0 absorbance unit. Y. pestiscells for infection were harvested by centrifugation and resuspended in PBS. Plate counts were performed to verify CFUs for the infectious doses. Infected animals were monitored for death for up to 19 days, after which survivors were euthanized by CO~2~inhalation, according to the guidelines of the Panel on Euthanasia of the American Veterinary Medical Association. All animal work for this project was reviewed and approved by UND\'s IACUC. Bacterial strains, growth and fractionation ------------------------------------------- Bacterial strains used were KIM8-3002 \[[@B24]\], ΔyscFan isogenic in-frame deletion of yscF(G. Plano, University of Miami), and KIM8.3002.2 ΔyopD\[[@B45]\]. Y. pestisstrains were grown in heart infusion broth or on tryptose blood agar base medium (Difco Laboratories, Detroit, MI) at 26°C for genetic manipulations. For physiological studies, growth of Y. pestiswas conducted in a defined medium, TMH, as previously described \[[@B46]\]. Bacterial cells were fractionated as previously described \[[@B24]\]. Briefly, bacterial cells were chilled on ice after growth, harvested by centrifugation, and washed in cold phosphate-buffered saline (PBS; \[[@B37]\]). Bacterial whole cell fractions were prepared by resuspending the washed cells in cold PBS and precipitating total proteins with 10% (vol/vol) trichloroacetic acid (TCA) on ice overnight. Secreted proteins were recovered from the bacterial growth medium by centrifuging the spent medium a second time, transferring the supernatant to a clean tube, and precipitating with 10% (vol/vol) TCA on ice overnight. The TCA-precipitated proteins were pelleted by centrifugation (20,800 × g at 4°C) for 20 min and resuspended in 2X sodium dodecyl sulfate (SDS) sample buffer \[[@B37]\]. Protein electrophoresis, visualization and immunodetection ---------------------------------------------------------- Proteins were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE), using 12.5 % or 15 % (wt/vol) polyacrylamide gels according to the method of Laemmli \[[@B47]\]. Samples were boiled 3--5 min before loading on the gels. Samples were loaded such that lanes containing different culture fractions represented equivalent amounts of the original cultures. Proteins were visualized in gels using GelCode Blue stain (Pierce Chemical, Rockford, IL) according to directions. For immunoblots, proteins resolved by SDS-PAGE were transferred to Immobilon-P membranes (Millipore Corp., Bedford, MA) using carbonate transfer buffer (pH 9.9) \[[@B48]\]. Specific proteins were visualized using mouse or rabbit polyclonal antibodies specific for YopE (rabbit α-YopE; gift from G. Plano, University of Miami, Miami, FL), YscF (mouse α-YscF, this study), Elk (rabbit α-Elk, Cell Signaling Technology, Beverley, MA) and PO~4~-Elk (rabbit α-PO~4~-Elk, Cell Signaling Technology). Hexahistidine tagged YscF was visualized using a penta-histidine specific antibody (mouse α-Penta-His, Qiagen, Valencia, CA) Alkaline phosphatase-conjugated secondary antibody (goat anti-rabbit immunoglobulin G or goat anti-mouse immunoglobulin G; Pierce) was used to visualize proteins by development with nitroblue tetrazolium and 5-bromo-4-chloro-3-indolylphosphate (NBT-BCIP; Fisher Scientific, Fair Lawn, NJ). Antibody characterization and isotyping --------------------------------------- Flat-bottom, 96-well Nunc Maxisorp immunoplates (Fisher Scientific, Pittsburgh, PA) were coated with 100 μl of HT-YscF solution (4 μg/ml in Binding solution (0.1 M NaH~2~PO~4~, ph 9.0) at room temperature for 2 h (or overnight at 4°C). The wells were blocked with 200 μl/well blocking buffer (1% bovine serum albumin in TTBS (tris-buffered saline \[[@B37]\] + 0.5% Tween 20) and washed with TTBS. Test sera were serially diluted in blocking buffer and 100 μl of each dilution was added to duplicate wells that were incubated for 2 h at RT (or overnight at 4°C). The plates were washed and incubated for 2 h at RT with alkaline-phosphatase-conjugated anti-mouse secondary antibody. For quantitation of YscF-specific immunoglobulin isotypes and subclasses the plates were coated with alkaline-phosphatase-labeled anti-mouse isotype-specific antibody (1:400 in blocking buffer; Southern Biotech, Birmingham, AL). The wells were washed and 75 μl 3 mM para-nitro phenyl phosphate (p-NPP) was added to each well. The plates were incubated for 15 min at RT and the reaction was stopped by the addition of 50 μl of 1.5 M NaOH to each well. A~405~was measured to monitor the cleavage of p-NPP. Antibody titers were determined as reciprocal numbers of the highest serum dilution that displayed values for optical density twofold higher than the value of the control serum. Infection assays ---------------- Infection of eukaryotic cells was performed as described previously \[[@B24]\]. Prior to infection, eukaryotic cells were subcultured into 35-mm-diameter six-well tissue culture plates in RPMI-FBS and incubated at 37°C under 5% CO~2~for 48 to 72 h to a density of 5 × 10^5^to 8 × 10^5^cells per well. Cells were washed twice with warm L15 lacking FBS immediately prior to infection. Bacteria were cultivated at 26°C in HIB and used at an OD~620~of \~1.0 for tissue culture infections. Bacteria were added (at a multiplicity of infection (MOI) of 5 to 10) directly to prewarmed medium in the wells of the six-well plates. Plates were then centrifuged at 200 × g at RT for 5 min to achieve contact between the bacteria and the target cells and incubated at 37°C for 4 h. Translocation of YopE --------------------- Detection of Elk-tagged YopE from pYopE~129~-Elk was performed as described \[[@B43]\]. Y. pestisstrains carrying plasmid pYopE~129~-Elk were used to infect HeLa cells. After 4 h, the culture supernatants were removed, and the infected adherent cells were lysed by the addition of 100 μl of 2X SDS-PAGE lysis buffer containing Pefabloc (Roche Molecular Biochemicals, Indianapoli, IN) and phosphatase inhibitor (P-2850) cocktail (Sigma, St. Louis, MO). Samples were boiled for 5 min and loaded onto 12.5 % SDS-PAGE gels, immunoblotted to PVDF membranes and probed with Elk-1 (\#9182) or phosphospecific Elk-1 (\#9181) antibody preparations (Cell Signaling Technology). Anti-sera specific for HT-YscF (titer of 1:100,000, this study) or the Pseudomonas aeruginosaLcrG homolog, PcrG (titer of 1:20,000, Matson and Nilles, unpublished), were used at dilutions of 1:10 or 1:25 in the infection medium to assess the ability of α-YscF to effect YopE translocation. Authors\' contributions ======================= J .S. M. cloned yscFfor overexpression, purified HT-YscF, characterized the reactivity of antiserum to YscF, assisted with immunization and infection of mice, assisted with the LD~50~calculation, and wrote the draft of the manuscript. K. A. D performed immunizations and infections of the mice and performed the antibody isotyping. D. S. B. helped to design the immunization protocol and edited the manuscript. M. L N. conceived of the study, supervised the work, calculated the LD~50~and edited the manuscript. Acknowledgements ================ The author\'s would like to thank Deanna O\'Bryant and Jennifer Lamoureux for assistance with mouse experiments, Jennifer Miller for help with antibody ELISAs and Gregory Plano (University of Miami, Miami, FL) for YopE antiserum and the ΔyscFstrain of Y. pestis. This work was supported by the UND Faculty Research Seed Money program. J. S. M was supported by a pre-doctoral fellowship from ND-EPSCoR. Work in M. L. N.\'s laboratory is supported by NIAID grants R01-AI051520 and U01-AI54815. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Purified HT-YscF was run on SDS-PAGE gels (12.5% for Panels A and B, 15% Panel C) to analyze purity of the recovered protein. Panel A, separated proteins were stained with GelCode Blue stain (Pierce Chemical). Panel B, purified HT-YscF was immunoblotted to PVDF and probed with antibody specific for penta-histidine to identify which bands contained the His-tag. Panel C, purified HT-YscF was immunoblotted to PVDF and probed with post-immune serum from immunized mice. ::: ![](1471-2180-5-38-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Derivatives of Yersinia pestisKIM8-3002 (KIM5 pPCP1-minus, Sm^r^) were grown in a chemically defined medium at 26°C for 2 h in the presence (lanes 1, 3, and 5) or absence of calcium (lanes 2, 4, and 6). Lanes 1 and 2 contain Y. pestisKIM8-3002. Lanes 3 and 4 contain Y. pestisKIM8-3002 ΔyscFexpressing YscF from pBAD18-YscF. Lanes 5 and 6 contain Y. pestisKIM8-3002 ΔyscFgene. After the 2 h growth, the culture was shifted to 37°C to induce expression of the Ysc type III secretion system and the Low Calcium Response. Following 4 h of growth at 37°C cultures, were centrifuged to obtain whole cell fractions and cell-free culture supernatant fractions. Total proteins from each fraction were precipitated with 10% trichloro acetic acid. Dried proteins were resuspended in SDS-PAGE sample buffer and electrophoresed in a 15% SDS-PAGE gel. Proteins were then transferred to a PVDF membrane and immuno-blotted with pooled mouse serum used at a 1:20,000 dilution. Mouse serum was obtained by bleeding mice subsequent to immunization with HT-YscF, serum from several mice was pooled to control for animal specific variation. The position and sizes for the molecular weight markers are indicated and the position of YscF is shown. ::: ![](1471-2180-5-38-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Y. pestisstrain KIM8-3002 (wt) (lane 1 and lanes 3--6) and KIM8-3002.2 (ΔyopD) (lane 2 and lanes 7--10) both containing plasmid pYopE~129~-Elk were used to infect HeLa cells at an MOI of 10. 4 h following infection the culture supernatant containing non-adherent bacteria were removed the remaining adherent HeLa cells were solubilized in 2X SDS-PAGE buffer. Following solubilization proteins were separated and immunoblotted to triplicate PVDF membranes. The triplicate blots were probed with α-YopE, α-Elk, or α-PO~4~-Elk, followed by incubation with an alkaline phosphatase conjugated secondary antibody and developed using NBT/BCIP. To some samples anti-sera specific for HT-YscF (lanes 3--4 and lanes 7--8) or the Pseudomonasprotein, PcrG (lanes 5--6 and lanes 9--10), were added at dilutions of 1:10 (lanes 3, 5, 7, and 9) or 1:25 (lanes 4, 6, 8, and 10). ::: ![](1471-2180-5-38-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### IgG response to HT-YscF vaccination and LD~50~determination. ::: Immunogen anti-YscF GMT\* LD~50~ Fold increase in survival ----------- ----------------- -------- --------------------------- PBS \< 1:400 159 \- HT-YscF 1:40,000 21,344 134 \Geometric mean titer for total IgG, determined from 22 HT-YscF immunized mice. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Antibody isotype titers from mice\ immunized with HT-YscF. ::: Antibody isotype Titer ------------------ -------------- Total Ig 1:100,000 IgM \< 1:800 IgG1 1:100,000 IgG2a 1:100,000 IgG2b \> 1:100,000 IgG3 1:20,000 IgA \< 1:800 \*Sera from 22 mice were pooled for this analysis. :::	PubMed Central	2024-06-05T03:55:58.244478	2005-6-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168899/", "journal": "BMC Microbiol. 2005 Jun 24; 5:38", "authors": [ { "first": "Jyl S", "last": "Matson" }, { "first": "Kelly A", "last": "Durick" }, { "first": "David S", "last": "Bradley" }, { "first": "Matthew L", "last": "Nilles" } ] }
PMC1168900	Background ========== Patients are increasingly urged to become involved in treatment choices, including watchful waiting (surveillance but no active treatment), by working through medical evidence to decide on a course of action with their physicians. Guidelines and evidence summaries are commonly available to physicians and frequently include a \"summary for patients\", as do some major clinical journals. Decision aids have proliferated in recent years yet most require high levels of literacy to access them and have not been designed according to the principles of plain language. This paper describes one of the first attempts to make a decision aid accessible to consumers with lower levels of literacy and education using principles from reading research and document design. To develop a prototype plain language decision aid, and to describe the design process, we designed a decision aid for treatment decision making in localized prostate cancer. Three standard therapies, surgery, radiation therapy, and watchful waiting, are available to patients and paid for by most insurers in the US. The 10-year mortality rate for each of the standard treatments is equivalent, but the side effects and the processes are quite different. This conundrum is made even more difficult because specialty physicians tend to recommend treatments performed by their own specialty \[[@B1],[@B2]\]. Even inter-disciplinary case conferences can expose patients to evidence and professional opinions that appear to conflict. The aim of the research reported here was to develop a plain language decision aid for localized prostate cancer initial management following a positive biopsy. Three decision aids (booklet, Internet, and audio-tape formats) were designed to contain equivalent information. Formative evaluation was designed to test acceptability to men with both highly functional and limited literacy skills. We wanted to make sure that the DAs were not perceived as \"dumbed down\" either in tone or content. A further aim was to identify missing information, and information that might not adequately represent the cancer experience, The aim was to include the perspective of community dwelling men in the appropriate age group and men who had had treatment for their prostate cancer. Evidence-based patient choice ----------------------------- \"Evidence-based patient choice\" is a term coined by Hope in 1996 and elaborated in a book by the same title \[[@B3]\] to articulate the addition of the patient perspective to the evidence based paradigm laid out by Sackett \[[@B4]\]. Adding \"patient choice\" to evidence-based medicine is an effort to locate the need for evidence squarely in a patient-centred approach to patient-provider encounters. The theoretical roots are in economics, ethics, psychology and medicine. There are also political influences, including the consumer choice movement and struggles over resource allocation and health care organization. The synthesis of ethical approaches to shared decision making, and rational choice models provoke both theoretical and political concerns, and is far from a perfect resolution. The practical task of supporting patients\' access to information they can use led us to develop this new prototype. We also desired to have the information available in real time to reach the most satisfying resolution of treatment dilemmas. By making our assumptions transparent, and testing the results with patients and scholars we hoped to improve both theory and practice. Approaches to evidence tools for patients ----------------------------------------- Decision aids exist in modest numbers, specific to particular clinical conditions. A recent meta-analysis identified 131 different decision aids across a variety of clinical problems. Sixty-five of these fit the Cochrane criteria and were evaluated in a randomized trial \[[@B5]\]. Decision aids have expanded the scope of traditional patient education materials and draw on several perspectives. These may be classified conceptually as 1) traditional medical, 2) technology assessment (TA), and 3) patient-centred. Traditional patient education materials mirror the medical model didactic elements. They include descriptions of: a) the disease, b) its epidemiology, c) the morbidity and mortality outcomes of the disease in the population, and d) screening and treatment recommendations for either public information campaigns or clinical encounters. This model has been criticized for its emphasis on diseases, rather than on patients and their concerns, and for being paternalistic \[[@B6]\]. Decision aids have addressed the limitations of the traditional medical approach by integrating evidence-based medicine with patient choice and by encouraging shared decision making. It has largely accomplished this with a technology assessment (TA) approach. TA focuses on the quality of the medical evidence, levels of rigor of evidence and providing symmetry in describing risks and benefits of all patient options. The most widely available examples of TA-based decision aids use the Ottawa Decision Support Framework and BMJ Best Treatments approach. The Ottawa Framework asks the decision-maker to choose among treatment or screening choices demonstrated to be effective \[[@B7]\]. It emphasizes clarifying values for outcomes and emphasizes weighing the pros and cons of alternatives in the choice process to arrive at a tentative choice that the patient discusses with the physician. Consistent with evidenced-based health care, it grades the medical evidence for each alternative according to its quality. The Best Treatments approach likewise grades the evidence, teaches patients about uncertainty in data, but does not consistently provide numerical data for side effects \[[@B8]\]. The TA approach is a major step forward in providing patients with accurate evidence for making comparisons among alternatives, but it may suffer from speaking to the professional health care community better than it speaks to patients. To develop a plain language prototype, we accepted most of the premises of the TA approach, but redesigned the presentation of results, using plain language and document design strategies. Translation of evidence into plain language ------------------------------------------- The rationale for \"re-constructing\" decision aids in plain language is that it works for patients with a wide range of literacy abilities. Plain language includes three major elements: 1) The use of everyday language and other clear writing strategies, 2) well-structured, logically sequenced, and focused information, 3) effective design and layout. It uses medical evidence as a base and attempts to lead readers through the relevant elements to provide an opportunity for synthesis and drawing conclusions about which therapy best suits the values, fears and expectations of the patient. A plain language approach incorporates social marketing principles and is often used by pharmaceutical companies, patient advocacy groups, and others to deliver compelling messages. The materials these groups produce, however, sometimes minimize or eliminate numbers in the interest of being non-threatening and engaging and accommodating limited quantitative literacy skills\[[@B9],[@B10]\]. Plain language approaches have also been used to communicate public health messages to patients. They frequently attempt to persuade them to undertake an action (e.g., engaging in cancer screening) rather than engage in shared decision making. Highlighted by the recent Institute of Medicine report \[[@B11]\], many groups have published plain language document design guidelines \[[@B12]-[@B17]\]. As stated by Hibbard and Peters \[[@B18]\], the challenge is \"how to present\\ldotsinformation so that it is actually used in decision making.\" They note that three information presentation strategies assist consumers, all of which are reflected in a plain language approach: lowering the cognitive effort required to use information; linking information to real-life situations to increase the emotional connection; and presenting numbers in frequencies (e.g. 10 in 100). We addressed each of these in developing the prostate cancer DA. To address the immediate world of patients in the midst of being told of a disease and making a decision, it is important to describe the relevant risks and benefits of treatments without losing the precision and realistic risk communication capacity of numbers. Unpacking technical jargon actually adds substantial detail about what will happen to the patient in non-technical, everyday terms. It also explains laboratory tests and other medical information that doctors will use in making recommendations. The purpose is to provide patients with tools and encouragement to engage with the doctor in discussing technical aspects of the information doctors use to make a decision. Along with risk and benefit data, a decision aid should also explain the treatment process to the patient, in order that she or he can anticipate the personal experience of undergoing treatment. This approach does not abandon or delete technical language. Rather it defines it and provides examples, and phonetic pronunciations of medical technical terms in order to improve communication between patients and health professionals. In the prototype, we introduce the idea that there is variability among patients in their values for potential outcomes. Unlike some decision aids that provide a rational recommendation based on formal utility assessment \[[@B19]\], this approach does not make a recommendation. The resolution of the values dilemma is left to patients, their families and friends, and health care providers. Needs assessment ---------------- Prior to developing a new decision aid, we performed a two-part needs assessment. In the first part, we searched the literature for surveys providing information about knowledge levels of newly diagnosed patients following biopsy and discussion with their doctors but before treatment. In the second part, we evaluated existing patient education materials for localized prostate cancer. Since we found no surveys of men actually facing the localized prostate cancer treatment decision, we interviewed 184 men newly diagnosed with localized prostate cancer from both community and tertiary care urology practices in Michigan \[[@B20]\]. This sample served as the historical controls for this paper. The survey included demographics, patient knowledge about their prostate cancer, treatment options, side effects and satisfaction with care. Bivariate and multivariable analyses were performed to adjust for potential confounding and to determine if sociodemographic factors affected understanding of key issues. Average age of the men was 64 years and 19% were African American. Seventy-one percent completed some college and 51% of patients were currently employed. White men reported higher rates of knowing their PSA, stage and grade (92%, 93% and 99% respectively) than African American men (69%, 88% and 85%). Ninety-three percent of all respondents stated that they had discussed treatment options with a physician. Only 62% knew that radical prostatectomy, external beam radiation therapy, brachytherapy and watchful waiting were standard options for localized prostate cancer. Significant associations between sociodemographic factors and a patient\'s understanding of prostate cancer treatment issues were identified. Despite objective gaps in knowledge, men showed high satisfaction with information received from their physicians. Prostate cancer has been the subject of several decision aids \[[@B21],[@B22]\] Considerable research has identified patient and health professional concerns about treatment. In part 2 of the needs assessment, we used our previously developed generic evidence template to provide evaluation guidelines, but made it specific to prostate cancer \[[@B23]\]. We used the template to comprehensively survey patient education materials publicly available from US government agencies, charities, drug companies and others for inclusion or absence of the evidence elements. The full report found key evidence to be missing from most patient education materials (PEMs) \[[@B24]\]. We found that standard treatment options were almost never explicitly compared to one other. PEMs frequently did not describe the chances of dying from prostate cancer or of experiencing other side effects of each treatment. We found most did not reflect consideration of the user in the way materials were written, structured or designed. Finally, most did not encourage patients to formulate a treatment preference to discuss with their doctors. Based on the needs assessment, we developed the decision aid, described in this report, in three formats (booklet, audio tape, and Internet) with particular attention to plain language and to potential concerns of African American men. Three major challenges lay at the heart of the plain language translation: 1) introducing the social and emotional context, 2) translating the outcome and side effect rates into plain language, and 3) encouraging shared decision making. The three formats used the same words and content organization to facilitate evaluation of the impact of print, audio and Internet media on patient knowledge and patients\' assessment of the usefulness of the different media in preparing to talk with the doctor. Methods ======= The decision aid prototype: design method and assumptions --------------------------------------------------------- Based upon our earlier analysis of good and bad aspects of the prostate cancer PEMs and an extensive search of the clinical literature, a draft of the decision aid was prepared in what might be called the standard medical parlance. This material was iteratively modified to emphasize shared decision making and incorporate the principles of plain language. The research team who designed the project was strongly guided at this point by two health literacy/plain language experts. Similarly, document design and web design were strongly guided by graphics experts. The prototype was presented to the Michigan Prostate Cancer Action Committee (PCAC), a sub-committee of the Michigan Cancer Consortium (MCC), a consortium of hospitals, universities and cancer centrss in the State of Michigan. Final edits were made based on the formative evaluation by patients, clinicians, and the development collaborators and the final document produced. The prototype booklet and Internet version that resulted from the evaluation process can be viewed at the Michigan Cancer Consortium (MCC) website \[[@B25]\]. Table [1](#T1){ref-type="table"} lists the design decisions made for the prostate cancer prototype. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Information and Information Architecture Issues and Solutions ::: Concern Solution -------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------- Complex information • Organize and structure document for ease of understanding and use, including use of summary tables • Create affective appeal using individual vignettes • Describe all elements of treatment process patients experience Engage the reader in the social context • Use pictures captioned with comments from men of varying ethnicities • Suggest connection with trusted others and other patients • Recognize and address feelings of fear No mandatory autonomy • Suggest that decisions may be made by the patient or delegated to the doctor, or trusted individual. • Suggest shared decision making is compatible with keeping or delegating Anticipate cultural issues • African Americans are likely to be sensitive to potential to withhold treatment in the name of watchful waiting. • African Americans frequently are diagnosed with more severe disease relative to whites. • Sources of authoritative advice vary among ethnic groups. These can be acknowledged subtly in text Translate medical words, but do not eliminate technical information • Define words and include a glossary Creating/choosing appropriate pictures • Simplify anatomical drawings to include only necessary visual information • Include realistic picture of torso, with head included to provide conceptual context • Include photographs of men who look serious, but not devastated Graphic Design and Layout: Use evidence-based document design principles • Spacing to emphasize different points and topic changes • Wide margins to make text less dense • Adequate page size to keep information on a topic on one page • Graphic design that draws the reader through the booklet • Text design that keeps related information in close proximity Plain language • Use conversational words and sentence structure, even if not standard grammar • Keep the writing tight. Eliminate unnecessary words. • Keep the tone friendly and personal, speaking directly to the reader. • Explain technical language even though it adds words. Assumptions behind technical words must be included • Make scientific references available on the website, but do not include in the text ::: Connecting to patients\' social and emotional world --------------------------------------------------- To connect with the concerns of men and their families and friends about prostate cancer treatment, the decision aids embed a number of emotional \"hooks\", showing photographs of middle-aged men of varying races. Quotes from different men state varying preferences for their preferred degree of autonomy: making the decision themselves, delegating it to family or the doctor, sharing it with clergy, friends, or health professionals. These statements are designed to legitimize a range of appropriate alternatives for the problem of \"who decides\". They are meant to avoid a \"mandatory autonomy\" position, allowing the patient to hand off the decision to the physician if he does not want to decide himself \[[@B26]\]. The central message promotes active patient participation, followed by making, sharing, or delegating the decision. Further, an attempt is made to imbed in the text items that speak to culture-specific issues. For example, the statement is made that watchful waiting does not mean that the doctor \"just does not want to treat me.\" This statement anticipates the common belief, particularly among African American patients, that doctors may withhold treatment for cost or other reasons. The emotional context of the booklet is also invoked by explicitly noting in the text that cancer is scary, and that treatment side effects may have heavy emotional impact. The emotional \"threat\" level of how the disease is discussed is planned to be at a moderate level. The decision aid acknowledges inevitable fear, supplies concrete understanding of disease and treatment, and encourages a sense that patients are competent to make the decision. Presenting quantitative evidence -------------------------------- The goal of an evidence-based decision aid is to make the information easy to read, remember, and use. The prostate cancer prototypes use specific textual, organizational, and design elements enhanced by appropriate message framing. They do not assume the reader knows either the technical meaning of words or the background context in which they are used. This filling in of the \"why\" and \"what for\" of laboratory tests, treatment regimens, and other medical information is essential for the millions of adults with limited literacy skills and limited experience with navigating the medical system. It is also preferred by most people when learning new information, especially in a highly charged emotional context. Specific guidelines for the prostate cancer decision aids were: 1) aim for low to average reading level (grade 7 level), 2) present quantitative risk information as simply as possible, both positively and negatively framed \[[@B27]\], and 3) use a minimum of technical medical terms, and provide a glossary, called \"explanation of medical terms\". The objective is to invite all patients to grapple with the decisions they face with as little intervening struggle with technical medical language as possible. The result, ideally, is sophisticated ideas in plain language. To invite the patient into the content requires attention to \"information architecture\" and the layout of information on pages to guide the reader through the text. Thus, the density of information on a page, the choice of a booklet or an electronic tool, and the length and \"feel\" of the information tool become important. The organizing principle was to describe each treatment option in a complete synopsis on its own on two facing pages of the booklet, or on sequential web pages under one button. The audiotape is organized in \"chapters\", like a \"book on tape\". For each treatment, the text describes, \"What happens\", \"How this treatment can help\", and \"How this treatment may cause problems.\" Each of the individual synopses tells the whole story about a specific treatment. The treatment options are followed by two comparison tables, one narrative and one quantitative. The summary tables explicitly compare results across treatments. Very brief directions guide the reader in how to use the tables. Additional information, not related explicitly to the choice (stage and grade of cancer, and PSA tests), is described in the same amount of detail as a single treatment option (two facing booklet pages). The objective is to demystify what will happen, show how doctors decide what is best, describe the degree of uncertainty about test outcomes, and indicate the medical information on which decisions are based. Side effects and mortality are presented as mean values, not including the ranges. This is a pragmatic solution to the simplification task, and should be tested further. Explaining that diagnostic test results are imperfect information, and that treatments have only a probability of success is a new idea to most patients. The design task was to present uncertainty, but keep it simple. Another quantitative principle that is critical to informed consent is presentation of absolute, rather than relative risks in terms of specifying a number out of 100 rather than a percentage. Absolute rates have been demonstrated to provide patients with improved estimates of their own chances of experiencing a particular outcome \[[@B28]\]. Shared decision making ---------------------- A third challenge was to encourage shared decision making without specifying the form of the clinician-patient conversation or the values discussion. The message on the cover of the DA first introduces the concept of choice, that \"there is no right answer\" to this treatment problem. The general statement, \"Your treatment decision is a shared one between you and your doctor\" is later stated directly. A \"things to think about\" section at the end of the booklet asks a patient to identify his most important goal for treatment, such as curing cancer, ameliorating symptoms, having the best possible sexual performance, or good bowel and bladder control. Patients are also asked to write down the best and worst thing for them about each treatment option. These sections are designed to facilitate an open discussion with the physician without directing it. The booklet was provided to each person with a request to provide written or verbal feedback. The booklet form of the decision aid was finished first, and provided the text and order of presentation that were duplicated in the audiotape and Internet versions. Qualitative methods were used to evaluate and revise the booklet before survey testing was performed with a new sample of men newly diagnosed with prostate cancer. Results ======= Formative evaluation -------------------- Evaluation was performed in two phases. Phase 1 was a focus group evaluation of the draft language. Phase 2 was a survey follow-up of 60 men newly diagnosed with prostate cancer, using the DA in real-time decision making. The Michigan State University and University of Michigan human subjects IRBs approved all study materials and procedures. Verbal and written consent for participation was obtained at the beginning of interviews and patient encounters. Phase 1: focus groups --------------------- ### Design and procedure Focus group methods were used to identify unacceptable or confusing language and to identify missing information or information needed to fully represent the important issues for decision making. This information was obtained from the perspective of men who had not experienced prostate cancer, and also those who had. The analytic techniques used were similar to ethnographic techniques, although cross-sectional rather than longitudinal qualitative reports were used \[[@B29]\]. Homogeneous groups by race were formed to encourage participants to feel comfortable and to express concerns about bias. A moderator\'s guide addressed the acceptability of content and language, perceived understandability, and usability of the format. We inquired specifically about text passages that described risk numbers, results of a randomized clinical trial, the meaning of laboratory tests and their implications for mortality and treatment. Open-ended questions asked men their perception of the DA main message. Four focus groups were conducted. Each session lasted between 1.5 and 2 hours and was facilitated by a physician. Booklets were mailed to participants two weeks prior to the focus group and participants were encouraged to review the booklet in detail to provide feedback during the focus group session. Confidentiality ground rules were established and agreed upon by participants. After the session, a twenty-five dollar honorarium was provided to compensate participants for time and travel. Immediately following participants\' departure, a debriefing session was held between the facilitator and assistant to develop and amend observation notes and record suggestions for subsequent focus groups. All focus groups were tape-recorded. ### Sample Post-treatment men who had undergone surgery or radiation therapy for prostate cancer were recruited from a tertiary care prostate cancer clinic. Community men at risk for prostate cancer due to age were recruited from a study of decision making in benign prostatic hyperplasia. Actual patients at the point of decision making were not included in the focus group, due to their potential vulnerability to inaccurate, misleading, or offensive content. Twenty-five men participated in four focus groups. Mean age was 64.5 years (range 49--80). Twenty-four had health insurance; approximately half were not college educated. ### Focus group results The men found the prototype booklet \"encouraging\" and \"comforting.\" The general consensus was that prostate cancer appeared to be something \"that can be dealt with.\" The DA was described as \"professional,\" the \"right size\" (8 1/2 × 11), and containing the right balance between the size of white space and of print. The summary table of treatments and risks and the comparison table were viewed as especially helpful and men stated that these clarified individual treatment descriptions. The level of detail was described as being \"just right\", with the exception that the decision aid did not include detailed description of non-standard treatments (cryosurgery and nerve-sparing surgery). These were added, as was a recommended diagram showing cancer spreading to the surrounding lymph nodes and nerves. ### Positive and negative framing of statistical information The booklet stated each side effect as both a number out of 100 chance of occurrence and that number subtracted from 100 as the chance of avoiding occurrence. However, the obvious redundancy was interpreted by some readers as talking down to them and insulting their intelligence. To make it less repetitive, the wording was changed to vary the outcome named and acknowledge that this was another way of stating the outcome. Table [2](#T2){ref-type="table"} shows major changes made based on feedback. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Booklet Improvements based on formative evaluation ::: Information item identified in focus groups and approved by investigators Draft 1 Draft 2 ------------------------------------------------------------------------------- --------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Emotional impact None Hearing that you have prostate cancer may shock or frighten you, your family, and your friends. These feelings are natural. They may change over time, as you learn about your diagnosis, make treatment decisions, deal with symptoms, and go on with your life. Men are often afraid to share their feelings or get help from a counselor if needed. If strong feelings are hurting you or your family, ask your doctor to suggest help. Positive and negative framing \"If 45 men out of 100 experience impotence, this means that 55 do not.\" \"\...about 45 men out of 100 have permanent impotence. This means that 55 men out of 100 will have their original level of sexual activity.\" African American differences None \"African American men are often diagnosed at a younger age than white men and with more advanced prostate cancer. However, treatment may be equally successful for both groups if given the same care.\" Specific drug name, \"like Viagra\" \"medicine that helps with erections\" Treatment detail No mention of cryosurgery or nerve-sparing surgery Included under \"newer treatments\". Drawing Three inset pictures Only two Watchful Waiting description Few disadvantages listed Added disadvantages regarding potential progression of disease ::: ### Perspectives of African American men vs. White men While all found the booklet professional and a good source of information, there were differences in what issues the groups discussed. For African American men, the multicultural photos were important in that they illustrated that prostate cancer affects men of any race. The participants did suggest inclusion of more specific mortality information by race. (See Table [2](#T2){ref-type="table"}). African American men discussed several aspects of fear, including fear about acknowledging health problems, and the absence of fear in the faces of the men in the photos. They characterized the overall message of the booklet as not to be afraid. They also cited fear and distrust as major obstacles preventing their acknowledging a health problem. African American men did not feel the text talked down to them and felt the words were manageable -- \"there are no big words in this booklet.\" One commented that, \"I didn\'t feel like you were treating me like a 3^rd^grader.\" Lastly, the African American men read the overall message of the booklet as one of \"don\'t be afraid,\" whereas the White men read the message as one of \"informed decision making\". Several men across groups commented that they rarely read health related materials, but that they read every word before coming to the focus group. \"I normally only read sports. But I read this straight through\". ### Post-treatment men vs. community men Men who had been treated for their prostate cancer appeared to find discussing the DA more difficult than did the men of the same age who did not have prostate cancer. While they appreciated the information, some indicated that they were not convinced that they really had a choice when deciding on actual therapy, either because of the doctor\'s presentation or their own concerns about cancer. They also indicated that the consequences of the treatment decision were more emotionally difficult than they anticipated. These men appeared to struggle with the emotional and psychological consequences of their own treatment decisions. Post-treatment White men suggested that the emotional impact of treatment be more strongly addressed as illustrated by the following comments: \"But this doesn\'t deal with\... your first issue is to stay alive and the second issue is the quality of your life afterwards and that may be whether you are incontinent or maybe whether you don\'t have erections\.....But to the fact that, uh, you need to talk to your surgeon about the psychological consequences of the expected outcome of this surgery. Maybe you only think about living first, but looking at it in retrospect, I was like\.... Whoa, why didn\'t you tell me that ahead of time? Not that I would have made a different decision. I just want to be well informed. I didn\'t get that\". Some post-treatment men felt that the booklet was biased toward watchful waiting as a treatment choice. This was addressed by adding specific negative aspects to watchful waiting in the summary. For example: \"But this book, uh, had in mind a definite cast toward watchful waiting as opposed to more active interventions and I agree that, uh, that in fact for many of us that was not a very viable option I mean intellectually or emotionally or whatever\" and \"But as I said before, I thought there was a subliminal message there that said that watchful waiting was probably a preferred choice.\" Phase 2: survey --------------- To test the usefulness of the revised decision aid to men at the point of decision, a new convenience sample of 60 men newly diagnosed with prostate cancer on the basis of biopsy was recruited to test the usefulness and influence of the DA in real-time decision making. Participants were recruited and provided a decision aid after receiving their prostate cancer diagnosis at a post-biopsy visit to their urologists. The urologist introduced a research assistant to patients at the end of the visit in which he was told of his cancer diagnosis. The patient was asked if they would like to be in a research study evaluating the effectiveness of the three different decision tool formats. Patients were recruited sequentially until 20 had been accrued to use the booklet. After the Internet and audiotapes were developed, 40 additional patients were recruited to evaluate these formats. Due to the necessity for computer access for use of the Internet version, these men were offered a choice of the audiotape or Internet decision aid until 20 men were recruited for Internet and 20 for audio-tape use. A telephone survey lasting approximately 20 minutes was conducted by trained survey researchers from the Institute for Public Policy and Social Research of Michigan State University. The survey included knowledge items taken from the previous needs assessment. DA evaluation questions assessing balance, clarity, and length of the DA were modeled after those of Barry et. al. \[[@B30]\]. Demographic (age, race, income, education and geography) and disease (prostate specific antigen (PSA), cancer stage, and grade) characteristics were ascertained during the interview. Knowledge questions included knowing one\'s own test results (PSA, stage, grade), treatment options and side effects. ### Sample The sample of 60 men recently diagnosed with prostate cancer had an average age of 62 years (SD = 7.9). Ninety percent reported their race as White (8% as African American), 5% had not graduated from high school, 20% had a high school dipoloma or GED, 32% had some college training, 19% had received a bachelors degree and 24% had received training post-baccalaureate. Sixty percent were employed, 33% were retired, and 5% were unable to work. Twenty-two percent had an income of less than \$25,000 and 77% reported Internet access. ### Survey results The responses of men using the three different media (internet, audio, booklet) were virtually identical across all survey items with the exception that those receiving the audio version were less likely to share it with family and friends (data not shown). Table [3](#T3){ref-type="table"} shows that the sample of men utilizing the DA and historical controls had similar knowledge of their own pathology results, with the exception that the DA sample were less likely to be informed of the stage of their cancer (Fisher\'s Exact Test P-value = 0.003). Knowledge of treatment options was not significantly different between the two groups. However, 12% more men believed watchful waiting to be a standard treatment in the DA group (87% vs 75% in controls, p = 0.073). In addition, 6% fewer men believed external beam radiation to be a standard treatment in the DA group (88% vs 94% in controls, p = 0.159). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Knowledge: DA sample vs. historical controls ::: DA sample (N = 60) Historical controls (No./Total) Fisher\'s Exact Test P-value -------------------------------------------------------------------------------------------------------- ------------------------ ------------------------------------- ---------------------------------- Knowledge of personal pathology results Knew PSA 56 (93%) 160/181 (88%) 0.34 Informed of stage of cancer 53 (88%) 178/181 (98%) 0.00 Knew grade of cancer 57 (95%) 178/184 (97%) 0.69 Knowledge of treatment options Knew all of age, grade, stage, health, and PSA are at least somewhat important for treatment decision. 52 (87%) 142/184 (77%) 0.14 Believed watchful waiting to be a standard treatment 52 (87%) 138/184 (75%) 0.07 Believed external radiation to be a standard treatment 53 (88%) 173/184 (94%) 0.16 Believed brachytherapy to be a standard treatment 47 (78%) 143/184 (78%) \>0.99 Believed surgery to be a standard treatment 57 (95%) 177/184 (96%) 0.71 Discussed watchful waiting with physician 28 (47%) 72/184 (39%) 0.36 Discussed external radiation with physician 53 (88%) 166/184 (90%) 0.63 Discussed brachytherapy with physician 38 (63%) 121/184 (66%) 0.76 Discussed surgery with physician 59 (98%) 163/184 (89%) 0.02 Knowledge of side effects Knew surgery associated with incontinence 59 (98%) 172/184 (93%) 0.20 Knew surgery associated with impotence 58 (97%) 172/184 (93%) 0.53 Knew surgery associated with painful bowel movements 33 (55%) 82/184 (45%) 0.18 Knew radiation associated with incontinence 48 (80%) 113/184 (61%) 0.01 Knew radiation associated with impotence 56 (93%) 145/183 (79%) 0.01 Knew radiation associated with painful bowel movements 46 (77%) 118/184 (64%) 0.08 ::: Discussion of treatment options with the physician showed a significant increase in surgery discussions (98% in DA group, 89% in controls, p = 0.019). Discussions of watchful waiting increased by 6% to 47% in the DA group, and discussions of radiation decreased by 2% (external beam) and 3% (brachytherapy). While not significant, and very small, these changes are consistent with the increase in knowledge of options and side effects. Table [4](#T4){ref-type="table"} shows that the DA was universally found to be clear and helpful. A substantial percentage of men reported that the DA influenced their decision making. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Clarity and usefulness of DA (n = 60) ::: ------------------------------------ -------------------------- ------------------------------- ------------------------- ------------------------------ ------------------------- ------------------ Amount of information Much less than needed 5% A little less than needed 14% About right 77% A little more than needed 4% DK or refused 7% Length Much too short 0% A little too short 11% About right 84% A little too long 5% DK or refused 5% Clarity of words All clear 44% Mostly clear 51% Some clear/ not 5% Most unclear 0% All unclear 0% DK or refused 2% Difficulty of numbers Very easy 64% Somewhat Easy 24% Somewhat difficult 12% Very difficult 0% DK or refused 3% Treatment description balance Complete balance 80% Slanted to surgery 9% Slanted to radiation 4% Slanted to WW 7% DK or refused 8% Recommend DA to a friend? Definitely would 78% Probably would 20% Unsure 0% Probably would not 0% Definitely Would not 2% DK or refused 2% DA improved understanding Definitely did 49% Probably did 42% Unsure 2% Probably did not 2% Definitely did not 5% DK or refused 2% Numbers influenced my decision Definitely did 12% Probably did 41% Unsure 2% Probably did not 22% Definitely did not 22% DK or refused 3% DA helped decision making A lot 14% Quite a bit 30% Moderate 32% A little 16% Not at all 9% DK or refused 5% ------------------------------------ -------------------------- ------------------------------- ------------------------- ------------------------------ ------------------------- ------------------ DK = Don\'t know ::: Eight-six percent of men reported that they shared the DA with a spouse or partner; 22% shared it with other family members; 14% shared the DA with friends. Seventy-two percent reported that they were more likely to take an active role in their treatment decision. Discussion ========== This attempt to marry evidence-based shared decision making and plain language appears to have been informative and helpful in decision making among a diverse set of men. The design process we used was found to translate medical language while retaining quantitative information. The resulting DA was almost universally found to be helpful in decision making. Men in the DA group were more likely to discuss surgery with their physicians and to know that radiation therapy has side effects. Men reported sharing the DA with family and reported that the DA increased the likelihood of taking an active role in decision-making with the physician. The comparisons reported are limited by the use of historical controls and by the unavailability of data about actual decisions made. However, the two cohorts were surveyed using exactly the same questions. Both cohorts were very well informed about their own test results and about side effects of surgery. Side effects of radiation where less well known and somewhat responsive to the decision aid. In qualitative analyses, patients identified several principles of the plain language approach as particularly important to them. Those included \"translations\" of medical language, the emotional difficulties of choices, and the use of what they called attractive layout and illustrations. Some men objected to numbers shown both as a percentage and 100 minus that percentage to rigorously frame each in both positive and negative. However, the provision of numbers to show frequency of side effects of treatments was considered essential. We did not find that patients perceived plain language as \"talking down\". It is important to ask what is gained and what is lost in translating evidence-based TA approaches to plain language. Decision aids have been demonstrated to improve knowledge and assist patients in coming to stable decisions earlier in the treatment process \[[@B31],[@B32]\]. In the face of that success, what improvements in decision aid effectiveness may be expected with the plain language integration and what may be lost? The major deletion we have made from the TA approach is the grading of the medical evidence studies used as source material. We presented rates of outcomes but did not indicate whether they came from randomized clinical trials or descriptive studies. Our reasoning is that grading actually distracts from the simple presentation of rates of mortality and side effects of treatments, and that absorbing what the rates mean is the core of what is needed to make an informed choice. References were made available separately. Health literacy research shows that even highly educated patients who are worried and stressed by a difficult health decision prefer simple, every-day language that is easy to read quickly. No patients asked for references. The credibility of the DA may have been enhanced by their physician\'s offering it to them and by the endorsement of the Michigan Cancer Consortium (MCC). MCC represents major research universities, the State health department, and 75 other major Michigan medical organizations. Conclusion ========== The plain language DA presenting medical evidence in text and numerical formats appears acceptable and useful in decision-making about localized prostate cancer treatment. The gains lie in the potential to improve on earlier gains in patient knowledge. If the actual content is not diluted, but the language, organization and design are more useful for patients, it is possible that mean knowledge scores may improve and that the median and the mean will converge. Parker et al \[[@B9]\] argue that health literacy is central to multiple health system priorities, including quality, cost containment, safety, and patients\' involvement in health care decisions. They suggest that without attention to literacy, the move toward increased patient participation in health care decisions will exacerbate disparities in access and outcomes. This could mean that people with poor health literacy cannot function successfully in an environment designed for active, informed consumers. It should be noted that among our historical controls, there were large and significant deficits in knowledge among African Americans compared with White patients. We did not have sufficient numbers of African Americans in the intervention sample to test the ability of the DA to improve this deficit or to gauge the effect the DA had on willingness to discuss treatment options with physicians. Further testing should evaluate the DA among men with historically low levels or information, as well as among very low literacy men and non-readers. Testing of this DA in a trial against usual care is needed, as is development of the plain language approach in other conditions. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors collaborated on conception and design of the study. MHR was responsible for drafting the article, the design of the decision aid evaluation and drafting the template. DRR provided the original expert draft of the DA. MHR, JTW, and AF contributed early changes. SS and JOF drafted the plain language translation. AF and JTW designed, drafted and performed the survey. KKB analyzed focus group data. RLD performed the statistical analyses. All contributed to editing and revising and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6947/5/16/prepub> Acknowledgements ================ The authors are indebted to Dr. W. Underwood, and M. Velade for their assistance in conducting focus groups. The University of Michigan Comprehensive Cancer Centre, Health Media Research Laboratory (HMRL) performed design and execution of the booklet. Supported in part by the Michigan Public Health Institute, the Michigan Department of Community Health, and the Agency for Healthcare Research and Quality with funding from the Centres for Disease Control and Prevention. The contents of the booklet described in this report do not necessarily represent the official views of the Centres for Disease Control and Prevention.	PubMed Central	2024-06-05T03:55:58.247038	2005-6-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168900/", "journal": "BMC Med Inform Decis Mak. 2005 Jun 20; 5:16", "authors": [ { "first": "Margaret", "last": "Holmes-Rovner" }, { "first": "Sue", "last": "Stableford" }, { "first": "Angela", "last": "Fagerlin" }, { "first": "John T", "last": "Wei" }, { "first": "Rodney L", "last": "Dunn" }, { "first": "Janet", "last": "Ohene-Frempong" }, { "first": "Karen", "last": "Kelly-Blake" }, { "first": "David R", "last": "Rovner" } ] }
PMC1168901	Background ========== The martial arts have their beginnings in the Orient but more specifically the common styles seen in Western society are from Japan, China, and Korea. Taekwondo, which originated in Korea more than 1000 years ago, is more sport than self-defense oriented \[[@B1]\]. In 2000, Taekwondo became recognized as an official sport at the Sydney Olympics. Taekwondo is a full contact free-sparring sport which awards points for head contact. As such, there has been increased interest in injury rates in the sport, especially relating to head injuries \[[@B2]-[@B5]\]. Although much of the research focuses on injury rates, very little examination into pre-competition habits and training has been conducted. The current authors felt that certain key areas needed to be addressed. These included training habits, injuries, dietary practices, and social support. Training habits and injury -------------------------- Most martial arts athletes practice between two to four times per week \[[@B1]\]. However, like in any sport, the frequency and hours of martial arts training can vary widely depending on athletic and competitive level of the individual. Training may be defined as a routine or process undertaken by athletes to further enhance their skill. Specific training may vary among each athlete, but there is usually a general format which is followed. Training classes often begin with a brief warm-up or stretching routine. This may then be followed by kicking drills, self-defense drills, training in patterns (forms) and sparring \[[@B1]\]. Taekwondo athletes have a wide variety of protective equipment available, although its use varies greatly, and concerns have been raised that the equipment often protects the attacker more than the defender \[[@B2]\]. Besides the regular training routine, students also participate in full-contact tournaments. At the time of data collection at the tournament in question, one point was awarded for any strike from the waist upward. The World Taekwondo Federation (WTF) rules and regulations for Olympic competition were used for this tournament \[[@B7]\]. For the purpose of this study, an athlete was considered injured if the following occurred: 1) any circumstance which forced the athlete to leave the competition or training session; 2) any circumstance for which the referee or athlete had to stop the competition; or 3) any circumstance for which the athlete requested medical attention \[[@B6]\]. Injuries occur in both training and competition, and trauma to the lower extremity and head are the most common sites reported \[[@B6]\]. A distinction can be made between overuse and traumatic forms of injury; although in reality they could be considered as points along one continuum. Overuse injuries may occur following continued or accumulated microtrauma to a structure or body area \[[@B8]\]. Traumatic injuries are the result of physical trauma or external force to a certain region leading to a diminished functional ability \[[@B8]\]. In the current pilot study no distinction was made between these forms of injury although it would be an interesting feature to examine in future research. The majority of research about Taekwondo injuries has examined injuries sustained during competition. Even so, it has been noted that up to 60% of injuries are not reported \[[@B2]\]. There are a variety of explanations for this, including poor recall, lack of importance placed on the injury, and unwillingness to disappoint trainers. Weight cycling -------------- Weight cycling is a term used to describe rapid weight loss following self-induced food limitation and/or dehydration. Both gradual (seasonal) and rapid (weekly) weight reduction cycles are used by athletes, and have been investigated for potential effects on nutrition and performance \[[@B9]\]. These cycles are used in various sports such as judo, rowing, wrestling, and boxing in order to make a weight category. Like many of these sports, Taekwondo consists of repeated-effort, high intensity physical demands. In addition to this, Taekwondo competition is structured in a similar fashion to boxing and rowing in that athletes are required to meet weight requirements in order to compete. Although there is no known reported research about weight cycling in Taekwondo, it is the primary author\'s experience (holding a fifth degree black belt in WTF Taekwondo and practicing Taekwondo for more than twenty-five years) that it is widely practiced in the sport. The WTF has various weight classes depending on competition level \[[@B7]\]. Table [1](#T1){ref-type="table"} illustrates the four WTF weight classes per gender for Olympic competition, which were also the categories used at the tournament for the present study. Although the World Taekwondo Federation has eight distinct weight classes per gender for all competitions and championships except for the Olympic Games, no rulings have been implemented to address weight cycling in the sport. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### World Taekwondo Federation Olympic Weight Classes for Men and Women ::: Males Females --------------------- --------------------- Less than 58 kg Less than 49 kg Between 58 to 68 kg Between 49 to 57 kg Between 68 to 80 kg Between 57 to 67 kg Over 80 kg Over 67 kg ::: To date there has been no research investigating the perceived benefit of weight cycling among Taekwondo athletes and this is an area in which much work should be undertaken. Due to the similarities between boxing and Taekwondo, with respect to competition weigh-ins, it may be possible to infer that Taekwondo athlete\'s perceptions of this technique may be similar. One study examining weight cycling among boxers reported that all the subjects felt it necessary to lose weight prior to competition and that it improved their performance \[[@B10]\]. Athletes using this weight control technique may be mistaken in thinking that an advantage will be gained over the opponent competing at his/her natural weight. To this point the research findings into the effect of food and fluid restriction has been equivocal \[[@B10]\]. There is also a belief that nutrients and strength can be restored by eating and drinking in the period between the weigh-in and the competition. Several authors have reported various techniques for rapid weight loss. A few strategies include dieting, restricting food and fluid intake, diuretic use, long runs, skipping, cycling, saunas, and exercising in rubber/plastic suits \[[@B10]-[@B12]\]. Psychological state / support ----------------------------- Despite athletes\' perceptions of the benefits of weight cycling, there are both physiological and psychological side effects. Using the Profile of Mood States-A (POMS-A), anger, confusion, depression, fatigue, tension, and vigor were measured among weight cycling amateur boxers \[[@B10]\]. The study reported that rapid weight loss was associated with significantly higher scores on anger, fatigue, and tension, with decreased vigor. The authors concluded that weight cycling resulted in negative mood and debilitated performance among their respondents. It is important for parents, coaches, and significant others to recognize these signs and address them appropriately. The idea of participating in competitive situations can be daunting for some individuals. Intense pressure, anxiety, and somatic manifestations may result. For those in athletic competition, it is vital to recognize and address these sources of stress in order to produce more successful outcomes. To date there is a lack of research in the areas of weight cycling and its perceived benefits among Taekwondo athletes. There is also limited research in the areas of social support and injury profiles in Taekwondo athletes outside of competition. This pilot study is an initial step towards increasing our knowledge in these areas. The purpose of the present study was to assess training characteristics, competition preparation habits and injury profiles of Taekwondo athletes. Methods ======= Subjects -------- Sixty Canadian male and female respondents were recruited for the study. Participants were Taekwondo athletes competing at a national-level tournament. A total of 28 respondents with an age range of 16 to 29 years returned the distributed questionnaire. Four females and 18 males completed the questionnaire. An additional six participants did not indicate their gender on the returned questionnaires. The mean age of the competitors was just over 22 years, with a mean height and weight of 68.6 inches and 148 pounds, respectively. Instrumentation --------------- A twenty-one-item questionnaire (See [Additional file: 1](#S1){ref-type="supplementary-material"} was used to obtain a general profile of the athletes. Areas of focus included: amount of Taekwondo practice; training satisfaction; protective gear used; pre-competition eating habits; competition preparedness; social support for the sport; and injury profiles. The questionnaire was modelled after one that had been developed for Dragon Boat racers. Neither questionnaire has been tested for validity or reliability. Procedure --------- The first author was working at the national carding tournament as a member of the health care team. As they entered the facility, potential participants were invited to fill out the survey by the current author and his assistants. Only card-carrying athletes competing in the tournament were given questionnaires. Prior to participation, informed consent was obtained by the participants or their guardians. At that time, any questions regarding the study or survey were addressed. Sixty questionnaires were distributed. When participants were given the questionnaire, they were asked to complete it immediately and then return it to the current author or assistants. Statistics ---------- The Statistica Release 6 statistical package was used for all analysis. Descriptive statistics and Pearson\'s chi-square test were used. When inputting data, it was noted that certain variables had missing responses. In these instances, the number of participants who completed the questions was used to calculate the results. Results ======= Training Habits --------------- Training time, measured by number of practices per week, and number of hours per training session, is outlined in Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}. Specific activities during training were also examined. The frequency of sparring was reported. Twenty-five percent (n = 7) of the competitors reported sparring one to two times per week. Over 53% (n = 15) of respondents sparred three to four times per week, and over 21% (n = 6) sparred five or more times per week. Pre- and post-training stretching was also reported. Over 40% (n = 11) of respondents reported only stretching prior to their training sessions, and close to 60% (n = 16) of respondents stretched both before and after training. When examining the use of warm-up and cool-down exercises, over 57% (n = 16) of participants noted always warming up prior to training, while almost 43% (n = 12) reported only warming up occasionally. Only six of 28 respondents (21%) reported they always engaged in post-training cool-down exercises, other than stretching. Over sixty-four and fourteen percent (n = 22) of respondents reported occasionally and never using cool down exercises, respectively. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Training Time -- Number of practices per week (n = 28) ::: Practices Per Week ------------------------- ------------- Number of Practices Percent 2 7.1 3 14.3 4 25 5--6 39.3 7 or more 14.3 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Training Time -- Number of hours per practice (n = 28) ::: Hours Per Practice ------------------------ ------------- Number of Hours Percent 1 17.9 2 53.6 3 17.9 4 7.1 5 or more 3.6 ::: The use of protective gear was also examined. Over 60 percent (n = 17) of respondents reported always using protective gear, while 39.3% (n = 11) only used it occasionally. Table [4](#T4){ref-type="table"} lists the type of gear and its percentage use by the respondents. When examining the frequency of missed practices, five respondents reported never having missed a practice even though they were injured. More athletes missed one to two practices when injured, although there were a few respondents who missed a substantial number of practices regardless of the injury frequency. Less time and practices were missed as injury number increased. Eight respondents reported that they had not experienced any injuries. Four cases had incomplete data and were subsequently not used in the calculation of missed practices. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Type and frequency of protective gear used by Taekwondo athletes (n = 28) ::: Gear Used Elbow Pads Shoes Shin Pads Gloves Head Gear Instep Pads Chest Protector Mouth Guard --------------- ------------ ------- ----------- -------- ----------- ------------- ----------------- ------------- % of Use 57.1 35.7 92.9 3.6 57.1 10.7 78.6 14.3 ::: Table [5](#T5){ref-type="table"} reports the number of missed practices for the first to fifth injuries. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Number of missed practices affected by injury frequency (n = 16) ::: \# of Missed Practices 1^st^Injury (n) 2^nd^Injury (n) 3^rd^Injury (n) 4^th^Injury (n) 5^th^Injury (n) ---------------------------- --------------------- --------------------- --------------------- --------------------- --------------------- None 3 0 1 0 1 1 to 2 5 1 0 1 0 3 1 0 0 0 0 5 to 9 1 1 0 0 0 10 to 14 1 2 0 0 0 20 to 24 2 0 0 0 0 25 to 29 0 0 1 1 0 30 to 39 1 0 0 0 0 40 to 49 1 1 0 0 0 50 to 99 1 0 0 0 0 150+ 0 1 0 0 0 ::: Injury profile -------------- At this level of competition, 75% (n = 21) had six or more years of Taekwondo experience, with over 57% (n = 16) of individuals having eight or more years experience. Of the 28 participants, only 6 (21%) reported never having experienced an injury. At a value of 46.5% (n = 13), the lower limb was reported as being the region most injured in first-time injuries. The upper limb and back had an injury rate of 18% (n = 5) and 10.8% (n = 3), respectively. Over 3% (n = 1) of participants reported experiencing head injuries. Data for participants suffering injuries are reported in Table [6](#T6){ref-type="table"}. Seventy nine percent (n = 22) of athletes reported they had an injury. Of those, 47 percent (n = 13) had a lower limb injury, 18 percent (n = 5) had an upper limb injury, and 11 percent (n = 3) had a back injury. Thirteen athletes (46%) reported they had experienced a second injury. Of those, 29 percent (n = 8) had a lower limb injury, 11 percent (n = 3) had an upper limb injury, and seven percent (n = 2) had a back injury. Injury rates decreased substantially for those who reported experiencing a third injury. Only fourteen percent (n = 4) of the athletes reported suffering from a third injury. Of those, seven percent (n = 2) had a lower limb injury, while upper limb (n = 1) and back injuries (n = 1) each accounted for four percent of the injuries. Four athletes (14%) reported experiencing a fourth injury while 2 athletes (7%) reported they had a fifth injury. Of those, only lower limb injuries were reported. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Injury rates and location of injuries in Taekwondo athletes (n = 24) ::: Number of Injury Lower Limb Injury (%) Upper Limb Injury (%) Back Injury (%) Other Injury (%) ---------------------- --------------------------- --------------------------- --------------------- ---------------------- 1^st^ 13 (46.5) 5 (17.9) 3 (10.8) 1 (3.6) 2^nd^ 8 (28.5) 3 (10.8) 2 (7.2) 0 3^rd^ 2 (7.2) 1 (3.6) 1 (3.6) 0 4^th^ 4 (14.4) 0 0 0 5^th^ 2 (7.2) 0 0 0 ::: The frequency of injuries in training versus competition was also examined. Out of a total 22 responses, 13 respondents reported experiencing their first injury during training, while nine respondents experienced their first injury in competition. Training was most frequently reported as the time of injury, with eight out of thirteen respondents reporting second injuries occurring during training, versus five out of thirteen during competition. No competition injuries were reported for the third to fifth injuries. In order to better understand the injury data, injury rates were calculated using the basic rate formula: (\#injuries / \# athlete-exposures) × 1000 = \# injuries per 1000 athlete-exposures (A-E). Due to problematic data, only 24 of the 28 respondent\'s data were used. The overall rate of injuries was 520/1000 A-E. The injury rate was 354.2/1000 A-E for training, and 166.7/1000 A-E for competition. The injury rate for training per hour was 32.5/1000 A-E/ hour. A variety of care was sought by the athletes following injury. Of those respondents completing the survey, 25% (n = 7) did not seek any form of treatment. Another 10.7% (n = 3) of the athletes were treated by medical doctors, 10.7% (n = 3) were treated by physiotherapist, 10.7% (n = 3) were treated by chiropractors, and 3.6% (n = 1) received acupuncture. A variety of treatment combinations were also reported by 14.4% (n = 4) of the athletes. These combinations included chiropractic care as well as various sources of therapy listed above. Weight cycling -------------- Pre-competition habits are an important factor to examine in all sports. Due to weight classifications in Taekwondo, the athletes are very conscious of their weight. Certain trends were reported by participants in order to achieve the desired weight. Over 53% (n = 15) of participants reported fasting prior to the competition. Of these individuals, 33.3% (n = 4) neither ate nor drank, 50% (n = 6) only drank, and 17% (n = 2) ate but did not drink. Aerobic exercise was another method used by competitors in order to reach the desired weight category. In addition to dieting, 83%, or 10 of the 15 fasting participants reported doing aerobic activity prior to competition. Social support -------------- Support is often key to athletes at higher levels of competition. The current author examined athlete support for the sport by significant others. Seventy-eight percent (n = 22) of athletes reported they had parental support, while 14% (n = 4) reported no parental support was given, and 7% (n = 2) of respondents responded that this support did not apply. Spousal or partner support was reported by 32% (n = 9) of the athletes, while 18% (n = 5) did not receive this support, and 50% (n = 14) of respondents noted that the category did not apply to them. Because the survey was completed prior to competing, participants were asked to record if they felt prepared for the upcoming event. Fifty percent (n = 14) of respondents responded that they were prepared, 39% (n = 11) felt prepared but nervous, and 11% (n = 3) did not feel adequately prepared for the competition. Several comparison analyses were performed using Pearson\'s chi-square test. None of the values were of statistical significance, and thus not reported. The rationale for only reporting frequencies is due to the small sample size of the study, making the use of other analyses like Pearson\'s or Fisher\'s Exact Test erroneous. Discussion ========== The objective of this retrospective investigation was to assess training characteristics, competition preparation habits, and injury profiles of taekwondo athletes. By having the athletes complete a survey, several areas of concern regarding competition preparation and injuries were highlighted. Training and injuries --------------------- When examining the training habits of taekwondo athletes, the current study reviewed several components of performance. Respondents had significant experience in the sport, with over 75% having six or more years of involvement. Training time, measured by number of practices per week and number of hours per practice, was also high. Over 53% of practices were two hours, with over 45% of athletes practicing between two to four times per week. Of importance is the relationship of how training time and competition is affected by injury. Unfortunately, our sample size was too small to have meaningful comparisons. Other authors have reviewed this relationship. In a study by Feehan and Waller \[[@B2]\], competitive performance affected by previous injury was examined. On the day of the competition, 35% of respondents had a current injury affecting performance. Some of these required strapping or support in order to perform. Seventeen percent reported continuing to train/compete against medical advice. Even with these injury rates, the authors noted that fight outcome was not significantly associated with current or previous injuries. One conclusion which might be drawn is that the injuries reported were not severe enough to negatively impact the athletes\' performance. It can also be assumed that many of those with severe injuries would have chosen to withdraw from or not enter the competition until an appropriate level of health was reached. Practice activities among the respondents varied. A large proportion of respondents warmed up prior to kicking drills, while less than 25% cooled down. One possible explanation for warm-up participation may be that it is encompassed within the class. On the other hand, cool downs may be left to the discretion of the athlete once training is finished. In the current study, stretching was considered a separate activity from warm-ups and cool-downs. Over 60% of respondents stretched both before and after training, while just over 40% stretched prior to training only. In future studies, it would be interesting to note if stretching occurred after warm-up, which is a newer trend of thinking in the prevention of muscle injury \[[@B13]\]. Within the questionnaire, it was specified that cool down exercises did not include stretching. By doing this, the authors intended to eliminate the overlap between the stretching and cool down items. Future studies should allow subject to specify the various types of cool down activities used, such as light jogging or light-paced jumping jacks. Future studies should also examine the relationship between injury rates and the use of stretching, warm-ups, and cool downs. Due to a limited sample size the current study was not able to make these comparisons. The final training activity examined in the current study was sparring. This was an integral part of taekwondo training, with over 50% of respondents sparring three to four times per week. The current study attempted to examine the relationship between frequencies of sparring when injured, but there were no statistically significant differences. Birrer \[[@B14]\] reported that most injuries occur during sparring, thus it is an area of training which deserves specific focus. Future studies should focus on both injury type and frequency occurring during sparring, as well as limitations in sparring due to injury. In the current study, training was most frequently reported as the time of injury and relatively few injuries occurred during competition. Even so, the overall reported injury rate was quite high, at 520/1000 A-E. The injury rates calculated for both training and competition are likely to be skewed. Firstly, respondents were asked to simply report if they had been injured during competition or training. A more accurate representation may have occurred if athletes were asked to report how many competitions they had participated in during the previous year and if they had suffered injuries during any of these. Training injury rates may have also been affected by athletes returning to play prior to complete resolution of their problem. This could make athletes more susceptible to subsequent injuries. When reviewing injury location reported in the current study, it was not surprising, that the lower extremity received the most injuries. This was also true for all subsequent injuries reported (up to five per athlete). These results are consistent with those of several other studies \[[@B2],[@B5]\]. The upper limb was the second most frequently injured region, with the head being the least frequently injured. Sprains and strains were the most common injuries, followed by contusions, which is similar to other research \[[@B2]\]. Other reports have listed contusions and concussions as the most common forms of taekwondo-related injuries \[[@B3],[@B6]\]. Zemper et al. \[[@B5]\] reported that contusions were the predominant type of injury in his study of injury rates recorded during the 1988 US Olympic team trials for taekwondo. In general, the number of practices missed decreased with subsequent injuries. Also, there were relatively few athletes who missed a large number of practices. This is perhaps attributable to injury severity. One explanation for fewer missed practices could be that once an athlete experienced one injury, s/he was more likely to increase the use of protective gear, thus avoiding or decreasing future injury severity. Future studies should examine if there is a relationship between increased uses of protective gear following an injury. Recently, there has been increased concern regarding head injuries in taekwondo. Koh and Watkinson \[[@B4]\] reported that when compared to other contact sports, competition Taekwondo had the highest incidence rate of concussions. This might be explained by the fact that athletes are awarded points for head contact. Disturbingly, it was also found that over 30% of concussed athletes suffered more than one significant head blow in the same match. Also, among 99% of the head blows, no evasive manoeuvres were attempted. This would suggest that athletes are poorly trained in blocking skills. Pieter and Zemper \[[@B3]\] also reported that contusions and cerebral concussions were the leading injury types among young male and female Taekwondo athletes. Again, unblocked attacks were a frequent occurrence in these injuries. Widespread safety education on head injuries, and more specifically concussions, is needed among Taekwondo athletes, trainers, and referees. Improved blocking skills and headgear are a priority in order to help avoid serious injury. Following injury, a variety of care was sought by the current study\'s athletes. Interestingly, one quarter of the athletes chose not to seek any form of treatment. This could perhaps be accounted for by the athlete\'s perception of the injury being relatively minor, or being able to manage it without medical advice. Several health professional were consulted by the injured athletes. These included medical doctors, physiotherapist, chiropractors, acupuncturists, and massage therapists. In addition, several athletes consulted multiple therapists. Athletes are generally anxious to return to their pre-injury status, and often become impatient with long-term therapy. This may explain why multiple health professionals were consulted. Also, some health professionals realize the benefit to a multidisciplinary approach, and use a network of referral sources when necessary. Weight cycling -------------- Not surprisingly, more than half of the competitors in the current study dieted prior to competition in order to make their weight class. Although the questionnaire did not specifically define fasting, the subsequent question provided several categories of fasting, such as \"did not eat and drink\", \"did not drink but eat\", and so on. Even with the lack of a clear definition for fasting, fifty percent of the participants reported having completely restricted food intake, while 33% fully restricted food and liquids. Because of the nature of this tournament setting, it was not feasible to weigh athletes prior to their competition. As such, the authors were not able to report actual weight loss occurrence among the athletes. Future studies should focus on intended and actual weight loss among Taekwondo athletes in order to better capture the occurrence of weight cycling in the sport. Rapid weight loss is a common practice among athletes in weight class sports. Hall and Lane \[[@B10]\] reported that their boxing subjects lost an average of 5.16% of their body weight within one week. Along with the weight loss, subjects reported higher anger, fatigue, and tension, as well as reduced vigour. Participants were able to maintain their baseline performance of circuit training when at the reduced weight, although the scores were significantly lower than the athletes expected. It can be postulated that athletes have a misplaced sense of improved strength and performance capabilities when weight cycling for competition. Unfortunately, these views may be reinforced if a weight cycling athlete wins a competition, thus increasing the likelihood of using the strategies in the future. In a study by Alderman et al. \[[@B11]\] examining the prevalence of and weight loss techniques used by high school wrestlers, more successful wrestlers engaged in rapid weight loss (RWL) versus less successful wrestlers. This further reinforces the use of RWL among young competitors. What is particularly striking are the methods used to induce rapid weight loss. Among high school wrestlers, excessive running was used by almost 92% of individuals practicing RWL. Exercising in rubber/plastic suits and using saunas are prohibited in American high school wrestling, but they continued to be used by 40--60% of wrestlers to achieve RWL \[[@B11]\]. Thirty-six percent of respondents in the current study did aerobic exercise in addition to dieting to make their weight, but specific activities were not asked for in the survey. Many short term and long term side effects have been reported with rapid weight loss. Alderman et al. \[[@B11]\] reported multiple symptoms experienced by collegiate weight cycling wrestlers. Over 46% of participants experienced headaches, while over 44% and 42% experienced dizziness and nausea, respectively. Other symptoms included hot flashes, nosebleeds, feverish sensations, disorientation, and increased heart rate. Wenos and Amato \[[@B15]\] reported that college-level wrestlers also experienced an increased perception of effort as muscle strength and endurance decreased with rapid weight loss. Fogelholm et al. \[[@B9]\] studied the effects of gradual versus rapid weight loss in national wrestlers and judo athletes on nutrient intake, micronutrient status, and physical performance (sprint, jump height, and anaerobic performance). A 5% to 6% reduction in body weight was reported in the gradual and rapid loss groups. Nutrient intake was significantly decreased in both groups in B1, B2, K, Ca, Mg, Fe, and Zn values, compared to baseline measures. Speed, vertical jump, and anaerobic performance were not impaired by either rapid or gradual weight loss. Other studies have also reported that despite nutrient depletion, performance of Olympic level amateur boxers during rapid weight loss was not significantly different versus times of normal dietary behavior. These authors concluded that despite reduced carbohydrate intake, there were other sufficient energy sources to meet performance demands \[[@B16]\]. In contrast, Filare et al. \[[@B12]\] reported that all mean micronutrient intakes were below recommended values, while triglyceride levels and free fatty acids were increased in weight cycling judo athletes. Left hand grip values and 30-second jump test output were decreased after seven days of food restriction. By reviewing the literature, some might argue that the evidence of health risks from weight cycling is equivocal. Even so, there are several possibilities that may help explain the lack of supporting data. One possibility is that there may be no effect. Another proposed by Waslen, McCargar, and Taunton \[[@B17]\], is that the duration, frequency, and severity of food restriction among the judo athletes in their study may not have been sufficient to have an effect. Even with a lack of strong support to illustrate the ill effects of weight cycling, monitoring dietary habits of athletes in weight class sports is recommended. It is more prudent to assume that larger weight losses and more frequent dieting could potentially result in negative physiological and performance consequences. Widespread regulations need to be implemented to control weight cycling practices among weight class sports. Athletes need to be educated regarding the negative effects of the practice on both their health and performance. Psychological state/support --------------------------- Support is often key to athletes at higher levels of competition. The current authors examined athlete support by significant others. The majority of athletes reported receiving support from either their parents, or spouse/partner. Unfortunately, the questionnaire used in this study did not delve into the various aspects of psychological state or support. In this pilot study, respondents were simply asked \"Are your parents supportive of your involvement in Taekwondo?\" and \"Is your spouse or significant other supportive of your involvement in Taekwondo?\" It is obvious that neither of these questions addresses the various components involved in support. Future studies need to be more specific in questioning the types and level of support provided to athletes, whether it be emotional, financial, or other various forms of support. As such, these results are of little contributive value. It should be noted that although a large percentage of the athletes felt prepared for the competition, they also reported being nervous. The significance of anxiety and other personality traits in competitive sport has long been studied. It has been reported that winning Taekwondo athletes had lower cognitive and somatic anxiety and higher self-confidence then their losing counterparts \[[@B18]\]. Others found no support for the relationship between competition trait anxiety and Taekwondo performance \[[@B19]\]. Even so, for ultimate personal success, athletes often require a strong support base. This encompasses a sense of understanding, trust, and support from the trainer, and significant others. In weight class sports, the potential effects of weight cycling must also be kept in mind. As noted above, several studies have reported deleterious effects associated with rapid weight loss. These effects may involve one\'s mental status. Filaire et al. \[[@B12]\] reported that confusion, anger, fatigue, and tension were significantly higher after weight loss. Vigor was also significantly lower after food restriction. Thus, when considering the psychological preparedness of an athlete, multiple factors must be measured. There are a few limitations in the present study which need to be addressed. The most obvious methodological issue in this study is that the questionnaire has not been validated. There is very little reported research regarding precompetition habits among Taekwondo athletes. As such, the authors felt it necessary to develop the questionnaire, knowing that there would be issues with its validity. Because this is a pilot study, the results from this study should be used with caution and as a means to enhance future studies in this area. In addition, the small sample size significantly affected statistical analysis. No correlations were significant and thus specific conclusions regarding associations between training behaviors and injuries could not be made. The response rate was low, likely due to the fact the participants were asked to complete the surveys upon entering the tournament building. Athletes may have neglected to complete or return the surveys because of lack of time or feeling that it was not a priority prior to their match. Also, a self-report retrospective survey may be affected by poor recall and perception bias. For example, the recall of more severe and painful injuries would likely be better than that of minor injuries/trauma. The survey was also completed at a competition, thus those who were injured and not participating were already selected out. With respect to the information gained regarding weight cycling, actual weights were not taken. It would have been more informative to weigh the athletes at the mat just prior to their match and compare the result with that of their tournament weigh-in. As mentioned previously, the questionnaire used in this pilot study was vague regarding several concepts. Key definitions were not provided on the questionnaire. Future studies should ensure that all concepts are clearly defined in order to reduce subject confusion and hopefully avoid missing responses or poor response rates. Conclusion ========== The results of this pilot study are primarily descriptive. Even so, they highlight specific training habits and injuries among Taekwondo athletes. Although this pilot study examined a variety of pre-competition habits, it is evident that there are several specific areas which require more in-depth investigation. In order for safety recommendations to be implemented, it is likely that clear relationships will need to be demonstrated. Specifically, the physical effects of weight cycling on performance and improved training to avoid injury need to be examined. Athlete\'s perceptions and belief systems surrounding weight cycling, social support, and injury reporting are all topics in need of further investigation. As such, follow-up research on the relationship of pre-competition habits and injuries in Taekwondo athletes is necessary. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MK analyzed the data, wrote the abstract and various components of the manuscript. HS wrote various components of the manuscript, as well as helped to edit it. YSC helped to develop, distribute and collect the questionnaire, and edit the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2474/6/26/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 There is also an additional file attached titled, \"Figure 1. Questionnaire\". This is the exact Questionnaire which was distributed to the athletes. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We would like to thank Dr. Judith Waalen for reviewing the manuscript and Dr. Cameron McDermaid for his help with the statistical analysis. We would also like to thank the Canadian Memorial Chiropractic College for its support.	PubMed Central	2024-06-05T03:55:58.252395	2005-5-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168901/", "journal": "BMC Musculoskelet Disord. 2005 May 27; 6:26", "authors": [ { "first": "Mohsen", "last": "Kazemi" }, { "first": "Heather", "last": "Shearer" }, { "first": "Young", "last": "Su Choung" } ] }
PMC1168902	Background ========== Pregnancy-induced hypertension (PIH), especially preeclampsia, is a major cause of maternal and perinatal morbidity and mortality worldwide \[[@B1],[@B2]\]. The impact of PIH on birth outcomes has been extensively studied. However, the potential long-term effect of PIH on infants born to PIH mothers has been less studied. PIH has been confirmed to increase significantly the risk of low birth weight by both increasing preterm birth as well as reducing fetal growth. On the other hand, PIH has been found to be associated with an increased rate of high birth weight and large-for-gestational age babies \[[@B3],[@B4]\]. These findings suggest that PIH, more specifically preeclampsia, is a heterogeneous syndrome and that preeclampsia may appear in two forms: restricted fetal growth preeclampsia and normal fetal growth preeclampsia \[[@B4]-[@B7]\]. PIH may have different short and long term effects on infant growth between these possible two types of preeclampsia by intrauterine growth restriction (IUGR) \[[@B5],[@B6]\]. Low birth weight or IUGR babies have been associated with the occurrence of several chronic diseases such as cardiovascular diseases in later life \[[@B8]-[@B10]\]. However, the hypothesis of the fetal origins of adult disease is still the subject of debate \[[@B11],[@B12]\]. One argument is that maternal risk factors in pregnancy (such as PIH) and environmental risk factors in the postpartum period can contribute to this life-long development of the chronic disorders. A change in infant growth of the IUGR baby itself (e.g. catch-up growth) such as in the critical early infant period may also have long-term effects on health later in life, and this change of postpartum growth may be influenced by PIH. Therefore, it is important to study postpartum infant growth patterns of babies born to mothers with PIH, and to determine if there are differences in infant growth between babies with and those without IUGR. However, to date, there have been few studies on the effects of PIH on infant growth. The objective of this study was to examine the effects of the various types of PIH on infant weight gain at 28 and 42 days postpartum. Methods ======= Database -------- We conducted a retrospective cohort study based on a population-based perinatal database from Suzhou, China. The database was previously described by Xiong et al \[[@B3],[@B13]\]. In brief, data used in this analysis were population-based, collected in all 10 hospitals in the city, including16,936 pregnant women from January 1, 1989 through December 31, 1990. We excluded patients with multiple pregnancies (212 cases), chronic cardiovascular disease (36 cases), chronic renal disease or a history of renal disease (303 cases), chronic hypertension or history of hypertension (59 cases), and abnormal blood pressure (diastolic pressure ≥ 90 mmHg or systolic pressure ≥ 130 mmHg) at the first prenatal visit before 21 weeks (850 cases) and abnormal blood pressure after 42 postpartum days (988 cases). There were 2,567 women with no or missing information on infant weight at 28 days or 42 days postpartum, including 105 women who had stillbirths, 138 women who had infant deaths up to 28 days postpartum and 2,324 women who were lost to follow-up. After exclusions, a total of 11,921 pregnancies were retained for the analysis. Definition of pregnancy-induced hypertension -------------------------------------------- PIH was classified as gestational hypertension, preeclampsia, severe preeclampsia or eclampsia. According to the Chinese criteria \[[@B14]\], gestational hypertension was defined as a blood pressure equal to or greater than 130/90 mmHg on more than two occasions greater than six hours apart without proteinuria after 21 weeks of gestation. Preeclampsia was diagnosed as hypertension of equal or greater than 130/90 mmHg but less than 160/110 mmHg with proteinuria of 1+ or 2+ on dipstick in two samples 6 hours apart or greater than 0.3 grams in a 24-hour urine collection. Severe preeclampsia was diagnosed when preeclampsia was complicated by a systolic pressure of ≥ 160 mmHg or diastolic pressure ≥ 110 mmHg and/or if proteinuria was greater than 2+ on dipstick or 5 grams in 24-hour urine collection. Eclampsia was defined as seizure occurred in patients with preeclampsia. Patients with eclampsia were grouped with the severe preeclampsia group for analysis. We restricted our focus to PIH, therefore, pregnancies complicated by chronic hypertension and preeclampsia superimposed on chronic hypertension were not studied. Definition of outcomes and confounding variables ------------------------------------------------ Birth weight was measured in grams at birth. Infant weight was measured at 28 and 42 days postpartum. Gestational age was determined by the obstetricians on the basis of the information on menstrual history, physical examination or early ultrasound examination. IUGR was defined as birth weight below the tenth percentile of expected weight for gestational age \[[@B15]\]. Infant weight at 28 days was measured as the standardised medical procedures during the third postpartum home visit by personnel from the community clinic near the newborn\'s home. Infant weight at 42 days postpartum was measured in hospital during the last day of perinatal care examination while registering for the beginning of the infant care. We developed an index of Infant Growth Percentage (IGP) to measure the infant growth rate. IGP at 28 days is defined as \[(infant weight 28 days - birth weight) / birth weight\] × 100% and so is the IGP at 42 days by \[(infant weight 42 days - birth weight) / birth weight\] × 100%. Infant complications were defined as at least one of respiratory distress, newborn fever, birth injury, jaundice, sclerema, cord infection and intra-cranial bleeding. Other potential confounding variables include maternal age, body mass index \[weight (kg)/height(m)^2^\], infant\'s sex, pre-pregnancy or gestational diabetes mellitus and maternal anemia. Statistical analysis -------------------- Mean birth weight and infant weights at 28 and 42 days postpartum were compared by a variance of analysis between different PIH groups. First, the three groups with PIH disorders were compared to the normotensive referent group. Then each PIH group was further divided into two subgroups according to the existence of IUGR and the same analysis performed using normotensive pregnancies without IUGR as the referent group. Post hoc pair-wise multiple comparisons, which test which specific means differ significantly from others, were performed by Tukey and Bonferroni procedures \[[@B16]\]. In order to assess the independent effect of PIH on infant weight at 28 and 42 days postpartum respectively, multiple linear regression analysis was performed to adjust for confounding variables \[[@B16],[@B17]\]. Infant weight at 28 days and 42 days were the dependent variables. Dummy variables of severe preeclampsia with and without IUGR, preeclampsia with and without IUGR, gestational hypertension with and without IUGR, normotensive status with and without IUGR, and other confounding variables were independent variables. The regression coefficients (β) are estimated by the method of least squares \[[@B16]\]. The statistical significance (p-value) of β was also tested. All statistical analyses were performed with SPSS 10.0 for Windows (SPSS Inc., Chicago, IL). Results ======= Table [1](#T1){ref-type="table"} summarizes the demographic and reproductive characteristics of the women classified in each of the categories of the three PIH groups. Among the11,921 pregnant women, PIH occurred in 1,252 women (10.5 %). Mean maternal age at delivery was 25.8 years, 95.3 % of women were nulliparous. Almost all women gave birth at hospital (99.8 %). Mean (± SD) gestational age at delivery was 39.30 weeks (± 1.55 weeks), with a shorter gestational age for severe preeclampsia. The incidences of preterm birth (\<37 weeks of gestation) and low birth weight (\<2,500 g) were significantly higher among babies born to mothers with preeclampsia and severe preeclampsia than normotensive women. Infant complications occurred in 7 % of babies and were more frequent in preeclampsia and severe preeclampsia groups. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic and reproductive characteristics of study population by pregnancy-induced hypertension, Suzhou, China, 1989--1990 ::: Subjects ------------------------------ ---------- ------ ------- ------- ------- ------- ------------------------------ Maternal age ≤30 11,230 94.2 94.6 90.3 91.5 94.3 0.000 \> 30 691 5.8 5.4 9.7 8.5 5.7 Parity Nulliparous 11,365 95.3 95.3 94.9 97.8 97.1 0.100 Multiparous 556 4.7 4.7 5.1 2.2 2.9 Sex Male 6,026 50.5 50.8 48.6 47.4 55.2 0.290 Female 5,895 49.5 49.2 51.4 52.6 44.8 Preterm birth (\<37 weeks) No 11,527 96.7 96.6 98.2 96.7 93.3 0.023 Yes 394 3.3 3.4 1.8 3.3 6.7 Low birth weight (\<2,500 g) No 11,620 97.5 97.6 97.3 95.1 92.4 0.000 Yes 301 2.5 2.4 2.7 4.9 7.6 Mean gestational age (Weeks) 39.30 39.29 39.44 39.30 38.69 0.000 (Analysis of variance) Place of delivery Hospital 11,900 99.8 99.8 99.7 100 100 0.981 Home 11 0.1 0.1 0.1 0 0 BMI \< 24 11,012 92.8 93.1 90.0 89.6 93.3 0.005 24--28 426 3.6 3.4 5.3 5.8 4.8 \> 28 432 3.6 3.5 4.7 4.6 1.9 Infant complications No Yes No 10,989 93.1 93.1 90.9 87.0 88.5 0.000 Yes 810 6.9 6.4 9.1 13.0 11.5 \* Excluding cases with missing information. ::: Table [2](#T2){ref-type="table"} presents the overall effects of PIH on infant growth at 28 and 42 days postpartum. The infant weights were statistically significantly lower in infants born to mothers with both preeclampsia and severe preeclampsia than infants born to normotensive mothers. However, IGPs were not different between groups. For example, at 42 days postpartum, weights were significantly lower for the preeclampsia and severe preeclampsia groups than for normotensive group (mean weight = 4,763 g and 4,679 g vs. 4,869 g, p \< 0.001), but no evidence of catch-up growth in infants born to mothers with preeclampsia and severe preeclampsia (IGP = 47.7% and 48.3% vs. 49.7%, p = 0.618). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Pregnancy-induced hypertension and infant weight at 28 and 42 days postpartum, univariate analysis ::: Infant growth 28 days postpartum Infant growth 42 days postpartum ------------------------- ------------------- ------- ---------------------------------- ---------------------------------- ------ ------------------- ------- ------ Normotensive (N=10,669) 3252.4 (Referent) 416,1 4088.8 (Referent) 496.2 25.7 4869.1 (Referent) 609.9 49.7 GH\# (N=782) 3288.8 448.9 4084.1 540.4 24.2 4878.6 696.2 48.3 PE¶ (N=365) 3224.9 465.8 4018.6\* 513.7 24.6 4763.8\\ 631.8 47.7 Severe PE (N=105) 3170.9 502.9 3941.9\* 564.9 24.3 4679.9\\ 727.3 47.6 Total (N=11,921) 3253.2 420.9 4085.0 500.7 25.6 4864.8 618.2 49.5 \* p \< 0.05, \\ p \< 0.01 ^1^\[(Infant weight 28 days - birth weight) / birth weight\] × 100 %, ^2^\[(Infant weight 42 days - birth weight) / birth weight\] × 100 % † SD: standard deviation, \#GH: gestational hypertension, ¶PE: preeclampsia. ::: Table [3](#T3){ref-type="table"} presents the effect of PIH on growth at 28 and 42 days postpartum, stratified for the presence or absence of intrauterine growth restriction (IUGR). Infant weights were not different between infants born to mothers with any types of PIH without IUGRand normotensive babies without IUGR, and IGPs were not different between any types of PIH and non-IUGR normotensive control. Conversely, infant weights were markedly lower in all infants with IUGRthan normotensive controls without IUGR. Infant weight gains and IGPs were higher in any types of PIH and normotensive control with IUGRthan in their counterpart groups without IUGR. For example, at 42 days postpartum, in the absence of IUGR, there were no differences in infant weight between any of the types of PIH (ranging from 4,801 g to 4,929 g) and non-IUGR normotensive group (4,905 g). The IGPs were not different between any types of PIH (ranging from 46.2 % to 47.0 %) and non-IUGR normotensive control (48.6%). However, in the presence of IUGR, infant weights were markedly lower in the three types of PIH (ranging from 3,882 g to 4,231 g) and in normotensive (4,255 g) groups than in non-IUGR normotensive control group (4,905 g), p \< 0.001. The IGPs were significantly higher (ranging from 59.8% to 71.3%) as compared to the reference group (48.8%), suggesting a significant catch-up growth. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Pregnancy-induced hypertension and infant weight at 28 and 42 days postpartum by newborn IUGR status, univariate analysis ::: Infant growth 28 days postpartum Infant growth 42 days postpartum ------------------------------------- --------------------------- ---------------------------------- ---------------------------------- ----------------- --------------------------- -------- ----------------- Normotensive / non-IUGR‡ (N=10,067) 3296.6 (382.0) (Referent) 4126.9 (471.3) (Referent) 830.3 25.2 (Referent) 4905.8 (592.9) (Referent) 1609.2 48.8 (Referent) Normotensive / IUGR (N=602) 2513.2 (219.9)\\ 3451.1 (466.9) \\ 937.9 37.3\\ 4255.8 (563.3)\\ 1742.6 69.3\\ GH\# / non-IUGR (N=725) 3353.1 (395.4)\\ 4143.6 (500.4) 790.5 23.6 4929.4 (665.8) 1576.3 47.0 GH / IUGR (N=57) 2469.8 (230.9)\\ 3327.4 (454.8)\\ 857.5 34.7\\ 4231.6 (754.3)\\ 1761.8 71.3\\ PE¶/ non-IUGR (N=326) 3314.9 (401.8) 4091.3 (463.8) 776.4 23.4 4853.1 (572.9) 1537.2 46.4 PE / IUGR (N=39) 2472.1 (215.6)\\ 3410.4 (512.9)\\ 938.3 37.9\\ 4025.4 (624.2)\\ 1553.3 62.8\\ Severe PE / non-IUGR (N=91) 3285.2 (434.8) 4037.4 (518.9) 752.2 22.9 4801.8 (682.8) 1516.6 46.2 Severe PE / IUGR (N=14) 2428.6 (171.8) \\ 3321.4 (459.0) \\ 892.8 36.8\\ 3882.1 (463.5)\\ 1453.6 59.8 Total (N=11,921) 3253.2 (420.9) 4085.0 (500.7) 831.8 25.6 4864.8 (618.2) 1611.6 49.5 \* p \< 0.05, \\ p \< 0.01 ^1^\[(Infant weight 28 days - birth weight) / birth weight\] × 100 %, ^2^\[(Infant weight 42 days - birth weight) / birth weight\] × 100 % \#GH: gestational hypertension, ¶PE: preeclampsia, ‡ IUGR: intrauterine growth restriction. † SD: standard deviation ::: Table [4](#T4){ref-type="table"} presents mean birth weight differences, infant weight differences at 28 days and 42 days postpartum between various types of PIH with and without IUGR and the normotensive control group without IUGR, as well as the corresponding regression coefficients (i.e., the weight differences after adjustment for confounders). The results of multivariate linear regression were consistent with those of univariate analysis (Table [3](#T3){ref-type="table"}). After adjustment for confounding variables, IUGR babies born to any of the PIH groups had significantly lower infant weights at 28 and 42 days postpartum. For example, there was a 903 g (1,023 g before the adjustment for confounders) reduction in infant weight at 42 days postpartum in IUGR babies born to severe preeclamptic mothers relative to non-IUGR babies born to mothers with normal blood pressure, although the IGP was higher in the IUGR babies. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Pregnancy-induced hypertension and infant weight at 28 and 42 days postpartum, multivariate analysis ::: Mean birth Infant weight 28 days postpartum Infant weight 42 days postpartum ------------------------------------- ------------ ---------------------------------- ---------------------------------- ------------------- ------------ ------------------- Normotensive / non-IUGR‡ (N=10,067) Ref. Ref. Ref. Ref. Ref. Ref. Normotensive / IUGR (N=602) 783.4\\ 776.2 (15.4)\\ 675.8\\ 640.0 (19.4)\\ 650.0\\ 619.4 (25.2)\\ GH\# / non-IUGR (N=725) -56.6\\ -34.0 (14.0) -16.7 -4.4 (18.6) -23.7 -12.8 (22.8) GH / IUGR (N=57) 826.7\\ 825.4 (49.0)\\ 799.5\\ 815.1 (61.8)\\ 674.2\\ 678.9 (80.1)\\ PE¶/ non-IUGR (N=326) -18. 4 -6.97 (20.5) 35.6 38.0 (25.8) 53.6 55.4 (33.5) PE / IUGR (N=39) 824.5\\ 795.9 (60.3)\\ 716.5\\ 654.1 (76.0)\\ 880.4\\ 835.2 (98.7)\\ Severe PE / non-IUGR (N=91) 11.4 -20.0 (38.8) 89.5 39.4 (48.9) 104.0 69.9 (63.5) Severe PE / IUGR (N=14) 868.0\\ 796.8 (92.4)\\ 805.5 674.7 (116.5)\\ 1023.6\\ 903.0 (151.2)\\ \* p \< 0.05; \\ p \< 0.01 §β, coefficient, SE, standard error; adjusted for maternal age, body mass index, diabetes, infant\'s sex, maternal anemia, gestational age \#GH: gestational hypertension, ¶PE: preeclampsia, ‡ IUGR: intrauterine growth restriction ::: Discussion ========== To our knowledge, there is no previous study that has examined the effect of PIH on infant growth in the early infant period. Our study found that when babies were born with IUGR, all PIH groups had lower infant weight and significant \'catch-up\' growth at 28 and 42 days postpartum. However, if babies were not born with IUGR, any type of PIH did not have detrimental effect on infant growth at 28 and 42 days postpartum. Our study suggests that PIH per se does not affect early infant growth. The finding of different effects of PIH on early infant growth according to the presence of IUGR further supports the hypothesis that preeclampsia is a heterogeneous disorder and present at least two subtypes: IUGR vs. normal fetal growth preeclampsia \[[@B4]-[@B6]\]. These two subtypes of preeclampsia may have different pathogeneses, clinical manifestations and short and long term effects on both infants and mothers \[[@B4]-[@B7]\]. Patients with preeclampsia who have IUGR babies may have placental dysfunction such as reduced utero-placental perfusion as the common pathogeneses. The present study shows this group of infants has sustained lower weight in the early infant period. In contrast, patients with preeclampsia who have babies with normal fetal growth may not have the significant placental dysfunction, thus have normal intrauterine as well as early infant growth. Newborns with poor intrauterine growth often demonstrate postnatal catch-up growth \[[@B18],[@B19]\]. Our study suggests such a catch-up growth is initiated soon after birth. The impact of catch-up growth on infant\'s long-term health remains unclear. There is a dilemma as to whether this catch-up growth, either inherent to IUGR babies or through growth promotion programs as promoted in many developing countries, is good or bad. In short term, such a catch-up growth enables infants to accelerate growth in order to reach their normal growth curve, resulting in reduced infant morbidity and mortality \[[@B19]\]. However, it has been suggested that catch-up growth could have a detrimental influence, leading to the onset of certain chronic diseases in later life \[[@B20]-[@B23]\]. The crucial time for the development of long-term consequences is the early postnatal period when catch-up growth occurs in around 80 % of IUGR babies \[[@B18],[@B22]\]. Some studies \[[@B8],[@B12],[@B20],[@B22]\] found that hypertension, cardiovascular diseases and mellitus diabetes in adult life were correlated with the existence of a low birth weight and a catch-up growth in the early postnatal period. The hypothesis of the fetal origins of adult disease \[[@B10],[@B24],[@B25]\] may be redefined to consider the impact of the early infant period on the risk of adult diseases \[[@B22],[@B23]\]. Our study indicates that for the infants born to mothers with various types of PIH, only those babies with IUGR showed a catch-up growth. Future long-term follow up studies are needed to examine if IUGR babies born to PIH mothers are at higher risk to develop certain adult diseases later in life than those non-IUGR babies born to PIH mothers. Several limitations of this study have to be addressed. First, our database does not have information on important factors such as maternal smoking and breastfeeding. The prevalence of smoking is high for men in China, however, reported smoking is rare among women (\<0.5%) \[[@B26]\]. Therefore, our results are not likely to be biased by not controlling for smoking. To our knowledge, almost all women breast-fed their infants at that time. However, it is possible that the presence of PIH or IUGR affected breastfeeding and thereby early infant growth. Second, the database has many cases with missing information on weight at 28 days and/or 42 days postpartum. We compared these missing cases with those pregnant women remained for analysis. The rate of preterm birth and low birth weight were higher in missing cases than in the remained study population. This is partly due to women who had perinatal deaths up to 28 days postpartum (most of those babies were preterm, IUGR, or low birth weight) and thus had no information on infant weight at 28 and 42 days postpartum were classified into missing cases. This is the reason why the study population has fewer preterm birth, low birth weight and intrauterine restriction infants. However, there were no significant differences in demographic and reproductive characteristics such as pregnancy-induced hypertension, maternal age, parity, number of prenatal visits, infant\'s sex and gestational age. Third, weight can be a good growth indicator but, ideally, should be correlated to height and head circumference \[[@B27]\], in order to differentiate between symmetrical and non-symmetrical IUGR. We only considered weight as a measure of growth pattern because it was the only available variable. Conclusion ========== The effect of PIH on early infant growth is dependent on whether or not a baby is intrauterine growth restricted. IUGR secondary to PIH is associated with significant catch-up growth in early infant period. Therefore, we speculate that IUGR babies born to mothers with PIH may be at higher risk to develop certain chronic diseases in their later life. More studies in North American or European populations are needed to confirm these results. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Each author has contributed to the concept, the design and the analysis on the study and to the writing of the draft. All authors read and approved the final version. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2393/5/10/prepub> Acknowledgements ================ We thank Dr. Zirong Guo, Department of Epidemiology of Suzhou University Medical College (China) for data preparation.	PubMed Central	2024-06-05T03:55:58.256392	2005-5-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168902/", "journal": "BMC Pregnancy Childbirth. 2005 May 20; 5:10", "authors": [ { "first": "Emmanuelle", "last": "Baulon" }, { "first": "William D", "last": "Fraser" }, { "first": "Bruno", "last": "Piedboeuf" }, { "first": "Pierre", "last": "Buekens" }, { "first": "Xu", "last": "Xiong" } ] }
PMC1168903	Background ========== Asthma continues to be a key challenge in the field of pediatric health. With nine million children diagnosed with asthma in the United States, it is the leading cause of school absences, emergency room visits and hospitalizations among American youth \[[@B1],[@B2]\]. Studies of asthma prevalence however, have tended to neglect the Asian-American population, leaving the burden of asthma faced by this population largely unknown \[[@B3],[@B4]\]. Epidemiological studies of asthma have tended to utilize written questionnaires in English or Spanish, focusing on symptoms \[[@B1],[@B5]\]. In the few studies that have included Asian-American children, lower prevalence of diagnosed asthma than in comparable populations has usually been found \[[@B1],[@B3],[@B4]\]. When included as part of \"other races\" in the Center for Disease Control report, Asian-American children, were found to have an asthma prevalence of 11.4% \[[@B1]\]. Such results come into conflict with results obtained using other, non-written methods. For example a study published in 2002 found 47.1% of Asian-American third-graders in New Jersey to have abnormal spirometry \[[@B6]\]. Recent studies suggest that lack of general asthma knowledge as well as a misunderstanding of the word \"wheeze\" may affect detection of asthma in Asian children \[[@B3],[@B7]\]. A study of parents of asthmatic children in Nanjing, China found 54.7% of parents had \"poor\" knowledge of the disease. Parents were unfamiliar with the triggers of asthmatic attacks, the pathology of wheezing and proper management of the disease in general \[[@B7]\]. In a large international study comparing written questionnaire responses and reaction to a video demonstrating wheezing \[[@B5]\], Cantonese speakers showed less correlation between the written and video questionnaire then did English or Spanish speaking subjects, with the inclusion of the video increasing the number of positive responses for Cantonese speakers. The \"gold standard\" in asthma diagnosis consists of history, physical examination and spirometry. As the history portion relies heavily on questions utilizing the word \"wheeze\" and other symptoms of asthma, language could interfere with proper diagnosis. The word \"wheeze\" has multiple Cantonese translations, each conveying a slightly different meaning. To our knowledge, previous studies of asthma among Chinese speakers have not reported which translation(s) they used. The translations that we have encountered are: 1) ![](1471-2458-5-48-i1.gif) which means a sound from the throat, 2) ![](1471-2458-5-48-i2.gif)(![](1471-2458-5-48-i3.gif)) which means a sound from difficulty breathing, 3) ![](1471-2458-5-48-i4.gif) which means a special sound from asthma and is the more professional expression which is used in medical books, and 4) ![](1471-2458-5-48-i5.gif) which literally means a gasping sound made after crying. A five-question instrument, the Brief Pediatric Asthma Screen (BPAS) has been validated as an additional means of screening for asthma. The BPAS also relies heavily on the use of the word \"wheeze,\" thereby being subject to the same limitation \[[@B8]\]. Two previous studies have used Chinese translations of the BPAS in attempts to determine the prevalence of asthma in Asian-American immigrant children, yielding conflicting results \[[@B3],[@B4]\]. Both studies offered the survey in English and Chinese. In the first study \[[@B4]\], conducted in 2002, parents of kindergarten through fifth grade students completed a self-administered written version of the BPAS. Using, ![](1471-2458-5-48-i4.gif), the professional term for wheezing, this study found the prevalence of diagnosed and possibly undiagnosed asthma to be 16% and 3%, respectively. In the second study, the survey was administered orally, with surveyors physically demonstrating what they believed to be wheezing \[[@B3]\]. It should be noted that the demonstration more resembled labored breathing and therefore was not specific for asthma. As such, the prevalence of diagnosed asthma in the second study remained statistically the same at 15%, however the prevalence of possible undiagnosed asthma increased over six-fold to 18.6%. This study seeks to improve the methodology for asthma screening in Chinese-American immigrant children. By use of the BPAS, a video demonstration of wheezing, and spirometry; the study compares the use of current methods to detect asthma in this population. As there were approximately one-half million hospitalizations for asthma in the US in 1995, effective detection of asthma is critical in patient care \[[@B9]\]. Methods ======= Sample ------ A convenience sample of 152 children ages 5--18, was recruited in the waiting room of the pediatrics department at South Cove Community Health Center in Boston, Massachusetts between June 16-July 23, 2004. Children with cough or fever, who were outside the specified age range, who did not speak either English or Cantonese or who had previously taken the screen were excluded from the study. Procedures ---------- The Tufts-New England Medical Center Institutional Review Board as well as the South Cove Community Health Center Board of Directors approved the study protocol. Consent was given orally at time of entry into the study from parents/grandparents/guardians of children less than 11 years of age and the child him/herself for years thereafter. Collected data was anonymous and de-identified. Families were provided with a written description of the study in their choice of English or Cantonese. The protocol was expanded mid-way through the study to include additional data collection. Data from both the \"basic\" and \"expanded\" protocol were analyzed and are presented in this paper. Questionnaire ------------- A written questionnaire was administered to all participants in the study as our primary tool for assessing asthma status. It was administered to the parent/grandparent/guardian if the child was younger than 11 years of age. Children ages 11--18, were asked to complete the questionnaire him/herself. The survey was written in English and translated into Chinese by one translator using traditional characters. A second translator then translated it back into English in order to check for accuracy. Respondents had the option of taking the survey in their choice of English or Cantonese. A bilingual English/Cantonese speaker (author ACL) orally administered the questionnaire and answered questions when prompted by the participant. The questionnaire consisted of 19 questions. The majority of questions were demographic in nature. Questions were asked regarding the child\'s age, sex, place of birth, and length of residence in the United States. Preferred language was inferred from choice of Chinese or English surveys. For children under 11, this meant that it was the parent\'s preferred language, whereas for children over 11, it was the child\'s preferred language. Information was collected on risk factors for asthma, including family smoking habits, highest level of education for parents (as a measure of socioeconomic status), allergies, asthma medication use and family history of asthma. The expanded questionnaire included questions relating to other medical conditions and non-asthma medication use. The remaining five questions in both the basic and expanded questionnaire were from the Brief Pediatric Asthma Screen (BPAS) \[[@B8]\] and were as follows: 1\. Have you/your child ever been diagnosed by a doctoras having asthma? 2\. Have you/your child ever had episodes of wheezing (whistling in the chest) in the last 12 months? 3\. In the last 12 months, have you had/heard your child wheeze or cough during or after active play? 4\. Other than a cold, in the last 12 months, have you/your child had a dry cough at night? 5\. In the last 12 months, have you/your child been to a doctor, an emergency room or a hospital for wheezing? This study used the translation ![](1471-2458-5-48-i5.gif) for wheeze. Though this version literally means a gasping sound made after crying, it was chosen by a bilingual speaker with extensive experience in the Boston Chinatown community to represent wheezing when used in the context of asthma. Categorization of asthmatic status ---------------------------------- All completed questionnaires were categorized for asthmatic status as described in Wolf et al \[[@B8]\]. An affirmative answer to the first question was automatically categorized as \"diagnosed asthma\". An affirmative answer to the last question or two or more of questions 2--4, was categorized as \"possible undiagnosed asthma.\" All other responses were considered \"probably not asthmatic.\" For analysis of association between asthma and demographic factors, persons categorized as \"possible undiagnosed asthma\" were grouped with \"probably not asthmatic\" to create the non-asthmatic group. The BPAS was administered prior to the respondent\'s viewing of the video and thus, children were categorized based on the respondent\'s prior understanding of the word \"wheeze.\" Video ----- We asked all parents/grandparents/guardians or children to watch the \"international\" version (AVQ 3.0) of the International Studies of Asthma and Allergies in Childhood (ISAAC) Phase One study \[[@B5]\]. Our goal was to provide participants with a visual representation of wheezing on the assumption that they might not understand the concept of wheezing. This brief, one minute, video consists of the following five sequences: 1\. \"A young person seated with clearly audible wheezing, but without breathlessness and no evidence of airway obstruction 2\. Exercise-induced wheezing 3\. Waking at night with wheezing 4\. Nocturnal coughing 5\. Severe attack of asthma\" Respondents were asked if their previous understanding of wheezing on the BPAS corresponded to what they had viewed in the video. An answer of \"yes\" or \"no\" was recorded, as were any qualitative responses in regards to the concept of wheezing. Patients screened with the expanded protocol were then asked to re-answer questions 2--5 of the BPAS, basing their answers on the depiction of wheezing in the video. These patients were than categorized again based on their second set of answers to the BPAS. Detailed analyses of the second BPAS scoring are not presented. Cantonese speaking participants taking the expanded protocol were then asked which of the four Cantonese translations best matched the video\'s depiction of wheezing. Spirometry ---------- Pulmonary function testing was performed on all participants willing to attempt it using the ndd Model 2000 EasyOne™ Frontline Spirometer (ndd Medical Technologies, Andover, MA), an instrument whose performance correlates well with office-based spirometry \[[@B10]\]. We sought to use spirometry as an objective measure of asthma status. Standard, calibrated scales and stadiometers were used to determine height and weight. Each child was then categorized as underweight (\<5^th^percentile), healthy weight (5--85^th^percentile), at risk for overweight (85^th^-95^th^percentile) or overweight (\>95^th^percentile) using the CDC BMI growth charts for boys and girls ages 2--20. Categorization was performed twice, to ensure accuracy. Spirometry was performed using standard procedures by trained technicians. The children\'s noses were sealed manually or by use of pediatric size spirometry nose clips. For analysis, we used only spirometry reported by the EasyOne™ software as having at least 2 acceptable maneuvers and FEV~1~readings within 150 ml and FVC within 150 ml (score of \"A\" or \"B\"; EasyOne™, manual, undated). The parameters recorded were: percent predicted forced expiratory volume during the first second (FEV1), forced vital capacity (FVC), and FEV1 to FVC ratio (FEV1/FVC). Percent predicted values are based upon the results of the NHANES III study as described in Hankinson et. al. without adjusting for ethnicity \[[@B11]\]. Criteria for diagnosis of obstructive lung disease were: FEV~1~\< 80% predicted and FVC \< 80% predicted FEV~1~/FVC \< 75% Numerical values correspond to those recommended by the National Asthma Education and Prevention Program of the National Institutes of Health: National Heart, Lung and Blood Institute \[[@B12]\]. Results of the spirometry were given to the patient and his/her physician so that they could be incorporated into his/her pediatric appointment. Data management & analysis -------------------------- Data was double entered into SPSS version 11.5 and crosschecked for errors by reference to the original hard copies of the surveys and data forms. Chi Square tests were used to generate odds ratios (OR), to demonstrate the likelihood of being diagnosed as asthmatic versus being non-asthmatic for the following variables: birthplace, preferred language of respondent, paternal education, maternal education, smoking in the home, overweight status, allergies and family history of asthma. Due to small numbers of values below 80%, Fisher\'s exact test was used to examine the relationship of being diagnosed with asthma versus being non-asthmatic with FEV~1~and FVC values. We then performed a forward step-wise binomial logistic regression to create a model predicting the likelihood of being diagnosed with asthma based on the following independent variables: preferred language of respondent, foreign born, paternal and maternal education, sex, age, country of origin, smoking in the home, allergy, family history of asthma and overweight as possible predictors of diagnosis of asthma. We again used Chi Square tests to elicit odds ratios for having an understanding of wheezing consistent with the video\'s representation versus inconsistent for the following independent variables: Birthplace, preferred language of respondent, asthmatic status, paternal and maternal education. We performed a forward step-wise binary regression to create a model predictive of consistent or inconsistent understanding of the word wheeze with that portrayed in the video using the following variables: preferred language of respondent, sex, age, overweight, country of origin, foreign born, smoking in the home, allergy, family history of asthma and diagnosis of asthma. For those participants of the expanded protocol we used Mc Nemar\'s test to examine pre-video BPAS scores with post-video BPAS scores. Statistical significance was set at the p = 0.05 level and borderline significance was set at p ≤ 0.10 for all tests. Results ======= Participant demographics ------------------------ Table [1](#T1){ref-type="table"} presents the demographic characteristics of the full study population (n = 152) as well as three subsets, those children completing the basic protocol (n = 63), those completing the expanded protocol (n = 89) and those with acceptable spirometry results (n = 67). Based on BPAS scores, diagnosed asthma prevalence for the full sample was 27.0% with another 5.3% having possible undiagnosed asthma. Children completing the basic and expanded protocols were similar in most ways, however the expanded protocol subset had statistically higher prevalence of family history of asthma (23.9% vs. 11.1%) and lower prevalence of paternal (50.0% vs. 77.1%) and maternal (52.1% vs. 74.5%) education at high school level or above. The subset of children that completed acceptable spirometry did not differ from the total study population demographically. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of the study population and subsets of the population used in the analyses in this study. ::: Basic Protocol (n = 63) Expanded Protocol (n = 89) Total (n = 152) Spirometry (n= 67) --------------------------------- ----------------------------- -------------------------------- --------------------- ------------------------ Age Less than 11 39.7% (25) 37.1% (33) 38.2% (58) 40.3% (27) 11 and older 60.3% (38) 62.9% (56) 61.8% (94) 59.7% (40) Sex Female 50.8% (32) 50.6% (45) 50.7% (77) 58.2% (39) Male 49.2% (31) 49.4% (44) 49.3% (75) 41.8% (28) Birthplace Native-born 57.1% (36) 50.6% (45) 53.3% (81) 59.7% (40) Foreign-born 42.9% (27) 49.4% (44) 46.7% (71) 40.3% (27) Preferred Language Cantonese 61.9% (39) 53.9% (48) 57.2% (87) 55.2% (37) English 38.1% (24) 46.1% (41) 42.8% (65) 44.8% (30) Smoking Smoker in the home 38.1% (24) 35.2% (31) 36.4% (55) 33.3% (22) Smoke-free home 61.9% (39) 64.8% (57) 63.6% (96) 66.7% (44) Allergies Yes 37.1% (23) 37.5% (33) 37.3% (56) 37.9% (25) No 61.9% (39) 62.5% (55) 62.7% (94) 62.1% (41) Family Asthma Yes 11.1% (7)\* 23.9% (21)\* 18.5% (28) 19.4% (13) No 88.9% (56) 76.1% (67) 81.5% (123) 80.6% (54) Paternal Education Did not complete high school. 22.9% (11)\* 50.0% (34)\* 38.8% (45) 39.0% (23) Completed high school or above 77.1% (37) 50.0% (34) 61.2% (71) 61.0% (36) Maternal Education Did not complete high school 25.5% (13)\* 47.8% (33)\* 38.3% (46) 41.4% (24) Completed high school or above 74.5% (38) 52.1% (36) 61.7% (74) 58.6% (34) BPAS Score Diagnosed asthma 22.2% (14) 30.3% (27) 27.0% (41) 26.9% (18) Possible undiagnosed asthma 7.9% (5) 3.4% (3) 5.3% (8) 3.0% (2) Probably not asthmatic 69.8% (44) 66.3% (59) 67.8% (103) 70.1% (47) \* p \<.05 ::: Descriptive analysis -------------------- Table [2](#T2){ref-type="table"} lists frequencies, ORs and p-values for associations between key characteristics of the study population along with prevalence of diagnosis of asthma. Being foreign born as compared to native born was associated statistically with a lower prevalence of diagnosed asthma (11.3% vs. 40.7%, p = 0.001). Reporting a family history of asthma was associated with a higher prevalence of asthma (53.6% vs. 20.3%, p = 0.001). Borderline statistical associations with diagnosis of asthma were found for lower maternal education, which was associated with lower asthma prevalence (17.4% vs. 31.1%, p = 0.09) and the presence of allergies in the child, which was associated with higher prevalence of asthma (35.7% vs. 21.3%, p = 0.05). There was no statistical association for preferred language of respondent (a variable that combines the parent/grandparent/guardian for children less than 11 and the child him/herself for those over 11 years of age), paternal education, smoking in the home, body mass index, or both FEV~1~and FVC below 80% of predicted. We could not test associations between asthma diagnosis and abnormal FEV1/FVC ratio because there were only 3 FEV1/FVC ratios below 75%. Associations between FEV1 and sex, foreign born, preferred language of respondent, smoking in the home, allergies, family history of asthma, BPAS score and paternal and maternal education were not statistically significant (not shown). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of prevalence of diagnosed asthmatics and non-asthmatics for the total sample (n = 152) by demographic categories. ::: Diagnosed Asthmatics (n = 41) Non-asthmatics (n = 111) OR & 95% CI P value --------------------------------- ----------------------------------- ------------------------------ ---------------------- ------------- Birthplace Native-born 40.7% (33) 59.3% (48) 5.41 (2.29 -- 12.80) \<0.001 Foreign-born 11.3% (8) 88.7% (63) Preferred Language Cantonese 24.1% (21) 75.9% (66) 0.72 (0.35 -- 1.47) 0.36 English 30.8% (20) 69.2% (45) Paternal Education Completed high school or above 25.4% (18) 74.6% (53) 0.75 (0.33 -- 1.72) 0.66 Did not complete high school. 31.1% (14) 68.9% (31) Maternal Education Completed high school or above 31.1% (23) 68.9% (51) 2.14 (0.86 -- 5.31) 0.09 Did not complete high school 17.4% (8) 82.6% (38) Smoking Smoker in the home 27.3% (15) 72.7% (40) 1.01 (0.48 -- 2.13) 0.98 Smoke-free home 27.1% (26) 72.9% (70) Body Mass Index (BMI) ≥ 85 percentile 34.1% (15) 65.9% (29) 1.61 (0.75 -- 3.51) 0.22 \<85 percentile 24.2% (24) 75.8% (75) Allergies Allergic 35.7% (20) 64.3% (36) 2.06 (0.98 -- 4.29) 0.05 Non-allergic 21.3% (20) 78.7% (74) Family Asthma Asthma 53.6% (15) 46.4% (13) 4.52 (1.91 -- 10.72) \<0.001 No asthma 20.3% (25) 79.7% (98) FEV~1~and FVC \>80% 25.8% (16) 74.2% (46) 0.89 (0.27 -- 2.95) 1.00\\ ≤80% 40.0% (2) 60.0% (3) \* Odds ratios were computed using Chi-Square tests and are reflective of the likelihood of being diagnosed with asthma for each of the stated demographic variables. \\Fisher\'s exact test. ::: Multivariate analysis --------------------- We considered preferred language of respondent, foreign born, paternal and maternal education, sex, age, country of origin, smoking in the home, allergy, family history of asthma and overweight as possible predictors of diagnosis of asthma in a forward stepwise binary logistic regression. Only being native born and having a family history of asthma were statistically significant in the model (p = 0.002 and p = 0.003 respectively), both being associated with higher prevalence of asthma. Because of substantial missing data on parental and maternal education, we also ran the regression leaving these variables out of the analysis. Doing so did not qualitatively change the result (not shown). Video screening --------------- Participants were asked to indicate whether or not the video portrayal of wheezing was different from their prior understanding (Table [3](#T3){ref-type="table"}). About thirty-six percent (35.6%) indicated that the video was different from their conception of wheeze. Cantonese speakers were more likely to report an inconsistency than were English speakers (47.6% vs. 20.0%, p = 0.001). In a forward stepwise binary logistic regression that included preferred language of respondent, sex, age, overweight, country of origin, foreign born, smoking in the home, allergy, family history of asthma and diagnosis of asthma, Cantonese speakers and overweight children were more likely to say that the video portrayal of wheezing was different from their prior understanding (p \< 0.001 and p = 0.045 respectively). For those completing the expanded protocol, post-video BPAS scores were compared to pre-video BPAS scores using Mc Nemar\'s test, the results were not statistically significant. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Consistency of understanding wheezing between the video and questionnaire for the total sample (n = 149; 3 missing). ::: Consistent Inconsistent OR & 95% CI\* P value --------------------------------- ---------------- ------------------ --------------------- ------------- Total 64.4% (96) 35.6% (53) Birthplace Native-born 68.4% (54) 31.6% (25) 1.44 (0.73 -- 2.82) 0.288 Foreign-born 60.0% (42) 40.0% (28) Preferred Language Cantonese 52.4% (44) 47.6% (40) 0.28 (0.13 -- 0.58) \<0.001 English 80.0% (52) 20.0% (13) Asthma Diagnosis Asthmatic 60.0% (24) 40.0% (16) 0.77 (0.37 -- 1.63) 0.494 Non-asthmatic 66.1% (72) 33.9% (37) Paternal Education Did not complete high school. 58.1% (25) 41.9% (18) 1.33 (0.61 -- 2.88) 0.756 Completed high school or above 64.8% (46) 35.2% (25) Maternal Education Did not complete high school 54.5% (24) 45.5% (20) 1.54 (0.72 -- 3.30) 0.130 Completed high school or above 64.9% (48) 35.1% (26) \*Odds ratios were computed using Chi-Square tests and are reflective of the likelihood of having an understanding of the word wheeze consistent or inconsistent with the video\'s depiction for each of the stated variables. ::: Translation of \"wheeze\" ------------------------- The forty-seven (47) Cantonese-speaking respondents completing the expanded protocol rated four choices of Cantonese translations of the English word \"wheeze\" (Figure [1](#F1){ref-type="fig"}). \"Sound from the throat\" was preferred by 46.8%, \"sound from difficulty breathing\" was preferred by 25.5%, the \"professional term for asthma\" was preferred by 10.6% and \"gasping sound made after crying\" was preferred by 17% of respondents. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Preferred translation of \"wheeze\" for respondents in the expanded protocol (n = 47). ::: ![](1471-2458-5-48-1) ::: Discussion ========== Limitations ----------- There are several limitations to our study. We did not ask about smoking by the children themselves, but limited tests of exhaled CO (not shown) suggested that at least a couple smoked. Our measure for socio-economic status (SES) was imperfect, as we had a high non-response rate for the questions of paternal and maternal education level. In most instances this was due to inability of children to recall their parents\' education level, but interpretation of educational level may also be limited by the fact that educational systems vary between the US and China. Our classification of preferred language by the parent\'s choice rather than the child\'s for children under 11 did not necessarily reflect the preferred language of the child. Our choice of Cantonese wording for \"wheeze\" turned out to be one of the less popular options, despite our having introduced it during the BPAS screenings. It is possible that if we had used the most popular choice that it would have improved our BPAS screening. In addition, qualitative reaction of respondents was that the asthma events depicted in the video were more extreme than what the respondents thought of as wheeze. We would suggest that the video might be best described as depictions of \"asthma attacks\" rather than solely wheezing. Inclusion of spirometry was not an effective tool for assessing asthma status. Possibly the addition of a bronchodilator would have improved its utility. Our results are not necessarily generalizable because we had a convenience sample from a health clinic waiting room and screened only families that spoke English or Cantonese. Interpretation -------------- The population of Chinese-American children that we enrolled in the study reported high prevalence of diagnosed asthma, but appeared to be clinically stable at the time of the screening. By screening out children who were coughing on the day of the screening, we may have underestimated overall asthma prevalence and prevalence of active asthma at screening. Few of their spirometry results were below 75% for FEV~1~/FVC and few were below 80% for FEV~1~and FVC. It is worth noting that abnormal spirometry is relatively uncommon, even among children diagnosed with asthma \[[@B13]\]. Only 41.5% of the children reported to have diagnosed with asthma would have been categorized as asthmatic by the other four questions in the BPAS screening. We cannot distinguish between the possibilities that there could be misdiagnosis of asthma as compared to prevalent mild or well-controlled asthma. For those children with diagnosed asthma, who would have been picked up by the BPAS, only 7.3% reported use of asthma medication on the day of their screening. Because we did not distinguish intermittent from persistent asthma and we asked only about medication use on the day of screening, we cannot state with certainty what percentage are not receiving optimal care. We found that asthmatic children were more likely to have allergies and more likely to have a family history of asthma. Both associations are well supported in the literature \[[@B14]\] suggesting that our study population was similar in this way to other populations. Our strongest finding was that being born in the US was highly predictive of having diagnosed asthma, which suggests that with the Asian-American immigrant population there may be a more vulnerable subpopulation visible only when data is disaggregated. We were unable to determine whether this is a function of true asthma prevalence or differential diagnosis. The BPAS screen showed only 5.3% \"possible undiagnosed asthma\" cases, suggesting that there is little undiagnosed asthma consistent with the study of Lee et al. \[[@B4]\]. However, difficulty in translating of the word \"wheeze\" into Cantonese and its centrality in the BPAS leaves ample room for uncertainty about the true prevalence of undiagnosed asthma. When screening Cantonese-speaking children for asthma we first would suggest using the translation of wheeze that literally means \"a sound from the throat\". However, we would urge caregivers to also try other translations, as there was not a single clear choice among our study population. Second, and related to the first, we would suggest that the concept of wheezing is not well-understood or popularized among Cantonese-speakers. This, combined with a low level of knowledge about asthma in general, \[[@B7]\] means that the provider needs to be careful when taking a medical history and rely less on the obvious symptom, wheezing. Even for providers treating diagnosed asthmatic children from Cantonese speaking families, we would suggest extensive follow-up to ensure proper understanding and compliance. Future studies -------------- There is a need to conduct asthma studies in the Chinese-American population that are free of the limitations of our data collection. This would include asking about children\'s smoking behavior, use of a better surrogate for socio-economic status (possibly parental occupation, which might be more easily recalled by children than education level), use of the more popular term for wheeze (\"sound from the throat\") in written or oral screening, and use of a video demonstration with less severe depictions of wheezing. Additionally, studies are needed to confirm and explain the dramatically different prevalence of asthma between foreign and US born Chinese-Americans that we found. Potential hypotheses for the difference include: differential diagnosis due to language and cultural differences and differences in environmental exposures. Interestingly, exposure to Hepatitis A, a virus more common in China, has been suggested to be protective against the future development of asthma via interaction with the TIM1 gene \[[@B15]\]. If this turns out to be the case, it would suggest a biological cause could explain at least part of the difference in asthma prevalence. Conclusion ========== We hope that this paper raises interest in and prompts examination of asthma among Chinese-American populations. There is a prevailing assumption that asthma is not a problem in this population. If nothing else, our study should bring the validity of such an assumption under scrutiny and raise awareness of the lack of general asthma knowledge in this community. In future studies examining prevalence among Chinese-Americans it is important to consider that prevalence among US born Chinese-Americans appears to be significantly higher than prevalence among foreign-born Chinese-Americans. Failure to desegregate data along these lines may mask the high prevalence of asthma within the sub-group. For both research and clinical practice purposes, the emphasis of cultural understanding needs to be explored. Because of asthma\'s commonplace stature within American culture at this time, it is easy to take one\'s level of understanding for granted, thereby missing a potentially life-threatening condition. While we raise this concern with respect to the Chinese-American population, it worth considering that it may hold true for other population groups as well. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= ROG participated in the design of the study, helped supervise the fieldwork, conducted data management, conducted the statistical analysis and took the lead in writing the manuscript and assisted in editing it. ACL conducted the field data collection, participated in data management, participated in the analysis and assisted with writing of the manuscript. RT assisted with design of the study, helped supervise the fieldwork, contributed to interpretation of findings and assisted with writing and editing the manuscript. DB provided overall direction to the study, supervised data collection, management and analysis, and participated in writing and editing the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/48/prepub> Acknowledgements ================ The authors wish to thank South Cove Community Health Center, Eugene Welch and the doctors, nurses and staff including: Sherrie Zhang, MD., Vivian Tsuei, MD., Ingrid Henar, MD., MPH, Chia-Mei Lu, Irene Chin, Yanty Leung, Chung He, Qi-Long Fun, Shu Lin, and Wendy Wong. We are indebted to the parents and children who participated in the surveys. We thank Weibo Lu for her expertise in translating our questionnaires and Karen Lee for providing oral translations. We thank Lois Doerr, PNP, of Boston Medical Center for instructing the research team in spirometry coaching techniques. We would also like to thank Cato Hiu for his assistance. William Rand, Ph.D., of Tufts University School of Medicine provided helpful advice on analytical approach. Suzanne Steinbach of Boston Medical Center made useful comments on an early draft of the manuscript. This project was funded by Tufts University College of Citizenship and Public Service and Tufts University School of Medicine, Department of Family Medicine and Community Health.	PubMed Central	2024-06-05T03:55:58.258939	2005-5-17	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168903/", "journal": "BMC Public Health. 2005 May 17; 5:48", "authors": [ { "first": "Robyn O", "last": "Greenfield" }, { "first": "Angela C", "last": "Lee" }, { "first": "Roland", "last": "Tang" }, { "first": "Doug", "last": "Brugge" } ] }
PMC1168904	Background ========== Polychlorinated dibenzo-p-dioxins (PCDDs), polychlorinated dibenzofurans (PCDFs), and polychlorinated biphenyls (PCBs) are persistent, lipophilic, halogenated aromatic chemicals that are developmental and male reproductive toxicants in laboratory animals \[[@B1]-[@B3]\]. These persistent chlorinated compounds are worldwide environmental pollutants that have been detected in areas as remote as the Arctic \[[@B4]\]. They are biologically concentrated and stored in human adipose tissue with prolonged half-lives. Current estimates of the half-life for 2,3,7,8-tetrachlorodibenzo-p-dioxin (2,3,7,8 TCDD) range from 3 to 10 years depending on age, gender, and serum concentration, with faster elimination in men, younger individuals, and those with higher peak exposures \[[@B5]\]. The general population is primarily exposed to these compounds through ingestion of contaminated food (fish, meat, milk, and their by-products), as well as through water sources, dermal contact with soil and house dust, and inhalation \[[@B6]\]. Human studies on the relationship of in utero and childhood (peri-pubertal) exposure to dioxin-like compounds (furans and PCBs) with growth and sexual development in boys are limited. A cross-sectional study of Belgian teenagers (15.8--19.6-year-olds) living in polluted and non-polluted communities demonstrated earlier male pubertal development (assessed by Tanner pubic hair and genital staging) in association with both living in polluted areas and higher serum PCBs as compared to boys living in clean areas and/or with lower PCB levels \[[@B7]\]. Lower testicular volume was associated with living in a polluted area but not with serum PCBs. Dioxin exposure assessed with a bioassay (CALUX) was not associated with male pubertal development in this study. Several additional epidemiologic studies have been conducted to investigate growth and development in relation to halogenated aromatic compounds. For example, higher prenatal exposure to DDE has been associated with greater male height and weight adjusted for height at puberty and increased weight for height in peri-pubertal girls \[[@B8]\]. In Taiwan, boys exposed in uteroto PCBs and PCDFs from maternal ingestion of contaminated rice oil had a shorter penile length than unexposed children \[[@B9]\]. Chapaevsk, Russia, is a town of approximately 83,000 residents, located in central Russia (1200 kilometers south-east of Moscow) on the bank of the Chapaevsk river, a tributary to the Volga. The town occupies an area of 187 km^2^, half of which is occupied by industries that are mostly of the military-industrial complex. In 1989, these industries were responsible for the production of the vast majority of the manufactured products from Chapaevsk, and almost half of the city\'s work force was employed there. One of the largest chemical factories in Chapaevsk is the Khimprom Chemical Plant (Middle Volga chemical plant), which before 1949 produced chemical warfare agents (such as lewisite and mustard gas). After 1949, there was a transition to the production of industrial and agricultural chemicals, such as gamma-hexachlorocyclohexane (lindane), and other chlorine-containing products such as liquid chlorine, dichloropropionic acid, methyl chloroform, vinyl chloride, and pentachlorophenol \[[@B10]\]. These processes produced PCDDs and PCDFs as industrial contaminants, which subsequently polluted the air, soil, water and food supply in the city \[[@B10]-[@B12]\]. Revich and coworkers \[[@B10]\] have found elevated levels of dioxin-like compounds in soil less than two kilometers from the Khimprom Chemical Plant (141 ng TEQ/kg). The plant has reduced production of these chemicals since 1991; however, since PCDDs and PCDFs persist in the environment, continued human exposure from contaminated air, soil, drinking water, as well as consumption of locally grown vegetables and locally raised animals remains a concern \[[@B10]-[@B12]\]. Moreover, a large proportion of the population lives in close proximity to the Khimprom complex. Given that children may be particularly susceptible to reproductive and developmental effects of organochlorines, and the paucity of information available regarding childhood exposures to dioxins in particular, we undertook a pilot study to describe the distribution of, and identify potential predictors of exposure to, dioxin-like compounds and dioxins among adolescent boys in Chapaevsk, Russia. The pilot study was also designed to guide the development of a large prospective cohort study on the relationship of exposure to PCDDs, PCDFs, and PCBs with growth and pubertal development in peri-pubertal Chapaevsk boys. Methods ======= Informed Consent ---------------- The study was approved by the Human Studies Institutional Review Boards of the Chapaevsk Medical Association, Harvard School of Public Health, University of Massachusetts Medical School, Brigham and Women\'s Hospital, and Centers for Disease Control and Prevention (CDC). All subjects signed informed consent prior to blood draw and study participation. Study Population ---------------- From 2579 boys, aged 10--16 years in 1999, enrolled in an earlier pilot study to generate growth and maturation curves for boys in Chapaevsk \[[@B13]\], a subset of 246 older boys (14.0 to 16.9 years) were identified for a sub-study in which blood samples and questionnaire information were obtained. Older boys were chosen for study because blood samples were required and participation rates were expected to be higher than among younger children. Of the 246 boys, 221 had blood samples collected, and of these samples, 30 bloods were initially sent to the CDC for chemical analysis of dioxins, furans and PCBs. By design, of the 30 blood samples, 15 were from children with cryptorchidism or hypospadias, and 15 were from children with neither condition (controls). The selection of the 15 cases and 15 controls was done blindly in relation to factors that may predict dioxin levels. Each of the 30 boys, with his mother, was asked to complete a nurse-administered detailed questionnaire on medical history, diet, and lifestyle. The diet questions were used to measure the current and lifetime consumption of locally grown or raised foods. The question was worded, \"Does your child eat any of the following foods from local sources (i.e. your own garden or farms or lakes in the Chapaevsk area)? Yes/No\". There were separate questions for current intake and lifetime intake of each food item. The distances the boys lived from the Khimprom factory at the time of the study and during pregnancy were assessed by questionnaire based on maternal self-report as \<2, 2--6, or \>6 kilometers, and the distance at the time of the study was also estimated using ArcView GIS 3.0 mapping of addresses. Dioxin, furan and PCB chemical analysis --------------------------------------- After blood collection and serum separation, the serum samples were stored at -20°C until shipment on dry ice to the US. The chemical analysis was performed by the laboratory at the National Center for Environmental Health, Centers for Disease Control and Prevention (NCEH, CDC), Atlanta, GA. Target analytes included polychlorinated dibenzo-p-dioxins (PCDDs), polychlorinated dibenzofurans (PCDFs), non-ortho substituted (coplanar) polychlorinated biphenyls (co-PCBs), mono-ortho substituted PCBs and other PCBs (nondioxin-like PCBs). Serum samples were spiked with a mixture of ^13^C~12~- labeled PCDDs/PCDFs and coplanar PCBs as internal standards and the analytes were isolated from serum by a C~18~solid phase extraction (SPE) followed by a multi-column automated cleanup and enrichment procedure \[[@B14]\]. Samples were processed in batches of 10, which included a method blank and two quality control samples that were aliquots from pooled bovine sera spiked with PCDDs, PCDFs and co-PCBs. The analytes were separated on a DB-5 MS capillary column and quantified using selected-ion-monitoring (SIM) high resolution (10,000 resolving power) mass spectrometry (HRGC-ID/HRMS) by the method described by Patterson et al. \[[@B15]\]. Quantification was by isotope dilution mass spectrometry using calibration standards containing ^13^C labeled, and unlabeled analytes. For specific PCDDs, PCDFs, and co-PCBs that lack their own labeled standard, a labeled congener with the same degree of substitution and a similar retention time was used. Mono-ortho and nondioxin-like PCBs were extracted from an aliquot (1 gram) of sample by SPE extraction \[[@B14]\]. Total Lipids were determined for each serum sample using a \"summation\" enzymatic method, which is based on the individual measurements of the four lipid groups (free or non-esterified cholesterol, cholesterol esters, triglycerides and phospholipids) \[[@B16]\]. The limits of detection (LODs) were calculated based on long-term standard deviations that were estimated from multiple measurements of low-level standards \[[@B17]\]. For an analyte with a positive blank, the average blank observed during the study was subtracted from individual calculated results \[[@B18]\]. The lipid adjusted detection limits (DLs) were based on observed LODs and corrected for individual sample weight and recovery. Since values below the DL are expected to have greater variability and uncertainty than values above the DL, some have recommended censoring such data. In the case of our serum analyses, a substantial proportion of our results would be censored at the DL. However, when analyzing group data (as opposed to individual data points), censoring data below the detection limit may lead to the loss of useful information \[[@B19]\]. Where comparisons between uncensored and censored datasets have been performed, the use of DL cut-offs compromised the statistical power of analyses and led to biased exposure estimates \[[@B20]\]. This bias was a greater limitation than any increased variability of low values \[[@B18],[@B21]\]. Thus, in the present study, we retained measurable values below the DL and assigned a value of zero only to samples in which the analyte was not detectable or was at or below the level of contamination in procedural blanks. The concentrations of PCDDs, PCDFs and co-PCBs were reported on a lipid-adjusted basis as picograms of analyte per gram of lipids (ppt). The concentrations of mono-ortho and nondioxin-like PCBs were reported as nanogram of analyte per gram of lipids (ppb). Toxic equivalents (TEQs) were also reported for PCDDs, PCDFs, co-PCBs and other \'dioxin-like\' PCBs based on Toxic Equivalency Factors (TEFs) for individual congeners as specified in the 1998 World Health Organization (WHO) system \[[@B22]\]. Statistical Analysis -------------------- Wilcoxon ranksum tests were used to compare dioxin and PCB levels between controls and boys with cryptorchidism or hypospadias. Generalized linear models were used to identify predictors of the log (base 10) of the sum of dioxin concentrations (PCDDs + PCDFs + co-PCBs) and log sum of dioxin TEQs (PCDDs + PCDFs + Co-PCBs) for the 30 boys. Predictors of dioxin concentrations were first modeled univariately and then adjusted for age. Univariate predictors with p-values below 0.15 were considered for inclusion in the multivariate models. Multivariate model selection was based on adjusted R-square values, Mallows C(p) criterion, and forward and backwards selection procedures; final models used a significance level of 0.05 for inclusion of predictors. Two final models were selected: one which considered all potential covariates, and a second model which considered only those covariates collected on all pilot study participants (i.e., excluding PCB measurements), in order to predict dioxin levels in the larger pilot study cohort. Dietary exposures were coded as \"Yes\" or \"No\" for lifetime consumption of food from local sources (own garden or livestock, friend\'s garden or livestock, or grown/obtained in Chapaevsk) within the following categories: (1) fruits and vegetables; (2) chicken; (3) non-chicken meats which included goats, cows, pigs, and other non-chicken meat; (4) milk and dairy products, including eggs; and (5) fish. Exploratory models included factors considered a priori to have a potential relationship with serum dioxin levels such as the child\'s age, body mass index (weight in kg/height in m^2^), and local food consumption habits; reproductive factors (gestational age at birth of the index son, maternal parity \[number of live births prior to the index son\], gravidity \[number of prior pregnancies\], and breastfeeding history \[cumulative number of weeks of breastfeeding previous children\]), residential information (duration in years that the index boy resided in Chapaevsk and distance that mother lived from the factory during pregnancy with the index son and at the time of the study), and socio-demographic factors including parental education, parental occupation, and monthly household income (1 = 300--1000 rubles \[\$10--\$35 U.S.\], 2 = 1000--1500 rubles \[\$35--\$50 U.S.\], 3 = \>1500 rubles \[\>\$50 U.S.\]). Results ======= Per our selection criteria, of the 30 boys with serum samples analyzed for dioxin-like compounds, fifteen boys had congenital abnormalities (six hypospadias and nine cryptorchidism). No boy had both malformations. The 15 control boys did not have either hypospadias or cryptorchidism. The distances the boys lived from the Khimprom factory at the time of the study, along with other demographic characteristics, are presented in Table [1](#T1){ref-type="table"}. The mean (sd) distance the boys lived from the Khimprom factories was 4.45 (1.76) km. During the pregnancy with her son, of the 29 mothers, 8 (27%) lived less than 2 km, 8 (27%) lived 2--6 km, and14 (45%) lived greater than 6 km from the Khimpron factory. One subject was missing questionnaire data, including distance from Khimprom, medical history, and local food consumption. Only three of the boys\' fathers worked at the Khimprom factory during the year prior to the child\'s birth. While pregnant, two of the boys\' mothers reported alcohol intake and one smoked. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of adolescent boys from Chapaevsk, Russia (N = 30) ::: Mean SD N % ----------------------------------------------------------- ---------- -------- ------- ------- Age (years) 16.1 0.81 Body Mass Index (kg/m^2^) 19.9 3.09 Weeks of Gestation^a^ 39.2 2.37 Breastfeeding duration of index son (weeks)^b^ 36.9 25.2 Prior maternal breastfeeding (weeks)^b^ 18.1 38.3 Birth weight (gm)^c^ 3433 439 Current residential distance to Khimprom factory (km) 4.45 1.76 Time boy resided in Chapaevsk(yrs) 14.0 3.70 Gravidity^d^ 0 14 48 1 8 28 2--3 7 24 Parity^d^ 0 17 59 1 10 34 2--3 2 7 Monthly family income level 300--1000 rubles (\$10--\$35 U.S.) 8 28 1000--1500 rubles (\$35--\$50 U.S.) 7 24 \>1500 rubles (\> \$50 U.S.) 14 48 Maximum parental education level 8--11^th^grade 2 7 Junior College 17 59 University 8 28 ^a^One subject was missing data on weeks of gestation of the index son. ^b^One subject was missing data on breastfeeding. Prior maternal breastfeeding is the cumulative number of weeks of breastfeeding children born prior to the index son. ^c^Two subjects were missing data on birth weight of the index son. ^d^Maternal gravidity is the number of pregnancies prior to the index son and parity is the number of live births prior to the index son. ::: Since the PCDD, PCDF and PCB distributions were skewed, percentile distributions for serum concentrations and WHO-TEQs for 7 PCDD congeners, 9 PCDF congeners, 3 non-ortho (coplanar) PCBs and 6 mono-ortho PCBs are shown in Table [2](#T2){ref-type="table"}. The median (25^th^, 75^th^percentile) concentrations for total PCDDs, PCDFs and coplanar PCBs were 95.8 pg/g lipids (40.9, 144), 33.9 pg/g lipids (20.4, 61.8), and 120 pg/g lipids (77.6, 157), respectively. The median (25^th^, 75^th^percentile) concentration for total mono-ortho PCBs was 47.7 ng/g lipids (39.2, 78.4); as expected more than two orders of magnitude larger than for the other compounds. For WHO-TEQs, the median (25^th^, 75^th^percentile) for total PCDDs, PCDFs, and coplanar PCBs were 0.29 (0.1, 9.14), 7.98 (5.27, 12.3), and 7.39 (4.51, 11.9), respectively. For total mono-ortho PCBs, the median (25^th^, 75^th^percentile) WHO-TEQ was 8.80 (7.16, 15.5). The total WHO-TEQ for dioxins, furans and PCBs had a median (25^th^, 75^th^percentile) of 30.9 (18.4, 46.8). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distribution of concentrations and WHO-TEQs of PCDD/Fs and PCBs in adolescent boys from Chapaevsk, Russia ::: Congener Average DL \# of samples above DL^a^ Concentrations WHO-TEF WHO-TEQ (pg TEQ/g lipid) ------------------------------------------- ---------------- ------------------------------- -------------------- ------------- ------------------------------ --------- ------ ------ ------ *PCDDs(pg/g lipid) 2,3,7,8-TCDD 4.1 2/29 0.00 0.00 0.00 1.000 0.00 0.00 0.00 1,2,3,7,8-PeCDD 4.8 12/30 0.00 0.00 8.30 1.000 0.00 0.00 8.30 1,2,3,6,7,8-HxCDD 7.1 1/30 0.00 0.00 0.00 0.100 0.00 0.00 0.00 1,2,3,7,8,9-HxCDD 6.9 3/30 0.00 0.00 0.00 0.100 0.00 0.00 0.00 1,2,3,4,6,7,8-HpCDD 14.6 15/30 9.40 14.9 25.1 0.010 0.09 0.15 0.25 1,2,3,4,6,7,9-HpCDD 9.0 0/30 0.00 0.20 1.70 0.0000 0.00 0.00 0.00 OCDD 73.1 15/29 32.9 75.0 100 0.0001 0.00 0.01 0.01 PCDFs(pg/g lipid) 2,3,7,8-TCDF 3.9 0/30 0.00 0.00 0.00 0.1000 0.00 0.00 0.00 1,2,3,7,8-PeCDF 4.7 1/30 0.00 0.00 0.00 0.050 0.00 0.00 0.00 2,3,4,7,8-PeCDF 4.5 29/30 8.50 12.6 17.4 0.500 4.25 6.30 8.75 1,2,3,4,7,8-HxCDF 4.2 27/30 5.50 9.15 20.1 0.100 0.55 0.92 2.01 1,2,3,6,7,8-HxCDF 4.5 21/30 0.00 5.70 8.70 0.100 0.00 0.57 0.87 1,2,3,7,8,9-HxCDF 4.2 0/30 0.00 0.00 0.00 0.100 0.00 0.00 0.00 2,3,4,6,7,8-HxCDF 4.5 1/30 0.00 0.00 1.80 0.100 0.00 0.00 0.18 1,2,3,4,6,7,8-HpCDF 16.8 0/30 3.00 5.30 10.5 0.010 0.03 0.05 0.11 1,2,3,4,7,8,9-HpCDF 6.0 0/30 0.00 0.00 0.00 0.010 0.00 0.00 0.00 Coplanar PCBs(pg/g lipid) 3,4,4\',5-TCB 81 11.0 2/30 0.00 0.00 1.60 0.0001 0.00 0.00 0.00 3,3\',4,4\',5-PeCB 126 25.3 30/30 43.3 68.4 116 0.1000 4.33 6.84 11.6 3,3\',4,4\',5,5\'-HxCB 169 7.7 28/30 27.3 44.2 54.3 0.0100 0.27 0.44 0.54 Mono-ortho PCBs(ng/g lipid) 2,3,3\',4,4\'-PeCB (105) 11.0 9/27 5.6 7.1 11.8 0.0001 0.56 0.71 1.18 2,3\',4,4\',5-PeCB (118) 11.0 27/27 20.7 28.0 51.6 0.0001 2.07 2.80 5.16 2,3,3\',4,4\',5-HxCB (156) 11.0 10/27 7.5 9.1 15.9 0.0005 3.75 4.55 7.95 2,3,3\',4,4\',5\'-HxCB (157) 11.0 0/27 2.3 2.6 4.3 0.0005 1.15 1.30 2.15 2,3\',4,4\',5,5\'-HxCB (167) 11.0 0/27 1.6 2.0 3.7 0.00001 0.02 0.03 0.04 2,3,3\',4,4\',5,5\'-HpCB (189) 10.9 0/25 0.4 0.6 0.8 0.0001 0.04 0.06 0.08 Total PCDDs (pg/g lipid)* 40.9 95.8 144 0.10 0.29 9.14 Total PCDFs (pg/g lipid) 20.4 33.9 61.8 5.27 7.98 12.3 Total coplanar PCBs (pg/g lipid) 77.6 120 157 4.51 7.39 11.9 Total PCDD/F/coplanar PCBs (pg/g lipid) 154 273 397 12.1 17.1 38.1 Total Mono-ortho PCBs(ng/g lipid) 39.2 47.7 78.4 7.16 8.80 15.5 Total TEQs 30.9 46.8 Abbreviations: DL, detection limit. Note: 1,2,3,4,7,8-HxCDD, OCDF, and 3,3\',4,4\'-TCB 77 were not reported in the present study due to interferences in method blanks. ^a^Results for some samples could not be reported because one or more of the analytical QA/QC criteria were not met. Therefore, the denominator may be less than 30. ::: Although means and standard deviations do not appropriately describe skewed distributions, to allow for comparisons with other studies we briefly present them. The mean (sd) of total PCDDs, PCDFs, and coplanar PCBs were 104 pg/g lipids (78.0), 63.8 pg/g lipid (89.9) and 123 pg/g lipid (53.1), respectively. The mean of the TEQs for PCDDs, PCDFs, and coplanar PCBs were 5.51, 13.9 and 8.35 pg TEQ/g lipid, respectively. PCDDs, PCDFs, and coplanar PCBs accounted for a mean percent of total serum dioxin concentrations of 34.3% (range 8.4 to 80.1%), 18.0% (range 4.4 to 57.5%), and 47.7% (range 15.5 to 73.9%), respectively. The relatively high concentrations of mono-ortho PCBs contributed significantly to TEQs and were therefore also included in the total TEQs. PCDDs, PCDFs, coplanar PCBs, and mono-ortho PCBs accounted for a mean percent of total TEQs of 11.9% (range 0.14 to 59.0%), 30.4% (range 3.9 to 71.3%), 25% (range 8.1 to 43.0%), and 32.8% (range 0 to 60.1%), respectively. Figure [1](#F1){ref-type="fig"} displays the percent contribution of each PCDD, PCDF and PCB congener to the total TEQ/g lipid for the thirty boys. Based on serum concentrations, the predominant dibenzo-p-dioxin congeners were OCDD and 1,2,3,4,6,7,8-HpCDD (mean percents of 73.8% and 22.2%, respectively), while 1,2,3,7,8-PeCDD and 1,2,3,4,6,7,8-HpCDD were the predominant congeners based on mean TEQs (mean percents of 56.2% and 37.0%, respectively). For PCDFs, 2,3,4,7,8-PeCDF and 1,2,3,4,7,8-HxCDF accounted for the majority of the total PCDF concentrations (means of 35.5% and 32.4%, respectively), and these same two congeners accounted for the majority of the total PCDF TEQs (75.2% and 16.7%, respectively). For coplanar PCBs, 3,3\',4,4\',5-PeCB (\#126) and 3,3\',4,4\',5,5\'-HxCB (\#169) accounted for the majority of the total coplanar PCB concentrations (means of 63.4 and 35.3%, respectively), and 3,3\',4,4\',5-PeCB (\#126) accounted for 94.1% of the total coplanar PCB TEQs. For mono-ortho PCBs, 2,3\',4,4\',5-PeCB (\#118) and 2,3,3\',4,4\',5-HxCB (\#156) accounted for the majority of the total mono-ortho PCB concentrations (means of 56.0% and 19.7%, respectively), while 2,3,3\',4,4\',5-HxCB (\#156) accounted for 48.3% of the total mono-ortho PCB TEQs and 2,3\',4,4\',5-PeCB (\#118) for 30.3%. The concentrations of six prevalent PCB congeners, or so-called indicator PCBs \[[@B23]\], are presented in Table [3](#T3){ref-type="table"}. Here the largest contributors to the sum of the 6 congeners were PCB 153 (34.6%) and PCB138 (20.4%). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Percent contribution of PCDDFs and PCBs to TEQs (pg TEQ/g lipid) among boys from Chapaevsk. ::: ![](1476-069X-4-8-1) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Concentrations (ng/g lipid) of six PCB congeners^a^in blood of adolescent boys from Chapaevsk, Russia ::: PCB congener (IUPAC \#) Average DL \# of Samples above the DL^b^ Median Concentration (ng/g lipid) Mean (SD) Concentration (ng/g lipid) ----------------------------- ---------------- ----------------------------------- --------------------------------------- ------------------------------------------ PCB 28 20.0 1/24 10.6 11.9 (6.33) PCB 74 11.0 14/27 10.8 14.8 (10.8) PCB 99 11.0 27/27 27.7 34.1 (22.5) PCB 138 18.0 21/27 29.6 39.4 (22.2) PCB 153 19.8 27/27 51.2 64.5 (30.7) PCB 180 11.0 23/27 17.7 22.8 (14.9) Sum of 6 congeners 136 168 (99.3) Abbreviations: DL, detection limit ^a^PCB congeners include 6 prevalent or indicator congeners (DIN 51 527, 1987) ^b^Results for some samples could not be reported because one or more of the analytical QA/QC criteria were not met. Therefore, the denominator may be less than 30. ::: Only two of the 30 boys had levels of 2,3,7,8-TCDD greater than the detection limits; levels for these two boys were 17.9 and 21.7 pg/g lipid. Because all samples had different weights (4.1 to 12.1 g), different percent lipids, and percent recoveries, the DLs for TCDD in our study samples ranged from 1.5 to 16.5 pg/g lipid. We were unable to identify unique characteristics of the two boys with detectable levels of 2,3,7,8-TCDD. There were no differences in sum of PCDDs, PCDFS, and PCBs between the group of boys with cryptorchidism or hypospadias and the comparison boys. The smallest difference we could detect however, in the sum of dioxin concentrations given the small size of this study is 175 pg/g lipid (with 80% power) or 200 pg/g lipid (with 90% power). In other words, we can only detect differences that are larger than the observed standard deviation in sum of dioxin-like compounds. Of the 29 boys with lifetime food consumption data, 86% (n = 25) had consumed local meat products (in particular, 83% had consumed non-chicken meat and 52% had consumed local chicken), 83% (n = 24) had consumed local fish, 93% (n = 27) had consumed locally produced dairy products or eggs, and 97% (n = 28) had consumed locally grown fruits and/or vegetables. Using regression models adjusting for age, associations were found between the log sum of dioxin concentrations and several characteristics of the thirty boys (Table [4](#T4){ref-type="table"}). Increasing age, local non-chicken meat and fish consumption, and log of lipid-adjusted PCB congener 118 were associated with increases in the sum of the dioxin concentrations. For instance, a one- year increase in age increased the mean sum of dioxin-like compounds (291 pg/g lipids) by 30% and consumption of local non-chicken meat increased dioxin-like compounds by 75%. Gestational age in weeks was inversely associated with the sum of the dioxin concentrations, multiplicative factor of 0.92. Although the coefficient for the relationship between log sum of dioxin-like compounds and current residential distance from the Khimprom factory was less than 1.0 (which would support a decrease in dioxin levels with increasing distance from the factory), it was not statistically significant. We also explored the relationship between the sum of dioxin-like compounds and categories of GIS distances from Khimprom plants (\<2 km, 2--6 km, \>6 km) and found an inverse association, though it too was not statistically significant. Duration of breast-feeding, distance from Khimprom during pregnancy, maternal parity, and parental occupation were unrelated to sum of dioxin-like compounds or sum of dioxin TEQs. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Individual predictors of log sum of dioxin-like compounds^a^, adjusted for age in years ::: Predictor Estimate p-value Multiplicative factor on dioxin (95% CI) R-Square Value -------------------------------------------------- -------------- ------------- ---------------------------------------------- -------------------- Age of index son (years) 0.26 0.057 1.30 (1.00--1.72) 0.13 BMI of index son (kg/m^2^) 0.03 0.40 1.03 (0.95--1.13) 0.15 Consumption of local non-chicken meat (y/n) 0.56 0.042 1.75 (1.05--2.92) 0.26 Consumption of local chicken (y/n) 0.13 0.54 1.14 (0.75--1.72) 0.14 Consumption of local fish (y/n) 0.48 0.079 1.62 (0.97--2.71) 0.23 Consumption of local eggs (y/n) 0.20 0.36 1.23 (0.80--1.88) 0.16 Consumption of local dairy (y/n) -0.19 0.65 0.82 (0.36--1.89) 0.13 Consumption of local fruits/vegetables (y/n) -0.12 0.84 0.89 (0.28--2.77) 0.13 Current residential distance from Khimprom (km) -0.06 0.37 0.94 (0.82--1.07) 0.15 Years resided in Chapaevsk 0.009 0.75 1.01 (0.95--1.07) 0.13 Monthly family income level^b^ -0.09 0.50 0.92 (0.71--1.18) 0.14 Maximum parental education^b^ 0.21 0.26 1.23 (0.86--1.76) 0.17 Log (PCB 118) (ng/g lipid) 0.64 \<0.001 1.90 (1.37--2.63) 0.46 Congenital abnormality of index son -0.015 0.95 0.99 (0.64--1.51) 0.13 Parity^b,c^ -0.011 0.94 0.99 (0.74--1.32) 0.13 Weeks of gestation -0.08 0.072 0.92 (0.84--1.00) 0.23 Weeks of breast-feeding index son (per 12 weeks) -0.004 0.94 1.00 (0.90--1.10) 0.13 Prior breastfeeding duration^d^(years) 0.09 0.53 1.10 (0.82--1.47) 0.14 Abbreviations: Log, logarithm base 10; BMI, body mass index; PCB 118, polychlorinated biphenyl 118 ^a^Sum of dioxin-like compounds includes PCDDs, PCDFs, and coplanar PCBs. ^b^Each variable was modeled using three ordinal levels. ^c^Parity refers to the number of live births prior to the index son. ^d^Prior maternal breastfeeding refers to the cumulative number of weeks of breastfeeding children born prior to the index son. N = 30 for age, BMI, and distance from Khimprom at time of blood draw; N = 29 for all food consumption and medical history predictors; N = 27 for model with PCB118. ::: Models predicting the log of the sum of dioxin TEQ\'s (Table [5](#T5){ref-type="table"}) were generally similar in magnitude and direction for the predictors shown in Table [4](#T4){ref-type="table"}, but suggested weaker effects of age (multiplicative factor = 1.30, p = 0.14) and local fish consumption (multiplicative factor = 1.43, p = 0.32), and a stronger effect of maximal parental education (multiplicative factor = 1.50, p = 0.086) and log of PCB 118 TEQ (multiplicative factor = 2.61, p \< 0.001). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Predictors of log dioxin TEQs^a^among boys in Chapaevsk, Russia (adjusted for age). ::: Predictor Estimate p-value Multiplicative factor on dioxin (95% CI) R-Square Value ----------------------------------------------------- -------------- ------------- ---------------------------------------------- -------------------- Age (years) 0.26 0.14 1.30 (0.93--1.81) 0.08 BMI (kg/m2) 0.04 0.41 1.04 (0.95--1.13) 0.10 Consumption of local non-chicken meat (y/n) 0.68 0.059 1.97 (1.01--3.87) 0.20 Consumption of local chicken (y/n) 0.06 0.82 1.07 (0.62--1.83) 0.08 Consumption of local Fish (y/n) 0.36 0.32 1.43 (0.71--2.88) 0.11 Consumption of local Eggs (y/n) 0.12 0.68 1.13 (0.64--1.98) 0.09 Consumption of local Dairy (y/n) -0.43 0.44 0.65 (0.22--1.89) 0.10 Consumption of local Fruits/Vegetables (y/n) 0.35 0.65 1.41 (0.32--6.17) 0.09 Current residential distance from Khimprom (km) -0.09 0.32 0.91 (0.77--1.09) 0.12 Years resided in Chapaevsk -0.014 0.71 0.99 (0.92--1.06) 0.09 Monthly family income level^b^ -0.14 0.41 0.87 (0.63--1.21) 0.10 Maximum parental education^b^ 0.41 0.086 1.50 (0.96--2.36) 0.18 log (PCB118 TEQ) 0.96 \<0.001 2.61 (1.81--3.75) 0.57 Sexual delay/congenital abnormality 0.21 0.46 1.23 (0.71--2.13) 0.10 Parity^b,c^ 0.06 0.77 1.06 (0.73--1.54) 0.09 Weeks of gestation -0.11 0.078 0.90 (0.80--1.01) 0.19 Breast-feeding duration of index son (per 12 weeks) -0.002 0.97 1.00 (0.87--1.14) 0.08 Prior breastfeeding duration^d^(years) 0.07 0.72 1.07 (0.73--1.57) 0.08 Abbreviations: Log, logarithm base 10; BMI, body mass index; TEQ, toxic equivalents ^a^Sum of dioxin TEQs include TEQs for PCDDs, PCDFs, coplanar PCBs and other \'dioxin-like\' PCBs ^b^Each variable was modeled using three ordinal levels. ^c^Parity refers to the number of live births prior to the index son. ^d^Prior maternal breastfeeding refers to the cumulative number of weeks of breastfeeding children born prior to the index son. N = 30 for age, BMI, and distance from Khimprom at time of blood draw; N = 29 for all food consumption and medical history predictors; N = 27 for model with PCB118. ::: The best fitting multivariate models for predicting the sum of dioxin concentrations are summarized in Table [6](#T6){ref-type="table"}. Even after accounting for the strong relationship between PCB 118 concentration and the sum of dioxin concentrations on the log scale, consumption of local non-chicken meat showed a significant association with increased dioxin levels. Similarly, the best fitting multivariate model for predicting the sum of dioxin TEQs also included PCB 118 and consumption of local non-chicken meat (R-square = 0.63). The best-fitting model for sum of dioxin concentrations, excluding log PCB 118, included age, weeks of gestation, income level, and consumption of local non-chicken meat and local dairy products, with an R-square of 0.56. The best fitting multivariate model for predicting the sum of dioxin TEQs, excluding log PCB 118 TEQ, included the same five predictors as those shown in Table [6](#T6){ref-type="table"} with similar magnitudes and significance of effects (R-square = 0.56). ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Multivariate models for predicting log sum of dioxin^a^concentrations among adolescent boys in Chapaevsk, Russia ::: Model \# Predictor Estimate p-value R-square value Adjusted R-square value -------------- ------------------------------------------------------- -------------- ------------- -------------------- ----------------------------- 1 Consumption of local non-chicken eat (y/n) 0.59 0.017 0.56 0.52 log (PCB 118) (ng/g lipid) 0.63 \<0.001 2 Age (years) 0.40 \<0.001 0.56 0.46 Weeks of gestation -0.11 0.005 Consumption of local dairy (y/n) -0.76 0.037 Consumption of local non-chicken meat (y/n) 0.90 \<0.001 Monthly family income level (ordinal, low, med, high) -0.27 0.020 Abbreviations: Log, logarithm base 10 ^a^Sum of dioxin-like compounds includes PCDDs, PCDFs, and coplanar PCBs N = 26 for model 1; N = 29 for model 2. Missing data for 3 boys on lipid-adjusted PCB 118, and one boy on questionnaire items (medical history, diet). Model 1: best fit multivariate model including PCB 118 data. Model 2: best fit multivariate model (forward and backward selection) not including PCB 118. ::: Discussion ========== Among adolescent boys in Chapaevsk, Russia, higher serum levels of sum of dioxin-like compounds and sum of dioxin TEQs were positively associated with increased age, consumption of fish, local meats other than chicken, and inversely with weeks of gestation. The age association was found despite a narrow age range of slightly over two years in our study. Although not statistically significant, the distance the boy lived from the Khimprom factory at the time of blood draw was inversely associated with serum levels of sum of dioxin-like compounds and sum of dioxin TEQs. As expected, serum PCBs, specifically PCB 118, were strongly associated with both sum of dioxin-like compounds and sum of dioxin TEQs. There was no association between the distance of the residence from the Khimprom plant during the pregnancy and subsequent serum dioxin levels. One potential explanation for the lack of association may include misclassification of distance since the mother was asked to recall a time period more than 14 years prior to the study. However, we would expect that the mother would be able to recall residential history at the time of the birth of their son. Mother\'s self-reported estimates of current residential distance from Khimprom was generally accurate; twenty-one of twenty-nine mothers correctly categorized their current residential distance from the Khimprom plants based on cross-referencing using GIS mapping. Other explanations include that prenatal exposure 14 or more years prior to the current serum sample is not as strong a predictor as are exposures resulting from present residential location. Although there was no association of case status (cryptorchidism or hypospadias) with dioxin levels, we did not have sufficient power to definitively assess this relationship. Perinatal history (e.g. weeks of breastfeeding) was generally not associated with exposure measures in this population; this may be a function of older age of the children. However, there are limited data on the relationship of perinatal factors with organochlorine exposures in this age group so a null finding is of interest given reports that differences in organochlorine levels among breastfed and non-breastfed are generally no longer discernable by early school age \[[@B24],[@B25]\] and, furthermore, that dietary intake after this age contributes significantly to total dioxin intake \[[@B26]\]. In prior studies, substantial emphasis has been placed on pre- and early postnatal (via breastfeeding) exposures because of particular vulnerability during fetal and early infant development. The exposure risk factors during peri-adolescence, another period of potential developmental vulnerability, has not been studied in-detail, therefore, the identification of exposure risk factors specific to this period will enhance our understanding of this critical period. Although data on levels of PCDD/PCDFs in children is limited, our results suggest that the mean total TEQs among Chapaevsk adolescents were higher than most values previously reported in non-occupationally exposed populations of comparable or even older ages. Figure [2](#F2){ref-type="fig"} shows a comparison of the mean PCDD/PCDFs TEQ levels in Chapaevsk boys with other populations (TEQ from dioxin-like PCBs was not included, since some of these studies did not report them). The mean TEQs of pooled blood samples from 10 year-old German boys in rural and urban settings was 8.2 pg TEQ/g lipid for an urban industrial area, 9.0 pg TEQ/g lipid for an industrial area within a rural setting, and 10.1 pg TEQ/g lipid for a rural area \[[@B27]\]. In comparison, the mean TEQ in the Chapevsk boys was 19.3 pg TEQ/g lipid. With the exception of children described by Wuthe et al. \[[@B27]\], subjects in the other studies in Figure [2](#F2){ref-type="fig"} were significantly older than the Chapaevsk boys. Despite age differences, the mean TEQ in Chapaevsk boys was comparable to or even higher than the mean TEQ in older populations from other countries. For example, they were higher than mean TEQs of 18.4 pg TEQ/g lipid in adults (40.6 years old average) from South Germany \[[@B27]\] or 16.4 pg TEQ/g lipid from 20 year old Japanese women \[[@B28]\] or pooled samples from randomly selected males and females 18--69 years of age from the Spanish city of Mataro \[[@B29]\], which were 12.5 and 14.7 pg TEQ/g lipid respectively. The mean levels in adult female (mean age 41 years) non-factory workers in the Russian city of Shelekhovo were also lower at 14.5 pg TEQ/g lipid \[[@B30]\]. Mean TEQs for the general population (mean age 44.2 years) in Germany, collected in 1997--98 \[[@B31]\], and median TEQs for long-term workers of pulp and paper mill and non-workers in the U.S. in 1996 \[[@B32]\] were similar to levels found in the Chapaevsk boys. The mean levels in the study in Germany were 20.71 pg TEQ/g lipid and in the U.S. the median levels were 19.1 pg TEQ/g lipid for community residents and 21.2 pg TEQ/g lipid for low exposure workers. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Mean PCDD/PCDFs TEQ levels in Chapaevsk boys in comparison with other populations. ::: ![](1476-069X-4-8-2) ::: Although TCDD was largely below the detection limits in this small pilot sample, the two boys with detectable values had high TCDD levels (17.9 and 21.7 pg/g lipid), suggesting that exposure for at least some portion of this population is substantially higher than typical of this age group. In comparison, in a cohort of adult (mean age of 58 years) fishermen from a polluted region of Finland, the mean TCDD concentration was 19 pg/g lipid \[[@B33]\]. In adult (mean age of 53 years) residents from Calcasieu Parish, Louisiana, which is near a chemical industrial complex, the mean TCDD level was 7.6 pg/g lipid \[[@B34]\]. Not only were the dioxin levels in these two children higher than those found in these studies, but the adults in the previous studies were several decades older and therefore would be expected to have higher dioxin body burdens than younger children \[[@B35]\]. Potential explanations for the large number of non-detectable samples for TCDD include the small sample volume and young age of the subjects. In our future studies in this population, we will collect larger volumes of serum for dioxin analysis. In one of the few studies on dioxin-like compounds in which children were included, Eskenazi and coworkers \[[@B36]\] evaluated the relationship between serum TCDD concentrations and age at exposure of female residents of Seveso, Italy. Residents near the ICMESA chemical plant in Seveso were exposed to some of the highest known residential levels of 2,3,7,8-TCDD as a result of an explosion at the plant. Archived serum collected near the time of the accident was used to measure exposures. Residents closest to the plant had a median 2,3,7,8-TCDD level of 272 ppt (IQR 92 -- 883 ppt). Residential proximity to the plant and younger age (up to 13 years old) were the strongest predictors of an individual\'s serum 2,3,7,8-TCDD level. Other predictors included being outdoors at the time of explosion and consumption of homegrown food. The higher levels found in children were most likely a result of increased exposure as a result of activity patterns and a greater proportionate consumption of food, water and air than adults \[[@B37]\]. Although the exposure scenario (an acute high exposure event) is different than the chronic low/moderate exposure occurring in Chapaevsk, the results from Seveso suggest that children may be at increased risk for high dioxin exposure from environmental contamination. In our study, although distance from the Khimprom plants was a weak predictor of serum dioxin-like compounds, consumption of local foods (specifically meat and fish), as in the Seveso study, was a strong predictor of sum of dioxin-like compounds and dioxin TEQs. This finding is notable given concerns regarding environmental dioxin contamination in the community and suggests that food may be one of the more, if not most, important routes of environmental contaminant exposure for residents in this setting. In other settings, contaminated food generally contributes much more substantially to human organochlorine burden than air or soil (which may be related to residential proximity to pollutant sources) \[[@B6]\]. We will investigate this issue in more detail in our ongoing study. Conclusion ========== Recently, data were published on dioxin levels among adult residents of Chapaevsk, Russia \[[@B38]\]. Twenty-four self-selected volunteers (12 men and 12 women) provided blood samples in 1998 for analysis of dioxin-like compounds. These serum samples were analyzed by the same laboratory at the CDC, which analyzed the serum samples for our study. None of the adult subjects worked at the Khimprom plant in Chapaevsk. The mean age of subjects was 44 years for men and 45 years for women. Among all subjects, the mean TEQ of total dioxins was 61.2 pg TEQ/g lipid (range 16.4 to 168.1 pg TEQ/g lipid). The adults had the same congeners as major contributors to TEQ of dioxin, furan and co-planar PCBs as the boys in our study. We could not compare the levels of mono-ortho-PCBs in the children to levels in the adult Chapevsk residents since they were not reported. In the adults in Chapaevsk, there were positive relationships of TEQs with age and BMI and an inverse relationship of TEQs with residential distance from the plant \[[@B38]\]. In both this adult and our pediatric Chapaevsk population, PCB congeners contributed significantly to TEQs and therefore it is important to assess these congeners vis-à-vis toxicological implications mediated through dioxin-like mechanisms. We have recently begun recruiting eight- and nine-year old boys and their families in Chapaevsk into a prospective cohort study on the relationship between dioxin exposure and somatic growth and pubertal development. Results of this exposure assessment study demonstrate relatively high TEQ exposures among Chapaevsk children compared to similar aged populations elsewhere and, as evidenced by the importance of local food consumption in determining this exposure, corroborate our hypothesis that local environmental contamination likely contributes to exposure risk in this setting. Our planned additional studies of the potential health consequences of Chapaevsk children\'s exposure is an important sequel to having characterized organochlorine exposure and exposure risk among this potentially vulnerable age group. List of Abbreviations ===================== IUPAC International Union of Pure and Applied Chemistry mono-orthoPCBs Includes PCB congener numbers 105, 114, 118, 123, 156, 157, 167, 189 non-orthoPCBs Includes PCB congener numbers 77, 81, 126, 169 OCDD Octachlorodibenzo-p-dioxin OCDF Octachlorodibenzo-furan PCBs Polychlorinated biphenyls PCDDs Polychlorinated dibenzo-p-dioxins PCDFs Polychlorinated dibenzofurans pg/g Picogram per gram, 10^-12^g. TEF Toxic equivalency factors TEQ Toxic Equivalents (WHO~98~) I-TEQ International Toxic Equivalents (NATO-CCMS, 1988) WHO World Health Organization Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= RH conceived the design of the study and took primary responsibility for drafting the manuscript. PW performed the statistical analysis, participated in the design of the study and the drafting of the manuscript. LA participated in the drafting of the manuscript and the interpretation of the dioxin results. SK participated in the design of the study and the drafting of the manuscript. LP participated in the statistical analysis. DGP carried out the dioxin analyses. WET carried out the dioxin analyses. MML participated in the design of the study and the drafting of the manuscript. BR participated in the design of the study. OS oversaw subject recruitment and data collection, participated in the design of the study and the drafting of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was funded by the US EPA, Grant number R-82943701-0 and NIEHS, Grant Number ES00002.	PubMed Central	2024-06-05T03:55:58.262895	2005-5-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168904/", "journal": "Environ Health. 2005 May 26; 4:8", "authors": [ { "first": "Russ", "last": "Hauser" }, { "first": "Paige", "last": "Williams" }, { "first": "Larisa", "last": "Altshul" }, { "first": "Susan", "last": "Korrick" }, { "first": "Lynne", "last": "Peeples" }, { "first": "Donald G", "last": "Patterson" }, { "first": "Wayman E", "last": "Turner" }, { "first": "Mary M", "last": "Lee" }, { "first": "Boris", "last": "Revich" }, { "first": "Oleg", "last": "Sergeyev" } ] }
PMC1168905	Introduction ============ Obesity is of major primary care concern and is targeted as an international health objective in Healthy 2000, which seeks to reduce the prevalence of obesity to less than 20%. In the last 10 years, the number of overweight people has increased from 26 to 34% \[[@B1]\]. Conventional dietary and behavioral treatment have failed in long-term management. Dietary strategies used to manage obesity include the use of high fibre, low carbohydrates and fats diet \[[@B2],[@B3]\]. Beneficial effect of dietary fibre in the management of obesity is not well established, since their mechanism of action is not known. The discovery of new medicinal plants has led to the creation of potential drugs that modify feeding behavior and metabolism and may therefore have application in weight management. The aim of the present study is to investigate the effect of Irvingia gabonensisextract on body weight. Subjects and methods ==================== A total of 40 obese subjects aged between 19 and 55 years were selected from a group responding to a radio advertisement. After physical examination and laboratory screening tests, diabetics, pregnant and lactating women were excluded. None of these subjects took any weight reducing medication and none was following any specific diet. The purpose, nature and potential risks of the study were explained to all patients and a written informed consent was obtained before their participation. The local research ethics committee approved the experimental protocol. Study design ------------ The study was as a randomised, double blind placebo-controlled crossover design, and consisted of a 4-week treatment period. Subjects were given two different types of capsules containing 350 mg of Irvingia gabonensisseed extract (active formulation) or oat bran (placebo). Three capsules were taken three times daily, one-half hour before meals (a total daily amount of 3.15 g of Irvingia gabonensisseed extract) with a glass of warm water. Capsules were identical in shape, colour and appearance, with neither patients nor researchers knowing what capsule they received. During the experimental period, subjects were examined weekly, with their body weight, body fat, waist and hip circumferences recorded each time. Subjective findings such as increased or decreased appetite, feeling of lightness and gastrointestinal pains were individually noted. Side effects of the active extract, if any were also solicited and noted during each visit. The subjects were also interviewed about their physical activity and food intake during the trial, and were instructed to eat a low fat diet (1800 Kcal) as well as keep a record for seven consecutive days (using household measurements). Anthropometric measurements --------------------------- Anthropometric measurements were done at each visit, with body weight and body fat (impedance measurement using a TANITA™ monitor Scale) measurements on fasted (12 hour) subjects wearing light clothing. Waist and hip circumferences were measured by soft non-stretchable plastic tape on the narrowest and the widest parts of the trunk. Laboratory methods ------------------ Blood samples were collected after a 12 h overnight fast into heparinized tubes at the beginning of the study, after two weeks and at the end (4 weeks) of treatment. The concentrations of total cholesterol, triacylglycerol, HDL-cholesterol, and glucose, in plasma were measured using a commercial diagnostic kit (Cholesterol infinity, triglycerides Int, EZ HDL™ cholesterol, EZ LDL™ cholesterol, Glucose Trinder, respectively) from SIGMA Diagnostics Statistical Analysis -------------------- Results are expressed as mean ± SEM except for anthropometric measurements. Paired Student\'s t-test was carried out on the start and end values of placebo and Irvingia gabonensiscapsules and also on the differences between the placebo and Irvingia gabonensiscrude extract. Results ======= Effect on body composition -------------------------- Irvingia gabonensisinduced a decrease in weight of 2.91 ± 1.48% (p \< 0.0001) after two weeks and 5.6 ± 2.7% (p \< 0,0001) after one month. Although the percentage of body fat was not significantly reduced with both placebo and IG, the waist circumference (5.07 ± 3.18%; p \< 0.0001) and hip circumference (3.42 ± 2.12%; p \< 0,0001) were significantly reduced by IG. A reduction of 1.32 ± 0.41% (p \< 0.02) and 2.23 ± 1.05% (p \< 0.05) was observed with the placebo after two and four weeks respectively of treatment. Effect on blood pressure ------------------------ From the second week of experimentation, the systolic blood pressure was significantly reduced by the active extract (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Effect Irvingia gabonensis on systolic (SBP) and diastolic (DBP) blood pressure ::: Treatment period (weeks) ------------ -------------------------- ---------------- ------------------ ------------------ SBP (mmHg) Active 136.41 ± 19.57 132.66 ± 18.48\* 132.83 ± 17.97\* Placebo 134 ± 5.05 121.5 ± 5.89 123.83 ± 2.92 DBP (mmHg) Active 98.5 ± 19.52 97.5 ± 22.80 94.08 ± 11.07 placebo 93.50 ± 10.31 93.83 ± 7.41 91.5 ± 6.53 Values are means ± sem. Significant differences were at \P \< 0.001 by comparison to the placebo group. ::: Effect on blood lipids components --------------------------------- The plasma total cholesterol cencentration was reduced by 39.21%, triglyerides by 44.9% (p \< 0,05) and LDL by 45.58%. This was accompanied by a significant increase in HDL-cholesterol of 46.852%. The CT/HDL ratio (p \< 0.05) and the blood glucose level were also reduced (32.36%; p \< 0.05). No significant change was observed in the placebo group. Discussion ========== The soluble fibre of the seed of Irvingia gabonensislike other forms of water-soluble dietary fibres, are \"bulk-forming\" laxatives. Irvingia gabonensisseeds delay stomach emptying, leading to a more gradual absorption of dietary sugar. This effect can reduce the elevation of blood sugar levels that is typical after a meal \[[@B4]\]. Controlled studies have found that after-meal blood sugar levels are lower in people with diabetes given glucomannan in their food \[[@B5]\] and overall diabetic control is improved with soluble fibre-enriched diets according to preliminary \[[@B6]\] and controlled \[[@B7],[@B8]\] trials. One double-blind study reported that glucomannan (8--13 grams per day) stabilized blood sugar levels in people with the insulin resistance syndrome \[[@B9]\]. Like other soluble fibers, Irvingia gabonensisseed fibre can bind to bile acids in the gut and carry them out of the body in the faeces, which requires the body to convert more cholesterol into bile acids \[[@B10]\]. This can result in the lowering of blood cholesterol as well as other blood lipids. Controlled double-blind \[[@B11],[@B12]\] studies have shown that supplementation with several grams per day of soluble fibre significantly reduced total blood cholesterol, LDL cholesterol, and triglycerides and in some cases raised HDL cholesterol, these being comparable with effects noticed with Irvingia gabonensis.* Considering the wide use of Irvingia gabonensisin the preparation of various dishes in Cameroon, its use should be further encouraged for the purposes of control of dietary lipids as well as for weight reduction. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Effect of Irvingia gabonensiscrude extract on body weight, body fat, waist and hip circumferences ::: Treatment period (weeks) -------------- -------------------------- ---------------- ---------------- ---------------- Weight (kg Active 105.10 ± 16.98 102.3 ± 17.06 101.01 ± 16.63 Placebo 79.43 ± 9.83 79.43 ± 9.83 79.33 ± 10.63 Active 46.11 ± 4.4848 46.5 ± 3.68 45.34 ± 3.52 Body fat (%) Placebo 40.58 ± 3.49 40.58 ± 3.9 40.3 ± 3.8 Active 112.76 ± 20.5 109.7 ± 20.4 106.6 ± 20.79 Waist (cm) Placebo 81.1 ± 7.1 81,91 ± 7,91 81.25 ± 7.52 Active 125.69 ± 11.34 122.92 ± 10.67 121.15 ± 10.39 Hip (cm) Placebo 122.2 ± 10.7 122.2 ± 10.7 121.5 ± 10.9 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The effect of Irvingia gabonensis on blood total cholesterol (TC), triglyceride (TRI), high density lipoprotein cholesterol (HDL-c), low density lipoprotein cholesterol (LDL-c) and glucose. ::: T-cholesterol TRI HDL-c LDL-c LDL/HDL T-cho/HDL GLUCOSE --------- --------- ---------------- ---------------- --------------- ---------------- -------------- ------------- ------------- Active Initial 215 ± 55.12 162 ± 33.15 61.23 ± 20.36 121.37 ± 36.3 1.98 ± 1.78 3.51 ± 2.70 3.8 ± 1.92 Final 130.68 ± 39.5 89.22 ± 55.63 89.9 ± 28.44 66.08 ± 34.27 0.735 ± 1.20 1.45 ± 1.38 2.57 ± 1.03 Placebo Initial 163.70 ± 25.32 130.65 ± 37.82 31.38 ± 25.21 105.06 ± 11.86 5.05 ± 3.94 6.44 ± 3.37 3.6 ± 0.41 Final 158.36 ± 30.46 100.52 ± 32.55 41.20 ± 19.53 98.55 ± 27.99 3.19 ± 1.85 4.51 ± 2.07 3.9 ± 0.74 :::	PubMed Central	2024-06-05T03:55:58.269865	2005-5-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168905/", "journal": "Lipids Health Dis. 2005 May 25; 4:12", "authors": [ { "first": "Judith L", "last": "Ngondi" }, { "first": "Julius E", "last": "Oben" }, { "first": "Samuel R", "last": "Minka" } ] }
PMC1168906	Introduction ============ The use of biomarkers in medicine lies in their ability to detect disease and support diagnostic and therapeutic decisions. New research and novel understanding of the molecular basis of the disease reveals an abundance of exciting new biomarkers who present a promise for use in the everyday clinical practice. The initial evaluation of a serum biomarker concerns its expression in patients with the disease and in normal individuals in order to define sensitivity and specificity. The sensitivity of a test is defined as the proportion of patients with disease having a positive test whereas the specificity is the proportion of patients without the disease who have a negative or normal test. Consequently the serum level of an ideal marker should: 1) increase pathologically in the presence of the disease (high sensitivity), 2) not increase in the absence of the disease (high specificity), 3) relate to the disease burden and extent, 4) change in accordance with the clinical evolution, reflecting the current status of disease, or better 5) anticipate clinical changes, i.e. indicating the presence of relapse before it becomes obvious at a clinical level and finally 6) possess constant serum levels (no major fluctuation) \[[@B1]\]. Additionally, a clinically suitable biomarker should fulfil the following requirements \[[@B2]\]: 1\. add independent information about the risk or prognosis 2\. account for a large proportion of the risk associated with a given disease or condition 3\. be reproducible (as determined by the low coefficient of variation) 4\. be sensitive, specific and should present with a high predictive value 5\. be of easy and cheap determination Very few markers present a threshold at which the risk suddenly rises. The interplay between sensitivity and specificity and the nature of the disease under prediction assigns suitable cut-off points. Sensitivity and specificity calculated at various cut-off points give rise to a receiver-operating-characteristic (ROC) curve \[[@B2]\]. A clinically useful biomarker will be one with the largest area under the ROC curve. A number of novel blood biomarkers of lung disease including cytokines, enzymes, adhesion molecules, collagen relevant products and products of type II epithelial cells, have been studied for their clinical applicability. The scope of this review is based on the fact that although there are numerous published papers investigating the utility of biomarkers in the clinical research field the number of review articles summarizing the current state of knowledge about the clinical applications of these molecules as diagnostic and prognostic tools in the research field relevant to acute respiratory distress syndrome (ARDS) and acute lung injury (ALI) still remains inadequately small. Serum biomarkers in Acute Respiratory Distress syndrome ======================================================= ARDS is a clinical and pathophysiologic entity characterized by severe acute injury, directly or indirectly via the blood, to the endothelial and epithelial surfaces of the lung leading to respiratory failure. The main characteristics of the syndrome are diffuse inflammation and increased microvascular permeability that cause diffused interstitial and alveolar oedema and persistent refractory hypoxemia \[[@B3]\]. Although a variety of insults may lead to ARDS, a common pathway may probably result in the lung damage \[[@B4],[@B5]\]. A complex series of inflammatory events have been recognized during the development of ARDS but the exact sequence of the events remains elusive. Immunological studies investigating bronchoalveolar lavage fluid (BALF) have shed further light into the pathogenetic mechanisms of ARDS \[[@B6]\] and formed the basis of concepts of its immunopathogenesis. A large variety of inflammatory mediators (Table [1](#T1){ref-type="table"}) have been found to be elevated in the early phase of ARDS, including lung-specific proteins, endotoxin binding proteins, tumor necrosis-alpha (TNFa), interleukins -- (ILs) -- 1, 2, 6, 8, 15, chemokines, ferritin, markers of endothelium activation (adhesive molecules and von-Willebrand factor antigen-VWF) as well as markers of neutrophil activation such as matrix metalloproteinases (MMPs) and their inhibitors and leukotrienes \[[@B5]-[@B7]\]. The majority of these molecules have features to recommend them as biologic markers in ARDS. Biomarkers have attracted a lot of attention in both ALI and ARDS since they can shed further light into the pathogenesis and pathophysiology of lung injury. Additionally, from a practical point of view, a clinical useful biomarker for ARDS must add information regarding the development of syndrome in at-risk patients that is not apparent from routine examination and investigation. The latter, could help the intensivist to monitor the disease and evaluate or modulate treatments before they have failed. Driven by this perspective idea, many studies have estimated their usefulness as early predictors of ARDS and accurate markers of lung injury before clinical changes can be detected. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### List of studied serum biomarkers in ARDS ::: --------------------------------------- -------------------------------------------------------- Lung epithelium-specific proteins Surfactant-associated proteins • SP-A • SP-B • SP-D Mucin-associated antigens • KL-6/MUC1 Cytokines • IL-1 • IL-2 • IL-6 • IL-8 • IL-10 • IL-15 • TNFa Other serological parameters Markers of endothelium activation • Adhesion molecules (E, L-selectin, I-CAM-1, V-CAM-1) • VWF Markers of neutrophil activation • MMP-9 • LTB4 Ferritin --------------------------------------- -------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, CC16: Clara-cell protein 16, IL: Interleukin, KL-6: Krebs von den Lungen-6, LTB4: Leukotriene B4 MMP-9: Metalloproteinase-9 MUC: Mucin, I-CAM-1: Intercellular Adhesion Molecule-1, sIL-2R: soluble interleukin-2 receptor, V-CAM-1: Vascular Adhesion Molecule-1, VWF: von Willebrand factor antigen ::: Cytokines (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}) ============================================================================ ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Studies measuring cytokines in patients with or at risk for ARDS ::: ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Investigator Patients Controls Biomarker / Summary ROC curve analysis Cut-off values Specificity -- Sensitivity PPV -- NPV Limitations ------------------------- ----------------------- --------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ------------------------------------------------ ----------------------------------------------------------------- Pinsky et al. ^13^ 52 at risk Relation of IL-6 and TNF plasma levels to multiple-system organ failure and mortality No Not estimated Small number of patients\ No ROC curve analysis / cut-off levels\ No serial measurement Takala et al. ^14^ 20 at risk\ IL-6 and IL-8 plasma levels predict organ failure in community-acquired septic shock No Not estimated Small number of patients\ 56 controls No ROC curve analysis / cut-off levels\ Poor discriminative value of serum biomarkers per se Calandra et al. ^15^ 70 at risk IL-6 plasma levels do not predict the outcome in at risk patients for ARDS No Not reported Small number of patients\ No ROC curve analysis Kiehl et al. ^16^ 19 at risk TNFa, IL-6, IL-8 plasma levels fail to associate with severity and course of ARDS in leukocytopenic patients No Not estimated Small number of patients\ No ROC curve analysis / cut-off levels\ Leukocytopenic patients Meduri et al. ^18^ 27 ARDS TNFa, IL-1β, IL-2, IL-4, IL-6, IL-8\ Yes\ TNFa: 89-50-85-57%\ Small number of patients\ Superiority of IL-1β and IL-6 plasma levels in monitoring disease activity over commonly applied clinicophysiologic parameters. TNFa: 400 pg/ml\ IL-1β: 78-83-83-78%\ Perspective study\ IL-1β: 400 pg/ml\ IL-2: 89-83-90-80%\ Overlap of cytokine levels between survivors and non-survivors\ IL-2: 200 pg/ml\ IL-4: 89-50-85-57%\ Heterogeneity of studied population\ IL-4: 200 pg/ml\ IL-6: 77-75-81-70%\ Definition criteria of ARDS IL-6: 400 pg/ml\ IL-8: 66-50-66-50% IL-8: 400 pg/ml Agouridakis et al. ^19^ 8 ARDS^a^\ Association between increased levels of IL-2 and IL-15 and outcome in patients with early ARDS Yes\ IL-2: 100-100-100-100% IL-15: 100-100-100-100% Small number of patients\ 26 at risk IL-2: 173 pg/ml\ Limited number of studied molecules IL-15: 250 pg/ml ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, BALF: Bronchoalveolar lavage Fluid, IL: Interleukin, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, sIL-2R: soluble interleukin-2 receptor, TNFa: Tumor Necrosis Factor-alpha ^a:^Use the American European Consensus Conference definitions ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Studies measuring cytokines in patients with or at risk for ARDS ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Investigator Patients Controls Biomarker / Summary ROC curve analysis Cut-off values Specificity -- Sensitivity PPV -- NPV Limitations --------------------- ----------------------- --------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ------------------------------------------- ----------------------------------------------------------------- Lesur et al. ^20^ 19 ARDS^a^\ Association of early low serum\ No Not estimated Small number of patients\ 14 at risk\ IL-2 levels with the patients\' survival No ROC curve analysis / cut-off levels\ 20 controls No serial measurement\ Discrepancies in serum and BALF IL-2 levels\ Limited number of studied molecules Parsons et al. ^23^ 77 at risk Association of serum IL-1ra, IL-10 levels with the disease outcome No Not estimated No ROC curve analysis / cut-off levels\ Poor predictive value for ARDS development\ Limited number of studied molecules Takala et al. ^24^ 52 at risk\ IL-8, IL-6, sIL-2R, E-selectin, procalcitonin\ No Not estimated Small number of non-survivors\ 9 ARDS^a^\ Persistent elevation of inflammatory markers in patients with ALI precedes its clinical diagnosis No ROC curve analysis / cut-off levels\ 45 controls Poor predictive value for ARDS development Bouros et al. ^25^ 32 ARDS^a^\ IL-4, IL-6, IL-6r, IL-8, IL-10\ Yes\ IL-4: 78-100-81-100%\ Small sample size\ 27 at risk High prognostic value of all the inflammatory markers in assessing the outcome in patients with or at risk for ARDS IL-4: 84 pg/ml\ IL-6: 59-96-69-94%\ No serial measurement\ IL-6: 160 pg/ml\ IL-6r: 78-76-76-78%\ Grouping and definition criteria IL-6r: 18 pg/ml\ IL-8: 93-96-92-96%\ IL-8: 2340 pg/ml\ IL-10: 96-92-96-93% IL-10: 98 pg/ml Schutte et al. ^26^ 30 ARDS^a^\ IL-6, IL-8, TNFa\ No Not estimated Small number of patients\ 44 at risk\ Serum levels of IL-6 and IL-8 in\ Weak correlations with clinical variables\ 17 controls ARDS and/or severe pneumonia, differentiate these entities from cardiogenic pulmonary oedema No definitive predictive value for outcome\ Overlap of cytokine levels between survivors and non-survivors\ No ROC curve analysis / cut-off levels Bauer et al. ^27^ 46 ARDS^a^\ IL-6, IL-1b, TNFa serum levels associate better with the degree of lung injury rather than clarify its specific aetiology No Not estimated Corticosteroid treatment\ 20 at risk\ Inconclusive prognostic value\ 10 controls No serial measurement\ Limited number of studied molecules ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, BALF: Bronchoalveolar lavage Fluid, IL-1ra: Interleukin-receptor antagonist, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, sIL-2R: soluble interleukin-2 receptor, TNFa: Tumor Necrosis Factor-alpha ^a:^Use the American European Consensus Conference definitions ::: Cytokines are widely recognized as mediators of an inflammatory response. Their discovery has stimulated multidisciplinary investigation to elucidate the role of these mediators in the injury and repair processes of human disease. In the lung, they are produced either by local resident cells such as alveolar macrophages, pneumocytes, endothelial cells and fibroblasts or by cells such as neutrophils, lymphocytes and platelets arriving to the lung in response to local or systemic injury \[[@B8],[@B9]\]. Cytokines are involved both in the early (TNFa, IL-1, 2, 6, 8, 15) and late phase (IL-4) of inflammation and have been shown unequivocally to be of crucial importance in the pathophysiology of septic shock, a condition frequently culminating to ARDS \[[@B10],[@B11]\]. Studies have demonstrated that in ARDS patients detectable cytokine serum levels are closely related to the disease severity and mortality \[[@B11],[@B12]\], suggesting a potential role in reflecting the severity of the lung injury. Moreover, humoral (IL-6, IL-8) and cellular markers (CD 11b) of systemic inflammation have been delineated to identify patients with septic shock at risk for organ failure, culminating to a fatal outcome \[[@B13],[@B14]\]. However, their monitoring and prognostic value in septic-shock patients still remains controversial. Calandra et al. \[[@B15]\] stated that serum cytokines could not be used as a routine laboratory test to predict the outcome in septic-shock patients. In addition, Kiehl et al. \[[@B16]\] failed to prove usefulness of plasma cytokines measurement for the evaluation of severity and course of ARDS in a small cohort of leukocytopenic septic-shock patients. In the same study, estimation of BALF levels appeared to differentiate between responders and non-responders to treatment before clinical differences become apparent. Nonetheless, it should be noted that the small sample size, the contradictive results, the lack of standardization techniques and uniform definitions for ARDS and at risk patients, together with the heterogeneity of the syndrome and the patients studied generate major concerns regarding the reproducibility and the reliability of the data presented. Consequently, there is a need for further large scale investigations in the context of appropriate clinical trials for any meaningful conclusions to be reached. Evidence from preceding studies \[[@B11]-[@B13],[@B16]\] give credence to the view that the inability of the lung to repair after ALI is the result of a persistent inflammatory stimulus that ultimately leads to an unfavorable outcome \[[@B17]\]. Fueled by this prospect, Meduri and coworkers \[[@B18]\] indicated a consistent, efficient and independent predictive value for IL-1β and IL-6 serological concentrations over time in a small cohort of patients with severe ARDS. They generated ROC curve analysis and demonstrated a clear superiority of inflammatory cytokines in monitoring disease activity over commonly applied clinicophysiologic parameters. However, this study exhibited substantial weaknesses including the retrospective analysis, the small sample size and the overlapping results between survivors and non-survivors. These observations coupled with the heterogeneity of the disease and the evidence that elevated serum cytokines may reflect the increased production or decreased clearance and not the disease activity pose major limitations to the aforementioned findings. In another study, Agouridakis et al. \[[@B19]\] evaluated both the prognostic and predictive significance of IL-2 and IL-15 for the development and outcome of patients at risk who developed ARDS or patients at risk who never developed ARDS, respectively. They applied ROC curve analysis and showed an excellent predictive value of cytokine plasma levels in terms of specificity and sensitivity compared to those observed in BALF. The most remarkable ascertainment of this study was the emergence of the discriminative usefulness of elevated IL-2 and IL-15 serological concentrations in patients with ARDS or at risk for ARDS. Nevertheless, the small number of patients enrolled combined with the lack of serial measurements throughout the clinical course of the disease and the causal diversity of the syndrome render major uncertainty to these findings. On the contrary with other analyses \[[@B16],[@B18]\], Lesur et al \[[@B20]\] found lower blood IL-2 levels in patients with ARDS compared to those that never developed the syndrome. In addition, evidence of this study regarding the strong association of early low serum IL-2 levels with the patients\' survival corroborated earlier findings \[[@B16]\]. Potential criticism of this study include the small number of patients recruited, the absence of multiple time-point evaluation of the cytokine plasma concentrations and more importantly opposite and disproportional fluctuations of IL-2 content in serum and BALF in patients with or without ARDS. The role of several inflammatory cytokines in monitoring the disease activity and predicting the survival in patients with ARDS has aroused increasing attention the past decade. One of the most intriguing aspects of the application of these biomarkers in the daily clinical practice is the early detection of patients admitted to the intensive care unit (ICU) that will develop ARDS. This approach will allow anti-inflammatory and other supportive treatments to be evaluated or eventually modified before they have failed. Predictive levels of inflammatory cytokines (IL-1, IL-2, IL-6, IL-8) for ARDS development in at risk patients have been extensively reported with controversial results \[[@B16],[@B17],[@B21],[@B22]\]. The importance of considering inflammatory constituents of serum in patients at risk for ARDS was initially raised by Parsons et al \[[@B23]\]. Authors conducted a large prospective analysis and demonstrated that although immunological parameters (IL-1ra and IL-10) were elevated in patients at risk for ARDS and exhibited a remarkable association with the disease outcome, none of these could predict the development of the syndrome. These observations were extended by the results of Takala et al. \[[@B24]\] who showed that serum levels of inflammatory mediators albeit their persistent elevation in patients with unresolving ALI, preceding its clinical diagnosis, were of poor discriminative value in patients with ALI that did or did not develop ARDS. Nevertheless, these findings provided us with useful knowledge about the inflammation marker profile on the days preceding diagnosis of ARDS, indicating a potential relation of sustained inflammatory response with a poor outcome. With this aim in mind, Bouros et al. \[[@B25]\] measured prospectively a slew of cytokines in the serum and BALF in ICU patients to identify predictive factors for the course and outcome of ARDS. The most remarkable result of this analysis was that almost all serum molecules studied showed a high prognostic value in assessing the outcome in patients with or at risk for ARDS. However, laboratory parameters failed to prove a positive correlation with the prediction of ARDS development evidence consistent with earlier studies \[[@B23],[@B24]\]. Moreover, major caveats that should be taken under consideration include the limited number of patients, the lack of sufficient follow-up serum data and the marked causal heterogeneity of the syndrome that could be a reason for the contradictive results reported in previous studies \[[@B20],[@B23]-[@B25]\]. Further prospective studies with sufficient statistical power are required to validate these results and ameliorate the predictive role of circulating inflammatory mediators in patients at risk for ARDS. Although the role of cytokines in the pathogenesis of ARDS has been extensively investigated, their importance in the differential diagnosis has not been clearly defined. It is widely accepted that numerous insults may lead to ARDS following a common pathway. Furthermore, a variety of conditions including severe pneumonia imitate clinical and radiological manifestations of ARDS and thereby it is often difficult to differentiate them. However, this would be a fruitful application because the treatment of these conditions differs considerably. Many groups of investigators have attempted to produce a discriminative systematic inflammatory profile and although much good work has been done towards this direction, the results still remain controversial. Schutte and co-workers \[[@B26]\] provided us with a really well done and heavily informative paper concerning the systemic cytokine profile in patients with ARDS, severe pneumonia and cardiogenic pulmonary oedema. Authors found remarkably and consistently elevated serum levels of IL-6 and IL-8 in ARDS and/or severe pneumonia, differentiating these entities from cardiogenic pulmonary oedema. Nevertheless, they were unable to separate the various entities of ARDS and states of severe pneumonia based solely on alterations in the immunomodulatory pattern. To streamline these observations, Bauer et al. \[[@B27]\] tested the potential of inflammatory markers (TNFa, IL-1β, IL-6) to differentiate between these two diseases. Results in harmony with the previous study \[[@B26]\] demonstrated higher TNFa serological concentrations in patients with ARDS from the remaining populations. However, they revealed the ability of immunological parameters to associate better with the degree of lung injury rather than clarify its specific aetiology. No clear relationship between serological data and patients\' survival was observed. In addition, this data exhibits major limitations, including the absence of uniform methodology (use of corticosteroid treatment in some of the patients), serial measurements and the lack of knowledge regarding serum alterations in other components of the inflammatory network (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). Other serological parameters ============================ Markers of endothelium activation (Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}) ---------------------------------------------------------------------------------------------------- ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Studies measuring markers of endothelium activation in patients with or at risk for ARDS ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Investigator Patients Controls Biomarker / Summary ROC curve analysis Cut-off values Specificity -- Sensitivity PPV-NPV Limitations ------------------------- ----------------------- -------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------- ---------------------------------------- ---------------------------------------------------------------------------- Donnelly et al. ^39^ 82 at risk\ E-selectin levels were not correlated with ARDS development and patients\' mortality. L-selectin levels exhibited a significant prognostic value No Not estimated Heterogeneity of studied population (trauma-sepsis)\ 14 ARDS^a^\ No ROC curve analysis / cut-off levels No serial measurement 62 controls Boldt et al. ^40^ 50 at risk Constantly lower E-selectin, ICAM-1 and VCAM-1 levels in survivors experiencing polytrauma than in nonsurvivors No Not estimated Small sample size\ Causal diversity of patient group\ No definitive association with patients\' mortality\ No ROC curve analysis / cut-off levels Cowley et al. ^41^ 40 SIRS\ Superiority of E-selectin plasma levels in predicting organ dysfunction and death patients with SIRS comparing to ICAM-1 No Not estimated Small number of patients\ 85 controls Causal diversity of patients studied\ No ROC curve analysis / cut-off levels Sessler et al. ^43^ 25 at risk\ Association of elevated ICAM-1\ Yes\ Not reported Small sample size\ 12 controls Sequential plasma levels with the severity of shock ICAM-1: 715 ng/ml (predicting survival) Heterogeneity of studied population\ Inconclusive association with disease Severity Kayal et al. ^44^ 32 at risk\ Cut off values of E-selectin, ICAM-1 and VWF serum levels predicted survival outcome Yes\ E-selectin: 73-80-67-85%\ Small number of patients\ 9 controls E-selectin: 128 ng/ml\ ICAM-1: 80-90-75-92%\ Most of the patients developed secondary ALI ICAM-1: 715 ng/ml\ VWF: 87-8080-87% VWF: 717% Agouridakis et al. ^45^ 23 ARDS^a^\ TNFa, IL-1, ICAM-1, VCAM-1\ Yes\ For ARDS development\ Small sample size\ 42 at risk ICAM-1 and VCAM-1 showed a high NPV for ARDS development\ TNFa: 325 pg/ml\ TNFa: 62-75-38-89%\ No serial measurement\ Correlation with the disease outcome\ IL-1: 225 pg/ml\ IL-1: 58-88-39-94%\ None of the studied markers was an independent factor for ARDS development None of the studied markers was an independent factor for ARDS development ICAM-1: 300 pg/ml\ ICAM-1: 69-75-42-90%\ VCAM-1: 260 pg/ml VCAM-1: 73-88-50-95% -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ALI: Acute Lung Injury, ARDS: Acute Respiratory Distress Syndrome, ICAM-1: Intercellular Cell Adhesion Molecule-1, IL: Interleukin, ROC: Receiver Operating Characteristic, SIRS: Systematic Inflammatory Response Syndrome, TNFa: Tumor Necrosis Factor-alpha, VCAM-1: Vascular Cell Adhesion Molecule-1, VWF: von Willebrand factor antigen, ^a:^Use the American European Consensus Conference definitions ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Studies measuring markers of endothelium activation in patients with or at risk for ARDS ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Investigator Patients Controls Biomarker / Summary ROC curve analysis Cut-off values Specificity -- Sensitivity PPV-NPV Limitations ----------------------- ----------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ---------------------------------------- -------------------------------------------------------------------------------------------------- Rubin et al. ^47^ 45 at risk Elevated plasma VWF is an early predictor of ALI in nonpulmonary sepsis syndrome Yes\ 77-87-80% Small sample size 25% of patients had already lung injury at the time sepsis was diagnosed\ VWF: 450% Exclusion of patients who developed\ ALI from a primary pulmonary source VWF levels measured by an old assay Ware et al. ^48^ 51 ALI/ ARDS^a^\ VWF is an independent predictor of hospital mortality in patients with ALI No\ 91-44-83-62% Inadequate sample volume\ 4 controls VWF:450% Heterogeneity of studied population\ No ROC curve analysis Ware et al. ^49^ 559\ Significant correlation of elevated VWF plasma levels with mortality, duration of unassisted ventilation and organ failures. No differences of VWF levels between septic and non septic patients No Not estimated Not definitive association with patients\' mortality\ ALI/ ARDS^a^ Lack of knowledge regarding the cellular source and the mechanisms of elevated VWF serum levels\ No ROC curve analysis / cut-off values Moalli et al. ^50^ 35 at risk\ VWF levels were higher in ARDS compared with at risk\ No Not reported Limited number of patients\ 10 ARDS\ VWF levels are not helpful in predicting ARDS development No ROC curve analysis / cut-off values 9 controls Moss et al. ^51^ 96 at risk VWF is not predictive of development of ARDS Yes\ 47-70%\ Causal diversity of patients studied\ VWF:273%\ 52-64% No definitive relation with disease severity VWF:399% Sabharwal et al. ^52^ 22 ARDS\ No significant association of VWF blood levels with patients\' mortality No Not estimated Small sample size\ 21 at risk Retrospective study\ No ROC curve analysis / cut-off values Bajaj et al. ^53^ 18 ARDS\ Serum VWF levels were non-useful markers for predicting ARDS in at risk patients Yes\ 71-62-34% Limited number of patients\ 15 at risk\ VWF: 300% No serial measurement\ 27 controls Coexisting multisystem organ failure\ Heterogeneity of studied population Moss et al. ^54^ 55 at risk\ ICAM-1, E-selectin, VWF\ No Not estimated Small sample size\ 14 ARDS\ Degree of endothelial activation varied in patients at risk for ARDS from different etiologic factors Heterogeneity of studied population 11 controls -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ALI: Acute Lung Injury, ARDS: Acute Respiratory Distress Syndrome, ICAM-1: Intercellular Cell Adhesion Molecule-1, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, VCAM-1: Vascular Cell Adhesion Molecule-1, VWF: von Willebrand factor antigen ^a:^Use the American European Consensus Conference definitions ::: The pathophysiologic sequence characterizing ALI involves apart from cytokine, free radical, proteases and aracidonic acid metabolites release, the endothelial and neutrophil activation which initiate a cascade of leukocyte-endothelium interactions and adhesions. This is followed by transendothelial migration of neutrophils and release of their cytotoxic products, ultimately resulting to microvascular and tissue injury \[[@B28]\]. Adhesion of neutrophils to the endothelium is regulated by at least three adhesion molecule families including selectins (E, L and P), integrins and the immunoglobulin superfamily (intercellular adhesion molecule- ICAM-1 and vascular cell adhesion molecule-VCAM-1) and by chemotactic signals \[[@B29],[@B30]\]. Initial interactions of leukocytes and the endothelium are mediated by members of the selectin family inducing (loose) contact with the endothelium also known as rolling, followed by firm adhesion requiring members of the integrin (β2) and immunoglobulin family (ICAM-1) \[[@B31],[@B32]\]. In recent years, soluble isoforms of some of these molecules {soluble-(s)-E-selectin, sICAM-1, sVCAM-1} have been detected in the circulating blood under various inflammatory conditions \[[@B33]-[@B35]\]. Mechanisms that could potentially explain an increase in circulating adhesion molecules include cytokine-induced (IL-1, TNFa) overexpression by the endothelial cells, increased proteolytic cleavage of endothelial-bound adhesion molecules secondary to endothelial damage or both \[[@B33],[@B35]\]. One attractive feature of these molecules and mostly E-selectin is that since their expression is almost restricted to stimulated endothelial cells \[[@B32]\] their presence in serum should potentially reflect the state of endothelium in disease and subsequently the disease severity in ALI. Other potential markers of endothelial cell injury that were delineated to shed further light into the pathophysiologic process of ALI include von-Willebrand factor antigen (VWF), a macromolecular antigen that is produced predominantly by endothelial cells and to a lesser extent by platelets and megakaryocytes \[[@B36]\]. Endothelial perturbation (as in at risk state) or injury (as in ARDS) results to the release of VWF from preformed stores into the circulation \[[@B37],[@B38]\]. Therefore, it appears that circulating VWF concentrations may serve as a suitable predictive marker for development of ARDS in patients at risk. So far, the potential usefulness of adhesion molecules and other markers of endothelial cell damage in reflecting the severity of endothelial damage and predicting the development or the final outcome of the disease is a subject of ongoing controversy. One of the first and most informative studies addressing this important issue was conducted by Donnelly et al \[[@B39]\]. Authors demonstrated in a large cohort of patients at risk for ARDS, that mean circulating levels of sE-selectin were not correlated with subsequent ARDS development and patients\' mortality. However, low values of sL-selectin exhibited a significant prognostic value. In contrast, Boldt et al. \[[@B40]\] studying the behaviour over 5 d of adhesion molecules (sE-selectin, sICAM-1 and sVCAM-1) in subjects experiencing polytrauma found constantly higher levels in nonsurvivors. In accordance to these findings, Cowley and colleagues \[[@B41]\] showed a superiority of sE-selectin plasma levels in predicting organ dysfunction and death in a group of patients with systemic inflammatory response (SIRS) comparing to sICAM-1 peripheral concentrations. This indicated that measurement of adhesion molecules could serve to advantage in the management of patients with sepsis. Nonetheless, the heterogeneity of patients studied (septic shock and polytraumatic) may justify these controversial results, since E-selectin expression has been found much greater in septic than in traumatic shock in experimental models \[[@B42]\]. Moreover, the relationship between the consequences of sepsis (organ failure, mortality) and blood levels of potential markers of endothelial-cell activation was strengthened by Sessler and co-workers \[[@B43]\]. Results from this study focusing on sICAM-1 sequential plasma levels, were suggestive of a strong association between the severity of shock (as determined by the presence of hypotension and the requirement of vasoactive drugs) and the circulating concentrations of the marker. The aforementioned observations were further confirmed by the study of Kayal et al \[[@B44]\]. Cut off values for three markers of endothelial activation were determined prospectively by ROC curve analysis and clearly predicted survival outcome with high sensitivity and specificity in a limited number of at-risk patients with secondary ALI. Despite that the role of soluble adhesion molecules in other inflammatory conditions strongly associated with ARDS \[[@B40]-[@B44]\] is well known, their value as markers of the disease progression and mortality has not been extensively studied in ARDS patients. To streamline these observations, Agouridakis et al. \[[@B45]\] scrutinized the role of two adhesion molecules (ICAM-1, VCAM-1) in parallel with proinflammatory cytokines in predicting the ARDS development and relating to the disease outcome. None of the studied mediators was found to be an independent factor for ARDS development, whereas both groups of molecules exhibited a considerable negative predictive value for ARDS development both in serum and BALF. Additionally, ROC curve analysis showed a clear superiority of plasma parameters in correlating with the disease outcome compared with BALF molecules. Further studies with serial BALF and serum measurements should be designed to elucidate the exact role of these markers over time in reflecting the disease behaviour and predicting the likelihood of progression. Elevated circulating concentrations of VWF in patients with ALI/ARDS were first reported in 1982 \[[@B46]\]. Their potential significance in predicting ARDS development was first demonstrated by Rubin and colleagues \[[@B47]\] who found that increased plasma levels of this marker exhibited a high predictive value both for the development of ARDS and for identifying patients with nonpulmonary sepsis who were unlikely to survive. However, authors did not use the uniform criteria for the definition of ARDS and risk state \[[@B3]\], evidence that poses major limitations to the results of the study. Furthermore, in an aforementioned study, Kayal et al. \[[@B44]\] in parallel with other findings reported a marked and independent association of circulating VWF with the disease severity as assessed by other commonly applied clinical variables. These findings were further supported by a single-center study from Ware et al \[[@B48]\]. They conducted the first comparative study of VWF concentrations in both plasma and edema fluids of patients with early ALI from a variety of causes and reported that serum VWF levels were an independent predictor of hospital mortality and were associated with longer duration of mechanical ventilation. Potential criticisms of this study was the implementation of high tidal volume ventilation that possibly increased systemic endothelial activation, the small sample size, the lack of sequential measurement and the fact that the studied biomarker appeared not to be endothelial-specific since it is produced in small amounts by platelets \[[@B36]\]. To streamline these observations and ameliorate potential hardships the same group of authors {Ware et al. \[[@B49]\]} carried out a multicenter study of 559 patients with ALI and ARDS which was recently published. In accordance with earlier studies \[[@B48]\], a significant correlation of elevated VWF plasma levels with adverse outcomes, including mortality, duration of unassisted ventilation and organ failures was pointed out. Intriguingly, authors demonstrated for the first time a negative association between markers of endothelial activation and presence or absence of sepsis, supporting the hypothesis that ALI might be an independent cause of systemic endothelial activation and injury. Finally, in the same study no modulation of plasma VWF concentrations by protective mechanical ventilation was observed. Despite the remarkable power of the presented findings, there are substantial weaknesses that deserve further investigations including the inconclusive analysis of the plasma VWF levels associated with patients\' mortality, the lack of definite knowledge regarding the source of VWF production, and the mechanisms leading to increased peripheral concentrations since the latter also could reflect decreased clearance from the circulation. Subsequent data derived from other studies \[[@B50]-[@B54]\] was rather contradictive and controversial. Even though, Moalli et al. \[[@B50]\] found a poor predictive value of serum VWF levels for the development of ARDS in a group of at risk patients, the biomarker concentrations were correlated with the disease severity,. Similarly, Moss et al. \[[@B51]\] plotted ROC curves and concluded that in patients at risk for ALI/ARDS from multiple causes, serum VWF levels failed to reliably discriminate which patients would develop ARDS. The evidence was further validated by Sabharwal et al \[[@B52]\]. Authors conducted the first study comparing plasma levels between survivors and nonsurvivors in a group of patients both at risk for and with established ARDS and observed no significant association of VWF blood levels with patients\' mortality ;predictive value of the marker was not reported. In agreement with the previous study, a study from the same group of scientists {Bajaj et al. \[[@B53]\]} using standard criteria for the definition of ARDS and at risk state \[[@B3]\] demonstrated the inability of three endothelial-specific proteins including VWF to predict the progression of ARDS in at risk patients. Nonetheless, major caveats that should be addressed include the fact that many at-risk patients had already some degree of ALI, the lack of serial measurement that could potentially show a trend towards prediction of ARDS development and the causal diversity of patients examined that could possibly affect the results of the study. The latter limitation was addressed by Moss et al. \[[@B54]\] who established that the degree of endothelial activation as determined by the plasma levels of VWF (higher in subjects with sepsis than patients with trauma) is not uniform in all patients at risk for developing ARDS. Accumulated evidence from the preceding studies suggest that the etiologic diversity of patients enrolled renders major uncertainty to the reliability of the results and highlights the necessity for further prospective studies using standard criteria for the definition of ARDS and analyzing well defined and uniform group of at risk patients in order to produce knowledge of high scientific rigidity. Direct comparison of the different studies is difficult and in a way meaningless because of the use of varying definitions for ARDS and at-risk patients as well as the inclusion of different patient populations, in which of them some degree of ALI was probably already present (Tables [3a](#T3){ref-type="table"} and [3b](#T3){ref-type="table"}). Markers of neutrophil activation (Table [6](#T6){ref-type="table"}) ------------------------------------------------------------------- ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Studies measuring markers of neutrophil activation and ferritin in patients with or at risk for ARDS ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Investigator Patients Controls Biomarker / Summary ROC curve analysis Cut-off values Specificity -- Sensitivity PPV -- NPV Limitations ---------------------- ----------------------- -------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- -------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Pugin et al. ^60^ 31 at risk\ IL-8, MMP-2, MMP-9\ No Not estimated Measurement of circulating proinflammatory cytokines without the appreciation of their inhibitors or receptor antagonists is misleading mainly due to a possible neutralization\ 23 ARDS^a^ Plasma levels of inflammatory activity are not useful markers in differentiating permeability from hydrostatic pulmonary edema Small number of patients\ No ROC curve analysis / cut-off values Amat et al. ^64^ 21 ARDS^a^\ Strong association of LTB4 and IL-8 serum levels with the patients\' survival. Yes\ LTB4 + IL-8: 88-70-85-75% (markers of mortality rate)\ Small sample size\ 14 at risk LTB4: 14 pmol/ml\ LTB4 : 85-72-20-98% (marker of ARDS development) Lack of adjustment with the disease severity\ IL-8: 150 pmol/ml Inability of LTB4 plasma levels to be an independent predictive marker Connelly et al. ^66^ 75 at risk\ Serum ferritin is a sensitive and specific predictor of ARDS development Yes\ 71-83-86-67%\ Limited number of ARDS patients\ 8 ARDS^a^ Ferritin (male):270 ng/ml\ 90-60-82-75% Heterogeneity of studied population Ferritin (female): 680 ng/ml Sharkey et al. ^67^ 42 at risk\ Correlation of ferritin plasma levels with the development of ARDS, multiple organ failure and severity of lung injury Yes\ 64-73-75-62%\ Inadequate sample volume\ 16 ARDS^a^ Ferritin (male):270 ng/ml\ 92-60-95-75% Not specific cut-off values\ Ferritin (female): 680 ng/ml Elevated serum levels may reflect a systemic response to a risk factor ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, IL: Interleukin, LTB4: Leukotriene B4, MMP-Metalloproteinase, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, ^a:^Use the American European Consensus Conference definitions ::: Generally, it is strongly believed that ARDS arises as a result of tissue injury secondary to sequestration of inflammatory cells, tissue invasion, and secretion of cytotoxic products. Neutrophils have received much attention as key part of this process. Although ARDS has been described in neutropenic patients \[[@B16],[@B55]\] there is increased evidence implicating neutrophils in most cases of ARDS. They have been reported by several studies \[[@B56],[@B57]\] to exert an important role in the early phase of ALI characterized by architecture remodeling, surfactant and epithelial toxicity. They use a wide array of enzymes during the process of transmigration through biological membranes such as alveolar-capillary barrier \[[@B58]\]. These enzymes include, metalloproteinases (MMPs) such as MMP-9 also called gelatinase B which is secreted from preformed neutrophil granules in response to a variety of stimuli including proinflammatory cytokines (IL-8, TNFa). MMP-9 is secreted as a zymogen, and then activated by a variety of other proteases such as elastase, and plays a crucial role in digesting basement membranes \[[@B58],[@B59]\]. Therefore, it has been speculated that metalloproteinases probing aspects of the inflammatory response could be utilized as markers of neutrophil activation and subsequently to reflect disease activity and severity, shedding further light into the pathogenesis of ARDS. Fueled by this prospect, Pugin et al. \[[@B60]\] compared the concentrations of proinflammatory cytokines and collagenases in serum and pulmonary oedema fluids in a small group of patients with ARDS and hydrostatic oedema from congestive heart failure. Authors concluded that elevated pulmonary oedema levels of these mediators could differentiate between these conditions, whereas plasma levels of proinflammatory and metalloproteinase activity proved to be of poor discriminative value. The latter observation mirrors the hypothesis that the inflammatory response characterizing ARDS patients is well compartmentalized, with little spillover into the circulation and that the measurement of circulating proinflammatory cytokines without the appreciation of their inhibitors or receptor antagonists is misleading mainly due to a possible neutralization. To gain a more comprehensive understanding on the role that neutrophils exhibit during the inflammatory cascade resulting to ALI, investigators scrutinized the utility of other chemotactic agents including leukotrienes (LTs). LTs (B4, C4, D4, E4) exert a synergistic role with IL-8 in the neutrophil influx and activation leading to a massive recruitment of neutrophils and to a catastrophic inflammatory response. Their BALF levels have been found elevated in patients with ARDS and their involvement in the alterations of microvascular permeability correlated with the accumulation of pulmonary oedema has been suggested \[[@B61]-[@B63]\]. Moreover, Amat et al \[[@B64]\] utilizing ROC curve analysis demonstrated that LTB4 plasma levels could serve as a valuable predictive marker of ARDS in terms of specificity and sensitivity. In the same study, authors performed serial measurement and reported a strong association of both LTB4 and IL-8 peripheral concentrations with the patients\' survival. However, the small number of patients enrolled, the lack of adjustment with the disease severity and the inability of LTB4 plasma levels to be an independent predictive marker arise major concerns whether they could monitor disease behaviour and predict ARDS development in at risk patients (Table [6](#T6){ref-type="table"}). Ferritin (Table [6](#T6){ref-type="table"}) ------------------------------------------- Ferritin is a 480-kDa iron-storage protein that sequesters iron in the ferric (Fe^3+^) state. It has been speculated that ferritin may serve as a crucial antioxidant mediator because free iron enhances the formation of highly toxic hydroxyl radicals from superoxide anion and hydrogen peroxide. On the other hand, oxidative stress is a condition commonly seen in disorders at risk for ARDS development such as sepsis. Hence, ferritin-derived iron may aggravate oxidative damage in critically ill patients, contributing to the pathologic abnormalities encountered in ARDS. Furthermore, proinflammatory cytokines such as IL-8, IL-6 and TNFa which are increased and presumably participating in the pathogenetic derangements of ARDS have been suggested to promote ferritin synthesis \[[@B39],[@B65]\]. Thereby, it can be concluded that elevated ferritin levels could result from oxidative stress, proinflammatory cytokines and the degree of lung injury, all conditions characterizing the pathogenesis of ARDS and subsequently can be used as prognostic and monitoring tool reflecting the likelihood of ARDS development and the disease severity. The first study attempted to prove such correlation was conducted by Connelly et al \[[@B66]\]. They plotted ROC curves to estimate the utility of ferritin levels as prognostic factors and produced clinically useful cut-off points which could predict the development of ARDS with high sensitivity, specificity, negative and positive predictive value, both in male and female predominantly septic subjects. However, the heterogeneity of the etiologic factors resulting to ARDS development in at risk patients renders major uncertainty to the rigidity and reliability of the results. To ameliorate this hardship, the same group of authors {Sharkey et al. \[[@B67]\]} generalized and extended the latter results in a homogeneous group of at-risk patients with multiple trauma demonstrating a strong correlation of initial ferritin plasma levels with the development of ARDS and multiple organ failure. In addition, an association of serum ferritin levels with the severity of lung injury as well as other markers of endothelial activation was also noted supporting the premise that elevated levels could reflect the inflammatory status encountering in ALI. Nonetheless, authors failed to detect specific predictive cut-off values suggesting that circulating concentrations of this biomarker are unable to predict per se the progression to ARDS. A possible explanation could arise from the hypothesis that elevated levels of this marker must reflect a systemic response to a risk factor, which may prove to reduce its specificity (Table [6](#T6){ref-type="table"}). Lung epithelium-specific proteins (Table [7](#T7){ref-type="table"}) ==================================================================== ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Studies measuring lung-specific proteins in patients with or at risk for ARDS ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Investigator Patients Controls Biomarker / Summary ROC curve analysis Cut-off values Specificity -- Sensitivity Diagnostic accuracy Limitations ---------------------- ----------------------- ------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ---------------------------------------------------- ------------------------------------------------------------------------------------------------------- Doyle et al. ^81^ 15 ARDS^a^\ SP-A is an acute indicator of lung function and alveolocapillary membrane injury No Not estimated Small number of patients\ 10 at risk\ No ROC curve analysis / cut-off values\ 10 controls No definitive relation with disease severity Doyle et al. ^82^ 22 ARDS^a^\ Superiority of SP-B compared to SP-A plasma levels as a marker of lung function and alveolocapillary membrane injury No Not estimated Only 3 case-control studies\ 10 at risk\ Inadequate sample size\ 33 controls Lack of adjustment with disease behaviour\ No ROC curve analysis / cut-off values Greene et al. ^83^ 41 ARDS^a^\ SP-A, SP-B, SP-D\ Yes\ Poor predictive value\ Limited number of patients\ 22 at risk\ Serum changes found to be neither sensitive nor specific in predicting the onset of ARDS and discriminating survivors from non-survivors. Not reported Low specificity/sensitivity Serial measurements for a short period of time/ Lack of serial measurement for the most severe forms\ 35 controls Heterogeneity of studied population\ Poor predictive value for serum levels Cheng et al. ^84^ 36 ARDS^a^\ SP-A levels were associated with severity of clinical lung injury and with disease outcome No Not estimated Small sample size\ 2 ALI Causal diversity of studied population\ No serial measurement Greene et al. ^85^ 51 at risk\ SP-A levels are predictive for at risk patients who developed ARDS from sepsis and aspiration but not trauma No Not estimated Small sample size\ 26 ARDS^a^\ No ROC curve analysis / cut-off levels 16 controls Bersten et al. ^86^ 54 at risk\ SP-B but not SP-A cut-off plasma levels predict ARDS development, particularly in at-risk patients suffering a direct lung injury Yes\ 78-85-85-78% Small number of patients\ 9 controls SP-B: 4.994 ng/ml Limited follow-up serum data\ Most of patients had already lung injury\ Exclusion of milder at risk patients Eisner et al. ^87^ 565\ SP-A, SP-D\ No Not estimated Only 2 serial measurements\ ALI/ ARDS^a^ Attenuation of SP-D plasma levels by lower volume ventilation strategies Heterogeneity of studied population Potential selection bias\ No ROC curve analysis / cut-off levels Ishizaka et al. ^95^ 35 at risk\ Association of optimal cut-off values of KL-6 serum levels with patients\' mortality Yes\ 100-87% Inadequate sample volume\ 27 ARDS^a^\ KL-6: 253 U/ml Heterogeneity of studied population 21 controls Sato et al. ^96^ 28 ARDS^a^\ Association of KL-6 serum levels with variables of lung injury severity and with mortality rates\ No Not estimated Small sample size\ 10 controls No correlation with ventilation strategies Heterogeneity of studied group\ No serial measurement\ No ROC curve analysis / cut-off levels\ Diversity of ventilatory treatment ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ALI: Acute Lung Injury, ARDS: Acute Respiratory Distress Syndrome, BAL: Bronchoalveolar Lavage, KL-6: Krebs von den Lungen-6, ROC: Receiver Operating Characteristic, SP: Surfactant Protein, ^a:^Use the American European Consensus Conference definitions ::: Beyond other important functions, the lung epithelium produces complex secretions, including mucus blanket, surfactant proteins, as well as several proteins important for host defense \[[@B68]\]. Sampling the epithelial lining fluid by bronchoalveolar lavage (BAL) represents the common means of studying the proteins secreted by the lung epithelium and investigating their alterations in lung disorders \[[@B69]\]. However, the past fifteen years pioneering studies \[[@B70]\] showed the presence of these proteins in the bloodstream as well, even though in small amounts. Because these proteins are mainly, if not exclusively secreted within the respiratory tract, their occurrence in the vascular compartment can be explained by several hypothetical mechanisms including, leakage from the lung into the bloodstream, increased production by the alveolar type II cells or diminished clearance rates from the circulation \[[@B68]\]. Surfactant-associated Proteins ------------------------------ Pulmonary surfactant is a complex and highly surface active material covering the alveolar space of the lung. Biochemically, surfactant is a molecular mixture composed mainly of structurally heterogeneous phospholipids. A major function of pulmonary surfactant is to reduce the surface tension at the air-liquid interface of the alveolus, thereby preventing alveolar collapse on expiration. It has also been demonstrated that the surfactant contains specific proteins \[[@B71]\]. Four surfactant-specific proteins with different structural and functional properties have so far been identified. They were named surfactant protein-(SP)-A, SP-B, SP-C and SP-D according to the chronologic order of their discovery \[[@B72]\] and have been divided in two distinctive groups, the low-molecular-weight hydrophobic SP-B and SP-C and the high-molecular-weight-hydrophilic SP-A and SP-D. The latter belong to the collectin subgroup of the C-type lectin superfamily and are produced by two types of non-ciliated epithelial cells in the peripheral airway, Clara cells and alveolar type II cells. Studies have demonstrated that SP-B and SP-C seem to play an essential role for the adsorption of phospholipids to the air-water interface resulting to a stable phospholipids film and for the dynamic surface-tension-lowering properties \[[@B73]\]. Additional functions of the alveolar surfactant system include prevention of alveolar edema \[[@B74]\] and a pronounced influence, especially of the collectins SP-A and SP-D in the innate immune system of the lung \[[@B75],[@B76]\] and have been used as useful markers for confirming the diagnosis and evaluation of disease activity of various ILDs since they reflect the epithelial damage and turnover \[[@B77]\]. Thus, it has been speculated that alterations of SPs in biological fluids could serve as valuable markers of the severity of the lung injury or clinical outcome in ARDS patients. Most of our knowledge regarding changes in SP concentrations that occur in patients with or at risk for ARDS and their value in reflecting the disease severity or the likelihood of ARDS development comes from BAL studies. Several reports in the literature have demonstrated the occurrence of low SP-A levels in BALF of patients with ARDS following trauma \[[@B78],[@B79]\] coupled with a strong relation of this biological marker to the severity of endothelial damage \[[@B80]\]. Moreover, a potential value of SP-A plasma levels in discriminating patients with ALI of various etiologic factors has also been shown \[[@B79]\]. On the other hand, little is known about changes in peripheral concentrations of surfactant-associated proteins in patients with ALI and whether these alterations can serve as markers of injury to the epithelial and endothelial barriers in the lungs. Doyle et al. \[[@B81]\] documented elevated circulating concentrations of SP-A in patients with ARDS and in those with acute cardiogenic pulmonary edema possibly resulting from increased alveolocapillary permeability due to excessively high pulmonary capillary pressures. In the same study, blood SP-A levels were inversely associated with blood oxygenation and static respiratory system compliance. These results were fully confirmed by the same group of authors {Doyle et al. \[[@B82]\]} who also illustrated a clear superiority of SP-B compared to SP-A plasma levels as a marker of lung function and alveolocapillary membrane injury. Another study by Greene et al. \[[@B83]\], evaluated the differences that occur in SPs in BALF and serum of a relatively small cohort of patients at risk for ARDS and during the course of the syndrome. Authors demonstrated that only SP-A and SP-D BALF levels were strongly related to outcome and likelihood of disease progression whereas serum changes found to be neither sensitive nor specific in predicting the onset of ARDS and discriminating survivors from non-survivors. These data were further confirmed by a small cohort observational study by Cheng et al \[[@B84]\]. Even though authors reported an association of elevated SP-A plasma levels with a high degree of lung injury, they failed to extend this correlation with the disease mortality. Moreover, serum SP-D levels exhibited weak relation to the disease severity. It should also be noted that the aforementioned results present low statistical power due to the limited number of patients, the causal heterogeneity of the studied group and the absence of serial measurement and therefore no meaningful outcome can be excluded. In harmony with the latter results, Greene et al. \[[@B85]\] found that plasma SP-A was weakly predictive for ARDS development in septic patients and were unable to detect at risk trauma patients that developed the syndrome. Additionally, authors raised the crucial issue whether circulating SPs can reflect pathophysiologic differences between direct and indirect causes of ARDS and subsequently detect biologic changes early after an insult. From a pragmatic clinical perspective the most important question to be answered is which ICU patients requiring ventilatory assistance will develop ARDS. To do so Bersten et al. \[[@B86]\] generated ROC curve analysis and identified practical thresholds for SP-B plasma levels that could be clinically useful in predicting ARDS development, particularly in at-risk patients suffering a direct lung injury. In consistency with earlier studies \[[@B83]-[@B85]\] SP-A blood levels added no significant information on the disease prognosis. Further, an increase of circulating SP-B concentrations was documented on study entry, before changes in commonly applied clinical variables for the assessment of lung injury become apparent. These findings emphasize the usefulness of surfactant-associated proteins for the early detection of ARDS pathophysiologic alterations preceding changes in clinical parameters such as respiratory dysfunction. Arguments that can be made include the small sample size, the limited sequential measurements and the exclusion from the study recruitment of at risk patients with milder pulmonary dysfunction. These caveats coupled with the evidence that a considerable number of patients studied had already lung injury, pose major limitations to the predictive capacity of SP-B plasma levels and raise the necessity for larger prospective studies. The only so far large multicenter randomized controlled trial was performed by Eisner et al. \[[@B87]\] who estimated the prognostic value of SP-A and SP-D levels in an overall of 565 patients with early ALI/ARDS. Authors conducted the first study with adequate statistical power to examine the impact of SPs on mortality and other clinical variables and clearly demonstrated a strong linkage of elevated SP-D levels with worse clinical outcomes such as greater risk of death, fewer ventilator- free and organ failure-free days. One of the most remarkable ascertainments of this study was the attenuation of SP-D plasma levels by the lower volume ventilation strategies which reduces patients\' mortality postulating for the first time a significant association of biological parameters with therapeutic approaches and subsequently emphasizing the role of this mediator as a marker of the disease severity and prognosis. However, this study exhibited substantial weaknesses including the lack of sufficient serial measurements, a potential selection bias of patients recruited and the diversity of predisposing factors for ARDS development. These observations are not to diminish their value as prognostic and monitoring tools but to highlight the need for further confirmation studies using independent and well-defined populations of ALI/ARDS patients. Mucin-associated Antigens ------------------------- Mucins are major components of the mucus layer covering the airway epithelium. They consist of high-molecular-weight glycoproteins belonging to a broad family of mucin peptides \[[@B68]\]. Mucins are either associated with membranes or secreted at the surface of the respiratory tract \[[@B68]\]. Krebs von den Lungen-(KL)-6 is mainly associated with cellular membranes. It was initially described by Kohno et al. \[[@B70]\] as a high-molecular-weight glycoprotein and was classified as human MUC1 mucin. Immunohistochemistry has mainly detected KL-6 in alveolar type II and epithelial cells of the respiratory bronchioles. KL-6 is predominantly expressed by airway cells; however, is not entirely lung specific, since it is also present on other somatic cells, such as pancreatic cells, eosophageal cells and fundic cells of the stomach \[[@B88]\]. Additionally, KL-6 is a sensitive indicator of damage to alveolar type II cells, which strongly express this mucin at their surface \[[@B70]\]. Type II pneumonocytes are regenerated over the alveolar basement membrane after the death of type I pneumonocytes over the first stage of lung injury. Therefore, its raise would theoretically represent the destruction of the normal lung parenchyma and architecture, the increased permeability of the air-blood barrier as long as the regenerating process as expressed by type II pneumonocytes\' activity. Towards this direction, the presence of KL-6 has been extensively used with great promises to monitor the severity of disease in idiopathic pulmonary fibrosis \[[@B89]-[@B91]\] and other interstitial lung diseases \[[@B92]-[@B94]\]. Since damage to, and disruption of, the alveolar epithelial lining coupled with loss of integrity of the air-blood barrier represent key features in the pathophysiology of ARDS, KL-6 serum levels could potentially serve as valuable indicators of the disease severity directly assessing the degree of epithelial damage and predicting the progression to ARDS. Nevertheless, only few studies so far, have evaluated their monitoring and prognostic efficacy in patients with or at risk for ARDS development. One of the first studies to do so was recently carried out by Ishizaka and co-workers \[[@B95]\]. Authors generated ROC curve analysis and documented a highly sensitive and specific association of optimal cut-off values of KL-6 serum levels with patients\' mortality. The latter, further supports the premise that disruption of the alveolar barrier represents a major determinant of prognosis of ALI and that serial measurements of KL-6 plasma levels might be helpful markers of the disease progression. Limitations that should be addressed include the limited number of patients, the retrospective analysis of the results and the diversity of the etiologic factors of ALI generate major concerns about the reproducibility and the reliability of the data. Recently, Sato et al. \[[@B96]\] sought to determine potential correlations of KL-6 circulating concentrations with disease severity, patients\' survival and different predisposing factors of ARDS. The most remarkable ascertainments of this study include strong associations of KL-6 peripheral levels with variables of lung injury severity and with the rates of mortality indicating possible relationship between the degree of epithelial damage and poor outcome in ARDS. Even though, authors attempted to show a modulation of KL-6 serum levels by ventilatory strategies, this relationship failed to reach a statistical significance. Despite substantial weaknesses exhibited such as the lack of serial measurement, the small sample size and the diversity of the applied treatment data derived from this study is highly informative and provides important knowledge regarding the biological impact of mechanical support strategies in this syndrome indicating the monitoring value of the epithelial damage markers. Further and sizeable prospective studies are required to validate the aforementioned hypothesis (Table [7](#T7){ref-type="table"}). Future challenges and limitations ================================= The ARDS represents an overwhelming inflammatory reaction to numerous insults within the pulmonary parenchyma resulting in life-threatening derangements in pulmonary vasomotion, alveolar ventilation and gas exchange. ARDS is a frequent disease with a devastating incidence between 13.5 and 75 per 100,000, thus affecting about 16--18% of all patients ventilated in the ICU \[[@B76],[@B97]\]. Hence, ALI/ARDS is a major public health problem encountered frequently by all physicians who care for critically ill patients. Despite the fact that research efforts over the past several years have provided a more comprehensive knowledge of the potential mechanisms comprising the immunopathogenesis of ALI/ARDS and led to the development of innumerable causative or symptomatic treatment approaches, the mortality rate of these patients remains unacceptably high at 30--40% \[[@B97]\]. Currently, the only therapy that has been proven to be effective at reducing mortality is a protective ventilatory strategy \[[@B98]\]. However, new therapies are still needed. One of the most fruitful applications is monitoring the disease activity and consequently the early identification of at risk patients with increased likelihood of non-response to treatment and progression to ARDS. Nevertheless, there are problems with the sensitivity, effort-dependability and ease of repetition of the current modalities being used for this purpose, including radiological and BAL techniques as well as clinical and physiological indices of pulmonary injury (Murray score), systemic illness (Acute Physiology and Chronic Health Evaluation- APACHE-II score and Simplified Acute Physiological Score -SAPS), ARDS severity (respiratory system compliance and PaO2/FiO2 ratio) and multiorgan system failure (Multi-Organ Dysfunction Score-MODS). Most of these clinical parameters have failed to be independent predictors of mortality in studies of adults with ALI/ARDS \[[@B99],[@B100]\]. Development of a prognostic index that combines clinical and biological determinants may be useful to ameliorate these hardships. On the basis of this conception, a large body of serum markers either cytokines and lung-specific proteins or markers of endothelium and neutrophil activation as well as other serological parameters probing different facets of the immunopathogenesis of ALI/ARDS has been delineated. The applications of these markers in the clinical setting created major expectations in terms of defining categories of patients for different therapies or prognosis for the purpose of counseling families and patients and/or possibly identifying novel therapeutic targets. The determination of a reliable serologic marker reflecting the disease behaviour and adding independent information regarding the development of the syndrome before it becomes obvious in clinical level, easily reproducible and feasible to be measured serially represents a major challenge. The early serial measurement of this biomarker may serve as an independent non-invasive prognosticator of the disease outcome even at the onset of the syndrome and therefore lead to an early detection of at risk patients with increased likelihood of progression to ARDS. The latter, if sufficiently accurate, could prove extremely useful in identifying and counseling families of patients at low or high risk for adverse outcomes and further, will allow ventilatory or other types of treatment to be evaluated or eventually modulated before they have failed in the high risk group. The presented data give credence to the view that multiple biomarkers can be used to measure the lung and systemic response to a protective ventilatory strategy and potentially to discriminate patients who are ineffectively treated and might be candidates for rescue therapies \[[@B49],[@B87]\]. More importantly, use of one or more of these biologic markers to select a group of patients at higher risk of adverse clinical outcome could be used to restrict or stratify enrollment in future clinical trials applying novel ventilator treatments such as high-frequency oscillatory ventilation leading to a better patient care. Thereby, a combination of clinical factors and biologic marker measurements could be crucial for the selection of more homogeneous groups of patients with ALI/ARDS for further studies producing evidence of high scientific rigidity \[[@B101]\]. Finally, understanding the relative roles of markers of systemic and pulmonary endothelial injury and other inflammatory mediators to the pathogenetic process of the syndrome is likely to lead to valuable insights into the final pathway resulting in diffuse alveolar damage and significant lung dysfunction, highlighting therapeutic targets for novel interventions. The aforementioned components can potentially compile a clinician\'s \"wish list\". However, the feeling of excitement arising from the expected clinical utility comes in contrast with important deficiencies exhibited by the new methodologies including non-standardization techniques, lack of knowledge of reproducibility and link to disease behaviour. Furthermore, most of the studies enrolled a limited number of patients, insufficient to extract any meaningful or statistically significant outcome. In addition, the heterogeneity of the studied population resulting from the causal diversity of the syndrome and the use of non-uniform criteria for the definitions of ARDS and at-risk patients (Tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}) render major uncertainty to the reproducibility and the scientific rigidity of these findings and may explain potential discrepancies between various studies investigating different groups of at risk patients. Moreover, many of the caveats arising from this data are generated by the origin disadvantages of the investigated serological parameters to serve as specific markers of the disease activity and severity. In particular, it is well known that assays of circulating cytokine concentration may be misleading, because they do not detect receptor or cellular bound cytokines, or they may fail to detect cytokines when inhibitors or receptor antagonists are present. Thus, measured cytokine concentrations may not reflect the disease activity or the state of inflammation but the increased production or decreased clearance from the circulation. In consistency with these limitations, it should be underlined that cytokines are part of an inflammatory cascade and biological effects are difficult even impossible to interpret without the appreciation of the entire network of the inflammatory response. Hence, data in most of the studies was inconclusive and incomplete since none of them analyzed serum alterations of a considerable number of inflammatory components. Additionally, it is of high importance to note that unfortunately only few studied molecules (VWF, IL-1β, IL-6, ICAM-1, VCAM-1) exhibited independent discriminatory power \[[@B18],[@B44],[@B48],[@B49]\] and associated with the mortality of patients with high sensitivity and specificity \[[@B45]\]. Finally, only the minority of the studies \[[@B18],[@B19],[@B25]-[@B27],[@B43],[@B45],[@B47],[@B51],[@B53],[@B64],[@B66],[@B67],[@B83],[@B86],[@B95]\] clarified the effectiveness and the diagnostic accuracy of the biomarkers by applying ROC curve analysis which is essential to estimate the sensitivity and specificity of a marker and to introduce clinically practical cut-off levels for the prediction of ARDS development in at risk individuals. Collectively, these findings highlight the necessity for further investigations in the context of large prospective studies analyzing homogeneous and well defined group of ARDS or at risk patients and the assessment of novel molecules to serve as diagnostic and prognostic tools, as well as markers of the disease activity and severity. Conclusion ========== Currently, the application status in routine clinical practice for most of these biologic markers is still in its infancy and remains exploratory. Unfortunately, they do not yield independent indications for therapy or mark the end of the inflammatory process and their prognostic value still needs to be established. Although the majority of them have not yet lived up to the \"great hype\" that was generated, markers of endothelium activation and mostly VWF and adhesion molecules (ICAM-1, VCAM-1, E-selectin) show the greatest promise in ARDS and ALI. On the contrary the majority of serum cytokines and ferritin appear to be not ready for routine monitoring since they may reflect an inflammatory response to a risk factor rather than lung injury and disease severity. Additionally, lung specific proteins have proven to be neither specific nor sensitive for the prediction of ARDS development and the disease outcome and moreover they have failed to associate with alterations in the ventilatory strategies in large clinical trials. Further prospective investigations, technical improvements and introduction of novel markers are warranted in order to elevate the association of serum biomarkers with the pathogenesis of ARDS in the same status as for tumour markers with lung cancer. Nevertheless, crossing the boundary from research to clinical application requires validation in multiple settings, experimental evidence supporting a pathophysiologic role, and ideally intervention trials showing that modification improves the outcome. The emergence of pioneering technologies including DNA microarrays which have already been applied with great success in the respiratory research field \[[@B102]\] can help scientists to circumvent this problem and bridge this boundary. In the interim, these markers can be quite useful to supplement the clinical, radiological and physiological monitoring of the disease and identify high-risk patients who would benefit from aggressive management of established risk factors. List of Abbreviations ===================== Acute Lung Injury (ALI) Acute Physiology and Chronic Health Evaluation (APACHE) Acute Respiratory Distress Syndrome (ARDS) Bronchoalveolar lavage fluid (BALF) ICU: Intensive Care Unit ICAM-1: Intercellular Cell Adhesion Molecule-1 ILs: Interleukins ILDs: Interstitial Lung Diseases Krebs von den Lungen-(KL)-6 LTs: Leukotrienes MMPs (Metalloproteinases) MODS: Multi-Organ Dysfunction Score Receiver-operating-characteristic (ROC) Simplified Acute Physiological Score (SAPS) SIRS: Systemic Inflammatory Response Syndrome Soluble-E-selectin: s-E-selectin Soluble IL-2 receptor (sIL-2R) Surfactant protein-(SP) TNFa: Tumor Necrosis Factor-alpha VWF: von Willebrand factor VCAM-1: Vascular Cell Adhesion Molecule-1 Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AT, IP and DB were involved with the study conception. AT and IP performed the data acquisition and interpretation. DB was involved in revising the article for important intellectual content. All authors read and approved the final manuscript. Acknowledgements ================ The authors are grateful to Stavros Anevlavis (M.D) for his valuable assistance in collecting the data and revising the article.	PubMed Central	2024-06-05T03:55:58.272073	2005-6-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168906/", "journal": "Respir Res. 2005 Jun 22; 6(1):62", "authors": [ { "first": "Argyris", "last": "Tzouvelekis" }, { "first": "Ioannis", "last": "Pneumatikos" }, { "first": "Demosthenes", "last": "Bouros" } ] }
PMC1168907	Introduction ============ Cystic fibrosis (CF) is the most common serious, monogenic, autosomal recessive disease in Caucasians and results from mutations in the gene encoding the cystic fibrosis transmembrane conductance regulator (CFTR). CF is characterised by variable phenotypic expression, which is not entirely explained by the allelic heterogeneity of the pathogenic CFTR mutations, and there is accumulating evidence that much of this phenotypic diversity is due to the effect of modifier genes \[[@B1]\]. Pseudomonas aeruginosa, an environmental organism, is the most important pathogen in patients with CF as chronic infection results in a more rapid decline in lung function and reduced survival \[[@B2]\]. CD14, a key gene of the innate immune system, functions as a receptor for lipopolysaccharide (LPS), a constitutive element of the P.aeruginosacell wall, and is a potential modifier of severity in patients with CF. CD14 is expressed in both a membrane-bound form on macrophages, monocytes and neutrophils (mCD14) and soluble form in serum (sCD14). A polymorphism in the CD14 gene promoter (C-159T) has an allele frequency of approximately 50% in Europeans \[[@B3]\]. The -159C allele is associated with lower circulating levels of sCD14 in healthy children. Higher constitutive levels of mCD14 and sCD14 have been shown to increase the magnitude of airway neutrophil response to LPS \[[@B4]\], whereas CD14 receptor blockade results in a reduction in the deleterious systemic responses that occur in sepsis, due to a reduction in pro-inflammatory cytokines but at the expense of an increased bacterial load \[[@B5]\]. Hence, CD14 may play a pivotal role in determining the balance of infection and inflammation in CF. Increased expression of CD14 may be associated with more inflammation but a reduced bacterial load, whereas reduced expression may result in less inflammation but a greater bacterial load. We hypothesised that P.aeruginosainfection would be acquired earlier in children with the -159CC genotype. Methods ======= A prospective, population-based, cohort of children with CF was recruited from the only tertiary paediatric hospital in Western Australia, which cares for all children with CF in the state. Annual bronchoscopy, performed through a laryngeal mask to minimise the risk of upper airway contamination, was done routinely from diagnosis until 7 years of age in all children with CF. Children diagnosed after the neonatal period, whose first BAL was positive for P.aeruginosawere excluded from this analysis, as in this situation it could not be determined when they had initially acquired this pathogen. All children with a positive BAL culture for P.aeruginosa, which was defined as \>10,000 CFU/ml, were admitted to hospital for further treatment. Blood was collected for genetic studies and subjects were genotyped for the CD14 promoter polymorphism as previously described \[[@B3]\]. Plasma samples taken at the time of the earliest BAL, when the children were not infected with P.aeruginosa, was available for measurement of sCD14 levels in 31 children. Plasma sCD14 levels were determined using a commercially available enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems, Minneapolis, USA). The Kaplan-Meier survival method was used to explore the difference in the P.aeruginosafree survival rates between CD14 genotype groups. As children in the study were not the same age, data was censored for individual children who were known to have not isolated P.aeruginosaat the time of their last BAL and this information was incorporated into the analysis. Breslow test (Breslow Generalized Wilcoxon Test) was employed to compare the difference in the survival curves of acquisition of P. aeruginosain the three CD14 genotype groups. Multivariate Cox regression was used to estimate the relative risks after adjustment for potential confounding factors including age, sex, CFTR mutation and nutritional status. Plasma sCD14 levels were analysed by ANOVA and all the statistical analyses were performed using SPSS for Windows (Version 11). Parents of all participants gave informed consent and the Ethics Committee of King Edward Memorial and Princess Margaret Hospitals, Western Australia, approved the study. Results ======= Forty-five children (22 male), aged 0.6--6.6 years (mean 3.25 years) were studied, of whom 25/45 (55%) were DF508 homozygous and 20/45 DF508 heterozygous. CF was diagnosed at a mean age of 0.23 yrs (95%CI 0.06--0.4 yrs), with 73% diagnosed in the newborn period. There was no significant difference between CFTR mutation (p = 0.74), gender (p = 0.38), nutritional status (weight for height z scores) (p = 0.39) or socio-economic status (p = 0.29) in those who did and did not isolate P.aeruginosa. CD14 genotype frequencies at position -159 were: CC 12/45 (27%), CT 20/45 (44%), TT 13/45 (29%). P.aeruginosawas isolated from thirteen children (29%) and the mean age of acquisition was 2.26 years (95% CI 1.29--3.25). Subjects with CD14 -159CC appeared to isolate P.aeruginosaat a younger age (mean = 1.1 years, 95%CI = 0.2--1.9 years) than -159CT (mean = 2.8 years, 95%CI = 1.3--4.2 years) and TT subjects (mean = 3.3 years), although this difference was not statistically significant (p = 0.19). Figure [1](#F1){ref-type="fig"} shows the P.aeruginosafree survival curves of the three CD14 genotype groups. The probability of children remaining free of P.aeruginosawith -159CC is consistently below that of children who are CT or TT and the curve for children with CT is below that of children with TT. For example, at 2 years of age, the % children remaining uninfected with P.aeruginosais 55% vs. 82% vs. 100% for CC, CT and TT respectively. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Kaplan-Meier estimates of proportion of children free from P.aeruginosaby CD14 C-159T. ^1^Breslow Test (Breslow Generalized Wilcoxon Test) ^2^\"Censored\" represents the censored observations that arise when the duration of a study is limited. ::: ![](1465-9921-6-63-1) ::: Compared with -159TT, children with the CC genotype had a 10-fold (95%CI = 1.09--92.30, p = 0.042) higher relative risk and children with the CT genotype had an intermediate 5.5 fold (95%CI = 0.69--44.63, p = 0.108) higher relative risk of being infected with P.aeruginosa. After adjustment for CFTR mutation and nutritional status, the estimated relative risk in children with CC or CT increased further (RR = 13.32, 95%CI = 1.37--129.13, p = 0.025 and RR = 6.0, 95%CI = 0.71--51.03, p = 0.101 respectively). This suggested that independent of CFTR mutation and nutritional status, those with the C allele had a significantly increased risk of being infected with P.aeruginosa. In addition, there was a significant linear trend across the three genotype groups, between increasing numbers of C alleles and increasing likelihood of being infected with P.aeruginosa(p = 0.015). Compared to children who subsequently isolated P.aeruginosa, those who remained free of P.aeruginosahad significantly higher plasma sCD14 levels: 1339.43 μg/ml (95%CI 1096.63--1635.98 μg/ml) vs. 492.75 μg/ml (95%CI 55.7--4359.01 μg/ml), p = 0.018. We found no significant association between plasma sCD14 levels and CD14 C-159T (p = 0.38). Discussion ========== This study showed an association between CD14 -159CC and early acquisition of P. aeruginosain children with CF. Many CF centres promote aggressive treatment regimes to eradicate first isolates of P.aeruginosa, and, thus, prospectively identifying a \"high risk\" group of infants with CF could have substantial clinical benefit. CD14 receptor activation results in a strong Th1 cytokine response, directed through IL-12, in an attempt to eradicate pathogens. The finding of lower plasma sCD14 levels in children who subsequently isolate P.aeruginosasuggests that an inadequate pro-inflammatory response to LPS may place these individuals with CF at greater risk of early P.aeruginosaacquisition. The lack of association between CD14 C-159T and plasma sCD14 levels may merely reflect the small number of subjects in this population, but may also indicate that in CF there are factors more critical than CD14 C-159T genotype that determine plasma sCD14 levels. Although children with higher sCD14 levels may be relatively protected from earlier acquisition of P.aeruginosa, when they become colonised they may paradoxically have a worse outcome, as a result of an over-aggressive and ineffectual inflammatory response. These latter patients might potentially benefit from aggressive anti-inflammatory treatment. Thus, the timing of anti-inflammatory therapy would need to be carefully considered, as inflammation might be beneficial in early life by delaying colonisation with P.aeruginosa, but harmful once infection is established, due to the over-exuberant inflammatory response. Therapy that modulates the inflammatory response might promote the acquisition of pathogens such as P. aeruginosain young children with different genetic susceptibilities. The results of this study are based on a limited number of subjects due to the stringent inclusion criteria and the prospective longitudinal design. In addition, the potential role of other single nucleotide polymorphisms in genes involved in innate immunity such as Toll-like receptor 4 Asp299Gly \[[@B6]\], Toll-like receptor 2 Arg753Gln \[[@B6]\] and mannose binding lectin \[[@B7]\], may influence the acquisition of P.aeruginosain children with CF and more detailed investigation of these pathways in CF is indicated. This study identified the potential association between a common genetic variant and the early isolation of P.aeruginosa. The CD14 promoter polymorphism may have a central function in determining the age of first isolation of P.aeruginosa, with the CC genotype conferring a higher risk of early isolation and the TT genotype being relatively protective, emphasising the importance of understanding the delicate balance between inflammation and infection that exists in CF. Acknowledgements ================ We would like to thank the children and families who participated in this study and acknowledge the technical assistance of Elizabeth Balding and Samantha Gard. This study was supported by The National Health and Medical Research Council and the Australian Cystic Fibrosis Research Trust.	PubMed Central	2024-06-05T03:55:58.279679	2005-6-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168907/", "journal": "Respir Res. 2005 Jun 23; 6(1):63", "authors": [ { "first": "AC", "last": "Martin" }, { "first": "IA", "last": "Laing" }, { "first": "G", "last": "Zhang" }, { "first": "S", "last": "Brennan" }, { "first": "K", "last": "Winfield" }, { "first": "PD", "last": "Sly" }, { "first": "SM", "last": "Stick" }, { "first": "J", "last": "Goldblatt" }, { "first": "PN", "last": "LeSouef" } ] }
PMC1168908	Background ========== Obstructive airway diseases, including asthma, bronchiolitis, obstructive sleep apnea, and chronic obstructive pulmonary disease (COPD), are common in children and adults \[[@B1]-[@B7]\]. Early recognition and accurate assessment of the severity of airway obstruction and the response to therapy are fundamental to the improvement of health for patients with these disorders. However, objective measures of airway obstruction currently available in the Emergency Department (ED) and other acute care settings have significant limitations. Spirometry is frequently not available in acute clinical settings, including the ED. Peak expiratory flow rate (PEFR) has been demonstrated to progressively underestimate airway obstruction with increasing air trapping, making it less reliable as airway obstruction worsens \[[@B8]\]. As well, the ability of a patient with moderate to severe airway obstruction to generate an erroneously normal PEFR and the inability to measure PEFR in young children render this test less useful in the setting of an acute asthma exacerbation \[[@B8]\]. Further, both spirometry and PEFR require patient coordination and cooperation. Validated, objective measures to determine severity of airway obstruction in bronchiolitis are nonexistent \[[@B9]\]. The pulse oximeter plethysmograph waveform reflects dynamic net changes in arteriolar inflow and venous outflow of tissue bed capillaries interrogated by the oximeter light emitting diodes \[[@B10]-[@B12]\]. Indeed, the oxygen saturation output of the device (Sp0~2~) depends upon isolation of the oxygenated, arterialized light signal from those light signals representing tissue, venous blood and other chromophobes \[[@B13]\]. At levels of arterial oxygen saturation (Sa0~2~) approaching 100%, the waveform is derived almost entirely from the infrared (940 nm) signal determined by oxyhemoglobin concentration and arterialized flow. Because oxyhemoglobin concentration is constant, dynamic changes in the waveform are a result of arterialized flow change \[[@B13]\]. Under these conditions the waveform represents a plethysmograph, a device measuring change in volume, in this case change in volume of arterialized blood \[[@B11],[@B12],[@B14]\]. As such, the plethysmograph waveform has been demonstrated to correlate with radial artery Doppler waveforms \[[@B12]\]. Changes in the plethysmograph waveform might be useful clinically to estimate the severity of perturbations in physiologic events influencing arterial flow \[[@B10]\]. Certain pathologic conditions, most notably airway obstruction, influence these physiologic events and result in the phenomenon known as pulsus paradoxus \[[@B15]\]. Although pulsus paradoxus cannot be readily measured directly from the plethysmograph waveform, changes in plethysmograph waveform variables might nonetheless correlate with the physiologic perturbations characteristic of pulsus paradoxus and be useful in assessing the severity of physiologic alterations resulting from airway obstruction. Changes in waveform curve or baseline height, one-dimensional parameters, have been used to estimate pulsus paradoxus \[[@B16]-[@B18]\]. Pulsus paradoxus represents change in left ventricular stroke volume, a three-dimensional parameter. As a two-dimensional parameter, area under the curve may more accurately reflect the physiologic events resulting in pulsus paradoxus. Additionally, the contribution of diastolic blood pressure changes to pulsus paradoxus have been noted, and AUC measurement might more completely and accurately incorporate these events \[[@B17],[@B19]\]. Finally, a general principle of signal analysis maintains that the signal-to-noise ratio improves at a rate proportionate to the square root of the number of data points obtained \[[@B20]\]. Area under the curve data may therefore be less prone to noise artifact than height data, and might provide a more optimal signal to noise ratio. With this in mind, changes in area under the waveform curve might represent a more accurate measure of waveform variability than changes in waveform height. Indeed, Hartert and colleagues have suggested evaluation of area under the waveform baseline during the respiratory cycle, rather than baseline height change, as a more accurate measurement of waveform variation\[[@B18]\]. There are limited data on the levels of intrapleural pressure generated in the presence of most obstructive airway diseases. However, levels of intrapleural pressure generated in adults in severe status asthmaticus have been demonstrated to be (-)24.4 ± 6.5 cmH~2~0 on inspiration and (+)7.6 ± 6.0 cmH~2~0 on expiration \[[@B15]\]. Mouth pressure reflects intrapleural pressure within 4 cmH~2~0 \[[@B21]\]. In this study our primary objective was to determine whether maximal change in area under the pulse oximeter plethysmograph waveform curve correlates with the degree of experimentally applied airway obstruction across a range of mouth pressures up to these levels of obstruction. A secondary objective was to determine whether maximal changes in height of the plethysmograph waveform curve similarly correlate with the degree of airway obstruction. Methods ======= Study Setting and Population ---------------------------- The study was approved by the University of Alabama at Birmingham Institutional Review Board as an expedited study. Informed written consent was obtained from each subject prior to enrollment. This study was conducted in the Pulmonary Function Laboratory of an urban children\'s hospital. A prospective convenience sample of healthy young adult subjects, twenty years of age and above, were recruited. Subjects with doctor-diagnosed asthma, a history consistent with asthma, or either FEV~1~or FEV~1~/FVC less than 80% predicted, were excluded from this study. The subjects underwent spirometry, performed by certified pulmonary function technicians according to American Thoracic Society protocol \[[@B22],[@B23]\]. Study Design and Protocol ------------------------- We utilized a closed airway circuit to generate airway obstruction, consisting of a Hans Rudolph 2600 two-way non-rebreather valve assembly with adjustable spring-loaded inspiratory and expiratory pressure relief valves and a mouthpiece pressure transducer port (Hans Rudolph, Kansas City, MO). Our experimental method was to allow each subject to experience increasing levels of inspiratory and expiratory airway obstruction corresponding to the levels of mouth pressure and to the estimated levels of intrapleural pressure noted previously \[[@B15],[@B21]\]. The pressure relief valves were adjusted accordingly at a minimum of five intervals and a maximum of ten intervals, to provide progressively increasing levels of mouth pressure from approximately (-)15 to (-)26 cmH~2~0 on inspiration and (+)2 to (+)9 cmH~2~0 on expiration. Each subject was allowed to rest for a minimum of one minute before testing at the subsequent, increased level of applied resistance in order to allow the plethysmograph waveform to return to baseline. Pulse oximeter plethysmograph waveform data was acquired for 60 seconds at each level of applied obstruction. Plethysmographic waveforms were acquired with a BioPac MP150 data acquisition system using a TSD123A transducer and an OXY100C pulse oximeter module (BioPac Systems, Santa Barbara, CA). This apparatus utilizes optical transmission at red (660 nm) and infrared (940 nm) wavelengths and employs Novametrix Medical Systems, Inc. artifact rejection and averaging algorithms that use an eight second pulse history signal to output Sp0~2.~The algorithm averages signal only for Sp0~2~calculation \[[@B24]\]. Plethysmograph waveform signal was acquired, processed and analyzed without averaging, smoothing or filtering. Mouth pressure waveforms were acquired with a BioPac TSD160C transducer. Transducers were calibrated according to manufacturer protocol. Waveform data were analyzed with BioPac Acknowledge software (version 3.7.2). The software algorithm calculates area under the curve (AUC) as the area encompassed by a waveform from the point of deflection from baseline to the point of return to baseline, and calculates height (HT) as height from the point of deflection from baseline to the waveform peak. Each subject was studied in the sitting position. A nose clip was applied, and the subject was instructed to exclusively mouth breathe through the airway circuit at a respiratory rate of approximately 10--16/min and at normal to slightly increased inspiratory and expiratory effort. Data were acquired at progressively increasing levels of applied inspiratory and expiratory obstruction for approximately 60 seconds at each level. Data collection and processing ------------------------------ Physiologic perturbations occurring during the respiratory cycle, such as airway obstruction, result in alterations of arterial flow and the phenomenon known as pulsus paradoxus \[[@B15]\]. It is these dynamic changes in arterial flow that we hypothesize might allow estimation of airway obstruction from oximeter plethysmograph waveform changes. Timing the measurement of these changes with the respiratory cycle is difficult in the clinical environment because patients with these disorders often have rapid respiratory rates. For this reason we chose to analyze data during specified time intervals. In order that at least one complete respiratory cycle and the corresponding maximum and minimum mouth pressure be included in each interval, the interval so chosen was 7.5 seconds. Data extracted for each 7.5 second interval consisted of maximum and minimum waveform area under the curve, maximum and minimum waveform height, and maximum and minimum mouth pressure. Maximum change in area under the curve and height were calculated as the difference between the maximum and minimum values of each parameter divided by the maximum value of the respective parameter during the specified 7.5 second interval. Maximum change in mouth pressure was calculated for the corresponding interval as the absolute difference between the maximum and minimum mouth pressure in cmH~2~0. These data were acquired using the Acknowledge software and entered into a spreadsheet program (Excel, Microsoft, Redmond, WA). Using the Excel formula function, mean values for maximal change in area under the curve, height, and mouth pressure for each level of applied obstruction were calculated from the multiple sequential 7.5-second intervals at the corresponding level of obstruction. This data was then entered into a statistical analysis program (SAS^®^v9.0, Cary, NC.) for analysis \[[@B25]\]. Outcome Measures ---------------- The primary outcome measure was the correlation of mean maximum change in area under the plethysmograph waveform curve with mean maximum change in mouth pressure at each successive level of applied obstruction. The secondary outcome measure was the corresponding correlation using mean maximum change in height. Data Analysis ------------- Subjects in this study contributed multiple observations to the dataset. Because of this the fundamental assumption of independence across observations was violated. Performing a separate analysis for each subject would reduce the number of observations in each analysis and increase the potential for Type II errors. On the other hand, if all of the observations were analyzed as independent, ignoring the inherent clustering within subjects, then the potential for Type 1 errors would increase. We utilized a repeated measures model that takes into account the clustering and correlation between subjects. In this analysis, the PROC MIXED procedure in SAS^®^was used to model the relationship between maximum change in area under the curve and maximum change in mouth pressure as well as the relationship between maximum change in height and maximum change in mouth pressure. Each subject contributed a single data point for each level of applied obstruction, representing the average of the 7.5-second intervals for that level of applied obstruction. Akaike\'s Information Criteria was used to compare the fit of the area under the curve vs. height models for mouth pressure \[[@B25]\]. An alpha level of p \< .05 was considered statistically significant. A total sample size of 30 subjects would allow us to construct a 95% CI for correlation and achieve a power of 0.8 and a two-tailed alpha of 0.05. Results ======= Forty-eight subjects were enrolled in the study; no subject experienced any known adverse event during or as a result of this study. Two subjects were found after enrollment to have asthma and were excluded from data analysis. Eight subjects experienced an uncomfortable sensation of dyspnea and could not use the closed airway circuit in accordance with study protocol. Data from these subjects was excluded from analysis. Seven subjects had recurrent electrical interference of the waveform baseline, the source of which could not be determined after consultation with software and hardware engineers (BioPac Systems, Santa Barbara, CA). Data from these seven subjects was excluded from analysis. Overall thirty-one subjects met inclusion criteria and had data included for analysis. Of these thirty-one subjects, eleven were male and twenty were female. The mean age was 29.9 years with a median of 28 years and range of 23 to 48 years. One subject had a prior history of cigarette smoking. No subject had heart or lung disease. One subject performed breathing maneuvers at five levels of applied obstruction, one subject at eight levels, six subjects at nine levels, and twenty-three subjects at ten levels. A total of 297 data points were available for analysis. Plethysmograph waveforms were noted to return to baseline during the period of rest (at least 1 minute) between sequentially increasing levels of applied resistance. Subjects were noted to generate plethysmograph waveforms visually significant for periodic changes with the respiratory cycle, similar to changes characteristic of pulsus paradoxus, when utilizing this apparatus (Figure [1](#F1){ref-type="fig"}). There was a significant relationship between plethysmograph waveform maximum change in area under the curve and maximum change in mouth pressure (P \< 0.0001) (Figure [2](#F2){ref-type="fig"}). The prediction equation for each cmH~2~0 maximum change in mouth pressure was 12.01 + 37.21 × (maximum change in area under the curve), 95% CI for coefficient = 30.56 to 43.87. Similarly, there was a significant relationship between maximum change in height and maximum change in mouth pressure (P \<0.0001). The prediction equation for each cmH~2~0 maximum change in mouth pressure was 16.10 + 35.94 × (maximum change in height), 95% CI for coefficient = 27.57 to 43.30. A comparison of Akaike\'s Information Criteria (AIC) between the models showed that the AIC statistic was smaller for the area under the curve model than the height model, indicating a better model fit for the area under the curve model. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Oximeter plethysmographic waveform (Pleth) generated with inspiratory and expiratory pressure relief valve apparatus. Corresponding mouth pressure indicates pressure at airway circuit mouthpiece. ::: ![](1465-9921-6-65-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Relationship between maximum changes in mouth pressure and area under the plethysmograph waveform curve. ::: ![](1465-9921-6-65-2) ::: Discussion ========== Pulse oximetry is widely available and applied in acute care settings. The device outputs a continuous plethysmographic waveform corresponding to flow of arterialized blood in the tissue bed to which the transducer is applied \[[@B10],[@B12]-[@B14]\]. It is plausible that, in the setting of airway obstruction, such changes in arteriolar flow might reflect alterations in left ventricular stroke volume resulting from the same physiologic perturbations that abnormally increase pulsus paradoxus. It is thus of interest whether the severity of airway obstruction might be estimated from changes in mathematic plethysmograph waveform variables. The study results indicate a correlation between maximum changes in area under the curve or in height of the plethysmograph waveform and the severity of airway obstruction. Analysis of both direct arterial waveform and oximeter plethysmograph waveform data for calculation of arterial flow have previously been explored in the laboratory setting. Cerutti and colleagues provide compelling data from conscious, freely moving Sprague-Dawley rats \[[@B26]\]. These investigators compared different models of central arterial line waveform analysis with simultaneously recorded cardiac output. A model using different waveform parameters identified by multiple linear regression analysis provided a reliable and precise estimation of cardiac output. Although these investigators did not use oximeter plethysmograph waveforms, their findings nonetheless support the principal of waveform analysis. Steele and colleagues performed an unblinded study on one healthy adult, breathing through a valve to which airway resistance was applied to artificially induce pulsus paradoxus. For this subject, the percent decrease in plethysmograph waveform height during the respiratory cycle correlated modestly with pulsus paradoxus calculated similarly from intra-arterial waveform (r = 0.59, 95% CI 0.32 to 0.78). This study was limited by the small subject size (n = 1) and did not measure the degree of airway obstruction generated by the resistance valves in use. The technique relied upon determination of phases of the respiratory cycle and capture of waveform indices in accordance with estimated peak inspiration and expiration \[[@B16]\]. In the clinical setting, variation of the oximeter plethysmograph waveform baseline has been noted to occur during the respiratory cycle and to represent fluctuations in local venous pressure \[[@B14],[@B18]\]. Hartert and colleagues hypothesized that this respiratory waveform variation might occur in response to pleural pressure changes and thus reflect changes in left ventricular stroke volume and pulsus paradoxus. This was studied in adult patients admitted to an ICU with obstructive airway disease, 46% of whom were receiving mechanical ventilation. Respiratory waveform variation was significantly correlated with manually measured pulsus paradoxus (R^2^= 0.88) as well as with auto-PEEP (R^2^= 0.96) \[[@B18]\]. Frey and Butt compared simultaneous 1 minute paper recordings of intra-arterial pressure and plethysmograph waveforms in 62 non-intubated children with and without respiratory disease. Correlation was noted (r = 0.85) between changes in plethysmograph waveform height and pulsus paradoxus determined from intra-arterial waveform height change \[[@B17]\]. Our study demonstrates that maximal change in height and in area under the plethysmograph waveform curve might provide a non-invasive, clinically relevant estimate of the severity of airway obstruction. A possible limitation to our study was the method of artificially inducing airway obstruction. The dynamic biomechanical changes occurring during an asthma exacerbation are not ideally simulated by externally applied resistance \[[@B27]\]. Also, in lieu of invasive, intra-arterial waveform analysis as the dependent variable and reference standard, the study protocol utilized change in mouth pressure as a surrogate measure of obstruction induced. The levels of progressive obstruction were not standardized, except insofar as the mouth pressure generated reflects intrapleural pressure \[[@B21]\]. As well, subjects were exposed to both inspiratory and expiratory obstruction during the test period. It is of interest whether correlations of waveform parameters may differ during isolated inspiratory or expiratory obstruction. Other variables that may influence the plethysmograph waveform, including hydration status, hyperinflation, and tidal volume, were likewise not controlled for in this study. Our method of using time intervals to measure changes in plethysmograph waveform AUC, HT and mouth pressure is unique. Pulsus paradoxus has traditionally been determined by noting the difference between the systolic pressure at which heart sounds are heard only during expiration and the point at which they are heard continuously \[[@B28]-[@B30]\]. However, in the tachypneic patient it is often difficult to correlate auscultation of heart sounds with the corresponding phase of the respiratory cycle. With this in mind, we chose to analyze data during specified time intervals that would encompass at least one respiratory cycle. The chosen interval, 7.5 seconds, was based upon the expected duration of the respiratory cycle in our subjects. We additionally chose to utilize the average values of data extracted from sequential intervals at each level of applied obstruction. Frey and Freezer demonstrated significant intrasubject variation of breath-to-breath measurement of pulsus paradoxus utilizing arterial waveform tracings, and averaging of pulsus paradoxus determined from multiple consecutive respiratory cycles was reported to be more accurate \[[@B19]\]. Pulse oximeters have incorporated an analogous technology for calculation of Sp0~2~, running weighted signal averaging, to minimize the effect of signal artifact and to thus enhance the reliability and validity of the calculated Sp0~2~\[[@B13]\]. Oxygen saturation is calculated 30 times per second with values averaged over a minimum of several seconds. Each instantaneous value is first compared with this moving average and assigned a weighted value based upon variation from the moving average. This weighted value then contributes to the moving average that in turn is displayed as the Sp0~2~value \[[@B13]\]. Our analysis may minimize the influence of individual waveform and respiratory cycle artifact and thus enhance the internal validity of the estimated airway obstruction. With these elements of waveform analysis in mind, our method of measuring waveform parameters may represent a strength of study design rather than a limitation. Conclusion ========== There is accumulating evidence that the plethysmograph waveform might provide clinically useful information. Our results suggest that analysis of oximeter plethysmograph waveform data may be feasible for real-time estimation of airway obstruction. To our knowledge this is the first investigation of area under the curve as a waveform parameter of potential value, and our results indicate that this parameter may achieve better correlation with airway obstruction than analyses based on waveform height. A non-invasive, real-time method to estimate the severity of airway obstruction, as well as other disorders involving pulsus paradoxus physiology, might enhance the ability of clinicians to identify and quantify the severity of such disorders \[[@B31]\]. An essential step in the development of such technology is to validate the physiologic relevance of estimating the severity of these pathophysiologic events from the oximeter plethysmograph waveform. Future study of patients with obstructive airway disease in the clinical environment, using a quantifiable, objective criterion standard such as FEV~1~will enable further assessment of oximeter plethysmograph waveform parameters to predict severity of airway obstruction. Should the accuracy and feasibility of such a tool be demonstrated in the clinical environment, development of this technology for routine clinical practice may be justified. Competing interests =================== Don Arnold has applied for patent protection for methods of waveform analysis discussed in this manuscript. Authors\' contributions ======================= DA was the principal investigator and participated in study concept and design, acquisition of the data, drafting of the manuscript and obtained institutional funding for this study to be conducted. DS was a co-investigator and participated in study concept and design, acquisition of the data, drafting of the manuscript and critical revision of the manuscript for important intellectual content. RD assisted in the statistical design and analysis and interpretation of the data, and provided critical revision of the manuscript for important intellectual content. JH participated in study concept and design, acquisition of the data, drafting of the manuscript, critical revision of the manuscript for important intellectual content, and supervised the study. Grants ====== This study was funded by a grant from The Research Institute at The Children\'s Hospital of Alabama. Acknowledgements ================ The authors are gratified for the assistance of Sheila S. Gibson, R.R.T., R.P.F.T. Johanna Kimbrough, R.P.T., and Bettye Mitchell, R.P.T. in the conduct of this study.	PubMed Central	2024-06-05T03:55:58.280752	2005-6-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168908/", "journal": "Respir Res. 2005 Jun 28; 6(1):65", "authors": [ { "first": "Donald H", "last": "Arnold" }, { "first": "David M", "last": "Spiro" }, { "first": "Renee' A", "last": "Desmond" }, { "first": "James S", "last": "Hagood" } ] }
PMC1168909	Background ========== Despite the standardization of the operative technique, improvement of preoperative risk assessment, and postoperative intensive care management, surgical therapy for esophageal carcinoma continues to be associated with a high incidence of operative complications, and a high mortality rate. Risk stratification in the selection of patients for surgery, and the choice of the surgical procedure are therefore important considerations. The present study was aimed to identify preoperative and intraoperative factors that could potentially influence morbidity and mortality after esophagectomy for esophageal carcinoma. Patients and methods ==================== Between September 1985 and March 2004, a total of 424 patients underwent esophagectomy for esophageal carcinoma in the Department of General and Abdominal Surgery of the Johannes Gutenberg-University Hospital Mainz. Of these, 186 (43.9%) patients had transhiatal esophagectomy, 231 (54.5%) patients underwent abdominothoracic esophagectomy with two-field lymphadenectomy (abdominal and mediastinal) and 7 patients had a proximal esophageal resection with a free jejunal graft. Reconstruction was accomplished by gastric tube pull-up in 384 (93.0%) patients, by colon interposition in 21 (5.1%), and small intestine interposition in 7 (1.7%) patients; the anatomic prevertebral esophageal bed was used for the majority of these procedures. Extra-anatomic reconstruction by the retrosternal route with cervical anastomosis after pull-up was carried out in patients with a high risk for loco-regional recurrence only (n = 42; 10.2%). Abdominothoracic esophagectomy was routinely performed for squamous cell carcinoma (n = 229). A transhiatal procedure was selected for tumors with a distal location, for malignancies without esophageal wall penetration, or in the presence of a high general risk. Transhiatal esophagectomy with abdominal and posterior mediastinal lymphadenectomy was carried out for the majority of adenocarcinomas (n = 150), the two-field procedure was performed in the presence of advanced tumor growth or extended lymph node involvement. Long-term results of this choice of the operative procedure adjusted to the histological tumor type had shown a significant prognostic advantage in patients undergoing transthoracic compared to transhiatal resection in squamous cell carcinoma whereas there was no survival benefit in patients with adenocarcinoma of the esophagus \[Junginger T, et al.; Gockel I et al.; unpublished data\]. Data were collected prospectively in a specially established database. Preoperative and intraoperative variables as well as postoperative morbidity and mortality were documented, in addition to routine demographic data. The following preoperative risk factorswere recorded: ASA-classification (I-IV) according to the preoperative anesthesiology evaluation, BMI (Body Mass Index) based on body weight and height in kg/m^2^, and the nutritional status on a scale from 0 (= no alcohol or tobacco consumption), 1 (= tobacco alone), 2 (alcohol alone) to 3 (= combined nutritional risk with tobacco and alcohol use). Among the preoperative diseases, cardiovascular risk factors were defined as a history of coronary heart disease, or myocardial infarction, arterial hypertension, valvular disease (\>II°), arrhythmia requiring therapy (\>III° according to the Lown-classification), congestive heart failure NYHA (New York Heart Association) \> Grade II, and peripheral occlusive arterial disease (\>IIb according to Fontaine). A history of chronic obstructive pulmonary disease (COPD), regular tobacco consumption and/or the use of bronchodilators were subsumed under pulmonary diseases. The preoperative assessment of the vital capacity (VC) and forced expiratory volume (FEV~1~= Tiffeneau-test) served to ensure a more accurate assessment. Prior cirrhosis of the liver (\>/= CHILD-Pugh A) was defined as hepatic disease, and determined on the basis of the assessment of serum albumin (g/dl), serum bilirubin (mg/dl), Quick-value (%), and the presence of ascites or encephalopathy. The evaluation of additional risk factors included the prevalence of diabetes mellitus (insulin-dependent or requiring oral antidiabetic therapy), and the history of a secondary carcinoma. Intraoperative variables included in the risk analysis were tumor location, the operative procedure, and the transfusion requirement of packed erythrocytes. The group of tumor characteristics comprised tumor size, R-classification, TNM-stage, and the number of dissected lymph nodes. Postoperative variables were not part of this risk analysis. Morbidity, surgical (anastomotic leakage, graft necrosis, mediastinitis, recurrent laryngeal nerve paralysis, chylothorax, tracheal fistula, bleeding requiring reoperation) and general complications (pneumonia, atelectasis, ARDS i.e. adult respiratory distress syndrome, myocardial infarction, cardiac failure, pulmonary embolism, renal insufficiency, pancreatitis, deep vein thrombosis), length of intensive care unit stay, 30-day mortality, and the mortality rate served as factors for assessment. Statistical analysis -------------------- The SSPS 10.0 software package was used for statistical data analysis (SSPS, Chicago, IL, USA: 1999). Data are expressed as median with ranges (minimum -- maximum), or as percentages (%). Factors with a possible influence on perioperative morbidity and mortality were calculated using the logistic regression model (univariate and multivariate). The χ^2^test with Pearson\'s correction and Fisher\'s exact test were used for comparison of the parameters for squamous cell carcinoma and adenocarcinoma. The Mann-Whitney U test served as the non-parametric method. A p-value of \<0.05 was considered statistically significant for all procedures. Results ======= Preoperative and intraoperative parameters, tumor characteristics ----------------------------------------------------------------- Median patient age at the time of surgery was 58 (28--84) years; the proportion of males was 83.0%. The median BMI (Body Mass Index) was 24.4 (13.8--39.3) kg/m^2^. A total of 36.9% patients were classified as ASA grade II, 58.2% as ASA grade III, and 4.9% as ASA grade IV. A combined nutritional risk (alcohol and tobacco use) was determined in 41.5% of patients. Tobacco use alone was found in 59.1%, and habitual alcohol consumption in 58.2% of patients. Preexisting cardiovascular diseases were noted in 28.2%, pulmonary disease in 13.5%, hepatopathy in 4.9%, and diabetes mellitus in 5.9% of patients undergoing surgery for esophageal carcinoma. The history of 9.8% of all patients showed the presence of secondary carcinoma. Median preoperative vital capacity (VC) ranged at 3.8 (1--7.2) l, FEV~1~(Tiffeneau-test) was 2.9 (0.5--9.0) l/sec. Out of all tumors, 56.3% were located in the lower, 34.6% in the middle, and 9.1% in the upper third of the esophagus. Squamous cell carcinoma was identified in 55.3%, adenocarcinoma in 35.9%, and an undifferentiated carcinoma in 7.6% of patients (1.2% with other malignant esophageal tumors, e.g. melanoma) who underwent esophagectomy. Operative time ranged at 300 (160--560) minutes. The median number of packed erythrocyte units used was 1 (0--38) (54.6% of all patients did not require packed erythrocyte transfusion). Median length of intensive care therapy was 10 (0--176) days, at a total postoperative hospital stay of 22 (0--189) days. Median tumor size was 4 (0.3--20) cm. A R0 resection was accomplished in 81.0% of all patients (R1 resection: 16.0%, R2 resection: 3.0%). Based on results of the pathological examination, the majority of tumors were assigned to the T3 category (57.5%), while 34.6% of all tumors were allocated to the T1 or T2, and 7.9% to the T4-category. Distribution for the N category was as follows: N0: 34.0%, N1: 66.0%. A M1-situation (M1-lymph node or M1-organ) was identified in 26.5% of all patients, which was treated with curative intent in the course of the same surgical procedure. Morbidity and mortality: Prevalence and influence factors --------------------------------------------------------- There was a 35.5% prevalence of surgical complications in the total population of 424 patients undergoing esophagectomy for esophageal carcinoma. General complications were observed in 36.0% of all patients. The 30-day mortality rate was 6.7%, at a mortality rate for the entire observation period of 11.5%. Anastomotic leakage was the most common surgical complication (18.2%), followed by recurrent laryngeal nerve paresis (15.7%). There was a similar incidence of graft necrosis (3.2%) and postoperative hemorrhage (2.9%) requiring surgical revision, while 1.3% of all patients developed a chylothorax. Univariate analysis showed tumor location (p = 0.0194) and the surgical procedure (p = 0.0116) to be significantly related to the incidence of surgical complications; none of the other pre- and intraoperative factors, including tumor characteristics, were of relevance in this model (Table [1](#T1){ref-type="table"}). Multivariate analysis identified the transthoracic surgical procedure (p = 0.0004), tumor location (upper third) (p = 0.02), and the transfusion of packed erythrocytes (p = 0.04) as factors significantly related to the incidence of surgical complications (Table [2](#T2){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Factors with a potential influence on surgical and general complications, as well as on the perioperative mortality rate after esophagectomy for esophageal carcinoma: univariate analysis (p values). ::: Surgical complications General Complications Perioperative Mortality --------------------------------------- ---------------------------- --------------------------- ----------------------------- [Preoperative factors:]{.underline} \- Age 0.8996 0.0073\* 0.1349 \- Sex 0.7367 0.3396 0.5658 \- BMI 0.3710 0.1200 0.0719 \- ASA 0.3992 0.0076\* 0.0274\* \- Nutritional status 0.5093 0.1897 0.2223 \- Cardiovascular PD 0.7803 0.0171\* 0.0172\* \- Pulmonary PD 0.2910 0.0405\* 0.0059\* \- Hepatopathy 0.1696 0.0282\* 0.0165\* \- Diabetes mellitus 0.8218 0.0635 0.9430 [Intraoperative factors:]{.underline} \- Tumor location 0.0194\* 0.0156\* 0.0391\* \- Surgical procedure 0.0116\* 0.0001\* 0.0057\* \- Packed erythrocyte transfusion 0.1045 0.0006\* 0.0022\* [Tumor characteristics:]{.underline} \- Tumor size 0.5041 0.1595 0.1502 \- R-classification 0.5573 0.3372 0.0439\* \- T-category 0.2017 0.0072\* 0.0036\* \- N-category 0.6805 0.9928 0.0584 \- M-category 0.0621 0.8128 0.4359 \statistically significant ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Significant factors with a potential influence on surgical and general complications, as well as on the perioperative mortality rate after esophagectomy for esophageal carcinoma: multivariate analysis (p value). ::: Surgical complications* General complications Perioperative Mortality --------------------------------------------- --------------------------------------------- --------------------------------------------- Surgical procedure (p = 0.0004) Surgical procedure (p = 0.0001) Surgical procedure (p = 0.0068) Tumor location (p = 0.0204) Packed erythrocyte transfusion (p = 0.0009) Pulmonary PD (p = 0.0096) Packed erythrocyte transfusion (p = 0.0493) Patient age (p = 0.0039) Packed erythrocyte transfusion (p = 0.0099) Nutritional status (p = 0.0284) ASA-classification (p = 0.0486) ::: Pulmonary complications were most common among the general complications (32.9%). Adult respiratory distress syndrome (ARDS) occurred in 6.5% of patients. Postoperatively, 3.7% of patients developed renal failure requiring dialysis, and myocardial infarction or pulmonary embolism occurred in 1.2%, respectively, of all patients. Postoperative pancreatitis was found in 0.5% of patients. Univariate analysis determined patient age (p = 0.0073), ASA-classification (p = 0.0076), preexisting cardiovascular (p = 0.0171), pulmonary (p = 0.0405), and hepatic (p = 0.0282) disease as significant preoperative variables in relation to postoperative general complications. Similarly significant among the intraoperative factors were tumor location (p = 0.0156), the surgical procedure (p = 0.0001), transfusion of packed erythrocytes, as well as the T-category in the group of tumor characteristics (Table [1](#T1){ref-type="table"}). On multivariate analysis, independent parameters with a significant influence on postoperative general complications were in decreasing order: the surgical procedure (p = 0.0001), transfusion of packed erythrocytes (p = 0.0009), patient age (p = 0.0039), nutritional status (p = 0.0284), and ASA-classification (p = 0.0486) (Table [2](#T2){ref-type="table"}). The cause of postoperative mortality (11.5%; n = 49 patients) in 19 patients was sepsis (accounted for by anastomotic insufficiency in 8, and in 3 patients by graft necrosis). Twelve patients died from pulmonary failure, and 5 patients died after myocardial infarction. Fulminant pulmonary embolism was the cause of death in 3 patients (one out of these occurred in a patient with suture dehiscence). Two patients developed hemorrhagic shock, and another 2 patients died in hospital from rapid progression of the malignant disease. Further causes of hospital death in 6 additional patients after esophagectomy were; hepatorenal syndrome, syncopal attack with cardiovascular failure, right ventricular heard failure, medial cerebral infarction, intraoperative cardiac arrest, and tracheogastric fistula. Preoperative variables as predictors of a fatal postoperative course found on univariate analysis comprised ASA-classification (p = 0.0274), preexisting cardiovascular (p = 0.0172) and pulmonary disease (p = 0.0059), and hepatopathy (p = 0.0165). Further factors exerting an influence on mortality were: tumor location (p = 0.0391), the surgical procedure (p = 0.0057), packed erythrocyte transfusion (p = 0.0022), R-classification (p = 0.0439), and T-category (p = 0.0036) (Table [1](#T1){ref-type="table"}). Multivariate analysis identified the surgical procedure (p = 0.0068), preexisting pulmonary disease (p = 0.0096), and the transfusion of packed erythrocytes (p = 0.0099) as the most significant predictors of mortality (Table [2](#T2){ref-type="table"}). Risk profile: comparison between squamous cell carcinoma and adenocarcinoma --------------------------------------------------------------------------- From among the group of preoperative factors, patients with adenocarcinoma were characterized by significantly more advanced age and a higher BMI (p \< 0.0001) compared to those with squamous cell carcinoma. Differences in gender distribution and ASA-classification were not significant. There was a significantly higher nutritional risk (alcohol and tobacco use) in patients with squamous cell carcinoma than in the comparison group of patients with adenocarcinoma undergoing esophagectomy (p \< 0.001). The incidence of cardiovascular and pulmonary disease, as well as preexisting diabetes mellitus, or secondary carcinoma in the patient history was similar in both groups. However, patients with squamous cell carcinoma had significantly poorer preoperative lung function (p = 0.0078), and a higher prevalence of hepatopathy (p \< 0.0001) (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of the preoperative risk profile between patients with squamous cell carcinoma and adenocarcinoma of the esophagus. ::: Parameter Squamous cell carcinoma (n = 229) Adenocarcinoma (n = 150) p-value ---------------------------- --------------------------------------- ------------------------------ ------------- \- Age (years) 56.0 (29--84) 61.0 (28--78) 0.0001\* \- Sex (% males) 79.8 87.4 0.071 \- BMI (kg/m^2^) 23.2 (13.9--34.9) 25.8 (15.8--39.3) 0.0001\* \- ASA-classification (%) -II 33.2 43.2 -III 62.4 52.1 0.072 -IV 4.4 4.8 \- Nutritional risk (%) -No nutritional risk (%) 16.2 43.2 0.0001\* -Alcohol use (%) 69.5 42.9 0.0001\* -Tobacco use (%) 73.5 38.1 0.0001\* -Combined risk (%) 83.8 56.9 0.0001\* \- Cardiovascular PD (%) 30.4 24.7 0.240 \- Pulmonary PD (%) 14.7 14.4 0.593 -VC (l) 3.7 (1.0--6.6) 3.8 (1.9--6.1) 0.3063 -FEV 1 (l/sec) 2.8 (0.5--5.2) 3.1 (1.3--5.9) 0.0078\* \- Hepatopathy (%) 8.1 1.4 0.0001\* \- Diabetes mellitus (%) 4.9 7.5 0.369 \- Secondary carcinoma (%) 11.2 8.2 0.382 \statistically significant ::: The surgical procedure used was different with respect to the significantly more frequent performance of transhiatal esophagectomy in patients with adenocarcinoma (p \< 0.0001). In the presence of this histological subtype, the gastric tube (p = 0.027) brought up in the existing esophageal bed (p = 0.001) was used more often as the interposed organ. The number of dissected abdominal lymph nodes was significantly higher (p = 0.0146) for adenocarcinoma, while a significantly higher number of thoracic lymph nodes (p = 0.0011) was removed in patients with squamous cell carcinoma. The operative time was significantly shorter (p \< 0.0001) at a lower packed erythrocyte transfusion requirement for adenocarcinoma (p = 0.0495) compared to squamous cell carcinoma (Table [4](#T4){ref-type="table"}). There was no significant difference between the two groups with regard to UICC stages and tumor characteristics (p \> 0.05). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Intraoperative Factors: Comparison between squamous cell carcinoma and adenocarcinoma of the esophagus. ::: parameter* Squamous cell carcinoma (n = 229) Adenocarcinoma (n = 150) p-value ------------------------------------------------ --------------------------------------- ------------------------------ ------------- \- Surgical procedure (% transhiatal) 30.6 68.7 0.0001\* \- Esophageal substitute (% gastric tube) 90.8 97.3 0.027\* \- Repositioning (% esophageal bed) 83.7 95.3 0.001\* \- Number of removed abdominal LN (n) 11 (0--55) 13 (0--51) 0.0146\* \- Number of removed thoracic LN (n) 11 (0--47) 6 (0--83) 0.0011\* \- Operative time (min) 305 (115--560) 270 (160--540) 0.0001\* \- Intraoperative blood loss (ml) 1000 (200--5000) 800 (0--7500) 0.2224 \- Units of transfused packed erythrocytes (n) 1.5 (0--38) 0 (0--14) 0.0495\* \statistically significant ::: Although there was no significant difference in surgical complications between the two groups, the rate of general complications (p = 0.012), and thus the total complication rate, (p = 0.039) was significantly higher in patients with squamous cell carcinoma. Comparable to the morbidity rate, the duration of postoperative intensive care therapy was significantly longer (p \< 0.0001) for this type of tumor than for adenocarcinoma. The 30-day mortality rate was identical for both groups, while the mortality rate in patients with squamous cell carcinoma was significantly higher (p \< 0.0001) (Table [5](#T5){ref-type="table"}). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Postoperative morbidity and mortality for squamous cell carcinoma compared with adenocarcinoma of the esophagus. ::: parameter* Squamous cell carcinoma (n = 229) Adenocarcinoma (n = 150) p-value ------------------------------- --------------------------------------- ------------------------------ ------------- \- Surgical complications (%) \- General complications (%) 37.9 33.7 0.447 -Total complications (%) 40.7 28.0 0.012\* \- Length of ICU -- stay (d) 68.1 57.6 0.039\* \- 30-day-moratlity (%) \- Mortality (%) 11 (2--176) 8 (1--107) 0.0001\* 6.1 6.1 0.8569 15.1 6.6 0.0001\* \statistically significant ::: Discussion ========== Although the morbidity rate after esophagectomy for esophageal carcinoma has been markedly reduced in recent years as a result of improvements in patient selection, surgical technique, and advances in perioperative management, the morbidity rate remains high \[[@B1]-[@B7]\]. Pulmonary function disturbances are the major contributing factor here \[[@B8]-[@B10]\]. Risk factors for the development of impaired pulmonary function include smoking, patient age (\>70 years), obesity, and preexisting COPD \[[@B11]\]. Parameters with an influence on perioperative mortality defined by Bartels et al, are in decreasing order: reduced general status of the patient, impaired cardiac and hepatic function, and respiratory function \[[@B12]\]. The evaluation of the preoperative function of the described organ systems using a scoring system developed by the authors enables the classification of risk groups, and leads to a decrease in the operative risk as a result of adequate patient selection \[[@B12]\]. The aim of our analysis of data collected prospectively in 424 patients undergoing surgery for esophageal carcinoma was to answer the question as to what extent the selection of the surgical procedure, i.e. the transthoracic or the transhiatal approach, and the operative course assessed on the basis of intraoperative blood loss, in addition to patient and tumor related factors, exert an influence on the postoperative course. The data were collected in 424 consecutive patients undergoing surgery for esophageal carcinoma over the period from 1985 to 2004. Surgical complications occurred in 35.5% and general complications in 36.0% of patients. The 30-day mortality rate ranged at 6.7%, at a mortality rate of 11.5%. Univariate analysis identified the selection of the surgical procedure as the main risk factor affecting the mortality rate. Transthoracic esophagectomy was associated with a higher complication and mortality rate than transhiatal dissection. Various authors have investigated the question as to the effectiveness of an extended radical procedure with an associated increased operative risk after the transthoracic technique. Results of two randomized studies did not demonstrate a difference between these procedures, although the meaningfulness of these findings is limited due to the small number of patients enrolled and the lack of information on the oncological radicality, especially the extent of lymph node dissection \[[@B13],[@B14]\]. In addition, only patients with early tumor stages \[[@B14]\] or with a distal location of the carcinoma \[[@B13]\] were taken into consideration, respectively. A higher pulmonary complication rate after the transthoracic compared to the transhiatal procedure for adenocarcinoma was found by a prospective randomized study \[[@B15]\]. Postoperative ventilation time, intensive care unit and postoperative in-hospital stay, reflecting perioperative morbidity, were significantly longer after transthoracic in contrast to transhiatal esophagectomy in this trial. However, the authors did not determine any differences in the long-term course \[[@B15]\]. This prospective study confirmed the outcome of a meta analysis revealing significantly higher early (pulmonary) morbidity and mortality after the transthoracic procedure published previously by the same author with 5-year survival rates of approximately 20% after both kinds of resection \[[@B16]\]. In our patient population, the long-term prognosis for patients with squamous cell carcinoma undergoing transthoracic surgery -- though exhibiting a higher perioperative morbidity and mortality -- was significantly better than that for patients after transhiatal resection \[Junginger T, et al.; unpublished data\]. Patients with adenocarcinoma did not differ in survival undergoing transhiatal or transthoracic esophagectomy \[Gockel I, et al; unpublished data\]. In concert with other authors, we therefore favor the transhiatal technique with posterior mediastinal and upper abdominal lymph node dissection for adenocarcinoma of the esophagus, and the transthoracic procedure with abdominal and mediastinal lymphadenectomy for squamous cell carcinoma. Thus -- long term results regarding the surgical technique of our own patient population have to be viewed critically, especially for adenocarcinoma, as the two groups differed significantly in UICC-stage and R-classification in contrast to patients with squamous cell carcinoma with a rather equal distribution \[[@B1],[@B2]\]. Independent of the operative procedure, surgical blood loss had a significant influence on the postoperative morbidity and mortality rates. This is in accordance with experiences reported by Whooley et al\[[@B8]\], and indicates that the selection of a limited resection technique can be crucial for the development of the postoperative course. Patient-associated parameters (age, nutritional status, ASA-classification) were of relevance only with respect to the occurrence of general complications. Preexisting pulmonary disease was an independent predictor of postoperative mortality. This confirms the findings of Chan et al, who identified impaired pulmonary function as a preoperative variable predictive of postoperative mortality \[[@B17]\]. The implementation of an appropriate preoperative therapy, discontinuation of smoking, more frequent use of epidural analgesia, and early bronchoscopy in the presence of the suspicion of postoperative pulmonary secretion impairment are essential factors for risk reduction \[[@B8]\]. In contrast to results of risk analyses by Law et al\[[@B18]\] and Lund et al\[[@B19]\], tumor characteristics as, e.g. TNM classification, were of no influence on the postoperative course in the patient population of this study. A different operative risk was determined for the two histological tumor types of the esophagus: while there were similar surgical complications in both groups, overall morbidity and mortality rates were significantly higher in patients with squamous cell carcinoma than in the group with adenocarcinoma. In accordance with reports in the literature \[[@B20]-[@B22]\], this reflects, on the one hand, the different preoperative risk profile of both entities, consisting of an increased nutritional risk, higher prevalence of hepatopathy, and poorer lung function in patients with squamous cell carcinoma. Additionally, in this study there was a significantly higher incidence of transthoracic esophagectomy with a higher complication rate in patients with squamous cell carcinoma than in those with adenocarcinoma (69.4 vs. 31.3%). Conclusion ========== The present analysis shows that the selection and the type of the surgical procedure are crucial factors for both the incidence of postoperative complications and the mortality rate. The transhiatal procedure is associated with a significantly lower morbidity and mortality rate and thus represents -- as long term survival does not favor the transthoracic approach -- the surgical technique of choice for adenocarcinoma of the esophagus. In contrast, our previous long-term experience and results obtained by this study advocate the performance of the transthoracic procedure for squamous cell carcinoma. The higher operative risk is justified with a view to a better long-term prognosis. Independent of the choice of the operative approach, a less-invasive surgical procedure and the implementation of measures designed to minimize the risk of pulmonary complications are essential to achieve a reduction in the morbidity rate of esophageal carcinoma. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= IG:study design, collection of data, statistical analysis, sequence alignment, draft of manuscript ChE:collection of data, statistical analysis ThJ:*conceived of the study, design and coordination of the study, draft and revision of the manuscript	PubMed Central	2024-06-05T03:55:58.283343	2005-6-21	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168909/", "journal": "World J Surg Oncol. 2005 Jun 21; 3:37", "authors": [ { "first": "Ines", "last": "Gockel" }, { "first": "Christoph", "last": "Exner" }, { "first": "Theodor", "last": "Junginger" } ] }
PMC1171969	Introduction ============ There is an impressively large body of physiological, imaging, and psychopharmacological data regarding the phasic activity of dopaminergic (DA) cells in the midbrains of monkeys, rats and humans in classical and instrumental conditioning tasks involving predictions of future rewards \[[@B1]-[@B5]\]. These data have been taken to suggest \[[@B6],[@B7]\] that the activity of DA neurons represents temporal difference (TD) errors in the predictions of future reward \[[@B8],[@B9]\]. This TD theory of dopamine provides a precise computational foundation for understanding a host of behavioural and neural data. Furthermore, it suggests that DA provides a signal that is theoretically appropriate for controlling learning of both predictions and reward-optimising actions. Some of the most compelling evidence in favour of the TD theory comes from studies investigating the phasic activation of dopamine cells in response to arbitrary stimuli (such as fractal patterns on a monitor) that predict the proximate availability of rewards (such as drops of juice). In many variants, these have shown that with training, phasic DA signals transfer from the time of the initially unpredictable reward, to the time of the earliest cue predicting a reward. This is exactly the expected outcome for a temporal-difference based prediction error (eg.\[[@B1],[@B2],[@B10]-[@B13]\]). The basic finding \[[@B7]\] is that when a reward is unexpected (which is inevitable in early trials), dopamine cells respond strongly to it. When a reward is predicted, however, the cells respond to the predictor, and not to the now-expected reward. If a predicted reward is unexpectedly omitted, then the cells are phasically inhibited at the normal time of the reward, an inhibition which reveals the precise timing of the reward prediction \[[@B10]\], and whose temporal metrics are currently under a forensic spotlight \[[@B14]\]. The shift in activity from the time of reward to the time of the predictor resembles the shift of the animal\'s appetitive behavioural reaction from the time of the reward (the unconditioned stimulus) to that of the conditioned stimulus in classical conditioning experiments \[[@B7],[@B10]\]. In a most interesting recent study, Fiorillo et al.\[[@B15]\] examined the case of partial reinforcement, in which there is persistent, ineluctable, prediction error on every single trial. A straightforward interpretation of the TD prediction error hypothesis would suggest that in this case (a) dopamine activity at the time of the predictive stimuli would scale with the probability of reward, and (b) on average over trials, the dopaminergic response after the stimulus and all the way to the time of the reward, should be zero. Although the first hypothesis was confirmed in the experiments, the second was not. The between-trial averaged responses showed a clear ramping of activity during the delay between stimulus onset and reward that seemed inconsistent with the TD account. Fiorillo et al.hypothesised that this activity represents the uncertainty in reward delivery, rather than a prediction error. In this paper, we visit the issue of persistent prediction error. We show that a crucial asymmetry in the coding of positive and negative prediction errors leads one to expect the ramping in the between-trial average dopamine signal, and also accounts well for two further features of the DA signal -- apparent persistent activity at the time of the (potential) reward, and disappearance (or at least weakening) of the ramping signal, but not the signal at the time of reward, in the face of trace rather than delay conditioning. Both of these phenomena have also been observed in the related instrumental conditioning experiments of Morris et al.\[[@B16]\]. Finally, we interpret the ramping signal as the best evidence available at present for the nature of the learning mechanism by which the shift in dopamine activity to the time of the predictive stimuli occurs. Uncertainty in reward occurrence: DA ramping ============================================ Fiorillo et al.\[[@B15]\] associated the presentation of five different visual stimuli to macaques with the delayed, probabilistic(p~r~= 0, 0.25, 0.5, 0.75, 1) delivery of juice rewards. They used a delay conditioning paradigm, in which the stimulus persists for a fixed interval of 2s, with reward being delivered when the stimulus disappears. After training, the monkeys\' anticipatory licking behavior indicated that they were aware of the different reward probabilities associated with each stimulus. Figure [1a](#F1){ref-type="fig"} shows population histograms of extracellularly-recorded DA cell activity, for each p~r~. TD theory predicts that the phasic activation of the DA cells at the time of the visual stimuli should correspond to the averageexpected reward, and so should increase with p~r~. Figure [1a](#F1){ref-type="fig"} shows exactly this -- indeed, across the population, the increase is quite linear. Morris et al.\[[@B16]\] report a similar result in an instrumental (trace) conditioning task also involving probabilistic reinforcement. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Averaged prediction errors in a probabilistic reward task(a) DA response in trials with different reward probabilities. Population peri-stimulus time histograms (PSTHs) show the summed spiking activity of several DA neurons over many trials, for each p~r~, pooled over rewarded and unrewarded trials at intermediate probabilities. (b) TD prediction error with asymmetric scaling. In the simulated task, in each trial one of five stimuli was randomly chosen and displayed at time t= 5. The stimulus was turned off at t= 25, at which time a reward was given with a probability of p~r~specified by the stimulus. We used a tapped delay-line representation of the stimuli (see text), with each stimulus represented by a different set of units (\'neurons\'). The TD error was δ(t) = r(t) + w(t- 1)·x(t) - w(t- 1)·x(t- 1), with r(t) the reward at time t, and x(t) and w(t) the state and weight vectors for the unit. A standard online TD learning rule was used with a fixed learning rate α, w(t) = w(t- 1) + αδ(t)x(t- 1), so each weight represented an expected future reward value. Similar to Fiorillo et al., we depict the prediction error δ(t) averaged over many trials, after the task has been learned. The representational asymmetry arises as negative values of δ(t) have been scaled by d= 1/6 prior to summation of the simulated PSTH, although learning proceeds according to unscaled errors. Finally, to account for the small positive responses at the time of the stimulus for p~r~= 0 and at the time of the (predicted) reward for p~r~= 1 seen in (a), we assumed a small (8%) chance that a predictive stimulus is misidentified. (c) DA response in p~r~= 0.5 trials, separated into rewarded (left) and unrewarded (right) trials. (d) TD Model of (c). (a,c) Reprinted with permission from \[15\]©2003 AAAS. Permission from AAAS is required for all other uses. ::: ![](1744-9081-1-6-1) ::: By contrast, at the time of potential reward delivery, TD theory predicts that on averagethere should be no activity, as, on average, there is no prediction error at that time. Of course, in the probabilistic reinforcement design (at least for p~r~≠ 0, 1) there is in fact a prediction error at the time of delivery or non-delivery of reward on every single trial. On trials in which a reward is delivered, the prediction error should be positive (as the reward obtained is larger than the average reward expected). Conversely, on trials with no reward it should be negative (see Figure [1c](#F1){ref-type="fig"}). Crucially, under TD, the average of these differences, weighted by their probabilities of occurring, should be zero. If it is not zero, then this prediction error should act as a plasticity signal, changing the predictions until there is no prediction error. At variance with this expectation, the data in Figure [1a](#F1){ref-type="fig"} which is averaged over both rewarded and unrewarded trials, show that there is in fact positive mean activity at this time. This is also evident in the data of Morris et al.\[[@B16]\] (see Figure [3c](#F3){ref-type="fig"}). The positive DA responses show no signs of disappearing even with substantial training (over the course of months). Worse than this for the TD model, and indeed the focus of Fiorillo et al.\[[@B15]\], is the apparent rampingof DA activity towards the expected time of the reward. As the magnitude of the ramp is greatest for p~r~= 0.5, Fiorillo et al.suggested that it reports the uncertaintyin reward delivery, rather thana prediction error, and speculated that this signal could explain the apparently appetitive properties of uncertainty (as seen in gambling). Both the ramping activity and the activity at the expected time of reward pose critical challenges to the TD theory. TD learning operates by arranging for DA activity at one time in a trial to be predicted awayby cues available earlier in that trial. Thus, it is not clear how any seemingly predictable activity, be it that at the time of the reward or in the ramp before, can persist without being predicted away by the onset of the visual stimulus. After all, the p~r~-dependent activity in response to the stimulus confirms its status as a valid predictor. Furthermore, a key aspect of TD \[[@B17]\], is that it couples prediction to action choice by using the valueof a state as an indication of the future rewards available from that state, and therefore its attractiveness as a target for action. From this perspective, since the ramping activity is explicitly not predicted by the earlier cue, it cannot influence early actions, such as the decision to gamble. For instance, consider a competition between two actions: one eventually leading to a state with a deterministic reward and therefore no ramp, and the other leading to a state followed by a probabilistic reward with the same mean, and a ramp. Since the ramp does not affect the activity at the time of the conditioned stimulus, it cannot be used to evaluate or favour the second action (gambling) over the first, despite the extra uncertainty. We suggest the alternative hypothesis that both these anomalous firing patterns result directly from the constraints implied by the low baseline rate of activity of DA neurons (2--4 Hz) on the coding of the signedprediction error. As noted by Fiorillo et al.\[[@B15]\], positive prediction errors are represented by firing rates of \~270% abovebaseline, while negative errors are represented by a decrease of only \~55% belowbaseline (see also \[[@B14],[@B18]\]). This asymmetry is a straightforward consequence of the coding of a signed quantity by firing which has a low baseline, though, obviously, can only be positive. Firing rates above baseline can encode positive prediction errors by using a large dynamic range, however, below baseline firing rates can only go down to zero, imposing a restriction on coding of negative prediction errors. Consequently, one has to be careful interpreting the sums (or averages) of peri-stimulus-time-histograms (PSTHs) of activity over different trials, as was done in Figure [1a](#F1){ref-type="fig"}. The asymmetrically coded positive and negative error signals at the time of the receipt or non-receipt of reward should indeed notsum up to zero, even if they represent correct TD prediction errors. When summed, the low firing representing the negative errors in the unrewarded trials will not \"cancel out\" the rapid firing encoding positive errors in the rewarded trials, and, overall, the average will show a positive response. In the brain, of course, as responses are not averaged over (rewarded and unrewarded) trials, but over neurons within a trial, this need not pose a problem. This explains the persistent positive activity (on average) at the time of delivery or non-delivery of the reward. But what about the ramp prior to this time? At least in certain neural representations of the time between stimulus and reward, when trials are averaged, this same asymmetry leads TD to result exactly in a ramping of activity toward the time of the reward. The TD learning mechanism has the effect of propagating, on a trial-by-trial basis, prediction errors that arise at one time in a trial (such as at the time of the reward) towards potential predictors (such as the CS) that arise at earlier times within each trial. Under the asymmetric representation of positive and negative prediction errors that we have just discussed, averaging these propagating errors over multiple trials (as in Figure [1a](#F1){ref-type="fig"}) will lead to positive means for epochs within a trial before a reward. The precise shape of the resulting ramp of activity depends on the way stimuli are represented over time, as well as on the speed of learning, as will be discussed below. Figure [2](#F2){ref-type="fig"} illustrates this view of the provenance of the ramping activity. Here, a tapped delay-line representation of time since the stimulus is used. For this, each unit (\'neuron\') becomes active (i.e., assumes the value 1) at a certain lag after the stimulus has been presented, so that every timestep after the stimulus onset is consistently represented by the firing of one unit. Learning is based on the (dopaminergically-reported) TD error, formalized as δ(t) = r(t) + V(t) - V(t- 1), with V(t) the weighted input from the active unit at time t, and r(t) the reward obtained at time t. Updating the weights of the units according to the standard TD update rule with a fixed learning rate, allows V(t) to, on average, represent the expected future rewards (see Figure [1](#F1){ref-type="fig"} caption). As each subsequent timestep is separately represented, TD prediction errors can arise at any time within the trial. Figure [2a](#F2){ref-type="fig"} shows these errors in six consecutive simulated trials in which p~r~= 0.5. In every trial, a new positive or negative error arises at the time of the reward, consequent on receipt or non-receipt of the reward, and step-by-step the errors from previous trials propagate back to the time of the stimulus, through the constant updating of the weights (eg.the error highlighted in red). When averaging (or, as in PSTHs, summing) over trials, these errors cancel each other on average, resulting in an overall flat histogram in the interval after the stimulus onset, and leading up to the time of the reward (black line in Figure [2b](#F2){ref-type="fig"}, summed over the 10 trials shown in thin blue). However, when summed after asymmetric scalingof the negative errors by a factor of d= 1/6 (which simulates the asymmetric coding of positive and negative prediction errors by DA neurons), a positive ramp of activity ensues, as illustrated by the black line in Figure [2c](#F2){ref-type="fig"}. Note that this rescaling is only a representationalissue, resulting from the constraints of encoding a negative value about a low baseline firing rate, and should not affect the learning of the weights, so as not to learn wrong values (see discussion). However, as PSTHs are directly sums of neuronal spikes, this representational issue bears on the resulting histogram. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Backpropagation of prediction errors explains ramping activity. (a) The TD prediction error across each of six consecutive trials (top to bottom) from the simulation in Figure 1b, with p~r~= 0.5. Highlighted in red is the error at the time of the reward in the first of the trials, and its gradual back-propagation towards the time of the stimulus in subsequent trials. Block letters indicate the outcome of each specific trial (R = rewarded; N = not rewarded). The sequence of rewards preceding these trials is given on the top right. (b) The TD error from these six trials, and four more following them, superimposed. The red and green lines illustrate the envelope of the errors in these trials. Summing over these trials results in no above-baseline activity on average (black line), as positive and negative errors occur at random 50% of the time, and so cancel each other. (c) However, when the prediction errors are asymmetrically represented above and below the baseline firing rate (here negative errors were asymmetrically scaled by d= 1/6 to simulate the asymmetric encoding of prediction errors by DA neurons), an average ramping of activity emerges when averaging over trials, as is illustrated by the black line. All simulation parameters are the same as in Figure 1b,d. ::: ![](1744-9081-1-6-2) ::: Figures [1b,d](#F1){ref-type="fig"} show the ramp arising from this combination of asymmetric coding and inter-trial averaging, for comparison with the experimental data. Figure [1b](#F1){ref-type="fig"} shows the PSTH computed from our simulated data by averaging over the asymmetrically-represented δ(t) signal in \~50 trials for each stimulus type. Figure [1d](#F1){ref-type="fig"} shows the results for the p~r~= 0.5 case, divided into rewarded and unrewarded trials for comparison with Figure [1c](#F1){ref-type="fig"}. The simulated results resemble the experimental data closely in that they replicate the net positive response to the uncertain rewards, as well as the ramping effect, which is highest in the p~r~= 0.5 case. It is simple to derive the average response at the time of the reward (t = N) in trial T, i.e.,the average TD error ![](1744-9081-1-6-i3.gif)δ~T~(N)![](1744-9081-1-6-i4.gif), from the TD learning rule with the simplified tapped delay-line time representation and a fixed learning rate α. The value at the next to last timestep in a trial, as a function of trial number (with initial values taken to be zero), is ![](1744-9081-1-6-i1.gif) where r(t) is the reward at the end of trial t. The error signal at the last timestep of trial Tis simply the difference between the obtained reward r(T), and the value predicting that reward V~T-\ 1~(N- 1). This error is positive with probability p~r~, and negative with probability (1 - p~r~). Scaling the negative errors by a factor of d∈ (0, 1\], we thus get ![](1744-9081-1-6-i2.gif) For symmetric coding of positive and negative errors (d= 1), the average response is 0. For asymmetric coding (0 \<d\< 1), the average response is indeed proportional to the variance of the rewards, and thus maximal at p~r~= 0.5. However, δ~T~is positive, and concomitantly, the ramps are positive, and in this particular setting, are related to uncertainty, because of, rather than instead of, the coding of δ(t). Indeed, there is a key difference between the uncertainty and TD accounts of the ramping activity. According to the former, the ramping is a within-trial phenomena, coding uncertainty in reward; by contrast, the latter suggests that ramps arise only through averaging across multiple trials. Within a trial, when averaging over simultaneously recorded neurons rather than trials, the traces should not show a smooth ramp, but intermittent positive and negative activity corresponding to back-propagating prediction errors from the immediately previous trials (as in Figure [2a](#F2){ref-type="fig"}). Trace conditioning: a test case =============================== An important test case for our interpretation arises in a variant of Fiorillo et al.\'s\[[@B15]\] task, as well as in the analogous instrumental task of Morris et al.\[[@B16]\], both involving trace conditioning. In contrast to delay conditioning (Figure [3a](#F3){ref-type="fig"}) in which the reward coincides with the offset of the predictive stimulus, here there is a substantial gap between the offset of the predictive stimulus and the delivery of the reward (Figure [3b](#F3){ref-type="fig"}). Clearly, in this case, uncertainty about the reward could only get larger, owing to noise in timing the interval between stimulus and reward \[[@B19]\], so under the uncertainty account, there should be comparable or even larger ramps. However, the experimental results show the ramping activity to be smaller, or even negligible (Figure [3c;d](#F3){ref-type="fig"}). Note, though, that the magnitude of the trial-average activity at the expected time of reward is maintained, pointing to a dissociation between the height of the ramp and the amount of positive activity at the expected time of reward. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Trace conditioning with probabilistic rewards. (a) An illustration of one trial of the delay conditioning task of Fiorillo et al.\[15\]. A trial consists of a 2-second visual stimulus, the offset of which coincides with the delivery of the juice reward, if such a reward is programmed according to the probability associated with the visual cue. In unrewarded trials the stimulus terminated without a reward. In both cases an inter-trial interval of 9 seconds on average separates trials. (b) An illustration of one trial of the trace conditioning task of Morris et al.\[16\]. The crucial difference is that there is now a substantial temporal delay between the offset of the stimulus and the onset of the reward (the \"trace\" period), and no external stimulus indicates the expected time of reward. This confers additional uncertainty as precise timing of the predicted reward must be internally resolved, especially in unrewarded trials. In this task, as in \[15\], one of several visual stimuli (not shown) was presented in each trial, and each stimulus was associated with a probability of reward. Here, also, the monkey was requested to perform an instrumental response (pressing the key corresponding to the side in which the stimulus was presented), the failure of which terminated the trial without a reward. Trials were separated by variable inter-trial intervals. (c,d) DA firing rate (smoothed) relative to baseline, around the expected time of the reward, in rewarded trials (c) and in unrewarded trials (d). (c,d) Reprinted from \[16\] ©2004 with permission from Elsevier. The traces imply an overall positive response at the expected time of the reward, but with a very small, or no ramp preceding this. Similar results were obtained in a classical conditioning task briefly described in \[15\], which employed a trace conditioning procedure, confirming that the trace period, and not the instrumental nature of the task depicted in (b) was the crucial difference from (a). ::: ![](1744-9081-1-6-3) ::: The TD model of DA readily explains these puzzling data. As shown in Figure [4](#F4){ref-type="fig"}, the shape of the ramp, though not the height of its peak, is affected by the learning rate. The size of the back-propagating prediction errors is determined, in part, by the learning rate, as these errors arise as part of the online learning of new predictions. Indeed, there is a continuous updating of predictions such that after a rewarded trial, there is a higher expectation of reward (and thus the next reward incurs a smaller prediction error), and conversely after a non-rewarded trial \[[@B18]\] (see Figure [2a](#F2){ref-type="fig"}). This updating of predictions is directly related to the learning rate -- the higher the learning rate, the larger the update of predictions according to the current prediction error, and the larger the fraction of the prediction error which is propagated back. In this way, with higher learning rates, the difference in expectations after a rewarded versus an unrewarded trial will be larger, and thus the prediction errors when the next reward is or is not available will be larger -- hence the larger and more gradual ramp. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Dependence of the ramp on learning rate.The shape of the ramp, but not the height of its peak, is dependent on the learning rate. The graph shows simulated activity for the case of p~r~= 0.5 near the time of the expected reward, for different learning rates, averaged over both rewarded and unrewarded trials. According to TD learning with persistent asymmetrically coded prediction errors, averaging over activity in rewarded and unrewarded trials results in a ramp up to the time of reward. The height of the peak of the ramp is determined by the ratio of rewarded and unrewarded trials, however, the breadth of the ramp is determined by the rate of back-propagation of these error signals from the time of the (expected) reward to the time of the predictive stimulus. A higher learning rate results in a larger fraction of the error propagating back, and thus a higher ramp. With lower learning rates, the ramp becomes negligible, although the positive activity (on average) at the time of reward is still maintained. Note that although the learning rate used in the simulations depicted in Figure 1b,d was 0.8, this should not be taken as the literal synaptic learning rate of the neural substrate, given our schematic representation of the stimulus. In a more realistic representation in which a population of neurons is active at every timestep, a much lower learning rate would produce similar results. ::: ![](1744-9081-1-6-4) ::: Indeed, compared to delay conditioning, trace conditioning is notoriously slow, suggesting that the learning rate is low, and thus that there should be a lower ramp, in accord with the experimental results. A direct examination of the learning rate in the data of Morris et al.\[[@B16]\], whose task required excessive training as it was not only a trace conditioning one but also involved an instrumental action, confirmed it indeed to be very low (Genela Morris -- personal communication, 2004). Discussion ========== The differential coding of positive and negative values by DA neurons is evident in all the studies of the phasic DA signal, and can be regarded as an inevitable consequence of the low baseline activity of these neurons. Indeed, the latter has directly inspired suggestions that an opponent neurotransmitter, putatively serotonin, be involved in representing and therefore learning the negative prediction errors \[[@B20]\], so that they also have full quarter. Here, however, we have confined ourselves to consideration of the effects of asymmetry on the trial-average analysis of the dopamine activity, and have shown that ramping DA activity, as well as an average positive response at the time of reward, result directly from the asymmetric coding of prediction errors. Apart from a clearer view of the error signal, the most important consequence of the new interpretation is that the ramps can be seen as a signature of a TD phenomenon that has hitherto been extremely elusive. This is the progressive back-propagation of the error signal represented by DA activity, from the time of reward to the time of the predictor (Figure [2a](#F2){ref-type="fig"}). Most previous studies of dopaminergic activity have used p~r~= 1, so making this back-propagation at best a transitory phenomenon apparent only at the beginning of training (when, typically, recordings have not yet begun), and potentially hard to discern in slow-firing DA neurons. Further, as mentioned, the back-propagation depends on the way that the time between the predictive stimulus and the reward is represented -- it is present for a tapped delay-line representation as in \[[@B6]\], but not for representations which span the entire delay, such as in \[[@B21]\]. Note that the shape of the ramp also depends on the use of eligibility traces and the so-called TD(λ) learning rule (simulation not shown), which provide an additional mechanism for bridging time between events during learning. Unfortunately, as the forms of the ramps in the data are rather variable (figure [1](#F1){ref-type="fig"}) and noisy, they can not provide strong constraints on the precise TD mechanism used by the brain. More recent studies involving persistent prediction errors also show activity suggestive of back-propagation, notably Figure 4 of \[[@B13]\]. In this study, prediction errors resulted from periodic changes in the task, and DA recordings were made from the onset of training, thus back-propagation-like activity is directly apparent, although this activity was not quantified. We expect the ramps to persist throughout training only if the learning rate does not decrease to zero as learning progresses. Pearce & Hall\'s \[[@B22]\] theory of the control of learning by uncertainty suggests exactly this persistence of learning -- and there is evidence from partial reinforcement schedules that the learning rate may be higher when there is more uncertainty associated with the reward. Indeed, from a \'rational\' statistical point of view, learning should persist when there is substantial uncertainty about the relationship between predictors and outcomes, as can arise from the ever-present possibility of a change in the predictive relationships. This form of persistent uncertainty, together with uncertainty due to initial ignorance regarding the task, have been used to formalize Pearce & Hall\'s theory of the way that uncertainty drives learning \[[@B23]\]. Thus, our claim that uncertainty may not be directly represented by the ramps, should certainly not be taken to mean that its representation and manipulation is not important. To the contrary, we have suggested that uncertainty influences cortical inference and learning through other neuromodulatory systems \[[@B24]\], and that it also may determine aspects of the selection of actions \[[@B25]\]. Various other features of the asymmetry should be noted. Most critical is the effect of the asymmetry on DA-dependent learning \[[@B26]\], if the below baseline DA activity is responsible by itself for decreasing predictions which are too high. In order to ensure that the learned predictions remain correct, we would have to assume that the asymmetric representation does not affect learning, i.e.,that a mechanism such as different scaling for potentiation and depression of the synaptic strengths compensates for the asymmetric error signal. Of course, this would be rendered moot if an opponent neurotransmitter is involved in learning from negative prediction errors. This issue is complicated by the suggestion of Bayer \[[@B14]\] that DA firing rates are actually similar for all prediction errors below some negative threshold, perhaps due to the floor effect of the low firing rate. Such lossy encoding does not affect the qualitative picture of the effects of inter-trial averaging on the emergence of ramps, but does reinforce the need for an opponent signal for the necessarily symmetric learning. Finally, the most direct test of our interpretation would be a comparison of intra- and inter-trial averaging of the DA signal. It would be important to do this in a temporally sophisticated manner, to avoid problems of averaging non-stationary signals. In order to overcome the noise in the neural firing, and determine whether indeed there was a gradual ramp within a trial, or, as we would predict -- intermittent positive and negative prediction errors, it would be necessary to average over many neurons recorded simultaneously within one trial, and furthermore neurons associated with similar learning rates. Alternatively, single neuron traces could be regressed against the backpropagation response predicted by their preceding trials and TD learning. A comparison of the amount of variability explained by such a model, compared to that from a regression against a monotonic ramp of activity, could point to the most fitting model. A less straightforward, but more testable prediction is that the shape of the ramp should depend on the learning rate. Learning rates can be assessed from the response to the probabilistic rewards, independent of the shape of the ramp (Nakahara et al.\[[@B18]\] showed in such a way, that in their partial reinforcement trace conditioning task, the learning rate was 0.3), and potentially manipulated by varying the amount of training or the frequency with which task contingencies are changed and relearned. Indeed, quantifying the existence and shape of a ramp in Nakahara et al.\'srecorded DA activity, could well shed light on the current proposal. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= YN, MD and PD jointly conceived and executed this study, and helped draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We are very grateful to H. Bergman, C. Fiorillo, N. Daw, D. Joel, P. Tobler, P. Shizgal and W. Schultz for discussions and comment, in some cases despite varying interpretation of the data. We are particularly grateful to Genela Morris for analyzing her own published and unpublished data in relation to ramping. This work was funded by the EC Thematic Network (YN), the Gatsby Charitable Foundation and the EU BIBA project.	PubMed Central	2024-06-05T03:55:58.286424	2005-5-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1171969/", "journal": "Behav Brain Funct. 2005 May 4; 1:6", "authors": [ { "first": "Yael", "last": "Niv" }, { "first": "Michael O", "last": "Duff" }, { "first": "Peter", "last": "Dayan" } ] }
PMC1173078	Introduction ============ In 1996, Salkeld et al. \[[@B1]\] found that screening the average-risk Australian population for colorectal cancer using a faecal occult blood test (FOBT), compared to existing practice, would cost \$24,660 per life-year gained (LYG). Due to uncertainty regarding the effectiveness of FOBT screening, they report a range of values, between \$12,695 and \$67,848 per LYG. Randomised controlled trials of population screening with FOBT conducted in the UK \[[@B2]\] and Denmark \[[@B3]\], but published after Salkeld et al.\'s study, have reduced this uncertainty. The cost-effectiveness analyses based on the UK trial data \[[@B4]\] suggest a cost per life-year gained between £1,371--£5,685 (approximately \$AU3,370--13,974) and the analysis of the Danish trial data \[[@B5]\] suggest a cost per life-year gained between 17,000--42,000DKK (approximately \$AU3,916--9,672). Three years before publication of Salkeld\'s study Bolin \[[@B6]\] discussed the advantages of using colonoscopy for population screening, and in 1996, he suggested colonoscopy was cost-effective \[[@B7]\]. In 1997, he asked whether the time had come to use colonoscopy for population screening in Australia \[[@B8]\]. Kermond \[[@B9]\] responded suggesting double contrast barium enema (DCBE) should not be overlooked arguing colonoscopy is 10 times more expensive, false negatives still occur and complication rates are higher. Bolin argued, in the same issue of the MJA, that the sensitivity of colonoscopy exceeds DCBE, the complication rate is only 0.1% and cost differentials are actually less than those suggested by Kermond \[[@B9]\]. Bolin also claimed that FOBT at one and three years and colonoscopy at 10 years, assuming a 10-year period during which time the cancer is detectable and curable (known as the dwell time), are cost-effective modes of CRC screening, probably referring to data subsequently published in 1999 \[[@B10]\]. For this research, the authors substituted Australian values for cost parameters into a US model of the cost-effectiveness of CRC screening \[[@B11]\] and generalised the results to the Australian population. They reported change in cost and change in life-years gained, as compared to existing practice, for competing screening strategies that encompass FOBT, colonoscopy, flexible sigmoidoscopy and DCBE \[[@B10]\]. By assuming that society is willing to pay up to \$US40,000 (approximately \$AU65,449 in 2002 prices) per LYG, Bolin proposed that annual FOBT, triennial FOBT, triennial DCBE, five-yearly DCBE, five-yearly colonoscopy and ten-yearly colonoscopy are all cost-effective and concluded that physicians have the option of offering individuals a range of screening alternatives, including colonoscopy \[[@B10]\]. Since publishing the research Bolin has argued, on four separate occasions that population screening with colonoscopy is cost-effective \[[@B12]-[@B15]\]. The last of these, in 2002 \[[@B14]\], provoked Macrae and Hebbard \[[@B16]\] to criticise Bolin\'s interpretation of epidemiological data. In 2004, O\'Leary et al. \[[@B17]\] also addressed the economic questions around population screening in Australia. They estimated the cost-effectiveness, compared to existing practice, of FOBT, flexible sigmoidoscopy and colonoscopy, and found flexible sigmoidoscopy and colonoscopy were cost-effective but FOBT was not. There are important differences in the way that Salkeld et al. \[[@B1]\], Bolin et al. \[[@B10]\] and O\'Leary et al. \[[@B17]\] report the Australian cost and effectiveness data. Salkeld et al. reports an incremental cost-effectiveness ratio, O\'Leary et al. reports both average and incremental cost-effectiveness ratios, however they draw their conclusions from an average analysis. Bolin only calculates average cost-effectiveness ratios. The correct ratio for decision-making is an incremental cost-effectiveness ratio: McMahon \[[@B18]\] argues the use of average ratios is not meaningful; Drummond \[[@B19]\] and Gold \[[@B20]\] both discuss why incremental rather than average cost-effectiveness ratios are relevant for decision making; and, both Torgerson \[[@B21]\] and Neuhauser & Lewicki \[[@B22]\] provide examples of how average analyses muddy the waters. In their much-cited 1975 paper, Neuhauser & Lewicki \[[@B22]\] reviewed data on screening for CRC. They illustrated that repeatedly testing a stool sample up to six times, when a previous test result was negative, would capture all cases of CRC, at an average cost per case of \$2451. They also did an incremental analysis, with the same data, and showed the incremental cost per case detected, from the fifth to sixth round of testing was \$47 million. This illustrates that average analyses can be grossly misleading. We have four objectives in this paper: first, to demonstrate why incremental, not average, cost-effectiveness ratios should be used for decision-making; second, to update the cost data reported by Salkeld et al. \[[@B1]\], Bolin et al. \[[@B10]\] and O\'Leary et al. \[[@B17]\] to 2002 Australian dollar prices; third, to calculate incremental cost-effectiveness ratios from the Bolin data; and fourth, to discuss the results of our re-analysis, comparing the outcomes from the three previous studies. This will provide readers with an up-to-date and appropriate assessment of the existing cost-effectiveness data for population-based CRC screening programmes in Australia. Defining average & incremental cost-effectiveness ratios ======================================================== Average cost-effectiveness ratios for health care interventions are the amounts by which costs change from a baseline comparator (ΔC) divided by the amount by which health benefits change from a baseline comparator (ΔE). The baseline comparator is often existing practice. We illustrate this method with hypothetical data in Table [1](#T1){ref-type="table"} by presenting the change in costs and health effects that arise from four competing health care alternatives. If we remain with Existing Practice, there is no change in cost or health effect. However, if we are committed to generating health effects and we wish to be efficient, then we should choose the cheapest option that improves health outcomes. The data in Table [1](#T1){ref-type="table"} show that Intervention 4 generates the best ratio of (ΔC) and (ΔE) when compared to Existing Practice. Cost changes by \$145,000 and health effects change by 150 and the cost per LYG is \$967. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### An illustration of average cost-effectiveness ratios for four competing hypothetical health care interventions ::: Change in cost (\$) Change in health effect (Life-years Gained) Average cost-effectiveness (\$) ------------------- --------------------- --------------------------------------------- --------------------------------- (ΔC) (ΔE) (ΔC) divided by (ΔE) Existing Practice 0 0 Intervention 1 200,000 12 16,667 Intervention 2 75,000 15 5,000 Intervention 3 300,000 250 1,200 Intervention 4 145,000 150 967 ::: We reject Interventions 1 and 2. The reasons lie in Figure [1](#F1){ref-type="fig"} (which is a graph of the data in Table [1](#T1){ref-type="table"}). Intervention 1 generates less health benefit and higher costs than Intervention 4, a situation described as \'simple dominance\'. For this comparison, Intervention 4 is preferred on both costs and outcomes. Intervention 2 also generates less health effect but differs from Intervention 1 in that it\'s cheaper than Intervention 4; however, the cost per LYG from Intervention 2 is greater than the cost per LYG from Intervention 4. This situation is known as \'extended dominance\' and is only relevant if the cost of Intervention 4 (\$145,000) exceeds the total amount of money available to the decision-maker. Rather than choosing Intervention 2 over Intervention 4, it would be better (more productively efficient) to choose some blend of existing practice and Intervention 4. This implies that some proportion of the population would receive Intervention 4 and the remainder would receive existing practice. This raises questions of equity of access and so poses another set of problems for decision-makers. If the available budget exceeds \$145,000, there is a further question to consider. Do we invest in the more costly but more effective Intervention 3? Some care is required when making this decision. The average cost-effectiveness ratio for Intervention 3 (\$1,200 per LYG), represented by the dashed line on Figure [1](#F1){ref-type="fig"}, is misleading. It\'s calculated by comparing Intervention 3 to existing practice; yet, the relevant decision is whether we should invest in Intervention 3 given that we have established Intervention 4 as the most cost-effective option. We must consider the incremental changes in cost and health effects compared to the next best alternative, Intervention 4. The incremental cost-effectiveness ratios for Interventions 4 and 3 are marked with solid lines on Figure [1](#F1){ref-type="fig"}. When a more effective alternative also costs more, then the decision-maker must compare the increased cost with the increased effects \[[@B19]\]. The only way to achieve this is to conduct an incremental analysis, which we illustrate in Table [2](#T2){ref-type="table"}. Investing in Intervention 3, as compared to 4, changes total costs by \$155,000 and LYG by 100, yielding an incremental cost-effectiveness ratio of \$1,550 per LYG not the \$1,200 per LYG as previously estimated using average cost-effectiveness ratios and illustrated in Table [1](#T1){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### An illustration of average and incremental cost-effectiveness ratios for the two remaining hypothetical health care interventions ::: Intervention Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ----------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- (a) (b) (c) (d) (a)/(c) (b)/(d) Existing Practice 0 0 Intervention 4 145,000 145,000 150 150 967 967 Intervention 3 300,000 155,000 250 100 1,200 1,550 Note: Interventions 1 and 2 have been rejected on the grounds of \'simple\' and \'extended\' dominance, see text for a discussion ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Change in cost and change in effect from four hypothetical health care interventions. ::: ![](1743-8462-2-10-1) ::: Imagine you have decided to take a holiday in a beach resort. You face a decision between a standard apartment for \$1,000 and a Penthouse apartment for \$1,300. Because you have decided to take the holiday (and so one of the apartments), it is the difference in cost (\$300) that you compare to the difference in benefit (penthouse vs. standard apartment). If you don\'t perceive the additional benefit to be worth the extra \$300, then you reject the penthouse. This simple example illustrates the importance of thinking about decisions in terms of incremental changes. An average analysis with both options compared to \'no holiday\' may lead to a bad decision. Methods for re-analysis and the results ======================================= We converted Bolin\'s \$US dollar estimates to Australian dollars with the exchange rate reported in the original article \[[@B10]\] and adjusted the estimates reported by Bolin, Salkeld and O\'Leary to 2002 prices using a health price index \[[@B23]\]. As Salkeld and O\'Leary reported incremental cost-effectiveness ratios, we only need to calculate incremental cost-effectiveness ratios from the Bolin data. We achieved this by inputting the reported estimates of (ΔC) and (ΔE) into decision analysis software \[[@B24]\]. Assuming a 5-year dwell time (the period during which cancer can be detected and cured), we include 13 screening strategies and the existing practice comparator. For a 10-year dwell time, estimates of (ΔC) and (ΔE)were reported for different, additional, frequencies of FOBT, flexible sigmoidoscopy and DCBE screening, resulting in 27 strategies and the existing practice comparator. We ranked all strategies by increasing cost, estimated incremental cost and effectiveness, and excluded all strategies for which other options prevailed on the basis of \'simple\' or \'extended\' dominance. Finally, we reported the strategies not excluded due to either \'simple\' or \'extended\' dominance and present the relevant incremental cost-effectiveness ratios. In Table [3](#T3){ref-type="table"}, we describe all strategies evaluated by the authors of the three studies. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Descriptions of the screening strategies included in the re-analysis ::: Screening strategy Description of screening strategy Salkeld \[1\] Bolin \[10\] (5-year dwell time) Bolin \[10\] (10-year dwell time) O\'Leary \[17\] -------------------- --------------------------------------------------------------------------- --------------- ---------------------------------- ----------------------------------- ----------------- Existing practice Existing screening practices X X X COL10 10-yearly colonoscopy X X X COL5 5-yearly colonoscopy X X COL one off screening colonoscopy at age 50 X X DCBE one off double contrast barium enema X DCBE10 10-yearly double contrast barium enema X DCBE15 15-yearly double contrast barium enema X DCBE20 20-yearly double contrast barium enema X DCBE3 3-yearly double contrast barium enema X X DCBE5 5-yearly double contrast barium enema X X FOBT10 10-yearly faecal occult blood test X FOBT15 15-yearly faecal occult blood test X FOBT2 2-yearly faecal occult blood test X X FOBT20 20-yearly faecal occult blood test X FOBT5 5-yearly faecal occult blood test X FOBT one off faecal occult blood test X FOBT1 annual faecal occult blood test X X X X FOBT1+DCBE3 annual faecal occult blood test and 3-yearly double contrast barium enema X X FOBT1+DCBE5 annual faecal occult blood test and 5-yearly double contrast barium enema X X FOBT1+FSIG3 annual faecal occult blood test and 3-yearly flexible sigmoidoscopy X X FOBT1+FSIG5 annual faecal occult blood test and 5-yearly flexible sigmoidoscopy X X FOBT3 3-yearly faecal occult blood test X X FSIG flexible sigmoidoscopy once only X FSIG10 10-yearly flexible sigmoidoscopy X X FSIG15 15-yearly flexible sigmoidoscopy X FSIG20 20-yearly flexible sigmoidoscopy X FSIG3 3-yearly flexible sigmoidoscopy X X FSIG5 5-yearly flexible sigmoidoscopy X X ::: In Figure [2](#F2){ref-type="fig"}, we illustrate the costs and effects of all strategies reported by Bolin, assuming a 5-year dwell time. The strategies to the left and above the cost-effective frontier, defined by the solid line, are excluded by either \'simple\' or \'extended\' dominance. The three remaining (un-dominated) strategies, that define the cost-effective frontier, are EXISTING PRACTICE, DCBE3 and FOBT1+DBCE3. In Table [4](#T4){ref-type="table"}, we report all corresponding costs, health benefits and cost-effectiveness ratios, and indicate the options that are \'simply\' dominated. In Table [5](#T5){ref-type="table"}, we report the incremental cost-effectiveness ratios for the options that survive the tests of \'simple\' or \'extended\' dominance; these options define the cost-effective frontier. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Updated costs and effects for Bolin et al\'s 13 tested strategies, assuming a five-year dwell time. ::: ![](1743-8462-2-10-2) ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Estimates of costs in 2002 prices, health benefits and cost-effectiveness from Bolin et al. \[10\], assuming a five-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- (a) (b) (c) (d) (a)/(c) (b)/(d) EXISTING PRACTICE 0 0 (Undefined) FOBT3 109,167,314 109,167,314 2,010 2,010 54,312 54,312 COL 195,987,062 86,819,748 1,166 -844 168,085 (Simply Dominated) FSIG5 205,570,434 96,403,120 3,365 1,355 61,091 71,146 FOBT1 230,156,355 24,585,921 4,447 1,082 51,755 22,723 DCBE5 253,881,621 23,725,266 5,050 603 50,274 39,345 FSIG3 271,052,169 17,170,548 3,909 -1,141 69,341 (Simply Dominated) COL10 278,701,522 24,819,901 3,718 -1,332 74,960 (Simply Dominated) DCBE3 307,911,416 54,029,795 6,184 1,134 49,792 47,645 COL5 360,264,079 52,352,663 6,181 -3 58,286 (Simply Dominated) FOBT1+FSIG5 364,595,167 56,683,751 5,849 -335 62,335 (Simply Dominated) FOBT1+DCBE5 373,803,843 65,892,427 6,573 389 56,870 169,389 FOBT1+FSIG3 420,786,416 46,982,573 6,032 -541 69,759 (Simply Dominated) FOBT1+DBCE3 424,911,339 51,107,496 7,020 447 60,529 114,334 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Incremental cost-effectiveness ratios in 2002 prices for the preferred (not dominated) strategies from Bolin et al. \[10\], assuming a five-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- (a) (b) (c) (d) (a)/(c) (b)/(d) EXISTING PRACTICE 0 0 (Undefined) DCBE3 307,911,416 307,911,416 6,184 6,184 49,792 49,792 FOBT1+DBCE3 424,911,339 116,999,923 7,020 836 60,529 139,952 Note: Other options rejected on the grounds of \'simple\' and \'extended\' dominance, see text for a discussion. ::: In Figure [3](#F3){ref-type="fig"}, we illustrate the costs and effects for all strategies reported by Bolin, assuming a 10-year dwell time. In this case EXISTING PRACTICE, FOBT2, DCBE5, DCBE3, FOBT1+DCBE5 and FOBT1+DCBE3 define the cost-effective frontier. In Table [6](#T6){ref-type="table"}, we report all corresponding costs, health benefits and cost-effectiveness ratios, and indicate the options that are \'simply\' dominated. In Table [7](#T7){ref-type="table"}, we report the incremental cost-effectiveness ratios for the options that survive the tests of \'simple\' or \'extended\' dominance; again, these options define the cost-effective frontier. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Estimates of costs in 2002 prices, health benefits and cost-effectiveness from Bolin et al. \[10\], assuming a 10-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- (a) (b) (c) (d) (a)/(c) (b)/(d) EXISTING PRACTICE 0 0 (Undefined) FOBT 24,429,089 24,429,089 352 352 69,401 69,401 FOBT20 31,633,953 7,204,864 558 206 56,692 34,975 FOBT15 37,620,230 5,986,277 681 123 55,243 48,669 FOBT10 47,472,852 9,852,622 951 270 49,919 36,491 FOBT5 75,561,379 28,088,527 1,674 723 45,138 38,850 FSIG 102,900,905 27,339,526 1,222 -452 84,207 (Simply Dominated) FOBT3 105,521,804 29,960,425 2,605 931 40,507 32,181 FSIG20 120,362,301 14,840,497 1,892 -713 63,616 (Simply Dominated) FSIG15 131,780,307 26,258,503 2,294 -311 57,446 (Simply Dominated) DCBE 134,697,657 29,175,853 1,896 -709 71,043 (Simply Dominated) FOBT2 140,117,791 34,595,987 3,551 946 39,459 36,571 FSIG10 148,850,170 8,732,379 3,127 -424 47,602 (Simply Dominated) DCBE20 158,390,199 18,272,408 2,939 -612 53,893 (Simply Dominated) DCBE15 173,448,383 33,330,592 3,566 15 48,639 2,222,039 COL 190,862,405 17,414,022 2,368 -1,198 80,601 (Simply Dominated) DCBE10 194,089,298 20,640,915 4,778 1,212 40,621 17,030 FSIG5 204,328,539 10,239,241 3,583 -1,195 57,027 (Simply Dominated) FOBT1 224,367,390 30,278,092 5,271 493 42,566 61,416 DCBE5 248,112,291 23,744,901 6,023 752 41,194 31,576 COL10 265,297,565 17,185,274 5,970 -53 44,438 (Simply Dominated) FSIG3 270,484,399 22,372,108 3,993 -2,030 67,740 (Simply Dominated) DCBE3 304,301,903 56,189,612 6,720 697 45,283 80,616 COL5 357,822,831 53,520,928 6,583 -137 54,356 (Simply Dominated) FOBT1+FSIG5 361,026,560 56,724,657 6,344 -376 56,908 (Simply Dominated) FOBT1+DCBE5 370,181,240 65,879,337 7,076 356 52,315 185,054 FOBT1+FSIG3 417,677,588 47,496,348 6,457 -619 64,686 (Simply Dominated) FOBT1+DCBE3 422,777,702 52,596,462 7,299 223 57,923 235,859 ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Incremental cost-effectiveness ratios in 2002 prices for the preferred (not dominated) strategies from Bolin et al. \[10\], assuming a 10-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- (a) (b) (c) (d) (a)/(c) (b)/(d) EXISTING PRACTICE 0 0 (Undefined) FOBT2 140,117,791 140,117,791 3,551 3,551 39,459 39,459 DCBE5 248,112,291 107,994,500 6,023 2,472 41,194 43,687 DCBE3 304,301,903 56,189,612 6,720 697 45,283 80,616 FOBT1+DCBE5 370,181,240 65,879,337 7,076 356 52,315 185,054 FOBT1+DCBE3 422,777,702 52,596,462 7,299 223 57,923 235,859 Note: Other options rejected on the grounds of \'simple\' and \'extended\' dominance, see text for a discussion ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Updated costs and effects for Bolin et al\'s 27 tested strategies, assuming a ten-year dwell time. ::: ![](1743-8462-2-10-3) ::: The incremental cost-effectiveness ratio derived from the Salkeld data, in 2002 prices, for FOBT, is \$30,556 per LYG, and the 2002 incremental cost-effectiveness ratios from the O\'Leary data are \$17,356 per LYG for FSIG and \$26,587 per LYG for COL. These results define the screening options that are preferred (ie, not dominated), by the measure of cost-effectiveness, for population-based CRC screening in Australia. However, the decision over which to choose depends on additional factors that we discuss next. Discussion ========== We defined average and incremental cost-effectiveness ratios and emphasise the latter are relevant for decision-making. We calculated incremental cost-effectiveness ratios, in 2002 prices, for a number of population screening strategies, for which Bolin had previously reported average cost-effectiveness ratios. For Bolin\'s estimates of (ΔC) and (ΔE), for a 5-year dwell time, we found only DCBE3 (\$49,792 per LYG) and FOBT1+DCBE3 (\$139,952 per LYG) were preferred (not dominated). For a 10-year dwell time, FOBT2 (\$39,459 per LYG), DCBE5 (\$43,687 per LYG), DCBE3 (\$80,616 per LYG), FOBT1+DCBE5 (\$185,054 per LYG) and FOBT1+DCBE3 (\$235,859 per LYG) were preferred (not dominated). Incremental cost-effectiveness ratios are useful for decision-making when a ceiling value for a LYG is specified. Bolin argued in 1996 \[[@B7]\] that \$US40,000 per LYG was the relevant cut-off (approximately \$AU65,449 in 2002 prices). We prefer a decision rule described by Garber & Phelps \[[@B25]\] that states a LYG is worth approximately twice the median annual per capita income. They derived this value from a model of optimal lifetime spending for medical care and explored its relationship to the cost-effectiveness criterion. They evaluated the model in terms of maximizing utility for individuals, with utility a function of income and health. Their rule implies, for Australia, a rational cut-off for one LYG is approximately \$AU39,000 \[[@B26]\]. If we apply this rule to our interpretation of Bolin\'s data, we wouldn\'t recommend any additional population screening activities for the 5-year dwell time, and for a 10-year dwell time, we would only recommend FOBT2. Based on the O\'Leary and Salkeld data we recommend COL and FOBT1, respectively. The results in Tables [4](#T4){ref-type="table"} and [6](#T6){ref-type="table"} illustrate the colonoscopy strategies, championed by Bolin, would cause between 33 and 1,322 years of life to be lost and between \$M17 and \$M87 to be wasted. We showed that all colonoscopy options were dominated by more cost-effective alternatives. Despite our re-analysis, the decision over which model of CRC screening is optimal for the Australian, average-risk population remains ambiguous. While the incremental cost-effectiveness ratios from two studies support annual FOBT \[[@B1]\] or biennial FOBT \[[@B10]\], the most recent study supports colonoscopy \[[@B17]\]. At least we now have incremental cost-effectiveness ratios in 2002 prices. While two of the studies agree that FOBT screening is preferred, we have not investigated why the O\'Leary \[[@B17]\] analysis leads to a different conclusion. The answer may be sought in a careful assessment of model structures, the particular perspectives adopted for each analysis and the values used for the parameters, which is beyond the scope of this commentary. In addition, we haven\'t attempted to model the effect of uncertainty on the conclusions. This would require access to the models, data and software used in each of the previous three studies. Policy-makers should review the three papers and make a judgement over which they believe produces the best estimates of change in cost and health benefit, identify their willingness to pay for the proposed health benefits and make their decision in the context of other logistic, social and political issues. Competing interests =================== The author(s) declare that they have no competing interests.	PubMed Central	2024-06-05T03:55:58.289772	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173078/", "journal": "Aust New Zealand Health Policy. 2005 May 18; 2:10", "authors": [ { "first": "Nicholas", "last": "Graves" }, { "first": "Loretta", "last": "McKinnon" }, { "first": "Barbara", "last": "Leggett" }, { "first": "Beth", "last": "Newman" } ] }
PMC1173079	Background ========== Coexistence of public and private health insurance, such as in Australia, has been the subject of intense debate among health economists and policy makers \[[@B1]\]. The main issue surrounding this debate has been how the mix of public and private health care financing influences the demand for private health insurance (PHI) and whether PHI takes pressure off the public system \[[@B2]\]. Falling PHI membership, observed since the introduction of Medicare in 1984, was thought to have increased the demand on the public system \[[@B3]\], prompting the federal government to implement policies aimed at encouraging possession of PHI to take the pressure off public hospitals and restore balance to the health care system \[[@B3]-[@B5]\]. Since 1995 three major policy reforms have been introduced in Australia \[[@B6]\]. Firstly, in 1995 selective contracting was introduced. The then federal Labor government passed legislation allowing private health plans to contract selectively with hospitals and doctors so as to improve competition on price and quality. Secondly, government subsidisation of PHI was introduced in 1997 by the Conservative coalition federal government as a means-tested rebate, capped at a flat amount irrespective of the cost of PHI, combined with an income tax surcharge for high income earners without PHI. The rebate part of the policy was subsequently replaced in 1999 by a non-means-tested 30% rebate for PHI available to everyone. Finally, in 2000, lifetime community rating was introduced. This policy relaxed the previous stringent community rating system by allowing the price of PHI to be varied according to the age at which a member joined \[[@B6]\]. The Australian Healthcare Agreement 1998--2003 committed the Commonwealth and states to review the relationship between PHI cover and the use of hospital services by private patients. The investigation of this relationship is becoming a priority as there is disagreement among commentators as to the financial efficiency of the 30% rebate \[[@B5],[@B7]-[@B10]\]. To date, analyses of the effects of policies aimed at supporting PHI in Australia have primarily centred on changes in the proportion of the population covered by PHI \[[@B3],[@B6],[@B11],[@B12]\]. However, changing the proportion of the population covered may not directly translate to increased utilisation and, therefore, reduce pressure on the public system. The relationship between PHI cover and type of hospital use is complex because the universality of Medicare in Australia means that anyone can be treated in a public hospital at no charge regardless of their insurance status. The right of individuals to choose between public and private insurance, regardless of the status (public or privately insured) of previous hospitalisations or the possession of PHI is protected by the principles of Medicare as set out in the Health Insurance Act of 1973. Thus switches between use of the public and PHI system are initiated solely by patients based on choice rather than being mandated by government, notwithstanding that this \'choice\' may be constrained in some instances by socioeconomic and locational access factors. We propose that since possession and utilisation of PHI are not equivalent, analysing the effectiveness of the recent government strategies in relieving the pressure on the public system cannot be accomplished by evaluations of changes in possession of PHI alone. Rather, changes in choice, as reflected by patient-initiated switching between the public and private insurance systems must be analysed. The aim of this study was to identify and measure changes in the behavioural patterns of switching between public and privately insured status for hospitalisation by the population of Western Australia using our novel couplet methodology for analysing longitudinal data. Method ====== Hospital morbidity data extraction and case selection ----------------------------------------------------- The WA Data Linkage System \[[@B13]\] was used to extract all hospital morbidity data system (HMDS) records from 1^st^January 1980 to 31^st^December 2001, containing encrypted patient identification and episode numbers, age, gender, date of admission, date of separation, payment classification (public, insured private, or \"other\"), and hospital type. The \"other\" payment category, which included the private uninsured (2.2% of the total episodes), workers compensation (1.8%), motor vehicle (0.7%), defence force personnel (0.3%) and Veteran Affairs (1.7%) classifications was removed from the data set, leaving only the categories of public and private insured. This was done because the study was concerned with elective shifts between PHI and public categories; not enforced payment classifications due to mandatory funding arrangements, or private episodes for which the patients paid the full cost. Data coverage ------------- This research has made use of records from the HMDS which is the inpatient information system for WA acute care hospitals. The data collected by the HMDS are patient identification, socio-demographic, services, administration and clinical diagnosis information. Every WA hospital defined as an acute care facility has to provide data via the Health Act or for private hospitals as part of their license and every free-standing day surgery unit must also provide data. The HMDS includes patient information for all acute A class hospitals, day surgery units, geriatric and psychiatric inpatient facilities in general hospitals and, as of July 1993, healthy newborn infants born in hospital. It does not include patients of stand-alone geriatric and psychiatric institutions, geriatric hostels or rehabilitation units; however, most metropolitan hospitals have geriatric and psychiatric units attached for which data are provided for the HMDS. Over the lifetime of this study there were no significant changes in the HMDS coverage. Formation of episodes of care and assignment of payment classification ---------------------------------------------------------------------- For each individual in the data set all eligible hospital records were grouped into episodes of care, using the separation and admission dates to define temporally contiguous periods of health care service utilisation. Thus one episode of care could have contained one or more inter hospital transfers. Each episode of care was assigned to one of the two eligible payment classifications (public or privately insured) on the basis of the initial payment classification at admission, where inter-hospital transfers were involved. Allocation of public or PHI status was based solely on payment classification and not on hospital type. This was done because the focus of this paper (consistent with the focus of public policy making in Australia) was on use of PHI rather than hospital type. Stratification by age group --------------------------- Stratification was based on age at admission of the first episode in each couplet. Each episode of care was assigned to one of four broad age categories (0--16 years, 17--39 years, 40--69 years and 70+ years) chosen to represent PHI market segmentation (children, young adults, middle age and old age) following consultation with a local private health insurer. Formation and classification of hospital couplets ------------------------------------------------- For each individual in the data set, eligible episodes of care were grouped incrementally, starting with the index (first episode of care) to form hospital couplets such that episodes 1 and 2 formed hospital couplet 1, episodes 2 and 3 formed hospital couplet 2 and so on (see figure [1](#F1){ref-type="fig"}). Hospital couplets were categorised depending upon the sequential combination of payment classifications of their two contributing episodes of care as follows: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The formation of hospital couplets. ::: ![](1743-8462-2-12-1) ::: \(i) Concordant public couplets containing only public episodes of care. \(ii) Concordant private couplets containing only privately insured episodes of care. \(iii) Discordant (mixed) public to private couplets had the first episode of care as public and the second as privately insured. \(iv) Discordant (mixed) private to public couplets had the first episode of care as privately insured and the second as public. Analysis of the mean intra-couplet interval ------------------------------------------- The intra-couplet interval was defined as the time in days between the final separation from the first episode of care and admission to the second episode of care. The mean intra-couplet interval was calculated for each couplet classification by the year of admission of the second episode of care (a term we subsequently refer to as the couplet year) and expressed as a ratio of the sum of the mean intra-couplet interval of all couplets in that year, regardless of classification. We subsequently refer to this measure as the grand mean. Thus a ratio greater than one for a particular couplet classification was indicative of an intra-couplet interval longer than the grand mean for that couplet year. The couplet year indicates the year the switch (choice) was made. Analysis of behavioural patterns in switching of payment classification ----------------------------------------------------------------------- The proportion of each type of discordant couplet relative to the total number of couplets having a first episode of care in the baseline payment classification was determined independently for all intra-couplet intervals (in whole years) in the data set. This analysis was performed separately for each age group for the whole observation period and two predefined time periods (1981 to 1989 and 1990 to 2001) chosen to represent the two main eras of health care policy in Australia. The first related to the removal and re-introduction of free public hospital care, while the second related to changes in federal health policies aimed at supporting PHI \[[@B5],[@B12],[@B14],[@B15]\]. Hospital couplets were partitioned into the two time periods using the year of admission of second episodes of care. The values obtained were plotted as segmented trend lines and the average rates of loss from each payment classification per year of intra-couplet interval were calculated using least squares fit. Results ======= Characteristics of the hospital couplets in the HMDS data file -------------------------------------------------------------- The HMDS data file contained data pertaining to 1,979,946 individuals of which 1,185,014 (60%) had at least one valid couplet. The 7,561,486 episodes of care recorded for these individuals formed 6,376,472 distinct hospital couplets (see table [1](#T1){ref-type="table"}). Of these, approximately one half (51%) had an intra-couplet interval of one year or greater. Details of the distribution of the hospital couplets by intra-couplet interval are shown in table [2](#T2){ref-type="table"}. Significant differences were observed in the characteristics of couplets having intra-couplet intervals of less than 1 year compared with those of one year or over. The largest differences were observed in relation to hospital type and couplet category. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of individuals, hospital episodes and hospital couplets ::: Characteristic Individuals Episodes Couplets(2^nd^Episode) -------------------- ---------------------- -------------- ---------------------------- --------- -------- --------- -------- Sex Male 522167 44.1 3156438 41.7 2634271 41.3 Female 662841 55.9 4405035 58.3 3742194 58.7 Indeterminate 4 0.0 13 0.0 7 0.0 TOTAL 1185014 100 7561486 100 6376472 100 Age Group 0--16 Years 313244 26.4 1045092 13.9 731848 11.5 17--39 Years 443602 37.4 2495039 33.1 2051437 32.3 40--69 Years 326974 27.6 2708572 35.9 2381598 37.5 70+ Years 100870 8.5 1292462 17.1 1191592 18.7 TOTAL 1184690 100^1^ 7541165 100^2^ 6356475 100^3^ Hospital Teaching 367964 31.1 2852257 37.7 2484293 39.0 Type Public Metropolitan 198456 16.7 902286 11.9 703830 11.0 Private Metropolitan 313563 26.5 1838354 24.3 1524791 23.9 Public Country 279315 23.6 1745527 23.1 1466212 23.0 Private Country 25097 2.1 208883 2.8 183786 2.9 TOTAL 1184395 100^4^ 7547307 100^5^ 6362912 100^6^ ^1^324 missing ^2^20321 missing ^3^19997 missing ^4^619 missing ^5^14179 missing ^6^13560 missing Missing records were caused by missing data in the relevant fields of the individual records and do not reflect deficiencies in the linkage process. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distribution of hospital couplets by characteristics and intra-couplet interval ::: Characteristic Intra-Couplet Interval (I-CI) ---------------------- ----------------------------------- --------- -------- -------- --------- -------- ------ Sex Male 1397957 44.6 53.1\* 1236314 38.2 46.9 Female 1739643 55.4 46.5\* 2002551 61.8 53.5 Indeterminate 5 0.0 71.4 2 0.0 28.6 TOTAL 3137605 100 49.2\* 3238867 100 50.8 Age Group 0--16 Years 318255 10.2 43.5\* 413593 12.8 56.5 17--39 Years 871234 27.9 42.5\* 1180203 36.5 57.5 40--69 Years 1271495 40.7 53.4\* 1110103 34.4 46.6 70+ Years 664439 21.3 55.8\* 527153 16.3 44.2 TOTAL 3125423 100^1^ 49.2\* 3231052 100^2^ 50.8 Hospital Type Teaching 1554708 49.6 62.6\* 929585 28.7 37.4 Public Metropolitan 249979 8.0 35.5\* 453851 14.0 64.5 Private Metropolitan 573729 18.3 37.6\* 951062 29.4 62.4 Public Country 672844 21.4 45.9\* 793368 24.5 54.1 Private Country 77205 2.5 42.0\* 106581 3.3 58.0 Other 9140 0.3 67.4\* 4420 0.1 32.6 TOTAL 3137605 100 49.2\* 3238867 100 50.8 Couplet Category Concordant Public 2104333 67.1 56.7\* 1605368 49.6 43.3 Private 887622 28.3 43.0\* 1176629 36.3 57.0 Discordant Public -- Private 74132 2.4 31.5\* 161140 5.0 68.5 Private -- Public 71518 2.3 19.4\* 296730 9.1 80.6 TOTAL 3137605 100 49.2\* 3238867 100 50.8 \* Statistically significantly different (at the 0.05% level) to the proportion of couplets with an intra-couplet interval 1 year or greater. ^1^12182 missing ^2^7815 missing ::: The distribution of first couplet episodes in each year of observation was more uniform in couplets with less than a one year intra-couplet interval compared with couplets longer intra-couplet interval or all couplets in the data file as shown in figure [2(A)](#F2){ref-type="fig"}. The reduction of first couplet episodes in couplets with intervals = one year was directly proportional to the number of years remaining in which a second episode (thus completing a couplet) could be observed. The lack of first episodes observed in 1980 was the result of a reduced volume of data in the original HMDS file for that year, most likely caused by an extraction error. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distribution of the proportion of first and second episodes in couplets by year. ::: ![](1743-8462-2-12-2) ::: Figure [2(B)](#F2){ref-type="fig"} shows the distribution of second episodes in couplets over the observation period. The proportion of second episodes increased over the observation period regardless of the duration of the intra-couplet interval. This was a function of the increased number of individuals eligible to complete a couplet with a second episode as time progressed. Mean intra-couplet interval --------------------------- The ratio of the mean intra-couplet interval observed for each couplet category relative to the grand mean by age group and couplet year is shown in figure [3](#F3){ref-type="fig"}. Discordant couplets had the longest intra-couplet intervals, having on average a ratio relative to the annual grand mean intra couplet interval of 1.35, while concordant couplets types had the shortest intra-couplet intervals, their ratio being 0.65. The overall pattern indicated that the longer the time between the first and second episode of a couplet, the more likelihood there was of a change in payment classification, especially where the first payment classification was private. The trends also indicated that, within each age group, individuals with public concordant couplets, on average, had shorter intervals between episodes (ratio 0.5) than individuals with private concordant couplets (ratio 0.8). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Ratio of the mean intra-couplet interval to the grand mean for each couplet category by age group and couplet year. ::: ![](1743-8462-2-12-3) ::: The removal / re-introduction of free public hospital care (1980 -- 1984) has previously been shown to have had a significant effect on use of in-patient insurance classifications \[[@B16]\]. In this study no significant difference was observed with regard to when the first episode of a couplet occurred (pre or post Medicare); however, the point in time of the second episode did correspond with behavioural change as can be seen in figure [3](#F3){ref-type="fig"}. Differences in trend were observed across the four age groups. The 0--16 years age group trended towards a reduction in the intra-couplet interval associated with discordant private to public couplets after 1985. This was not observed during the late 1980s and early 1990s in the other three age groups. The 40--69 years age group showed a trend pattern similar to the 17--39 years age group until 2000, when the average intra-couplet interval associated with discordant private to public couplets reduced sharply. In addition there was a sharp increase in the average intra-couplet interval associated with discordant public to private couplets (also observed in the 0 to 16 years age group) observed at that time. In the oldest age group (70+ years) the average difference in intra-couplet interval between public and private concordant couplets was much smaller than observed in any of the other three age groups. Over all age groups the trend in concordant couplets was that of a slowly reducing average intra-couplet interval, excluding the 1983 -- 1984 period, where the average intra-couplet interval for private concordant couplets increased in all age groups. Behavioural patterns in switching of payment classification ----------------------------------------------------------- Differences were observed in the rate of loss from the public and privately insured payment classifications across age groups as shown in table [3](#T3){ref-type="table"}. Across all age groups the largest losses, at all intra-couplet intervals, occurred from the privately insured payment classification. The largest differences in the rates of loss were observed in the 70+ years and the 17--39 years age groups, where an additional 2.39 and 1.78 percent of privately insured episodes, respectively, were lost for every year of intra-couplet interval. The rates of loss from the privately insured payment classification over shorter intra-couplet intervals (defined as 0 to 13 years) were greater than the rates of loss over longer intra-couplet intervals (defined as greater than 14 years) in all age groups with the largest difference being observed in the 40--69 years age group and the smallest in the 70+ years age group. The definition of short and long intra-couplet interval was based on an observed substantial change of slope (inflection) in the segmented trend lines. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Loss from each payment classification as a function of age and intra-couplet interval ::: Rate of Loss (percent per intra-couplet year) Difference in Rate (percent per intra-couplet year) --------------------- --------------------------------------------------- --------------------------------------------------------- ------ ------ ------ ------ 0--16 Years^\$^ 0.63 1.77 2.37 0.40 1.14 1.97 17--39 Years 0.49 2.27 2.39 0.55 1.78 1.84 40--69 Years 0.81 1.94 2.56 0.46 1.13 2.10 70+ Years 0.25 2.64 2.67 1.91 2.39 0.76 \* Averaged over all intra-couplet intervals in years. ^\$^Maximum intra-couplet interval 16 years as age determined at second episode. ::: Figure [4](#F4){ref-type="fig"} shows the degree of switching from private to public episodes over the two designated eras in health care policy. In all age groups the average rate of switching away from the private sector in 1991--2001 (era 2) was lower than that observed in 1981--1990 (era 1). The decrease in rate was small (average across all age groups -0.35% per intra-couplet year) but statistically significant. In addition, significance testing of the difference between each pair of proportions (era 1 versus era 2) indicated a significant difference for the majority as indicated in figure [4](#F4){ref-type="fig"}. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The proportionate discordance among hospital couplets with a private first episode by decade of the second couplet episode. \# Significant difference (p \< 0.01) between the percentage of discordant second episodes occurring in 1981--1990 versus 1991--2000. ::: ![](1743-8462-2-12-4) ::: We also found that as the intra-couplet interval increased, the difference between the proportions of discordant couplets having a private first episode versus a public first episode increased from 4.2% at one year to 34.3% at 18 years (data not shown). This indicated that overall, regardless of age, there was greater switching away from PHI than away from the public classification. Discussion ========== As expected, due to their greater capacity to move between respective payment classifications, private patients were found to be more likely to switch payment classifications in their next admission than public patients, irrespective of the length of time between the two episodes. There are a number of possible explanations for this including putative structural and cognitive reasons for the observed behaviour. Structurally, the average patient who begins with a public classification is likely to be of lesser socioeconomic means than the average patient who begins with a private insured classification. The option of the former patient to use PHI at the next hospitalisation will be dependant upon them taking out, or at the very least maintaining (if they had private cover at the initial episode but did not use it) private cover in the meantime. The same pre-requisite does not exist for an initially privately insured patient accessing a public classification on the next occasion. Also patients with private insurance experiencing trauma or an acute disease event may, in some circumstances, be admitted in an emergency as a public patient, thus to some extent disallowing the patient from exerting their preference. Cognitive explanations include the possibility that PHI may not be as entrenched culturally as Medicare (the public system) in this population and, as a consequence, use of PHI may be more dependant upon marketed value propositions than use of Medicare. In other words, PHI may be perceived by the populace as a market good, whereas the public system (Medicare) may be perceived as a fundamental right. We also found that the degree of switching from PHI towards the public system was inversly proportional to the length of time between episodes. Possibly, healthier individuals were more likely to switch to the public system than sicker individuals, assuming that a relatively short duration of intra-couplet interval can be taken as an indicator of increased morbidity. This phenomenon is consistent with reports that the decline in the proportion of the eligible population holding PHI since the introduction of Medicare in 1984 has been largely attributed to younger and healthier individuals dropping out \[[@B11],[@B12]\]. This finding indicated the presence of a substantial cross-subsidisation between low risk and high risk individuals further exacerbating the adverse selection price \'death spiral\' of PHI \[[@B11]\]. Improving the risk profile of PHI holders has been a major focus of recent federal government policy, with the introduction of a lifetime community rating in 2000 in an effort to encourage younger, healthier individuals to take out and remain in PHI funds. Our finding that in all age groups the overall rate of switching away from PHI was slightly higher in the period 1981 -- 1990 compared with the rate in the period 1991 -- 2001, suggests that these policies have had an effect on behaviour consistent with the government\'s intention. An alternative, but less likely explanation for our findings, may be systematic differences in the cognitive decision making process between long intra-couplet intervals compared with between shorter intra-couplet intervals. For example, individual historical preferences may play a role in short interval switching but may not be important over longer intervals, where decisions may be made in isolation. While it is important to consider this alternative, we feel that if the changes observed in switches away from PHI were largely due to differences in cognitive decision making, one would expect to see a similar phenomenon in switching away from a public classification. Such a phenomena was not observed in our data. In all age groups, our analysis indicated that the overall rate of switching from the private payment classification was slightly higher in the period 1981 -- 1990 compared with the rate in the period 1991 -- 2001, suggesting that the policies had an effect on behaviour consistent with the government\'s intention. Assumptions and limitations of the approach ------------------------------------------- This study made use of the WA Data Linkage Project which is unique in Australia and is one of only a small number of population-based record linkage systems in the world. The use of administrative data has strengths and weaknesses. For example data can be inaccurate due to recording or coding errors \[[@B17]\] or linkage errors. For this study individual patient records were linked by probabilistic matching, using an automated computer algorithm based on the probability of two records being from different people having the same identifier and two records from the same person having different identifiers. The probabilities were then aggregated into a score and checked against a threshold to determine if a match was made. This technique typically has been found to have a true positive predictive value of 95--99% and a negative predictive value of 98--99% \[[@B18]\]. Extensive validation of the quality of the performance of matching has been undertaken on the WA Record Linkage Project using sampling techniques and the proportions of mismatches and missed matches found were in the order of approximately 0.11% \[[@B18]\]. Linked data have the advantage of supporting a large and diverse research programme at relatively low cost, once the infrastructure is in place. They have the capacity to provide a population-based view of events experienced longitudinally by individuals across all institutions \[[@B17]\]. Given the objectives of this study, the latter point makes the use of linked data particularly appropriate. The approach we have taken in order to analyse the use of PHI and Medicare by the population of Western Australia is unique in at least two respects. Firstly the study was conducted at the population level, due to the use of hospital morbidity data. Therefore, the \"reference population\" was not merely an abstract concept as in conventional quantitative research, but an operationalised descriptor of the study sample. Secondly, our \'couplet methodology\' has not been reported previously in the literature and represents a new method of analysing longitudinal data on use of health insurance. The couplet methodology has enabled patterns to be measured based the behaviour of individuals rather than average shifts in private -- public mix generated from unlinked episodes of care. However, we recognise that since the analysis was based on episodes of care, patient initiated switching as part of a hospital transfer could not be analysed. This limitation may have affected our conclusions about intra-couplet intervals. This paper is dedicated to explaining the couplet technique for the first time and applying it to address an initial set of relatively descriptive questions (ie teasing out what is happening). The next stage of investigation is a more analytic analysis which recognises that a wide range of potential explanatory variables might predict the couplet-based phenomenon of switching (the why). For example hospitalisation rates in set periods before and after a couplet, stratified by different lengths of stay and/or different DRG weights. In addition, the admission types (emergency/elective) of the members of the couplet may also be important predictors. We feel that the couplet methodology described in this paper will enable significant inroads to be made into the investigation of such explanatory variables. Conclusion ========== Our study found that the population of Western Australia exhibited distinct behavioural patterns in the switching of payment classifications for inpatient hospitalisation between 1980 and 2001. Private patients were more likely to switch payment classification than public patients with shorter intervals between episodes corresponding to a greater probability of private-to-public switching. However, the average rate of switching from a privately insured classification was greater between 1981 and 1990 than between 1991 and 2001, indicating that recent health care policy reforms implemented by the federal government to promote uptake of PHI have had an impact on behaviour. Competing interests =================== Professor D\'Arcy Holman is an independent director of HBF Health Funds inc which is the largest provider of private health insurance in Western Australia. Authors\' contributions ======================= The manuscript has been read and approved by all authors and the requirements for authorship have been met as outlined below. REM was responsible for the conception and design of the study; analysis and interpretation of the data; and drafting and revising the paper. CDJH was responsible for conception and design of the study; interpretation of the data; and revising the paper. Acknowledgements ================ The initial construction of the Data Linkage System was funded by the Western Australian Lotteries Commission. We would like to thank the WA Department of Health for on-going support of the Data Linkage Unit.	PubMed Central	2024-06-05T03:55:58.294160	2005-6-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173079/", "journal": "Aust New Zealand Health Policy. 2005 Jun 27; 2:12", "authors": [ { "first": "Rachael E", "last": "Moorin" }, { "first": "C D'Arcy J", "last": "Holman" } ] }
PMC1173080	Background ========== In their recent review of protein sequence analysis in silico, Michalovich et al. \[[@B1]\] described the methodology for transferring functional annotation of known proteins to a novel protein. Computer-assisted technology is used to search for and assign the similarity from databases of well-maintained and previously annotated sources. Sequence-based and profile-based searches are conducted using BLAST and PSI-BLAST, respectively. Meanwhile, the Hidden Markov model is more efficient in searching for a distant family. Furthermore, structure-based annotation conducted by using a combination of PSI-BLAST and GenThreader (matching of substitution energy in evolution) may facilitate rapid functional annotation from structure \[[@B1]\]. However, proteins sharing the same fold can have different functions, and structure determination and analysis will not always mean that function can be derived \[[@B2]\]. There are examples of protein families, such as the four-helical cytokine and cytochrome super families, whose sequence similarities are either very low or not detectable \[[@B3]\]. Instead, their topography is more conserved than their sequences. This is rational, since protein functions are classified based on function per se, regardless of whether their sequences or 3D structures are similar or different. An example is the classification of a protein as possessing the function of lysozyme activity, as long as the protein possesses the ability of hydrolyzing peptidoglycans. Another direct approach for peptide QSAR has been simultaneously investigated in peptide sequence analysis \[[@B4]\]. A critical difference between those two approaches, namely bioinformatics and QSAR, is the prerequisite of 3D structure information on the basis of evolutionary conservation in the case of former; on the other hand, the 3D information is helpful but not always indispensable in the case of latter, by substituting with simpler steric parameters to account for the functional mechanism \[[@B5]\]. For example, Hellberg et al. \[[@B4]\] used altogether 29 properties of side chains of bioactive peptides. After dimension reduction using principal components analysis (PCA) the resultant three main PC scores, i.e., z~1,~z~2~and z~3,~representing hydrophobicity, molecular size and electronic parameter, respectively, were used as independent variables in regression analysis on the dependent variable of functionality \[[@B4]\]. Meanwhile, by using the homology similarity analysis (HSA), we have found the importance of functional segments within 15-residue sequences of lactoferricin derivatives to correlate with the minimum inhibitory concentration (MIC) \[[@B6]\]. Pattern similarity constant (a correlation coefficient) of the pattern of segments within a test derivative, in comparison to the reference pattern of the corresponding segment and the average of property values of the amino acid side-chains in the most potent derivative, was computed and correlated with MIC of the derivatives. In order to obtain the best (lowest) MIC, the pattern similarity should be close to 1.0 and the average property value should be close to that of the reference potent peptide (template). In the case of the above lactoferricin derivatives, higher correlation coefficients were obtained for log MIC predicted by HSA vs. measured log MIC computed as the output variables of regression ANN (artificial neural networks) than by sequence analysis based on the Hellberg approach \[[@B4]\]. More recently, a different approach, namely \"additive QSAR\" obtained by substituting with other amino acid residues at different positions in the same sequences, was reported to correlate well with peptide functions \[[@B7]\]. Lejon et al. \[[@B8]\] reported that in PCA analysis of peptide sequences, information of the positions of side chains in the sequence should be included for improving R^2^Xvalue compared to the results obtained by computing without side chain position data (0.99 vs. 0.60, respectively). Our HSA approach, by segregating segments with and without α-helix propensity, was in good agreement with theirs (R^2^Xof 0.90--0.94 compared to corresponding value of 0.60 but with a much larger number of derivatives). We have further extended this approach to infer the mechanism of emulsifying capacity of peptides with 10--32 residues as a function of hydrophobic periodicity \[[@B9]\]. For the study of emulsification function, a new homology similarity search (HSS) was introduced to plot similarity constants and average property values of segments (3--7 residues) by shifting the segment stepwise from N-terminus towards C-terminus of the sequences; the reference segment used was ELE, i.e., alternate cycle of charged (E), hydrophobic (L) and charged (E) residues. However, emulsification ability is a rather general function of peptides that is not dependent on specific active sites within the sequences; overall, the emulsification ability of peptides was highly correlated with hydrophobic periodicity of their entire sequences. There are cases of \"peptides\" which do not have definitive functional sites but requiring specific segments, or \"functions\" which do require neither specific sites nor segments. The lactoferricin derivatives described in the above study \[[@B4]\] are an example of the former since all of the mutants were prepared as derivatives of the corresponding wild-type lactoferricin 15-residue sequence, which has distinct helical and cationic segments. In contrast, the peptide emulsions \[[@B9]\] are an example of the latter. Typical examples of proteins with definitive functional sites are enzymes, for which the positions of active sites are critical to elucidate the functional mechanisms. Defective protein folding leading to amyloid fibril formation has been associated with various human diseases, such as Alzheimer\'s and Creutzfelds-Jacob diseases. In 1993, hereditary non-neuropathic systemic amyloidosis was reported to be caused by naturally occurring variants of human lysozyme that aggregated in the liver \[[@B10]\]. Similarly, cystatin C mutation in an elderly man was reported to be the cause of amyloid angiopathy and intracerebral hemorrhage \[[@B11]\]. The recent discovery of a new-fold enzyme named Cellosyl \[[@B12]\] led us to select the lysozyme family in this study as an important one to use for validating the HSS approach to search for functional sites \[[@B13]\]. Meanwhile, loss of papain inhibitory activity in recombinant human cystatin C was reported to be due to insolubilization \[[@B14]\]. Assuming that this loss was induced by an amyloidosis, changes of helix-to-strand in the inhibitory sites as well as the binding sites with papain could also be used for a rational example of application of the HSS approach in this study. The objective of this paper was to extend application of this new HSS approach to search for functional sites, such as active and substrate binding sites in lysozyme and amyloidosis of cystatin families, to verify the reliability of our new method. The intention was to validate the hypothesis that the evaluation of pattern similarity of short segments with 3--7 residues or even slightly longer in protein sequences is useful in predicting functionality, assuming that they are within allowable topographical units. Accordingly, it is not our intention to replace the 3D approach by the new peptide QSAR proposed in this study; rather, it is anticipated to be supplemental. Results ======= Lysozyme families ----------------- ### PCS classification Figure [1](#F1){ref-type="fig"} shows a scattergram derived from principal components similarity (PCS) analysis of 25 lysozyme sequences using the hydrophobicity index of side chains and hen lysozyme as a reference. A scattergram similar to this figure was also obtained when a charge index was used (instead of the hydrophobicity index) for classification. CH-type lysozymes used herein were from a fungus and Streptomyces globisporus. The v-type lysozymes were T4 and PA2 phage lysozymes, while g-type lysozymes were from goose, black swan, cassowary, ostrich and chicken G. Sixteen lysozymes (chicken C, human, horse, dog, rat, mouse, red deer, rainbow trout, pigeon, turkey, duck, California quail, Japanese quail, common bobwhite, fruit fly Drosophila, and tobacco hornworm) belong to the c-type family. The large deviations in slope (about 2) of CH-type lysozymes as seen in Figure [1](#F1){ref-type="fig"} are suggestive of explicit difference in their molecular structures from those of other lysozyme types. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### PCS scattergram of lysozyme families (C-, G-, V-, and CH-types) when hydrophobicity was used as property index.Hen lysozyme was used as the reference with \[coefficient of determination\] = 1.0 and \[slope\] = 1.0. ::: ![](1471-2091-6-9-1) ::: ### Segment pattern similarity search for active sites The HSS computer program was applied to the sequence alignment patterns shown in Figure [2](#F2){ref-type="fig"} using the active sites of hen lysozyme at positions 54--57(F^34^ESN) and 80--83(T^51^DYG) as the references. The position numbers in parenthesis are the positions in un-gapped sequences of individual lysozymes, while the immediately preceding numbers outside of parenthesis are the position numbers of the multiple sequence alignment (gapped) in Figure [2](#F2){ref-type="fig"}. Since the peptidoglycan-lysing activity is associated with glutamic 55(E^35^) and aspartic 81(D^52^) side-chains in the hen lysozyme \[[@B13]\], the segments flanking these residues in the sequences were the focus of pattern similarity search using the HSS. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Multiple sequence alignment of five lysozymes belonging to four families. ::: ![](1471-2091-6-9-2) ::: The search patterns illustrated in Figure [3A](#F3){ref-type="fig"} (charge) and 3B (turn propensity) are a trial run to validate the HSS approach, applied to human lysozyme vs. hen lysozyme, employing the five residues flanking E^35^(K^33^FESN of hen) as a search unit. The rules herein for selecting active segments are that the greater the similarity approaching to 1.0 and the nearer the average value to that of the reference, the more likely to be active sites in the test sequence. In Figure [3A](#F3){ref-type="fig"}, in addition to E^35^and D^53^, three residues of E^7^, S^80^and D^120^show about the same similarity constants (upper pattern) as well as similar average charge values (shown by arrows on the lower pattern). However, none of the above latter three positions are acceptable as the active site, as N^44^, I^59^, C^65^, Q^86^, and Q^126^in Figure [3B](#F3){ref-type="fig"} for β-turn search do not have matching peaks in Figure [2A](#F2){ref-type="fig"} for charge search. Therefore, only E^35^and D^53^are qualified as the active positions of human lysozyme. The same result was obtained when five residues flanking D^52^(S^50^TDYG) of hen lysozyme were used as an alternative reference segment. The two regions around E^35^and D^53^determined for human lysozyme are in good agreement with those being reported for the active site in the literature \[[@B13]\]. Despite the fact that the charge is of prime importance in defining active sites of lysozymes, the turn propensity values of the segment rather than its pattern similarity appear to play an important role in the enzyme, as exemplified in a low similarity value of 0.30 for D^53^compared to 0.98 for E^35^, whereas similar average turn values of 1.2 and 1.1, respectively, were computed (Fig. [3B](#F3){ref-type="fig"}). All these results may imply that the exposure of active sites to react with the substrate binding sites is essential. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### HSS search patterns for active sites of human lysozyme.Segment 53--57 of hen lysozyme was used as the reference. A: HSS search pattern based on charge. B: HSS search pattern based on turn propensity. ::: ![](1471-2091-6-9-3) ::: The same search was conducted for goose and T4 lysozymes as well as the new fold CH-lysozyme \[[@B12]\], with the assumption that it is an un-annotated sequence, to confirm validation of the HSS approach (Table [1](#T1){ref-type="table"}). Two positions of 79(E^73^) and 103(D^97^) were determined as the active sites of the goose lysozyme sequence. D^97^may not be essential for the catalytic activity of the goose lysozyme \[[@B15]\]. In comparison, 11(E^11^) and 20(D^20^) were noted for T4 lysozyme, which are in good agreement with those reported in the literature \[[@B13]\]. As shown in Figure [4](#F4){ref-type="fig"}, in the case of the sequence of Cellosyl (CH-type), 14(D^9^), 104(D^98^) and 106(E^100^) were identified as candidates to be the active positions of catalysis, which are in good agreement with Rau et al. \[[@B12]\]. These results would support the reliability of the HSS approach. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Determination of active sites in sequences of lysozymes in different families ::: Family Active sites Position Charge 1 Charge 2 Turn 1 Turn 2 -------- -------------- ----------- ---------- ---------- ---------- ----------- Hen 1 55(35)E 1.0/5.0 .98/5.0 1.0/1.12 2 81(52)D .98/5.2 1.0/5.2 1.0/1.21 Hman 1 55(35)E 1.0/5.9 .98/5.0 .98/1.17 2 81(53)D .98/5.2 1.0/5.2 1.01/1.22 Goose 1 79(73)E .67/6.9 .78/6.9 .06/1.03 2 103(97)D .83/7.3 .83/7.3 .67/,92 T4 1 11(11)E .61/4.5 .55/4.5 .76/1.16 2 20(20)D .55/5.4 .55/5.4 .53/1.07 CH 1 14(9)D .98/5.2 1.0/5.2 .95/.95 2 104(98)D .74/4.5 .65/4.5 .81/.92 3 106(100)E .95/5.6 .86/5.6 .87/.97 Position: the first number is the multiple sequence alignment position shown in Figure 2. Number in bracket is that in the sequence of each lysozyme. Charge 1 and 2 and Turn 1 and 2: Sites 54--57 and 80--83 (alignment position numbers in Fig. 2) were used as references. These position numbers correspond to 34--37 and 50--54, respectively in the sequence of hen lysozyme. The first and second digits n each cell of Charge and Turn are similarity constant and average of property value. ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### *HSS search patterns for active sites in Strep. coelicolorlysozyme.Segment 79--83 (STDYG) of hen lysozyme was used as the reference based on charge. ::: ![](1471-2091-6-9-4) ::: ### HSS search for substrate binding sites Compared to the active sites that are explicitly negative in charge at single positions, the substrate binding sites of lysozymes are rather loosely defined, mainly due to the greater complexity of determining the 3D structure of enzyme-substrate complexes than that of enzyme alone. It is generally agreed that the substrates locate inside the cleft formed between the helix lobe and the strand lobe of most lysozyme molecules, except for Cellosyl, the bacterial muramidase from Streptomyces coelicolor, which can be attributed to structural difference in the catalytic crevice \[[@B12]\]. The six-residue segment at alignment positions 84--89 (I^55^LQINS) of hen lysozyme was employed as the reference segment \[[@B16]\] using the hydrogen bonding scale (Table [2](#T2){ref-type="table"}). Based on the high pattern similarity constants and the average hydrogen bonding index values, eight potential sites were identified for hen and human lysozymes. The same rule as that for selecting active site was used herein for selecting of binding sites. Position 77(D^48^) and 188(W^111^) shown in \"Hen 1\" of Table [2](#T2){ref-type="table"} with lower similarity constant and hydrogen bonding value, respectively, may not be the potent substrate binding sites in the hen lysozyme. In the 3D structure \[[@B17]\], those two positions are far away from the catalytic cleft where the substrates snugly fit in. In the human lysozyme, 188(W^112^) is more likely to be the binding site than 102(A^73^), with higher pattern similarity and strength of hydrogen bonding of 0.89/0.45 than 0.80/0.32, respectively. Not only strong hydrogen bonding, but also high pattern similarity of a segment may be required to be qualified for substrate binding sites. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Determination of substrate binding sites in amino acid sequences of lysozyme families ::: Potential sites* --------- --------------------- ------------- --------------- --------------- ------------- --------------- --------------- --------------- --------------- Hen 1 55(35)E 70(42)A 77(48)D 84(55)I 90(61)R 102(72)S 117(83)L 118(111)W .81/.71 .79/.53 .34/.65 1.0/.41 .56/.58 .78/.56 .61/.42 .66/.35 Hen 2 36S 45R 49G 56L 62W 73R 90A 100S 112R .82/.71 .86/.44 .79/.47 1.0/.39 .59/.58 .91/.47 .98/.41 .55/.46 .76/.29 Human 1 54(34)W 70(42)A 77(49)D 84(56)I 90(62)R 102(73)A 116(83)A 118(112)W .48/.52 .42/.51 .34/.65 .91/.47 .50/.61 .80/.32 .61/.46 .89/.45 Human 2 27N 43T 50R 57F 63Y 74V 84L 101R 117Q .56/.39 .56/.51 .61/.47 1.0/.45 .55/.61 .72/.33 .48/.47 .58/.29 .79/.45 Goose 43(73)I 99(93)L 119(113)I 143(133)S 181(143)G 201(163)G .85/.31 .76/.34 .67/.46 .81/.55 .82/.36 .81/.58 T4 6(6)M 16(16)K 53(49)A 66(57)V 75(66)L 94(85)K 123(112)A 186(136)S 207(153)F .74/.34 .99/.38 .78/.30 1.0/.42 .52/.54 .67/.38 .73/.45 .58/.51 .78/.36 CH 40(35)T 47(42)D 55(50)T 80(74)A 92(86)W 124(118)T 153(147)C 186(180)T 1.0/.58 .63/.52 .80/.51 .51/.44 .73/.49 .83/.59 .51/.56 .56/.54 The first/second digits are similarity constants and average hydrogen bond index values of binding site with segments of sox residues beginning with the positions shown. Bold digits show more likely binding sites than non-bold digits. Hen 1 and Human 1: from the alignment shown in Figure 2; Hen 2 and Human 2; from the alignment of the lysozyme C family exclusively. ::: For the goose lysozyme, six potential binding positions were detected (Table [2](#T2){ref-type="table"}); I^113^and G^163^appear more likely to be the binding sites than other positions considering the location of the cleft in the molecule. Similarly, nine sites were found to be potential sites of T4 lysozyme, especially three positions, i.e. M^6^, L^66^and S^136^(Table [2](#T2){ref-type="table"}). In the case of Cellosyl, eight positions were identified as the potential binding sites (Table [2](#T2){ref-type="table"}). Probably due to the considerable 3D-structure difference of this lysozyme from those of other lysozyme families \[[@B12]\], the alignment positions 40--45 (T^34^EGTNY) instead of hen\'s 84--89 (I^55^LQINS) were used as a reference segment for obtaining more rational search results. Instead of hydrogen-bonding motivated interactions, less polar van der Waals interaction with the aromatic side chains in CH-lysozyme may be regarded as the second important stereochemical forces in the substrate binding \[[@B16]\]. In the literature, the most frequently cited substrate-binding sites in c-type lysozymes family have been W^62^and D^101^of hen lysozyme \[[@B13]\]. Since Figure [2](#F2){ref-type="fig"} includes the distant family of CH-type, the segment similarity search was repeated within the c-type lysozymes alone to restrict the search within similar fold. The results are shown as \"hen 2\" and \"human 2\" in Table [2](#T2){ref-type="table"}. Those results almost perfectly match to the substrate binding mechanism based on X-ray crystallographic analysis, e.g. D^101^, N^103^, N^104^, A^107^, V^109^, E^35^, N^46^, V^110^, E^52^, N^59^, and W^63^\[[@B15]\]. Almost all of these side-chains are very close or adjacent to the segments listed in \"Hen 1\" and \"Hen 2\" of Table [2](#T2){ref-type="table"}. ### Substrate binding sites reported by site-directed mutagenesis Among three mutants obtained by replacing W^62^with Y, F or H, the W62H mutant, and especially the double mutant W62H/D101G, reduced substrate binding drastically \[[@B15]\]. This change can be explained by a decrease in the hydrogen bond average value from 0.58 to 0.54 and from 0.46 to 0.29 in V62H and D101G, respectively, when the index values employed in this study were used in computation. The double mutant changed substrate-binding mode while maintaining the overall protein structure almost identical to that of the wild type \[[@B18]\]. An extensive cluster of hydrophobic structure is involved in distinct regions of the sequence, but is all disrupted by a single point mutation of W62G located at the interface of the two major structural domains in the native lysozyme \[[@B19]\]. Similar effects were observed in mutants Y63L and D102E of human lysozyme \[[@B20]\]. The double mutants R41N/R101S and V74R/Q126R of human lysozyme were better catalysts for lysis of Micrococcus lysodeikticus\[[@B18]\]. The average hydrogen bond value of both R41N and R102S was shown to increase in our HSS search, but similar effects could not be observed for V74R/Q126R. An interesting finding is that these two mutations have both resulted in the side chains being identical to those of hen lysozyme. R^41^and V^74^are near A^42^and A^73^, respectively. Importance of R^115^in substrate binding of human lysozyme was reported \[[@B21]\], which is in good agreement of W^112^within the same subsite F (Table [2](#T2){ref-type="table"}). Cystatins --------- ### PCS classification The PCS scattergram of the 17 cystatins using human cystatin C (HCC) as a reference and hydrophobicity as side-chain property index is shown in Figure [5](#F5){ref-type="fig"}; alteration of the index to α-helix and β-strand propensities did not appreciably change the grouping results. The three groups include human cystatins C, D, S, SA, SN and hen cystatin (Group I), human cystatins E, F and M (Group II), and human cystatins A and B (Group III) which are the stefin group cystatins that are smaller in molecular size and have slightly lower papain inhibitory activity than HCC \[[@B22]\]. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### PCS scattergram of cystatins when hydrophobicity was used as property index.Human cystatin C (HuC) was used as the reference. Human cystatins M, E, S, SA, SN, D, F, A and B are labelled as 1, 2, 7, 8, 9, 10, 15, 16 and 17. Labels 3, 4, 6, and 11--14 are for cystatins from mouse C, rat C, bovine, hen, rainbow trout, chum salmon and carp, respectively. ::: ![](1471-2091-6-9-5) ::: ### Active sites The HSS patterns using hydrophobicity to search for active sites of the four cystatins (HCC, EWC, HCA and HCB), when L^9^VGG of HCC has been employed as the reference, are shown in Figure [6](#F6){ref-type="fig"}. The active sites shown with arrows are at about the same location in all cystatins sequences used in this study. Similar results were obtained when bulkiness was used as side-chain index. In these cases, the similarity peaks are the major clue for identifying active sites. However, when similarity peaks appear in the neighbourhood, the average index values would become more reliable for identifying active sites as shown in Figure [6B](#F6){ref-type="fig"}; hen cystatin has two probable active positions side by side with the same similarity constants. The active sites, therefore, should be near the N-terminus with Prosite-type patterns of \[L,I,M\]-x(4)-G- \[G,A\]; the active sites are L^9^VGG and L^7^LGA in human cystatin C and hen cystatin, respectively. The similarity constants and the average hydrophobicity of the active sites are shown in Table [3](#T3){ref-type="table"}. Stefin B (HCB) shows much lower similarity constant and average hydrophobicity than those of other cystatins. This result is in good agreement with k~i~difference reported by Abrahamson \[[@B23]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Active and substrate binding sites of cystatins ::: HCC EWC HCA HCB ------------------------ ------------ ------------ ----------- ------------ Active site Segment L^9^VGG L^7^LGA I^2^PGG M^2^SGA SimConst/Av.Hydroph\* 1.00/1.35 0.87/1.80 0.97/1.00 0.55/-0.01 Binding site 1 Segment Q^55^IVAG Q^53^LVSG Q^46^VVAG Q^46^VVAG SimConst/Av.Hydroph 1.00/0.90 0.70/0.92 0.90/0.95 0.90/0.95 Binding site 2 Segment V^104^PWQG I^102^PWLN L^73^PGQN L^73^PHEN SimConst/Av.Hydroph 1.00/1.01 0.35/1.95 0.30/0.98 -0.16/0.42 SimConst/Av.Turn\\ 1.00/1.05 0.90/0.96 0.56/1.22 0.99/1.06 \* Pattern similarity constant / average hydrophobicity \\ Pattern similarity constant / average turn propensity Similar to the corresponding tables for lysozymes (Tables 1 and 2), Sim/Const is 1.0 for reference HCC, and the greater the Av.Property the greater the property strength. ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### HSS search patterns for active sites of cystatins against human cystatin C based on hydrophobicity.A: HCC (reference), B: EWC (egg white cystatin), C: HCA and D: HCB. ::: ![](1471-2091-6-9-6) ::: ### Substrate binding sites A HSA study similar to our previous paper \[[@B6]\] was conducted at the active and two binding sites of cystatins, yielding results (Table [3](#T3){ref-type="table"}) which are in good agreement with Turk et al. \[[@B22]\]. Substrate-binding site 1 had the pattern Q-x(3)-V- \[S,A\]-G, while substrate-binding site 2 had the pattern \[L,I,V\]-P-x(3)-x(3)- \[N,G\]. Similarity constants of binding loop 2 of egg white cystatin (EWC) and HCA are lower than that of HCC, whereas not only similarity but also average hydrophobicity are lower in HCB. Similarity constants at the active site (against 1.0 for HCC) using hydrophobicity index were \>0.8 for cystatins A, D, F and hen, \~0.5 for E and M, and 0.1--0.2 for cystatins B, S, SA and AN. Similarity constants at binding loop 2, when strand propensity was used for PCS computation, were lower for stefins A and B with values of 0.8 and 0.6, respectively, than \>0.9 for other cystatins. On the other hand, similarity constants for strand at binding site 1 were not much different among different cystatins, with values \>0.9 (not included in Table [3](#T3){ref-type="table"}). It is interesting to note that stefins A and B do not have the PW pair which is in the binding site 2 of HCC and EWC; instead they have PG and PH pairs, respectively (Table [3](#T3){ref-type="table"}). The W → G replacement increased strand propensity, while W → H replacement did so moderately. The values shown in Table [3](#T3){ref-type="table"} were almost inversely proportional to the equilibrium inhibition constant k~i~except for human cystatin S that was weak in the inhibitory activity, which might have been due to the difference in phosphorylation of serine at N-terminal region \[[@B23]\]. Although stefins A and B are classified differently from other groups on PCS scattergram (Fig. [5](#F5){ref-type="fig"}), the weak binding at the binding site 2 may not have considerable effects on the k~i~values. Turk et al. \[[@B22]\] have stated that the differences in the binding constants between cystatins and various cysteine proteases arise primarily from differences in the structure of enzyme active site clefts. The inhibition of endopeptidases, i.e. papain and cathepsins S and L, by cystatins is extremely tight and rapid, whereas the inhibition of exopeptidases, i.e. cathepsins B and H, is considerably weaker. The active site cleft of known endopeptidases is free to accommodate inhibitors, while in the case of exopeptidases, the active site cleft contains extra residues in it. In the N-terminal region of cystatins, it was observed that the affinity for target proteases decreased with both size and charge of substituting residues \[[@B22]\]. These observations are in good agreement with the results when bulkiness of side chains was used for the HSS computation for the binding site 1, \"SimConst /Av.bulk\" values were 1.00/12.3, 0.92/14.2, 0.98/11.4 and 0.91/11.34 for HCC, EWC, HCA and HCB, respectively. As expected, stefins A and B were less bulky. Furthermore, HCC and EWC include longer chains at the N-terminal sides with bulkier residues than those of stefins. These findings are in good agreement with the effect of bulkiness of G^4^in the stefin A sequence, implying that the bulkier the residue at position 4, the weaker the papain inhibitory activity \[[@B24]\]. Amyloidosis ----------- ### Lysozymes Similarity constants and average propensities of α-helix and β-strand were computed for G^54^IL and G^54^ILQIN of hen lysozyme (G^55^IF and G^55^IFQIN in the case of human lysozyme) as shown in Table [4](#T4){ref-type="table"}. The amyloidogenic mutant I55T showed increased strand propensity from 0.82 to 0.85, without a substantial change in the similarity constants. With regard to the helix structure, the similarity constant decreased from 1.00 to 0.93 while the average value increased from 1.08 to 1.14, thereby implicating a decrease in helix, since the helix index used herein is inversely related to the content of helix structure. These changes are favourable for amyloidosis. The fact that position 56 in human lysozyme is near its active position at D^53^may explain its dramatic effect on the enzymatic activity, more effective than other positions in the sequence. The six-residue computation did not show as clear a difference as was found in the three-residue computation. Nine double-site mutations in addition to I55T at other positions selected at random in the sequences of the amyloidogenic mutant I55T by using the RCG program did not restore the activity of wild-type lysozyme (unpublished). These results infer that the amyloid, once formed by detrimental mutation at the active site, cannot be restored by mutation at other locations in the sequence. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### HSA computation for I55T lysozyme ::: Hen Human I55T(Hen) ----------------- --------- ----------- --------------- Helix G^54^IL(F)QINS SimConst\* 1.000 0.917 0.952 Average 0.960 0.993 0.985 G^54^IL(F) SimConst 1.000 1.000 0.927 Average 1.082 1.082 1.141 Strand G^54^IL(F)QINS SimConst 1.000 0.978 0.997 Average 0.788 0.762 0.797 G^54^IL(F) SimConst 1.000 1.000 0.998 Average 0.822 0.822 0.845 L^56^for hen and F^57^for human. \* Similarity constant. ::: ### Cystatins Heat treatment of HCC induced its dimer formation at an early stage of separation, resulting in a complete loss of its activity \[[@B25]\]. Based on a dramatic decrease in the monomer form as shown by its CD spectrum, polymerization such as amyloidosis could be a cause of the loss of papain inhibitory activity of mutated HCC \[[@B14]\]. Of 35 residues (positions 1--35) of the helix domain of HCC, 17 residues were mutated in the 22 single-site mutants using the RCG program \[14 (Table [1](#T1){ref-type="table"})\]. When 33 mutants obtained by adding one extra residue each in both side of the original 17 residues after eliminating duplication were used for PCS computation, the resultant PCS demonstrated that helix propensity and bulkiness were playing important roles in thermostability (data not shown). Employment of three residues flanking the mutated residue was important in pattern similarity computation. This conclusion is in good agreement with Hall et al. \[[@B26]\] who did an exhaustive study showing that mutations at positions 8--10 enhanced thermostability of cystatin. With regard to the papain inhibitory activity, the importance of hydrophobicity and bulkiness was demonstrated (the PCS scattergrams, similar to Fig. [7](#F7){ref-type="fig"}, are not shown here). ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Effects of mutating the strand domain of human cystatin C (mutated 21 residues).A: Strand index, B: Helix index. The numbers show multiples of activity increases from wild-type. 1 shows no increase of inhibitory activity. 5\* is the reference mutant 12W86V with the highest inhibitory activity. ::: ![](1471-2091-6-9-7) ::: Of 86 residues in the strand domain (positions 36--121) of HCC, 21 residues (positions 36--120) were mutated in the 23 double-site mutants using the RCG program \[[@B14]\]. Thirty-seven residues were used for the PCS computation of single-site mutations as described above. Hydrophobicity appeared to be playing an important role in thermostability, while strand propensity was important for inhibitory activity (data not shown). Strand and helix propensities in the strand domain were influential to the papain inhibitory activity of HCC (Fig. [7](#F7){ref-type="fig"}). The figures show 12 data points only by eliminating data from single site mutation in the helix domain, which did not show distinct trends with broader scatter in these figures. The second mutations in addition to the above single mutations were conducted at the strand domain of the enzyme \[[@B14]\]. Coefficient of determination of 1.0 and slope of 1 indicate perfect match with the reference sample (5\) that is mutant G12W/H86V with the lowest strand propensity along with highest helix propensity in the strand domain among 23 double mutants. It is worth noting that the PCS is a classification program comparing pattern similarity without demonstrating quantitative relationships with functions but providing the information of the extent of involvement of side chain properties in the functions of interest. For mutant G12W/H86V that gained the greatest activity increase of 4.98 ± 0.09 times (mean ± SD at n = 3) that of recombinant wild-type \[[@B14]\], the strand propensity decreased from 0.78 to 0.69 (H86V) with a slight increase in the helix propensity (corresponding to decrease in the index values). The same was true for mutant D15P/H86I with 2.65 ± 0.30 times activity increase. A similar result was observed in mutant G4L/D40I with 2.11 ± 0.29 times activity increase, due to strand decrease along with almost no change in helix (D40I). However, mutant V10S/R93G with 4.50 ± 0.07 times activity increase behaved differently with increased strand and simultaneous decrease in helix. It is worth noting that the single mutation of V10S alone increased the activity 2.96 ± 0.06 times, therefore, changes in the helix domain may have a more predominant effect on the inhibitory activity than mutations in the strand domain. The activity change due to mutation helix → strand in the strand domain in the sequence may be slight in this case. It is well known that a single-point mutation of human lysozyme, namely I56T, has been identified as the origin of hereditary systemic amyloidosis \[[@B27]\]. The amyloidogenic nature of the lysozyme variants arises from a decrease in the stability of the native fold relative to partially folded intermediates. Accordingly, in a low population of soluble, partially folded species, the protein can aggregate in a slow and controlled manner to form amyloid fibrils. Similarly, sporadic amyloid angiopathy and intracerebral hemorrhage was reported in an elderly man due to cystatin C mutation \[[@B11]\]. In the case of human cystatin C, the decrease in strand along with an increase in helix might have prevented amyloidosis, despite the fact that helix change was not always as evident as in the case of lysozyme. Some inconsistency in the amyloidosis as a cause of inhibitory activity of human cystatin C in our mutation optimization \[[@B14]\] may be due to lack of the data of single-site mutation in the strand domain of the cystatin sequence. Unfortunately, the objective of that study \[[@B14]\] was for mutation optimization and not for investigation of the mechanism of amyloidosis. It has been reported that stefin B (HCB) readily formed amyloid \[[@B28]\], which may imply declined importance of the role being played by the binding site 2 in amyloidosis of HCC. Discussion ========== In a review on the quest to deduce protein function from sequences \[[@B29]\], the author stated that the searching of pattern databases would be more sensitive and selective than searching of sequence database. It was predicted that the sequence pattern databases, especially by comparing the pattern similarity, would play an increasingly important role, as the post-genome quest to assign functional information to raw sequence data gains pace \[[@B29]\]. Pattern similarity computation requires at least three residues in segments to represent a nonlinear curve, which is unlikely to be due to the effect of a single point mutation per se. With regard to an apparent effect of the single residue mutations of hen lysozyme on substrate binding, the structural analysis by NMR of the position-62 mutant of hen lysozyme \[[@B18],[@B30]\] found major changes in the chemical shift of back bone protons, especially in a loop region (positions 61--78), which contains W62 influencing the local folding. Similarly, Muraki et al. \[[@B20]\] reported that compared to the wild-type human lysozyme, the N-acetylglucosamine residue at subsite B of the L63 mutant markedly moved away from the 63^rd^residue, with substantial loss of hydrogen-bonding interaction. In Figure [5](#F5){ref-type="fig"} of Ref 17, involvement of not only Y63 but also W64 is evident. These results are supportive of the importance of pattern similarity of ≥ 3 residues, which are affected by single-residue mutation. The predictability of the active and binding sites solely on the basis of protein sequences \[[@B31]\] may be useful for investigating the underlying mechanisms of unknown functions of human genes after translation to protein sequences. Usually, two or three essential residues are directly involved in the bond making and breaking steps leading to formation of enzyme catalysis; however, the removal of an essential group often does not abolish activity, but can significantly alter the catalytic mechanism \[[@B32]\]. T4 lysozyme was cited by Peracchi \[[@B32]\] as an example of the alteration of catalytic mechanisms; the lytic activity of lysozyme changed to that of a transglucosidase. An approach utilizing the property of side chains in a sequence for identifying functional motifs has already been utilized in the computer-assisted selection of antigenic peptide sequences \[[@B33]\]. The authors stated that an antibody produced in response to a simple linear peptide with 7--9 residues in a protein would most likely recognize a linear epitope. Furthermore, this epitope must be solvent-exposed to be accessible to the antibody. In a large scale data mining study, Binkowski et al. \[[@B34]\] described the importance of local sequence and spatial surface patterns in inferring functional relationships of proteins. The general feature of protein structure that would correspond to these criteria could be turns or loop structures, which are generally found on the molecular surface connecting to other elements of secondary structure, and the area of high hydrophobicity, especially for those containing charged residues. Successful identification of active sites of new-fold of CH-lysozyme using the HSS approach in this study suggests that this approach could be applied to query proteins translated from unknown RNA segments of the human genes against templates with known functions, when their 3D structure information is still unavailable. It has been shown that the inhibition of the papain family by cystatin is due to a tripartite wedge-shaped structure with a good supplement to the active site clefts of the enzyme \[[@B35]\]. Todd \[[@B36]\] stated that despite highly homologous relationship as seen in Figure [8](#F8){ref-type="fig"}, lysozyme functions as an O-glycosyl hydrolase, while α-lactalbumin lacks this activity and instead regulates the substrate specificity of galactosyltransferase. The active site of peptidoglycan lysis is disrupted in α-lactalbumin. The ESS computation showed that although a pattern equivalent to hen\'s D^53^exists in the form of TEYG/YDYG, there is no residue equivalent to hen\'s E^35^around corresponding positions as in the form of HTSG/WESG. Two side-chain carboxyl radicals are required for the lysozyme activity within the crevice between the helix and strand domains of the molecule belonging to C-type family \[[@B16]\]. According to Alvarez-Fernandez et al. \[[@B35]\], the three parts of the cystatin polypeptide chain included in the enzyme-binding domain are the N-terminal segment, a central loop-forming segment with motif QXVXG and second C-terminal loop typically containing a PW pair \[[@B31],[@B37],[@B38]\]. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### An example of enzyme vs. non-enzyme*. Adopted from Figure 10.3b \[36\]. Fig. 8A: 1IWT human α-lactalbumin, Fig. 8B: 1B9O human lysozyme C using Swiss-PdbViewer (spdbv). Blast 2 sequences \[1\] showed that the identities and the positives between the two proteins were 35% and 55% respectively. ::: ![](1471-2091-6-9-8) ::: For multiple sequence alignment of an uncharacterized protein or peptide, many Web alignment servers are available for use \[[@B1]\], such as Blast and NPSA, as was done in this study. For classification of uncharacterized sequences, the PCS scatterplots are also useful as shown in Figures [1](#F1){ref-type="fig"}, [5](#F5){ref-type="fig"} and [7](#F7){ref-type="fig"}. The PCA demonstrated the classifying capacity superior to that of distance-based cluster analysis \[[@B39]\]. The PCS is more flexible than cluster analysis as different pattern similarity patterns can be drawn by rotating the reference segment for searching. It implies that similarity is not always \[1 -- dissimilarity\]. This difference resulted in the possibility of selecting outliers, which is critical in deriving true classes or ranking \[[@B40]\]. Most of the currently available peptide QSAR, such as the method of Hellberg et al. \[[@B4]\], intends to be based on whole sequence data. The new HSS approach reported in this study could be just the beginning of more detailed, reliable peptide QSAR to be developed in the future. Analysis of a variety of bioactive proteins contributing to human health is a potential future application of the HSS software package as well as multifunctional PCS. Considering the multifunctional nature of human diseases, the functionality of food proteins also can be manipulated based on combinations of bioactive segments in different or even single natural protein sequences. Therefore, for an uncharacterized protein or peptide, a new plan is proposed: (1) A reference sequence is chosen from multiple sequence alignment (MSA) as discussed above; PCS scattergrams would assist this selection in addition to BLAST search. (2) Based on segments with high similarity in MSA, segments to be used for search are selected within the reference sequence. Then, (3) HSS search is conducted to identify functional segments in the uncharacterized sequence. (4) From the above PCS computation, important PC scores are screened (PCA is a subroutine subprogram of PCS). (5) Regression neural networks are conducted using selected PC scores as input variables as exemplified in our lactoferricin derivative study \[[@B6]\]. (6) RCG would be useful for confirming the HSS data and also to find the best segment or sequence as exemplified in our HCC mutation \[[@B14]\]. One of the original purposes of our new approach in unsupervised data mining was to verify the hypothesis that there might be adaptability of different human cystatins to better inhibit different human cathepsins \[[@B41]\]. This hypothesis has not been fully pursued in the past, probably because of costly separation of pure cystatins and cathepsins. An advantage of our approach is to derive potential hypothesis for enzyme/substrate interactions exclusively from their sequence data. Although the verification of those hypotheses may need to await future 3D-structure study, it is important that most of the useful QSAR data could become available, which would promote the functional mechanism study based on 3D structure. However, we admit that more examples of application should be performed in the future to more thoroughly verify and establish this method for predicting functions based on sequences. This work is underway in our laboratory. Conclusion ========== Although the importance of pattern similarities of motifs with 20--30 residues as a whole has been reported for peptide QSAR in the past, the importance of a search for segments with three or more residues as functional sites of protein sequences has not been investigated. Lysozymes and cystatins were used as examples of proteins to demonstrate the capacity of segment pattern similarity analysis to predict functions, such as active and binding sites, amyloidosis and thermostability as a tool for quantitative functional sequence analysis. Methods ======= Amino acid sequences of proteins -------------------------------- Multiple sequence alignments of lysozymes were conducted using the Network Protein Sequence Analysis of Pôle Bio-Informatique Lyonnais \[[@B42]\] based on Clustal W. Similarly, multiple sequence alignments were obtained for human cystatins A (HCA), B (HCB) and C (HCC) and hen egg white cystatin (EWC) as well as for papain as host proteases of cysteine protease inhibitors, i.e. cystatins. For PCS analysis, a total of 17 cystatins were used: human A, B, C, D, E, F, M, S, SA, SN, hen (EWC), bovine, ratC, mouseC, Chum salmon, Rainbow trout and carp. Principal components similarity analysis of protein sequences ------------------------------------------------------------- The method described in the previous papers \[[@B9],[@B39]\] was followed. Principal components analysis (PCA) was modified to principal components similarity (PCS) by incorporating linear regression of PC scores to be able to account for more than three PC scores on a 2D scatter plot. The PCS was then modified to apply to peptide sequences. Homology similarity search -------------------------- Homology similarity search (HSS) was conducted as reported previously \[[@B9]\]. The similarity constant used in this study is eventually a correlation coefficient \[[@B43]\]. A preliminary study was carried out by changing the size of segment (normally 3--7) flanking the potential functional position to determine the most appropriate size of segment in differentiating the functional site from other segments within the sequence of lysozymes and cystains. The property indices used for amino acid side chains were hydrophobicity, charge, propensities of α-helix, β-strand and β-turn, hydrogen bonding, and bulkiness as reported previously \[[@B14],[@B44]\]. Segments with pattern similarity close to 1.0 and average values similar to that of the reference segment were sought within each gapped sequence. All software used in this study along with the instructions on how to use the computer programs are available in the form of ftp files on the Web \[[@B45]\] to download to PC computers. List of abbreviations ===================== EWC Egg white cystatin or hen cystatin. HCC Human cystatin C. HCA Human cystatin A or stefin A. HCB Human cystatin B or stefin B. HSA Homology similarity analysis: the PCS software was modified to compute pattern similarity constants and average side-chain property index values of segments in sequences \[[@B6]\]. HSS Homology similarity search: A step-wise search program initiated from N-terminus of query sequences by shifting the search unit (reference segment) towards C-terminus based on similar segments in terms of pattern similarity constant and average property values compared to those of template sequences \[[@B9]\]. MIC Minimum inhibitory concentration. PCA Principal components analysis. PCS Principal components similarity: PCA modified for multi-functional variables using linear regression of deviation of PC scores on the reference PC scores. Scatter plot is drawn as slope vs. coefficient of determination (r^2^) \[[@B44]\]. RCG Random-centroid optimization of site directed mutagenesis. QSAR Quantitative structure-activity relationships. Authors\' contributions ======================= EL participated in laboratory investigation to verify the hypothesis set in this study, while JD was taking care of the computer programming. SN was responsible mainly for the creation and application of software used in this study. All authors read and approved the final manuscript. Acknowledgements ================ This work was financially supported by a Multidisciplinary Network Grant entitled \"Structure-function of food biopolymers\" (Dr. Rickey Y. Yada of University of Guelph as the principal investigator) from the Natural Sciences and Engineering Research Council of Canada. The authors acknowledge the collaboration of all co-authors listed in our past publications as shown in the following references. Especially, the drawing of 3D structures to compare lysozyme and α-lactalbumin by Dr. Yasumi Horimoto is highly appreciated.	PubMed Central	2024-06-05T03:55:58.298066	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173080/", "journal": "BMC Biochem. 2005 May 18; 6:9", "authors": [ { "first": "Shuryo", "last": "Nakai" }, { "first": "Eunice CY", "last": "Li-Chan" }, { "first": "Jinglie", "last": "Dou" } ] }
PMC1173081	Background ========== Discovery of regulatory elements in human gene promoters is one of current bioinformatics challenges. Although transcriptional control mechanisms have been investigated in various organisms for at least three decades, it is still almost impossible to predict tissue-specific or developmental-stage-specific expression of a given gene by simply analyzing its promoter sequence \[[@B1]\]. The 5\' segment immediately adjacent to the TSS includes the core promoter and the proximal promoter, which usually extends about 200--300 nucleotides. This region is involved in the modulation of transcription. The distal part of a promoter is variable with respect to composition and length, which may encompass from 100 nucleotides to over 2 kb. There is no clear-cut defined 5\'-boundary for promoters \[[@B2]\]. Regulatory elements binding the same transcription factor can be found in different promoters as short DNA sequences, differing among them to some extent; they are, in general, from 5 to 25 nucleotides long \[[@B3],[@B4]\], often separated by un-conserved sequences. Control regions are modular in nature and expression of a given gene depends on specific combination of its regulatory elements and sometimes from their order and orientation \[[@B5]\]. Searching by computational methods for promoters and for regulatory elements in DNA sequences spanning several Kb, produces a large number of false-positive results. A possible solution to this problem is to identify a \"sheltered environment\" in which specificity of pattern discovery might be enhanced. Unknown binding sites for transcription factors might be detected by searching for common elements in upstream regulatory regions of genes with common biological function and/or expression. In fact, genes with similar expression are frequently co-regulated and genes with related function are often similarly expressed \[[@B6]\]. In this study, we attempted to detect putative regulatory elements in promoters of genes expressed in an adult human tissue (retina), by a multi-step approach involving computational analysis of large-scale expression data, selection of a subset of putatively co-expressed genes, retrieval of the upstream portion of their complete genomic sequence and application of pattern discovery on promoter regions. Results ======= Analysis of known regulatory sequence elements binding transcription factors ---------------------------------------------------------------------------- Before applying COOP software on a selected group of promoters, we attempted to exploit information on known regulatory sequences available in TRANSFAC \[[@B10]\], to establish some \"rules\" which could facilitate the discovery of novel regulatory elements. In particular, TRANSFAC matrix data were analysed in order to describe number, percent and localization of fixed and variable positions in consensus sequences. We first considered 385 matrices including information on mammalian regulatory elements. Average length of consensus sequences was 13.0 and mode 12; motifs of even length were more represented (even lengths two times more represented than odd lengths among consensus sequences of length ranging from 8 to 17). Less than 5% of the motifs showed only invariant positions (average and mode of length of completely invariant motifs were 10.3 and 9, respectively). About 33% of motifs showed more than 75% fixed positions (average length 11.5, mode 10), whereas about 73% showed more than 50% fixed positions (average length 12.3, mode 8). In general, the shortest the motif, the less variable appeared its consensus sequence. By separately considering three regions of consensus sequences (left, center and right), we observed that lateral positions are variable in 37% of sequences, whereas central positions are variable only in 20% of them. Most regulatory elements included in TRANSFAC seem to be symmetrical, being equally variable in their left and right sides. We obtained very similar conclusions from the analysis of the group of 610 eukaryotic matrices. Results of this analysis suggested that pattern discovery on mammalian promoter sequences might focus on patterns 10, 12 or 14 nucleotides long, showing from 0% to 25% variable positions, and possibly, less variable in the central region. COOP : Clustering Overlapping Occurrences of approximate Patterns ----------------------------------------------------------------- Since sequence signals with biological significance are frequently subtle, stringency of pattern discovery analyses in biological sequences cannot be set too high. This implies that results are often too numerous. A novel tool for Clustering Overlapping Occurrences of approximate Patterns (COOP) was implemented in Python (Figure [1](#F1){ref-type="fig"}). This software allows identification of tractable numbers of possibly interesting motifs, starting from large numbers of exact or approximate patterns. Selection of genes and retrieval of putative promoter regions ------------------------------------------------------------- Among 1,814 genes expressed in retina, statistical analysis of differential expression, by Audic and Claverie test \[[@B11]\], picked out 80 genes significantly more expressed in retina than in all other tissues. We selected as well a group of 59 known genes whose mutation is known to cause retinal diseases, recorded in OMIM and/or in RetNet databases, and/or encoding proteins for which a specific function in retina has been described. In total, 129 were selected. For each of these genes, the Reference Sequence or the longest sequence of the mRNA with complete CDS was compared by BLAT \[[@B12]\] to human genome sequence, for annotation of the intron/exon structure and for prediction of the most probable TSS (Transcription Start Site). We predicted with good confidence TSS of 90 genes (45 overexpressed UniGene clusters, 28 retinal disease genes, 7 both overexpressed and retinal disease-genes and 10 genes encoding proteins with a retinal function)\[[@B13]\]. Sequences from 90 selected genes, each corresponding to 1 Kb upstream the predicted TSS, were retrieved for further analyses. For one gene, USH3A (LL 7401) two alternative promoters (USH3A-A and USH3A-B) controlling transcription of messenger RNA encoding retinal products were found. Therefore, 91 gene promoters pertaining to 90 different retinal genes were considered for the study. Search for approximate patterns ------------------------------- ### Retinal datasets We considered datasets including sequences corresponding to the 500 bp upstream the predicted TSS of the selected genes (unmasked sequences, 500U91; partially masked, 500PM91; masked, 500M91) and the group of 91 fully masked sequences corresponding to 1 Kb upstream the predicted TSS (1000M91 dataset). Fully masking of sequences in datasets 500M91 and 1000M91 produced, on average, 435 and 788 unmasked nucleotides, respectively. In each group of sequences, approximate patterns of length ranging from 10 to 14 nucleotides, with at most two variable positions (10-2, 12-2, 14-2 patterns), were searched by SPEXS (Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}). For each dataset, patterns were ranked in different classes, according to the number of sequences in which they were represented (Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}, Figure [2](#F2){ref-type="fig"}). In the 500 bp datasets the number of sequences in which most represented patterns were found as relatively low, reaching a maximum of 44 (48%) for 10-2 patterns in unmasked sequences (500U91) and only 5 (5.5%) for 14-2 patterns in masked sequences (500M91). When considering sequences of length 1,000, the number of sequences with occurrences of most represented patterns slightly increased. For instance, most represented 10-2 patterns were found in 32 sequences in 500M91 and in 40 sequences in 1000M91. ### Negative control datasets Random groups of human promoter sequences were established as negative control. One thousand groups of 91 promoter sequences randomly selected among 17,156 (1 Kb long) human gene promoters were generated (RAN1000M91i, with iranging from 1 to 1,000). Each of these 1,000 groups included the same number of sequences of the 1000M91 set of retinal gene promoters and sequences fully masked and of the same length of retinal gene promoters. Moreover, TSS predictions were done by Promoser \[[@B14],[@B15]\] according to identical criteria adopted for TSS prediction of retina genes. Results of pattern discovery in the 91 retina gene promoters group (1000M91) and in the dataset of 1,000 groups of 91 human gene promoters randomly selected among 17,156 (RAN1000M91i) are shown in Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"} and in Figure [2](#F2){ref-type="fig"}. The number of patterns with different quorumin the retinal datasets and in the negative control groups of promoter sequences are reported. In the last two rows of Table [4](#T4){ref-type="table"} the comparison of pattern discovery results in the 1000M52 retinal gene promoters group and in the dataset of 1,000 groups of 52 sequences (RAN1000M52i) randomly selected among 17,156 human gene promoters is shown. The number of patterns found in at least 10 and in at least 20 out of 91 retinal promoter sequences is higher than the average of number of patterns found in at least 10 and in at least 20 out of 91 randomly selected promoter sequences, calculated over 1,000 different samples. Statistical significance (P-value) of number of patterns found in each number of sequences of the retina datasets was calculated as the proportion of negative control random datasets in which the number of patterns found was higher or equal to those found in retinal promoters. When this number was below 0.05, the difference was considered statistically significant. Over 1,000 random samples, 351.3 10-2 patterns were found in average in at least 20 promoter sequences, while 719 patterns were found in at least 20 retinal promoters (P-value = 0.017). Similarly, 714 12-2 patterns were found in 7 or more retinal promoter sequences, whereas 410.2 were found in average in random samples (P-value = 0.060). Statistical significance of this observed difference resulted just above 0.05. This effect could be due to possible heterogeneity of the class of 91 gene promoters. When considering the group of 52 promoters corresponding to genes significantly more expressed in retina than in all other considered tissues (1000M52) and the corresponding negative control dataset (RAN1000M52i), the difference between retinal and random datasets is more evident. For instance, 114 12-2 patterns were found in 7 or more retinal promoter sequences, whereas 33.7 in average were found in 1,000 random samples (P-value = 0.027). In Figure [2A](#F2){ref-type="fig"}, the cumulative number of patterns is plotted against the number of sequences in which they were found, among sequences of dataset 1000M52 and RAN1000M52i. In Figure [2B](#F2){ref-type="fig"} the comparison between the 1000M91 and the RAN1000M91i is shown. For negative control (random groups of promoters), the average value of 1,000 sets of sequences is given, with an interval of two standard deviations centred to the mean (e.g. 4,895 patterns were found in at least five sequences out of 91 in the retina dataset, whereas the average number of patterns found in at least five sequences out of 91 was 3,057 in the negative control dataset). It should be noticed that in both comparisons the number of patterns found in the retinal dataset is always considerably higher than average in the negative control dataset. Putative novel regulatory elements in retina gene promoters identified by COOP ------------------------------------------------------------------------------ We considered the group of 716, 12-2 patterns detected by SPEXS in at least 7 out of 91 promoter sequences of genes expressed in human retina (dataset 1000M91). We clustered the 6,611 occurrences of 716 selected patterns by using different sets of parameters, in order to identify combinations maximizing the biological meaning of resulting motifs. Distance parameters ranging from 2 to 5 nucleotides were used, each associated with ovalues of 0.6 or 0.7. The number of clusters decreases when dincreases and increases with o. For instance, when ois set to 0.6, the increase of dfrom 2 to 5 changes the total number of clusters from 211 to 183, changing the number of clusters containing only the occurrences of one SPEXS pattern (\"single clusters\") from 136 to 116, with maximum number of patterns per cluster increasing from 119 to 180. We kept ohigh, in order to cluster patterns according to their \"very frequent\" overlapping occurrences. The change of ofrom 0.6 to 0.7 changes the number of clusters obtained with d = 3 from 195 to 279, while the number of \"single clusters\" changes from 123 to 183. The number of clusters containing the occurrences of different SPEXS patterns is quite stable in respect to variation of o(72 and 97, respectively), whereas the number of \"single clusters\" increases considerably with o. In all cases, the number of extracted motifs, ranging in these experiments from 169 (d= 5, o= 0.6) to 300 (d= 2, o= 0.7), is considerably lower than the number of patterns considered. Starting from 6,611 occurrences of 716 selected patterns found in at least 7 sequences with dset to 3 and oto 0.7 COOP produced 279 clusters with an average number of occurrences per cluster of 13.9 (median 9 and mode 7, maximum 150). Clusters are associated to 279 sequence alignments, 89 of which (31.9%) are longer than 12 nucleotides, and to position-specific scoring matrices. In average, 13.9 occurrences per motif were observed. These 279 motifs occurred in 87 sequences out of 91. The mean number of motifs per sequence is 26.7, whereas the mean number of motif occurrences per sequence is 42.6. Subgroups of these motifs could be similar or overlapping. Motifs occurrences were ranked into four classes according to their position in promoter sequences (bp distance from the predicted TSS). The observed distribution significantly deviated from expectation under assumption of randomness (1,558 motif occurrences from -1 to -250; 1,002 in -251/-500; 626 in -501/-750 and 690 in -751/-1000; chi squared test: P = 3.03\10--121). There is a positive correlation between density of motif occurrences and proximity to TSS. In fact, over 40% of total motifs occurrences are concentrated in 250 nucleotides close to the predicted TSS and the two thirds of the total number of occurrences fall within 500 nucleotides adjacent to the first exon. Regions of promoters sequences corresponding to the 500 bp proximal to TSS were in average less masked than the upstream regions. When normalizing the number of motifs occurrences to the percentage of unmasked nucleotides in the different regions, a strong difference remains, with significant deviation from a random distribution. P value of chi squared test for the comparison among two groups (from -1 to -500 and from -501 to -1000) is 3.3\10--48. From each sequence alignment pertaining to a cluster, a consensus sequence representing the motif was built. Different sets of parameters were used to this purpose. The choice of different stringency of parameters for building consensus sequences influences the length of obtained motifs and the fraction of variable positions included in them. Threshold iranged from 0.1 to 0.5 and f~l~ and f~c~ ranged from 0.6 and 0.8 (data not shown). If f~l~ and f~c~are sufficiently stringent, a low value for i could be used, in order to maximize information extracted. When a i= 0.1 is applied for construction of consensus sequences, the average length of motifs remains almost unchanged (12.4), with 89 motifs of length over 12. Part of the information about each motif is lost when a matrix is built from the alignment of pattern occurrences and when frequency data are converted into a consensus sequence. We used moderately stringent thresholds for the minimum frequency of a single nucleotide determining \"fixed\" positions. In particular, f~l~ was set to 0.5, whereas f~c~, referring to the core regions of consensus sequences, was set to 0.7. Out of 279 motifs, only 62 \"most informative\" motifs were selected, which showed a completely conserved consensus or a consensus showing variable positions only in the side regions (Table [5](#T5){ref-type="table"}). The average length of this group of sequence motifs was 13.0, with about 55% of them longer than 12 nucleotides. Two pairs of motifs were represented by the same consensus sequence. The resulting group of 60 motif consensus sequences, representing putative functional elements in retinal promoters, were compared with TRANSFAC data, by TESS program \[[@B16]\]. In particular, we used the \"Filtered String-based Search Query\" tool \[[@B17]\] for comparison only with known regulatory sites of mammals, with no mismatches allowed and by using the entire length of known sites instead of their core positions. Out of 60 motifs, 53 (88%) exactly matched at least one sequence known to bind a mammalian transcription factor. Sequences corresponding to common or general transcription factors (Sp1, Sp3, MAZ, GCF, CUP or Yi) were matched by 47 out of 60 consensus sequences (78%), 22 of which matches also additional factors (AP-1, AP-2, WT1, Krox-20, GR, PPUR or ER). In total, 26 consensus sequences resulted similar to sequences recognized by AP-1, AP-2, WT1, Krox-20, GR, PPUR or ER. Analysis of positive control datasets ------------------------------------- To the purpose of both analysing COOP efficiency with selected conditions and of indirectly comparing the performances of the method with those of different software, we analysed by COOP a group of 26 human positive control datasets prepared by Tompa and colleagues for the assessment of computational tools for the discovery of transcription factor binding sites \[[@B18],[@B19]\]. This benchmark included 26 groups of promoters, for which it is known which regulatory signals should be detected. The number and the length of promoter sequences and of known signals per dataset are reported in Table 6, along with the adopted quorum \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. The number of signals per group of promoters which are shorter than the length of approximated patterns taken as input by COOP (12 nucleotides) and which were, in principle, very difficult to find by adopted settings, is also reported. For each experiment, among different motifs predicted we selected only the one corresponding to the motif represented in the highest number of sequences. For each dataset the nucleotide-level and site-level \[[@B18]\] overlap between pattern occurrences belonging to the cluster (i.e. the selected motif) and known signals, evaluated by different measures (see Methods), are included in Table 6 \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Moreover, the \"combined\" statistics summarizing COOP performance over the collection of human datasets, was compared with the same statistics calculated for the 14 different programs tested by Tompa and colleagues on the same datasets \[[@B18]\]. Results are presented in Figure [3](#F3){ref-type="fig"} and in Table 7 \[see [Additional file 2](#S2){ref-type="supplementary-material"}\]. The comparison showed that the performances of COOP on the \"very difficult\" human dataset are in line with those of the top rated tools. In comparison with other software, COOP resulted to be in the best 20% of evaluated tools according to four measures (nSn, nPC, sSn, sASP) and in the best one third according to six measures (nSn, nPC, nCC, sSn, sPPV, sASP). Discussion ========== Pattern discovery in sequences of putative and sometimes incomplete promoters is a considerably complex problem \[[@B20]-[@B25]\]. It may be reasonably assumed that some regulatory regions of a group of co-regulated genes share similar sequence elements. In yeast, pairs of genes showing over 0.84 Pearson correlation between their expression profiles, have over 50% probability of sharing at least one common transcription factor binder \[[@B26]\]. Since patterns with biological significance could be subtle \[[@B27]\], a main difficulty in pattern discovery approaches is a priori establishing a \"quorum\" and defining search parameters (e.g. pattern length, number of allowed wildcards or distance of occurrences from the model). By increasing the distance and/or decreasing the quorum, the number of false positives becomes excessively large. A possible solution to the problem of output explosion is to use biological knowledge both before and after application of automated pattern discovery. We analysed TRANSFAC data to obtain information about frequent properties of regulatory elements. It should be noticed that although not all TRANSFAC matrices are based on high-quality data and on large samples of sequences, this database represents the largest existing collection of known regulatory elements in different organisms. In this study we developed a novel tool, called COOP, for analysing promoter sequences of putatively co-regulated genes, aiming at extraction of sequence motifs with possible regulatory function. The motif extraction method is based on Clustering of Overlapping Occurrences of approximate Patterns, which allows identification of tractable numbers of possibly interesting motifs, starting from large numbers of exact or approximate patterns. Our method is somehow related to two approaches proposed by van Helden \[[@B28]\] and by Caselle \[[@B29]\], although these studies considered yeast promoter sequences and dealt with exact patterns. The originality of our approach, mainly resides principally in adopting a new similarity measure between patterns, based on the frequency of pattern co-occurrences, and in designing a flexible procedure, with seven parameters which could be varied in order to modulate stringency of different analysis steps. Motifs reconstruction was designed to maximize information included in each extracted sequence motif avoiding generation of spurious elements, given that clustering parameters (dand o) are appropriately set. Each obtained motif is represented by a consensus sequence, derived from the alignment of strings grouped in a specific cluster by adjustable criteria. In particular, the ithreshold affects the length of the consensus sequence, whereas the l, f~c~and f~l~determine the number and the distribution of variable positions in the consensus sequence. In order to evaluate the performance of the method, we analysed positive control datasets, such as all the human benchmark groups of promoter sequences, containing known signals at known positions, proposed by Tompa for a systematic assessment of motif discovery tools. COOP analyses were carried out with the same settings used for analysing retina gene promoter. The quorum was established by using an unique criterion for different datasets, based on the total number of sequences in the sample. In the first analysis phase, approximated patterns of 12 nucleotides, with at most 2 variable positions, were searched. Benchmark datasets were 26 groups of different number of promoters sequences. Each set of sequences contained a group of known signals. Lengths of signals ranged from 4 nucleotides to 71 nucleotides. Thus, several datasets contained a number of very short signals, which were very hard to find by a motif discovery approach designed to find motifs of length equal or higher than 12 nucleotides. We predicted a motif for all but two human datasets, which included only two sequences. COOP performed comparably well than the tools which were top rated in the Tompa assessment. It should be noticed that some of the other tools gave no predictions for a number of datasets, thus being advantaged from the averaging nPPV, sPPV and nCC scores when calculating combined statistics over all the human datasets \[[@B18]\]. The method we developed was applied, in a case study, to a collection of human promoter sequences pertaining to a group of 91 putatively co-regulated genes expressed in the retina. One Kb long promoter sequences were identified by predicting the most probable TSS according to the consensus of information about cDNA and ESTs alignments with the genome sequence. Even neglecting the possible presence of alternative promoters, definition of exact(s) TSS it is a still open problem, because of low sensitivity of promoter prediction programs and of incomplete cDNA coverage of 5\' exons. However, cDNA coverage of the majority of genes selected for this study is almost complete, since genes considered are well known and/or highly expressed. The adequacy of our method for selecting gene promoter sequences is supported by results obtained by Trinklein and colleagues \[[@B30]\]: over 90% 152 human 600 bp promoter sequences, randomly selected among 10,276 TSS predictions (based only on alignment with full-length cDNA clones from Mammalian Gene Collection) resulted active in at least two cell lines. Although the size of 1 Kb upstream the TSS might be insufficient to cover all possible regulatory regions for all genes, it could reasonably include both core and proximal promoters and at least part of the distal promoter. A very high number of approximate patterns of length from 10 to 14 with at most 20% variable positions was found in retina promoter sequences. This is partially due to the fact that they include similar sequences or sequences shifted only by few positions. By using COOP for clustering approximate patterns, on the basis of their frequent overlapping occurrences, we identified a number of interesting sequence motifs, often longer than the original patterns. In order to test the significance of the group of sequence patterns frequently found in retinal promoters and used as input for COOP, we generated and studied different negative control datasets corresponding to a thousand groups of sequences, randomly selected among a very large group of human promoters. The average pairwise level of pattern sharing in these groups of promoters was expected to reflect the general level of pattern sharing between human gene promoters. We observed that the selected group of retinal promoters (pertaining to a sample of genes putatively co-expressed, co-regulated and/or with similar function) is enriched in common patterns as compared to random groups. Sequence motifs produced by COOP resulted more frequently in the regions close to the TSS. Moreover, the group motifs consensus sequences selected according to very low variability in their central region was compared with sequences which are known to bind mammalian transcription factors under stringent criteria. Signals similar to those for general and widespread transcription factors, such as Sp1 or MAZ, are the most represented. Moreover, a number of selected motifs resulted to be similar to signals recognised by transcription factors expressed in tissues of ectodermal origin and relevant to development and function of retina (AP-1, AP-2, WT1, Krox-20, GR, PPUR or ER). For instance, AP-1 elements were found in a number of retinal gene promoters including cGMP-phosphodiesterase beta subunit \[[@B31]\] and hydroxyindole-O-methyltransferase \[[@B32]\] whereas the WT1 zinc finger factor is essential for normal development of retina and specifically involved in regulation of retinal genes \[[@B33]\]. Conclusion ========== We developed a method to detect sequence motifs corresponding to putative regulatory elements in gene promoters, starting from lists of approximate patterns with occurrences in promoter sequences. This method could be profitably applied to different datasets, including promoter sequences of different groups of genes in humans or in other Eukaryotes, for which co-regulation could be demonstrated or inferred. The method could be used to investigate on different kinds of regulatory sequences, such as intronic enhancers, or other sequence motifs with non-regulatory function. Availability and requirements ============================= COOP can be downloaded free-of-charge from the web page <http://telethon.bio.unipd.it/bioinfo/Retina/suppl_material.html>. COOP was developed in Python. The software works under Linux and requires Python 2.3 or higher, BioPhyton 1.40b and ClustalW. COOP is provided \'as is\' with no guarantee or warranty of any kind and it is freely available for all. Methods ======= COOP : Clustering Overlapping Occurrences of approximate Patterns ----------------------------------------------------------------- COOP takes as input a FASTA file of nucleotide sequences and a list of patterns with their number of occurrences in sequences or in their reverse complement. We used the SPEXS program \[[@B7]\] for producing the list of approximate patterns frequently represented in selected groups of promoter sequences. SPEXS code is freely available and it provides a number of advantages in terms of execution time and flexibility of parameters determining search conditions and output appearance. Seven COOP parameters can be varied in order to select stringency at different stages of the analysis (Table [1](#T1){ref-type="table"}). In the first step, patterns represented in more than qpromoters are searched by COOP in promoter sequences. In particular, both direct and reverse complement sequences of each pattern are compared against promoter sequences in order to collect pattern occurrences. Then, pattern occurrences (strings) are clustered according to a similarity measure based on frequency of their co-occurrences and by a joining algorithm derived from the so called \"quick-find\" algorithm \[[@B8]\]. In order to be included in the same cluster, two different strings must occur in promoter sequences much more frequently together than separately. Given the physical distance between pattern occurrences, measured as nucleotide distance between the 5\'-ends of two corresponding sequences, the threshold ddefines the maximum value for the distance between two pattern occurrences to be considered overlapping. Threshold oindicates the minimum ratio between observed overlapping occurrences of two strings and their average number of occurrences, allowing their inclusion in a unique cluster. The total number of clusters obtained in this way is influenced by the number of pattern occurrences to be clustered, depending on the qparameter, and depends as well on values selected for the minimum distance between patterns and for threshold o. Once clusters are obtained, all sequence elements corresponding to pattern occurrences belonging to each given cluster are multi-aligned by ClustalW \[[@B9]\], with Gap Opening Penalty set to 100. Each alignment is then analysed in order to build up a matrix describing nucleotide counts in alignment positions. The cell x~A1~of the matrix contains the number of times the A nucleotide has been observed in the first alignment position. Later, consensus sequences are built from matrices (Table [2](#T2){ref-type="table"}). The maximal number of adjacent matrix positions fulfilling the established threshold for i(minimum ratio between number of nucleotides per alignment position and total number of lines in the alignment) is further analysed to determine the mnucleotides long motif consensus sequence. Once the extension of lateral regions (l, ranging from 0 to m/2) is fixed, motif positions are considered invariable if the frequency of a single nucleotide exceed selected thresholds (f~l~ for lateral regions and f~c~ for the core region of the motif). Once the length of motif and of lateral regions is known, the extension of the core region is fixed. Moreover, an additional procedure is available, which uses IUB/IUPAC nucleic acid single letter, double-degenerate codes (M = \[A\\|C\], K = \[G\\|T\], S = \[G\\|C\], W = \[A\\|T\], R = \[G\\|A\], Y = \[T\\|C\]) and four-degenerate code (N = \[A\\|G\\|;C\\|T\]) and follows IUPAC rules for string consensus reconstruction: i. A single nucleotide is shown if its frequency is greater than 50% and at least twice as high as the second most frequent nucleotide; ii. A double-degenerate code indicates that the corresponding two nucleotides occur in more than 75% of the underlying sequences (but the criteria for a single nucleotide assignment are not met); iii. All other frequency distributions are represented by the letter \"N\". The output of the program is a collection of sequence clusters, each one representing a sequence motifs. Each cluster is associated to an alignment, to a matrix describing nucleotide counts in alignment positions and, ultimately, to a consensus sequence. Moreover, information about promoter sequences and nucleotide positions in which each cluster string occurs is given. Selection of genes and of putative promoter regions --------------------------------------------------- ### Analysis of genomic expression data For the study, genes significantly more expressed in retina than in all other tissues were identified by analysis of genomic expression profiles of several human tissues. Genomic expression profiles were reconstructed in silicoby using 41 unbiased (un-subtracted and/or un-normalized) UniGene cDNA libraries pertaining to 11 adult human normal tissues (retina, bone, hyppocampus, liver, lung, marrow, melanocyte, muscle, pancreas, prostate and testis) for which at least 6,000 ESTs per tissue were available \[[@B34]\]. The whole dataset included 270,871 ESTs, corresponding to 27,924 UniGene \"clusters\". The expression profiles were merged in an expression data matrix, which was then analysed by the Audic and Claverie test of differential expression in order to identify genes significantly more expressed in retina than in all other tissues considered. Significance threshold was set to a = 0.05. ### Disease genes and genes encoding proteins with a specific function in the retina By searching in OMIM \[[@B35]\], Retinal Information Network \[[@B36]\], LocusLink \[[@B37]\] and GeneCards \[[@B38],[@B39]\] we selected a group of known genes whose mutation causes retinal diseases and genes encoding proteins which play specific functions in retina. ### Retrieval of putative promoter regions Reference Sequences of selected genes (RefSeq) \[[@B37]\] were extracted from corresponding LocusLink entries. When RefSeq for a gene was unavailable, the longest mRNA sequence with complete CDS was used. Sequences were then searched by BLAT \[[@B40]\] against release 15 of human genome sequence, for prediction of Transcription Start Site (TSS), obtained by analysis of mRNA/genomic DNA alignment, 5\' ESTs placement and Acembly gene boundaries \[[@B41]\] annotation. Genomic sequences of 1 Kb upstream the predicted TSS were retrieved. These sequences were masked by RepeatMasker \[[@B42]\], in order to remove repetitive DNA. Negative controls ----------------- Negative control groups of promoter sequences were established as 1,000 sets of promoter sequences, sampled at random among 17,156 human gene promoters. Reference sequences of 27,427 human mRNA were obtained from GenBank. A group of 20,315 reference sequences was obtained, after exclusion of all sequences referring to unknown genes (chromosome open reading frames, hypothetical or predicted proteins) or to genes including in their sequence record words referring to vision, eye or retina. Retrieval of promoter sequences corresponding to 1 Kb upstream the predicted TSS of genes was done by PromoSer \[[@B14],[@B15]\]. TSS prediction options were set in order to retrieve for each gene 1 Kb upstream the most 5\' TSS, with the same criteria used for retrieval of retinal gene promoters. In addition, exclusion options of PromoSer were set to extract at most one promoter per gene and to avoid retrieval of overlapping sequences. By this way, 17,156 promoters were localized and retrieved. A Python script was developed for iterative random extraction of groups of m sequences from a list of N sequences, with N \> m, without repetition. Analysis of known regulatory sequence elements binding transcription factors ---------------------------------------------------------------------------- We used the TRANSFAC version available through BIOBASE. TRANSFAC \[[@B43]\] is a database of eukaryotic cis-acting regulatory DNA elements and trans-acting factors containing information on transcription factors, regulated genes, regulatory sites and nucleotide distribution matrices for binding sites of transcription factors. By using Perl scripts developed to the purpose, flat files pertaining to matrices data were parsed in order to extract information about consensus sequences length and about number, percent and localization of fixed and variable positions. Analysis of positive control datasets ------------------------------------- We analysed by COOP groups of promoters for which it is known which regulatory signals should be detected. Analyses were done of the complete set of groups of human sequences included in public benchmarks prepared by Tompa and colleagues \[[@B18]\], including three different types of sequences: 9 groups of real genomic promoter sequences containing known transcription factors binding sites, 9 groups of randomly chosen human genomic promoter sequences in which the binding sites were planted and 8 groups of sequences randomly generated according to a Markov chain of order 3 (that was constructed from human promoter sequences) in which the binding sites were planted. We analysed such datasets by using the same methodology applied for the identification of putative novel regulatory elements in retinal gene promoters, including searching for 12-2 patterns by SPEXS and motifs reconstruction by using COOP (clustering parameters: d= 3 and o= 0.7). The quorum (q) was set to the highest integer equal or less than one third of the total number of sequences in the dataset. When with the selected quorum no results were obtained, qwas set to the highest integer giving results. When analysing datasets composed of from 3 to 6 sequences, qwas set to 2. We made no motif predictions for the two datasets of two sequences each. For each group of promoters, among different clusters obtained, we selected only the one corresponding to the motif represented in the highest number of sequences. For each dataset out of the 26 considered, we checked the overlap between pattern occurrences belonging to the cluster (i.e. the motif) and known signals. The efficiency of COOP was evaluated according to different measures, defined as follows. At nucleotide-level nTP (number of nucleotide positions in both known sites and predicted sites), nFN (number of nucleotide positions in known sites but not in predicted sites), nFP (number of nucleotide positions not in known sites but in predicted sites) and nTN (number of nucleotide positions in neither known sites nor predicted sites). A predicted site overlaps a known site if they overlap by at least one-quarter the length of the known site. Thus, at site-level we calculated: sTP (number of known sites overlapped by predicted sites), sFN (number of known sites not overlapped by predicted sites) and sFP (number of predicted sites not overlapped by known sites). We then calculated the following measures of accuracy. At either the nucleotide (x = n) or site (x = s) level: Sensitivity, xSn = xTP/(xTP + xFN); Positive Predictive Value, xPPV = xTP/(xTP + xFP). At the nucleotide-level: Specificity, nSP = nTN/(nTN + nFP); Performance Coefficient, nPC = nTP/(nTP + nFN+ nFP); Correlation Coefficient, ![](1471-2105-6-121-i1.gif); Average Site Performance, sASP = (sSn + sPPV)/2 \[[@B18]\]. In addition, the statistics (nSn, nPPV, nSp, nPC, nCC, sSn, sPPV, sASP) summarizing COOP performance (with selected settings) over the collection of human datasets, were computed with the \"combined\" method \[[@B18]\] and compared with the same statistics calculated for the 14 different programs tested by Tompa and colleagues on human datasets. Authors\' contributions ======================= SB and GAD conceived the study. SB and AC carried out pattern discovery analyses, developed and tested COOP software and drafted the manuscript. AB and CP participated to the early phases of the work, contributing to gene promoter selection and to the development of the COOP algorithm, respectively. GAD revised the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table 6. Results of the analysis of 26 human positive control datasets with COOP. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Table 7. Comparative evaluation of COOP performance on 26 human positive control datasets. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ Financial support of the Italian Ministry for Technological and Scientific Research and of Padova University to G.A.D and of Italian Association for Cancer Research (AIRC) to S.B is gratefully acknowledged. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow-chart of COOP program. Input, output and main steps are shown. ::: ![](1471-2105-6-121-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Comparison of patterns discovery results in retinal gene promoter sequences and in 1,000 negative control datasets. Plots of number of patterns (12 bp long, with at most two variable positions) vsnumber of sequences in which they were found, in retinal gene promoter sequences (open squares) and in 1,000 negative control datasets (filled diamonds). For negative control datasets, the average value of 1,000 sets of sequences is given, with a two standard deviations interval. Statistically significant differences (0.05 threshold) are marked by stars. (A) Comparison between the 1000M52 dataset (52 promoter sequences of genes overexpressed in the retina) and the RAN1000M52i dataset (1,000 groups of 52 randomly chosen human promoters); (B) Comparison between the 1000M91 dataset (91 retinal gene promoter sequences) and the RAN1000M91i (1,000 groups of 91 randomly chosen human promoters). ::: ![](1471-2105-6-121-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Statistics comparing the accuracy of COOP and of 14 different motif discovery tools on 26 human positive control datasets. Combined measures of correctness over all 26 human datasets, as defined in Methods. The number of datasets (out of 26) for which no motif was predicted by each tool is reported in brackets, following the name of the tool. ::: ![](1471-2105-6-121-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Description of COOP parameters. ::: Step Parameter Description Range Example -------------------------------- --------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ----------- --------------------------- Search for pattern occurrences q Quorum(minimum number out of N input sequences in which a given pattern must be represented) 1 - N ≥ 20 out of 100 sequences Clustering d Physical distance between 5\'-ends of occurrences of patterns of length p 0 - \\|p\\| ≤ 2 nucleotides o Ratio between observed overlapping occurrences of two patterns and their average number of occurrences 0 -- 1 ≥ 0.8 Consensus building i Ratio between the number of nucleotides per alignment position and the total number of lines in the alignment. The maximal number of adjacent positions exceeding the threshold i(m) is further analysed for determining the consensus sequence of the motif 0 -- 1 ≥ 0.5 l Nucleotide length of the lateral region of the motif 0 - m/2 3 bp (out of 10) f~l~ Frequency of a single nucleotide in each position of the lateral region to be considered specified 0 -- 1 ≥ 0.6 f~c~ Frequency of a single nucleotide in each position of the core region to be considered specified 0 -- 1 ≥ 0.8 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Procedure for building a consensus sequence starting from a matrix of nucleotide counts, according to selected parameters. Rows from two to five represent the matrix of nucleotide counts in different positions of an alignment associated to a cluster of pattern occurrences. The sixth row contains, for each alignment position, the ratio between number of sequences in the position and the total number of lines in the alignment. Out of 11 positions of the matrix, positions from one to ten (shaded in grey) fulfil the minimum i(0.5) and are considered for building the consensus. If the lateral region length is set to 3 nucleotides, a 3-4-3 motif is obtained. The f~l~(0.6) threshold is applied to the positions in the lateral regions, whereas the f~c~(0.8) is applied to positions in the core region. Cells containing values fulfilling the condition reported on the left are in bold. In the last row, the derived consensus sequence is shown. ::: 1 2 3 4 5 6 7 8 9 10 11 -------------------- --------- --------- ------- --------- ------- ------- ------- ------- ------- --------- -------- A 0 0 0 4 0 0 0 0 0 0 0 C 0 0 5 0 5 2 0 0 0 0 2 G 0 4 0 0 0 3 5 5 0 4 0 T 3 0 0 1 0 0 0 0 5 0 0 i(0.5) 0.6 0.8 1 1 1 1 1 1 1 0.8 0.4 f~l~(0.6) 1 1 1 1 1 1 f~c~(0.8) 0.8 1 0.6 1 Consensus sequence T G C A C N G G T G - ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Number of sequences in which most represented patterns were found in different retinal datasets. ::: Pattern 500U91 500PM91 500M91 1000M91 ------------- ------------ ------------- ------------ ------------- 10-2 44 42 32 40 12-2 19 19 10 12 14-2 11 11 5 7 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Statistics about patterns found in different groups of retina gene promoter sequences and in the corresponding negative control random datasets. ::: *Quorum* Pattern Obs. Exp. P-value ---------------------------------------------------------------- -------------- ------------- ---------- ---------- ------------- 10-2 719 351.3 0.017 20 12-2 0 \- \- 14-2 0 \- \- 1000M91(91 promoters of retinal genes vsRAN1000M91~i~) 10-2 18683 12846.4 0.016 10 12-2 41 35.4 0.324 14-2 0 \- \- 7 12-2 714 410.2 0.060 1000M52(52 promoters of retinal genes vsRAN1000M52~i~) 5 12-2 1537 429.4 0.001 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### List of 60 consensus sequences corresponding to selected motifs showing most conserved central regions. For each motif, consensus sequence, length and total number of occurrences in the 1000M dataset are reported, along with LocusLink symbols of corresponding genes. In the last column, for each consensus, the list of mammalian transcription factors recognising similar DNA sequences is reported. ::: Consensus sequence Length Total occ. Genes in which promoter the motif was found Transcription factors ------------------------ ------------ ---------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------ AAAAAAAAAAAAAA 14 151 EFEMP1, CCNI, CNGB3, KCNV2, IMPDH2, SLC24A1, DHRS3, G2AN, RTP801, MGC15WIF11, USH3A, CRX, 18, HMGA1, SLC24A2, RDS, TULP1, DC-TM4F2, OPN1SW, RP1, MGAT4B, GAPD, ELOVL4, RRAD, ARR3 NGGCCCCGCCCCCN 14 114 EEF1G, HMGA1, EFEMP1, CYBA, KRT18, OPA1, DPYSL4, RAX, FLJ1415, MGC15WIF11, FLJ1415, ALMS1, EIF3S8, G2AN, ALMS1, DC-TM4F2, MSH6, RCV1, KRT19, DHRS3, PITPNC1, RRAD, HPCAL1, MGAT4B, SLC38A3, IMPDH2, CNGB1, RDH5, EFEMP1, CRABP1, C7orf20, CCNI, GNB1, CRX, GAPD, ARF4L, AIPL1, DKFZP564K0822 AP-1, GCF, Sp1, Sp3, TFIID GCACCCCCAGCCCCN 15 101 RHO, G2AN, EFEMP1, SLCO4A1, CYBA, HPCAL1, KIFC3, RCV1, NK4, KRT18, CRX, ARR3, PPP1R3F, MGAT4B, NRL, RRAD, CCNI, SAG, ALMS1, MGC15WIF11, DKFZP564K0822, VMD2, DPYSL4, GNAT1, GAPD, OPN1SW, RAX, DHRS3, COPEB, SLC38A3, TMEM16B, SLC24A1 Sp1 NGAGGGCAGGGGCNN 15 94 GNB1, KRT19, ELOVL4, VMD2, MSH6, HMGA1, RHO, NK4, SLC38A3, LRRCGUCA1B, CYBA, RCV1, RRAD, GUCY2D, MGC15WIF11, AIPL1, MGAT4B, KIFC3, CRX, CRABP1, G2AN, ALMS1, RTP801, EEF1G, COPEB, OPA1, EFEMP1, KCNV2, PDE6A, AOC2, RLBP1, FLJ1415, RAX, DPYSL4, WIF1, DC-TM4F2 Sp1 CCTCCCTCCCTCCC 14 76 ARF4L, COPEB, RHO, SLC38A3, FLJ1415, WDR17, ELOVL4, DHRS3, KCNV2, OPA1, CCNI, GUCA1B, RDH5, RAX, ALMS1, DKFZP564K0822, NK4, RGS19IP1, RRAD, KIFC3, KRT19, SLCO4A1, HPCAL1, DPYSL4, TNFRSF6, CNGB1, DC-TM4F2 MAZ NCTCCCCCTCCCCC 14 43 CNGB1, GAPD, RPE65, ALMS1, COPEB, MSH6, RRAD, CRABP1, TNFRSF6, CRX, WIF1, FLJ1415, DKFZP564K0822, PDE6A, RDH5, SLC38A3, CYBA, GNB1, MERTK, WDR17 Sp1, AP-2, MAZ GNNTGGGGGAGGGGN 15 41 CYBA, RLBP1, KCNV2, CNGB1, COPEB, KIFC3, RDH5, CCNI, FLJ1415, MGC15WIF11, AIPL1, NK4, HPCAL1, CNGB1, GUCA1A, ALMS1 MAZ, Sp1 CNCCCCCACCCCCACC 16 40 RCV1, SLC38A3, HPCAL1, KIFC3, RLBP1, RPE65, DHRS3, RTP801, CYBA, DPYSL4, RDH5, RRAD, COPEB AP-2alphaB, Sp1, WT1 CTCCCCCTCCCCNNC 15 26 CNGB1, CRX, GAPD, RHO, CNGB1, COPEB, CYBA, AIPL1, RAX AP-2, MAZ, Sp1 CCCCAGCCCCNCA 13 23 CCNI, EFEMP1, SLCO4A1, MGC15WIF11, ARR3, CYBA, HPCAL1, KIFC3, RAX, RLBP1, MGAT4B, AIPL1, RGS19IP1, ALMS1 Sp1 NNGGCCCCTGCCCN 14 23 HMGA1, NK4, LRRCGUCA1B, FLJ1415, GNB1, KRT19, AIPL1, GUCA1A, DHRS3 Sp1 NCCCCCTCCACCN 13 22 ARR3, HMGA1, KRT19, VMD2, DHRS3, ARF4L, RAX, CCNI, SIRT3, GUCA1B, DC-TM4F2 Sp1 NCNGGGCTGGGGN 13 22 CYBA, HPCAL1, RRAD, GAPD, GUCA1A, RHO, G2AN, EFEMP1 Sp1 NNTCCCCCTCCCNN 14 22 TNFRSF6, CNGB1, CRX, EEF1G, GAPD, RPE65, ALMS1, DKFZP564K0822, COPEB, AIPL1 AP-2alphaB, MAZ, Sp1, WT1 -KTS NNCCCAGCCCCCAN 14 20 RDH5, SLC38A3, EFEMP1, ARR3, CYBA, GAPD, HPCAL1, NK4, PPP1R3F Sp1 NTGGGGGAGGGGNA 14 20 COPEB, CYBA, RLBP1, PITPNC1, CNGB1, CRX, GAPD, MERTK, CCNI MAZ, Sp1, Sp3 CCNGCCCTGGCCT 13 18 GUCA1A, GUCY2D, RCV1, VMD2, EFEMP1, LRRCGUCA1B, C7orf20, 4, RRAD, UNC119, MERTK Sp1 GCNGCCCCTGCCN 13 18 CRX, CYBA, GNB1, HMGA1, RHO, SLC38A3, MGAT4B, FLJ1415, KRT18 NCNGGGGGCGGGG 13 18 CYBA, RRAD, FLJ1415, HMGA1, RDH5, RGS19IP1, G2AN, RTP801, DC-TM4F2 AP-1, ER, Sp1 CTNCCCCTCCCC 12 17 RLBP1, AIPL1, PITPNC1, CNGB1, GAPD, RHO, CNGB1, EFEMP1, COPEB, CYBA, GNB1, PDE6A AP-2alphaB, MAZ, Sp1 GGGGTGGGGNTG 12 17 GUCY2D, FLJ1415, AIPL1, RDH5, CRABP1, HPCAL1, KIFC3, DHRS3, RTP801, CYBA, RLBP1 AP-2alphaB, Sp1, Sp3 CCCGCCCCTGNCC 13 16 GNB1, HPCAL1, KRT19, MGAT4B, G2AN, Sp1 NGGGGGTGGGGGN 13 16 HPCAL1, RRAD, DHRS3, FLJ1415, CYBA, GNB1, DPYSL4 Sp1 NNCCCCCGCCCCNN 14 16 GNB1, RGS19IP1, LRRCGUCA1B, ALMS1, DC-TM4F2, KRT18, SAG AP-1, AP-2alphaB, ER, Krox-20, Sp1, WT1, WT1 I, WT1 I -KTS AGNGGGAGGGGCN 13 14 CYBA, EFEMP1, RAX, MGC15WIF11, ARF4L, CRX, SLCO4A1 MAZ, Sp1, Sp3 CCCTGTCCCTGGAN 14 14 ARR3, HPCAL1, FLJ1415, DC-TM4F2, KRT19, LRRCGUCA1B, TMEM16B GR CGGGGCCGCCNCN 13 14 FLJ1415, DC-TM4F2, MGC15WIF11, COPEB, MGAT4B, SLCO4A1, RAX CUP, Sp1 CTCTCTCTCCNTN 13 14 GAPD, GUCA1A, NRL, RRAD, FLJ1415, GNAT2, KCNV2 NANCTCTGCACCC 13 14 LRAT, TNFRSF6, CYBA, KIFC3, DPYSL4, G2AN, RTP801 NCCGCCCCCGCCN 13 14 GNB1, IMPDH2, SLC38A3, COPEB, CYBA, KRT18, SLCO4A1 AP-1, ER, Kxox-20, Sp1, WT1 I -KTS, WT1-del2 NGGCCTCTGGNCN 13 14 CYBA, GAPD, KRT19, RDH5, DPYSL4, HPCAL1, MGAT4B NGGGAGGGGGAAG 13 14 GAPD, AIPL1, FLJ1415, EEF1G, RPE65, ALMS1, WDR17 AP-2alphaB, MAZ, Sp1, WT1 I -KTS NGNCCCCAGCCCC 13 14 GAPD, GUCA1A, RHO, ARR3, CYBA, NK4, PPP1R3F AP-2, Sp1 NNCCCAGCCCAGNN 14 14 GAPD, RHO, ARR3, CRABP1, CYBA, RRAD, MGAT4B Sp1 TGGGGGTGGGGGN 13 14 HPCAL1, RLBP1, DHRS3, CYBA, HMGA1, RRAD, DPYSL4 Sp1 NGGCGGGGGCGGGG 14 13 EFEMP1, KRT18, RRAD, SLCO4A1, IMPDH2, EFEMP1, COPEB AP-1, Krox-20, Sp1, WT1 I -KTS, WT1-del2 GGNAGGGGCGGG 12 11 ELOVL4, REA, G2AN, GNB1, MSH6, GUCY2D, RGS19IP1, LRRC21, SLCO4A1, PITPNC1 MAZ, Sp1 CCCGCCCGCCCC 12 9 GNB1, RGS19IP1, WIF1, PITPNC1, DC-TM4F2, HMGA1, DPYSL4, KRT18, RAX Sp1 GGGCGGGGCNGG 12 9 CYBA, DPYSL4, MGAT4B, MSH6, RCV1, ALMS1, FLJ1415 ER, GCF, Sp1 GGGCTGGGGGTG 12 9 CYBA, HPCAL1, KIFC3, RCV1, RHO, G2AN, DKFZP564K0822 Sp1 GGGGAAGGGNGG 12 9 TULP1, CRX, MSH6, KRT19, CNGB1, SLC38A3, AIPL1, HMGA1, FLJ1415 GGGGCGGGCNNG 12 9 EEF1G, KRT19, DC-TM4F2, GUCY2D, RGS19IP1, PITPNC1, C7orf20, RTP801 ER, Sp1 GGNGCGGGCGGG 12 9 HMGA1, KRT19, DPYSL4, DC-TM4F2, RGS19IP1, WIF1, PITPNC1, FLJ1415 AP-2, ETF, Krox-20, Sp1, WT1 I -KTS GNNGGGGCTGGG 12 9 GAPD, HPCAL1, KIFC3, RCV1, RAX, COPEB, RDH5 WT1 -KTS CAGGGGGCGGGG 12 8 CYBA, EFEMP1, HPCAL1, FLJ1415, GAPD, HMGA1, G2AN, DC-TM4F2 AP-1, ER, Sp1, Yi CNCCCCCACCCC 12 8 CYBA, HMGA1, RCV1, SLC38A3, HPCAL1, RLBP1, DHRS3 AP-2alphaB, CACCC-binding, factor, Sp1, WT1 GAGTGGGGGAGG 12 8 DHRS3, KCNV2, COPEB, CYBA, HMGA1, WIF1, FLJ1415, MGC15WIF11 GCCTGGGGGAGG 12 8 CYBA, SIRT3, KIFC3, CCNI, DKFZP564K0822, DC-TM4F2, MGC15WIF11 AP-2 GGGCAGGGGCNG 12 8 CYBA, GNB1, HPCAL1, HMGA1, RHO, SLC38A3, MGAT4B, G2AN Sp1 GGGCGGGGCTGG 12 8 CYBA, HPCAL1, RAX, MSH6, RCV1, ALMS1, DC-TM4F2 ER, GCF, Sp1 CCCTGTCCCTGG 12 7 CNGB1, GNB1, FLJ1415, KRT19, ELOVL4, TMEM16B, FLJ1415 GR CCTTCCCCCNGC 12 7 GNB1, SLC38A3, AIPL1, SLCO4A1, RDH5, TULP1, NK4 MAZ CNCCTCCTGCNC 12 7 CRABP1, GUCA1A, PDE6A, RGR, DPYSL4, WIF1, HPCAL1 PPUR, Sp1 CNGCCCCCAGNC 12 7 RHO, EFEMP1, DC-TM4F2, CNGB1, CYBA, NK4, MERTK Sp1 GCNCCCCTCCCC 12 7 COPEB, CRX, HPCAL1, RGR, CNGB1, MERTK, RAX MAZ, Sp1 GGGCAGGGGCGG 12 7 ELOVL4, HMGA1, HPCAL1, RHO, SLC38A3, MGAT4B, G2AN Sp1 GGGGCTGGGGNC 12 7 ARR3, CYBA, HPCAL1, NK4, RAX, PPP1R3F, RLBP1 AP-2alphaB, Sp1 GNAGGGGGCAGG 12 7 GAPD, NK4, GUCA1B, SLC38A3, WIF1, G2AN, EFEMP1 Sp1 TGGGGGAGGNNA 12 7 KCNV2, COPEB, HMGA1, KIFC3, RDH5, CCNI, FLJ1415 MAZ, Sp1 TTTTTTTTTNTA 12 7 IMPDH2, G2AN, SLC24A2, RTP801, KCNV2, USH3A-PROMB, CCNI TBP :::	PubMed Central	2024-06-05T03:55:58.302732	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173081/", "journal": "BMC Bioinformatics. 2005 May 18; 6:121", "authors": [ { "first": "Stefania", "last": "Bortoluzzi" }, { "first": "Alessandro", "last": "Coppe" }, { "first": "Andrea", "last": "Bisognin" }, { "first": "Cinzia", "last": "Pizzi" }, { "first": "Gian Antonio", "last": "Danieli" } ] }
PMC1173082	Background ========== Protein modelling and structure prediction as well as binding and interaction prediction have become very valuable instruments for researchers in biology and medicine. In order to build reasonable and useful models, as much information as possible has to be incorporated into the protein modelling process. To refine protein models, chemical as well as spatial information about ligand structures can be considered, specifically to optimise side-chain conformations around binding-sites \[[@B1]\]. Several databases delivering structures and different additional information about ligand molecules from the Protein Data Bank (PDB) \[[@B2]\], \[[@B21]\] have been provided on the Internet. Ligand Depot \[[@B3]\], \[[@B22]\] comprises chemical and structural information for small molecules found in the PDB and also provides a graphical interface for performing chemical substructure searches. Idealized three-dimensional structures and additional information about PDB ligands can be retrieved via the search interface of the E-MSD macromolecular structure relational database \[[@B4]\], \[[@B23]\]. Besides many other features, Relibase \[[@B5]\], \[[@B24]\] allows for two-dimensional similarity and substructure search among the ligands as well as for sequence similarity search among the corresponding proteins. LigBase \[[@B6]\], \[[@B25]\] is a database of ligand binding sites aligned with related protein structures and sequences. Various information about ligands bound to macromolecules deposited in the PDB can be retrieved from many further sources like HIC-Up \[[@B7]\], \[[@B26]\], PDBsum \[[@B8]\], \[[@B27]\] and the IMB Jena Image Library of Biological Macromolecules \[[@B9]\], \[[@B28]\]. The latter can be searched after geometrical properties of the ligand binding sites. Information contained in these databases can help identifying ligands which are likely to bind to a given protein structure. The opposite question, namely to find target proteins for a certain ligand, was addressed in \[[@B10]\], where a collection of protein active sites was extracted from the PDB and scanned with aid of a docking algorithm. Further data collections emphasize the link between binding affinities and structures of the protein-ligand complexes and, inter alia, provide experimentally measured binding data, e.g. PLD \[[@B11]\], \[[@B29]\] LPDB \[[@B12]\], \[[@B30]\], PDBbind \[[@B13]\], \[[@B31]\]. For modelling and simulation purposes, chemical and spatial information about protein ligands is vitally important. Addressing this fact, SuperLigandsis a collection of small molecule structures contained in the PDB, facilitating comparison of the molecules regarding their two-dimensional similarity. As spatial comparison of compounds can deliver valuable information in addition to this, SuperLigandsalso allows for three-dimensional superpositions. Spatial coordinates of the compounds can be retrieved as MDL Mol files, which include information about multiple bonds. Construction and content ======================== Native conformations of small molecules contained in the PDB and additional information were collected from the PDB \[[@B2]\], \[[@B21]\], Ligand Depot \[[@B3]\], \[[@B22]\] and MSD \[[@B4]\], \[[@B23]\] and deposited in the database SuperLigands. The database has been designed as a MySQL relational database and supplemented with a user-friendly web interface. Database queries are performed and HTML pages are generated via PHP scripts. The freely available MDL^®^Chime plug-in is used to display molecules and allows the user some manipulations of the view and to store the displayed molecule in the MDL Mol file format. In order to enable fast two-dimensional searches, 960 bit binary fingerprints (MDL MACCS Keys \[[@B14]\]) were calculated and stored in the database for all ligands. Tanimoto coefficients are calculated via a PHP script. Here, all 960 MDL keys are included and equally weighted. The Tanimoto coefficient for two structures a and b is defined as follows: T(a,b)=Nab/(Na+Nb-Nab), where Na and Nb are the numbers of bits set in the fingerprint of structure a and b, respectively, and Nab is the number of bits which are common to both fingerprints. Three-dimensional superposition of two different PDB ligands is performed in the following way: each conformation of one molecule occurring in the PDB is superposed with each conformation of the second molecule. Those two instances matching best are displayed. The best match is defined by maximizing the scoredefined by ![](1471-2105-6-122-i1.gif) where RMSD is the Root Mean Square Deviation of the superposed atoms. For completion, PDB codes, chain identifiers and positions in the PDB files of the matched conformations as well as the atom numbers of both ligands, the number of superposed atoms, the number of superposed atoms of the same type and the RMSD of the superposition are returned. For detailed information regarding the superposition algorithm see \[[@B15]\]. Utility ======= SuperLigandscan be searched by hetero-ID (i.e. the three-letter PDB code for hetero-compounds), name, molecular formula or PDB identifier. In the results table, hetero-ID and names of the compounds are given. Moreover, the molecular structure is displayed in one cell of the table where it can be rotated by the user and different displaying options can be chosen. More information like molecular formula, atom numbers and occurrence in the PDB can be retrieved additionally. The database SuperLigands contains compounds defined by \'HETATM\' records in PDB files. Some of these molecules may be bound to pseudopeptides but can also be separate ligands. Coming across such a molecule, the user is given a hint and is provided with a list of pseudopeptides in which this molecule is bound. The user can search the database for molecules that have a significant two-dimensional similarity to a given ligand or assess the three-dimensional similarity of two compounds by superposing them with each other. Such similarity queries can be performed starting from the search results tables or directly using separate forms. Using such a form, the Tanimoto coefficient of two given structures can also be retrieved. A typical example for a query to SuperLigandsis a search for tobramycin, known as antiinfective and antibacterial drug, starting in the main form. Searching the database for similar compounds in the next step supplies the drug kanamycin as PDB ligand with the highest Tanimoto similarity (98.6%). Now, a three-dimensional superposition of all instances of tobramycin (32 atoms, two instances) and kanamycin (33 atoms, four instances) occurring in the PDB can be performed. The best fitting structures are superposed and then displayed. In this case, best fitting are the instances of tobramycin in PDB entry 1m4d and kanamycin in 1m4i with an RMSD of 0.14Å (32 atoms superposed). Navigating through the website, the topological and spatial similarities of PDB ligands can be obtained easily. For example, tobramycin and ribostamycin (31 atoms, four instances) have a Tanimoto coefficient of 96.5%, the RMSD of their best superposition is only 0.95Å (25 atoms superposed). In turn, geneticin (34 atoms, five instances) delivers a Tanimoto coefficient of only 81.1% and a much better RMSD (0.16Å, 30 atoms superposed). As an additional feature of SuperLigands, similarity of PDB ligands to known drugs can be assessed in a comfortable manner. Starting with a ligand, a two-dimensional similarity search as described above can be initiated, not only among the PDB ligands but also in a database containing the structures of known drugs (SuperDrug Database\[[@B16]\], \[[@B32]\]). The drug structures found can be superposed spatially (for an example, see Figure [1](#F1){ref-type="fig"}). Discussion ========== Statistics: comparison of PDB ligands with drugs ------------------------------------------------ Recently, a database containing 2396 drug molecules and having the same design as SuperLigandshas been created (SuperDrug Database\[[@B16]\], \[[@B32]\]). To answer the question, how many drugs or drug-like molecules are bound to PDB structures, Tanimoto coefficients have been calculated for all pairwise combinations of molecules from SuperLigandsand the SuperDrug Database. A set of 5,040 PDB ligands has been incorporated into these calculations. Considering two molecules having a Tanimoto coefficient of 100% (or greater than 95% ; 90%) identical or very similar, this analysis reveals that 413 (771 ; 1,457) of 5,040 PDB ligands are drugs or drug-like compounds. Furthermore, some chemical properties of PDB ligands and drugs have been compared (see Figure [2](#F2){ref-type="fig"}). The distributions of numbers of hydrogen bond donors for PDB ligands and drugs differ most significantly. A bigger percentage of the drugs (26%) have no hydrogen bond donor, the largest fraction of the PDB ligands (19%) have two of them. About half of the drugs have no or only one hydrogen bond donor, which applies for only a quarter of the PDB ligands. About one third of the drugs have three or four hydrogen bond acceptors, the fractions of drugs with nine or more hydrogen bond acceptors drop below 3%. For the PDB ligands, the distribution is more flat: only 22% of them have three or four hydrogen bond acceptors, still over 3% of them have 11 hydrogen bond acceptors. Most drugs have a logP value around 3, and the logP values of the PDB ligands accumulate around the negative value -1. Approximately the same fraction of PDB ligands and drugs are \"drug-like\" according to the Lipinski \"Rule of five\" \[[@B17]\]: 92 and 91%, respectively, have a logP value less than 5, although altogether the logP values of the drugs are closer to this critical value. A majority of the PDB ligands have very low molecular weights in comparison to the drugs, which supposedly is be caused by the fact that in proteins often very small solvent molecules are bound. Nevertheless, slightly more (5%) drugs than PDB ligands fulfil the Lipinski \"Rule of five\" regarding the molecular weight. The same applies for the numbers of hydrogen bond donors (and acceptors): 7% (5%) more drugs fulfil the Lipinski \"Rule of five\". Compounds violating more than one of the Lipinski Rules are assumed to have problems with bioavailability and are therefore presumably not suitable as drugs. Table [1](#T1){ref-type="table"} shows the percentages of PDB ligands and drugs violating the Lipinski Rules. From this table can be seen that a total of approximately 19% of the PDB ligands and 10% of the drugs, respectively, violate more than one of the Lipinski Rules. This analysis reveals that there are only marginal differences between PDB ligands and drugs regarding single chemical properties. But, not surprisingly, from a general point of view, PDB ligands are significantly less drug-like than drugs. Discussion ---------- SuperLigandsis a collection of PDB ligands freely accessible via a user-friendly web site. Molecular coordinates can be retrieved as MDL Mol files, supplementing the connectivity records contained in PDB files with bond types, which are necessary for modelling and simulation purposes. The database can be searched for compounds similar to a given ligand by comparison of Tanimoto coefficients. As stated in \[[@B15]\] and shown in the example in the section Utility, spatial comparison of small molecules can reveal more similarities, and thus similar kinds of interaction, than a pure two-dimensional topology comparison. With aid of SuperLigands, such three-dimensional comparisons can be performed easily. Moreover, the topological similarity of PDB ligand structures to known drugs can be assessed by calculation of Tanimoto coefficients. Conclusion ========== The database presented here supplements the set of existing resources of information about small molecules bound to PDB structures. As novel features, three-dimensional comparison of molecules as well as topology comparison of PDB ligands with known drugs are made possible. Thus, SuperLigandsrepresents a valuable means of analysis and prediction in the field of biological and medical research. Availability and requirements ============================= The database is publicly accessible at <http://bioinf.charite.de/superligands/>. For visualisation, the free browser plug-in MDL^®^Chime is required. Chime runs on Windows systems with Microsoft Internet Explorer (6.0 or 5.5 SP2) or Netscape 4.75, 4.79 or on Mac OS 9.0 or 8.6 with Netscape 4.75 (please see <http://www.mdlchime.com/products/framework/chime/sys_req.jsp> for detailed information). Authors\' contributions ======================= EM designed the database and the web site and finished its functionality, was responsible for data acquisition and processing and drafted the manuscript. MD delivered the basic part of the website functionality and contributed to database conception and data processing. AG provided the tool for three-dimensional superposition and helped to draft the manuscript. RP conceived of the project, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by the BMBF (German Federal Ministry of Education and Research). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *Usage of the web interface of SuperLigands**. From the main menu, the form Compound searchcan be reached. Here, a PDB ligand can be searched after hetero-ID, name, molecular formula or PDB identifier. In the first column of the results table, two buttons can be found to retrieve more information. The FULL infobutton delivers detailed information about the selected PDB ligand like molecular formula, atom numbers and occurrence in the PDB. After clicking the DRUGSbutton, a two-dimensional similarity search among the drugs in the SuperDrugdatabase \[16\] is performed. The best hits are displayed in a new window. From here, they can be spatially superposed. In the figure, this procedure was carried out for celecoxib, a COX-2 inhibitor which was recently categorised as problematic (see the \"Pfizer Statement on New Information Regarding Cardiovascular Safety of Celebrex\" \[18\]). The two-dimensional similarity search in the SuperDrugdatabase delivers only hits below 72% Tanimoto similarity. A following spatial superposition of the best hits reveals a further COX-2 inhibitor, namely valdecoxib, (RMSD 0.26Å and 21 of 22 atoms superposed) as very similar to celecoxib. The Tanimoto similarity of celecoxib and valdecoxib is only 65% and there are two drugs more similar to celecoxib: Sulfaphenazole (71% Tanimoto similarity, spatial superposition with 0.65Å RMSD and 15 of 22 atoms superposed) and Sulfamazone (67% Tanimoto similarity, spatial superposition with 0.32Å RMSD and 17 of 26 atoms superposed). Nevertheless, the three-dimensional comparison here proves to be very important to reveal molecular similarities in addition to topological comparison (as also shown in the example in the section Utility), which is confirmed by the fact that valdecoxib was categorised as toxic \[19\] and sales of this drug were suspended recently \[20\]. ::: ![](1471-2105-6-122-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Statistics: comparison of PDB ligands with drugs*. Chemical properties of 5040 PDB ligands and 2396 drug molecules are compared: histograms for numbers of hydrogen bond donors and acceptors, logP value and molecular weight are shown. Molecular weight within \[100,200) means that the molecular weight is greater than or equal to 100 and less than 200. Those areas for which the Lipinski \"Rule of 5\" is fulfilled, are highlighted in grey. ::: ![](1471-2105-6-122-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Percentage of PDB ligands and drugs violating certain numbers of Lipinski Rules ::: Number of violated Lipinski Rules PDB ligands Drugs ----------------------------------- ------------- ------- 0 64.4 75.7 1 16.9 14.0 2 10.3 5.7 3 8.3 4.7 4 0.1 0.0 :::	PubMed Central	2024-06-05T03:55:58.309870	2005-5-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173082/", "journal": "BMC Bioinformatics. 2005 May 19; 6:122", "authors": [ { "first": "Elke", "last": "Michalsky" }, { "first": "Mathias", "last": "Dunkel" }, { "first": "Andrean", "last": "Goede" }, { "first": "Robert", "last": "Preissner" } ] }
PMC1173083	Background ========== The introduction of a microarray based comparative genomic hybridization method (arrayCGH) in 1997 paved the way for higher resolution detection of DNA copy number aberrations \[[@B1]\]. ArrayCGH is founded on the same principles as metaphase CGH, but uses mapped reporters instead of whole chromosomes. One of the major challenges in arrayCGH studies remains the accessibility, management, and interpretation of the vast amount of data generated in single experiments, and parallel comparison of multiple experiments. Typically, these arrays contain 3,000 to 30,000 reporters, each of which has multiple biological annotations (chromosomal position, sequence information, gene name, biological and molecular function,\...) as well as physical (grid layout) and quality control (sequence verification, FISH mapping information,\...) annotations. In addition, the description of the DNA samples under investigation and the applied lab protocols should be easily accessible. For classical CGH, several commercial software packages are available to analyze and interpret the data of a CGH experiment. Also for arrayCGH there are a number of separate software systems that individually address some of the needs, such as databases for data storage (BASE \[[@B2]\]), applications for clustering and visualization of microarray data (seeGH \[[@B3]\], M-CGH \[[@B4]\], CGHAnalyzer \[[@B5]\], aCGH-smooth \[[@B6]\] and CGH-Miner \[[@B7]\]), public genome databases that contain reporter information, commercially available Laboratory Information Management Systems (LIMS), and various storage methods for recording biomaterial annotations. However, none of these software packages or databases combine all these features (see Supplemental Table). In this paper, we present the development of a web based open source arrayCGH analysis platform, arrayCGHbase, that combines all these features and on top provides additional unique aspects making the analysis and sharing of arrayCGH data easily implementable for both research and routine purposes. Implementation ============== MIAME compliant database ------------------------ arrayCGHbaseruns in Windows, Linux, Macintosh, and Unix environments. Particular attention was paid to the use of open source software for the development of arrayCGHbase. The software was developed in the PHP scripting language, with all data being stored in a relational, MIAME \[[@B8]\] (Minimal Information About a Microarray Experiment) supportive, MySQL database and communicated to the user through an Apache Web server (Figure [1](#F1){ref-type="fig"}). After installation on a private server, experiments can be shared by different users over the internet or a local intranet. ArrayCGHbaseintegrates DNA sample information, lab protocols, extracted data, and contains a plug-in architecture for data transformation, analysis, and graphical display, allowing users to develop their own modules. Reporters can be directly linked to the Ensembl \[[@B9]\] or UCSC \[[@B10]\] genome browsers, providing additional up-to-date information on each reporter. Reporters can also be manually imported into the MySQL database with the ability to update all linked experiments. The structure of arrayCGHbasewas designed to follow the laboratory workflow and is compatible with all types of arrayCGH experiments and data formats (dual colour genomic clone, cDNA \[[@B11]\], or oligonucleotide \[[@B12]\] arrays spotted on any substrate, physical layout, type of array, as well as single channel hybridizations such as the Affymetrix SNP chips \[[@B13]\]). With a personal account and administrated access levels, a user can enter new DNA samples, annotate these, and append all relevant sample information such as quantity, quality, and applied lab protocols at each step. Each user can group experiments together into projects and, in a uniform and streamlined fashion, apply filters and transformations and run analyses. Data is exportable in several formats for offline analysis using other (dedicated) software tools, for publication or for sharing data with the research community. For advanced users, an SQL query window allows interrogation of the underlying MySQL database. Data processing and visualization routines ------------------------------------------ A first and important step in data analysis of arrayCGH experiments is the processing of large, possibly noisy data sets to identify the specific reporters that are differentially hybridized and hence show an aberrant copy number. Data processing is performed in a streamlined four-step manner: (1) the local noise or background associated with the experiments is removed, (2) the quality of the experiment is assessed and poor quality features are removed, (3) ratios are calculated, transformed to log~2~scaled ratios, and normalized, and finally (4) reporters that show altered ratios are identified and hence, reporters with aberrant copy number are identified. In the past, this normally required the sequential processing of data by different, often incompatible programs. Using established and widely used microarray (CGH) data processing procedures, arrayCGHbasewill automatically correct the signal intensities, filter out unwanted poor quality features (based on signal to noise ratio, image processing software related flags, or other user defined filters), normalize the fluorescence intensity ratios, score levels of differential hybridization, combine the results of replicate experiments and assess the quality of individual and replicate experiments. All these steps are user adjustable. Input data and local background correction ------------------------------------------ The experimental input data for arrayCGHbaseconsists of export files generated by image analysis software. Currently, the program recognizes files from GenePix Pro versions 2.0--4.0, Scanalyze version 2.0, UCSF SPOT version 2.0, Imagene versions 4.0 -- 5.5 and the Affymetrix Chromosome Copy Number Tool. The program can easily be updated for the recognition of other data input formats upon request. Moreover, arrayCGHbasehas an interactive import wizard, which makes it possible to import data at your own desire. The processing steps may be changed by altering the parameters at the input stage. By default, the results for each feature are defined as the median foreground minus background intensities for each dye (as determined by the image processing software). The ratio of each feature is determined as the relative background corrected signal between the two dyes or in the case of single color experiments as the corrected signal intensity. Poor quality flagging --------------------- Nearly every experiment contains features of poor quality, comprising features that have unusual morphology (e.g. doughnut patterns), exhibit uneven hybridization, or have saturated signal intensity. After background corrections, arrayCGHbasecan automatically flag features of inferior quality using different criterions (e.g., the standard deviation between replicates), by a manually set signal or signal-to-noise threshold, or using image processing generated flag annotations. Normalization ------------- Following calculation of the corrected signal intensities and filtering for good quality features, the relative contributions of the fluorescence intensities are compared. To go from a multiplicative space to an additive space, ratios are log~2~transformed. Ideally, the signals of the two dyes should be equal for nucleic acid reporters that have equal amounts in the test and reference samples (i.e., the log~2~transformed ratios of the two corrected signals should approach zero for reporters hybridizing to an equal degree in both fluorescence channels). However, in practice the ratio of the corrected signal intensities deviates from the expected ratio due to the different molecular and physical characteristics of the dyes, the different amounts of DNA used for labeling with the different dyes, the spatial heterogeneity in the hybridization conditions across the slide, and many other factors. Normalization compensates these effects by applying a data transformation such that ratios of reporters with unchanged copy-number are close to zero. In the normalization step, an appropriate term is added or subtracted from the log~2~transformed ratio for each feature. The program allows normalization in several ways, either by global normalization or subgrid (or pin) normalization, or by a combination of different normalization procedures. A major issue in microarray normalization is the definition of the set of constant probes to which the data are normalized. The most widely accepted method employs the \'constant majority\' method, which assumes that the majority of reporters do not change in ratio. This method, which is implemented in arrayCGHbase, is generally applicable to most experiments as it is valid even in cases where up to 50% of reporters have altered ratios, it does not require prior knowledge of which features remain constant, and allows for intensity and spatial variation. Hence, this method calculates a scaling term from the median of all ratios, excluding all outliers. In this way the distribution of all ratios is transformed so that it centers around zero. Quality control --------------- ### Percentage of good quality spots This first quality assessment is a basic calculation of the number of reporters (or features) that are not flagged based on quality measures (user defined parameters and thresholds, see above). ### Intra- and inter-array hybridization quality Three other major quality parameters can be determined with arrayCGHbasefor each experiment. The first assesses the variation between reporters present in replicates on the array (typically duplicates or triplicates). An increased variation typically reflects lower quality hybridizations resulting in less reliable ratios. A second quality parameter is the standard variation between the different reporters on the array that show a normal (unaltered) copy number. This quality measure is only applicable in experiments with few reporters with aberrant copy number. The third quality measure is the average ratio for reporters with aberrant copy number. This ratio should significantly differ from zero to allow identification of differentially hybridized reporters. This last quality measure is only applicable in experiments where DNA copy number aberrations are known or validated. These parameters provide an objective quality measure and can also be helpful to compare different experiments. In addition to these parameters, different graphical displays, such as ratio-intensity plots (usually referred to as MA plots), dual channel intensity scatter plots, and ratio histograms give an idea of the quality of an individual experiment or series of experiments (Figure [2](#F2){ref-type="fig"}). In all these visualizations, thresholds for gains and losses are displayed and can be adjusted. The slide viewer generates a virtual spatial view of all features on the array using the ratio, or signal and background intensities; this viewer allows the identification of problematic regions or artifacts on the slide surface. Clicking on an individual feature shows specific data associated with this feature (e.g., reporter name, signal intensities, and data quality flags). Scoring chromosomal regions with aberrant copy number ----------------------------------------------------- The final step in arrayCGH data processing is the identification of reporters that exhibit differential hybridization, corresponding to chromosomal regions that have altered copy number. The major issue is to identify those reporters whose relative ratios stand out from the experimental noise with sufficient statistical significance. arrayCGHbasecurrently incorporates two scoring methods. The most widely used approach is to define a ratio threshold and identify the probes that exhibit ratios greater or smaller than this threshold. Another, statistically more sound approach, is to use a floating threshold based on the standard deviation of all reporters in a given experiment. Reporters that exhibit ratios greater than this threshold will be defined as differential \[[@B14]\]. Both methods are implemented in arrayCGHbaseand can be applied on each individual feature, or on the mean value of replicates. Besides the aberrant feature scoring methods, two other algorithms are available: a universal data smoothing algorithm, as well as a breakpoint-identification algorithm, which both consist of a moving window along the chromosomes and hence make use of the spatial \"along the chromosome\" distribution of the reporters. With these algorithms, chromosomal breakpoints can be easily identified in more noisy datasets. By writing custom plug-ins (in PHP or R), sophisticated algorithms that use segmentation methods (e.g. Cluster Along Chromosomes, CLAC \[[@B7]\]) or others, can be implemented by any user in a straightforward way. Chromosome visualization ------------------------ A wide variety of result viewers are available. The results can be mapped upon standard ISCN (International Standard on Cytogenetic Nomenclature) ideograms in an electronic karyotype, or visualized per chromosome or zoomed in on a region of interest (Figure [3](#F3){ref-type="fig"}). Moreover, various CGH profile views provide the user with a tool to compare different experiments and to identify regions with relevant copy number alterations. Views are returned to the user either as PNG (Portable Network Graphic) or as SVG (Scalable Vector Graphic) files, with the ability to scale images according to screen width. Data export ----------- Processed data can be exported as MIAME compliant text files and figures; these include the original feature signal and background intensities, the normalized ratio value, a list of reporters that are differentially hybridized, and the data quality parameters. Additionally, a file can be generated for submission of arrayCGH results directly into Progenetix \[[@B15]\], a comprehensive collection of published cytogenetic abnormalities in human neoplasms. Lastly, BED files can be created to map results and visualize the experiment from within the Ensembl or UCSC genome browser. ArrayCGHbase at work ==================== In several publications from our research group, arrayCGHbasehas been successfully used to analyse arrayCGH data to identify and delineate copy number aberrations \[[@B16]-[@B19]\]). At the demo site, users can explore the data published in Hellemans et al. \[[@B16]\], a small \~5 Mb deletion in chromosome 12q identified using SNP chips), the results of a case report of the identification of an unbalanced X-autosome translocation by arrayCGH in a boy with a syndromic form of chondrodysplasia punctata brachytelephalangic type \[[@B17]\], a distal 9p trisomy and distal Xp nullisomy caused by an unbalanced X;9 translocation: 46, Y, der(X)t(X;9)(p22.32;p23) detected with a 1 Mb BAC array), and the copy number profile of a cancer cell line NGP.1A.TR \[[@B18]\]). It is possible to look at the raw data of these hybridizations and more importantly, test the performance of the program using different settings. Conclusion ========== We present arrayCGHbase, a versatile web based, platform independent data storage and analysis tool for processing microarray CGH data. Routines were implemented for feature flagging, data normalization, data quality assessment and the identification of chromosomal regions with aberrant copy number. A zoomable graphical interface allows immediate identification of altered genomic regions and the underlying gene content by several database links. A multitude of export functions allow the user to further process the results. The easy plug-in architecture makes it possible for each user to add custom algorithms for data analysis and visualization and share these with the user community. This webtool and database will enable investigators to interpret single experiments and compare large data sets efficiently throughout different array platforms and provides all of the essential features and links for further investigation of the genomic regions of interest. Future developments =================== arrayCGHbasewill continually be updated to incorporate new processing methods that will be developed both within and outside our laboratory. Immediate plans include the addition of export and import functions to R \[[@B20]\] or Bioconductor \[[@B21]\] to be able to apply several available mathematical algorithms such as two-dimensional LOWESS normalization \[[@B22]\]. Immediate export functions to the DECIPHER web site \[[@B23]\] to link phenotypical data to actual experiments will also be included. The arrayCGHbasesource code is freely available under a Creative Commons License, to encourage others to develop new analysis methods and utilities that will further improve its capabilities. Availability and requirements ============================= An arrayCGHbasedemo site is available at <http://medgen.ugent.be/arrayCGHbase/>. At this site, all quality control features and other features can be tested for several experiments with BAC arrays as well as SNP chips (see \'arrayCGHbaseat work\'). At the same site, the complete package can be freely downloaded for local installation on a private hosted web server. For local use, additional software is required such as the MySQL database \[[@B24]\], a web server (e.g. Apache \[[@B25]\]), and PHP hypertext preprocessor \[[@B26]\]. These software packages are freely available and are key parts of LAMP (Linux, Apache, MySQL, PHP), an open source web platform. Enquiries for arrayCGHbaseshould be made to <arrayCGHbase@medgen.ugent.be>. Glossary ======== Reporter: any DNA fragment (BAC, PAC, cosmid, fosmid, cDNA clone, oligonucleotide, genomic PCR product) used for hybridization Feature: physical reporter spotted, printed, or otherwise linked to a substrate at a specific location PHP: Hypertext PreProcessor (server-side scripting language) MIAME: Minimal Information About a Microarray Experiment MySQL: My Structured Query Language ISCN: International System for human Cytogenetic Nomenclature BED: Browser Extendable Data Authors\' contributions ======================= BM was the principle programmer of arrayCGHbase. FP, KDP, PR and SVV contributed ideas for different features and display requirements. JV oversaw the project; all other authors have reviewed the manuscript and FS and JV were the final editors of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Comparison between different already available arrayCGH software programs and arrayCGHbase for the analysis and visualization of arrayCGH data. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ Jo Vandesompele and Katleen De Preter are supported by a grant from the Flemish Institute for the Promotion of Innovation by Science and Technology in Flanders (IWT). Filip Pattyn is a Research Assistant of the Research Foundation -- Flanders (FWO -- Vlaanderen). This study is supported by GOA-grant 12051203, FWO-grant G.0185.04, G.0200.03 and G.0106.05 and VEO project 011V1302, research grant of Kinderkankerfonds vzw (a non-profit childhood cancer foundation under Belgian law). This text presents research results of the Belgian program of Interuniversity Poles of attraction initiated by the Belgian State, Prime Minister\'s Office, Science Policy Programming (IUAP). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **arrayCGHbasescheme. The data is stored in a MIAME supportive MySQL database (red) and reporter info is updated using the NCBI, Ensembl and UCSC genome database. Data and results are presented to the user through a web browser via the PHP scripting language. Data-normalization and other analysis or result visualization methods can be integrated using the plug-in architecture. Further data processing using the R statistical scripting language will be implemented in the near future. Results can be exported to a Progenetix or MIAME compatible format, or visualized on the genome using the Ensemble or UCSC genome browser. ::: ![](1471-2105-6-124-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Quality control graphs. Graphical displays to assess the quality of an experiment, such as a dual channel intensity scatter plot, ratio histogram, ratio-intensity plot and a virtual slide view. ::: ![](1471-2105-6-124-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Selected result viewers*. Graphical displays of arrayCGH*results of neuroblastoma cell line NGP.1A.TR1: line view (all reporters ordered by chromosome and chromosomal position on one line), karyo view (al reporters mapped on their chromosomal position on a standard ISCN ideogram), chromosome view (zoom on one chromosome or chromosomal region) with breakpoint identification algorithm, and genome browser view (neuroblastoma cell line IMR32), with all reporters and their copy number status displayed in the UCSC genome browser. ::: ![](1471-2105-6-124-3) :::	PubMed Central	2024-06-05T03:55:58.310909	2005-5-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173083/", "journal": "BMC Bioinformatics. 2005 May 23; 6:124", "authors": [ { "first": "Björn", "last": "Menten" }, { "first": "Filip", "last": "Pattyn" }, { "first": "Katleen", "last": "De Preter" }, { "first": "Piet", "last": "Robbrecht" }, { "first": "Evi", "last": "Michels" }, { "first": "Karen", "last": "Buysse" }, { "first": "Geert", "last": "Mortier" }, { "first": "Anne", "last": "De Paepe" }, { "first": "Steven", "last": "van Vooren" }, { "first": "Joris", "last": "Vermeesch" }, { "first": "Yves", "last": "Moreau" }, { "first": "Bart", "last": "De Moor" }, { "first": "Stefan", "last": "Vermeulen" }, { "first": "Frank", "last": "Speleman" }, { "first": "Jo", "last": "Vandesompele" } ] }
PMC1173084	Background ========== Eukaryotic gene transcription is largely known to be orchestrated by protein factors like activators, co-activators and co-repressors \[[@B1]\]. However, nucleosomal organisation, non-passive structural scaffolds and global structure of chromatin are increasingly being recognised as major players in the regulation of gene expression. The ability of sequences to position nucleosomes and to be anchored to the nuclear matrix to provide a spatial context for regulation of expression are measurable parameters that may influence the interactions with transcription machinery \[[@B2],[@B3]\]. This level of regulation may be distinctly different for genes whose expression is constitutive in comparison to genes that exhibit tissue specific expression. The latter would demand an open chromatin configuration in certain tissues and repressive organisation in others. In this study, we examined whether the potential to utilise global regulatory mechanisms to control gene expression through chromatin organisation varies between housekeeping and tissue specific genes (Hkg and Tsg respectively) by virtue of their organisation. An in-silicocomparison of chromatin related organisational differences in the 5\' and 3\' regulatory regions of housekeeping and tissue specific genes was carried out to shed light in this direction. Results and discussion ====================== Chromatin landscape of a region plays a major role in determining and modulating the expression status of its neighbouring genes \[[@B4]\]. The role played by chromatin in the 5\' regulatory regions of genes in transcriptional regulation has been extensively studied \[[@B5],[@B6]\]. In the present study, we have taken 2 distinct sets of genes differing predominantly in their spatial expression aspect, namely, housekeeping and tissue specific, to understand the various attributes of the regulatory role played by chromatin organisation in the 5\' region. Analysis of scaffold/matrix associated sequences ------------------------------------------------ Scaffold/matrix attachment regions (S/MARs) are defined as sequences, which can attach themselves to the nuclear matrix and hence help in the formation of independent chromatin loops \[[@B7]\]. Transcriptional regulation of gene expression is known to involve formation of dynamic chromatin loops mediated by S/MAR attachment to the nuclear matrix \[[@B3]\]. The attachment of a DNA sequence to the matrix will place the neighbouring genes in proximity of the transcription factors. The abundance of S/MARs in the 5\' cis-regulatory regions of genes further demonstrates their role in transcriptional regulation \[[@B8]\]. We have analysed the predicted S/MAR sites in the 5\' and 3\' flanking regions of human Hkg and Tsg (Table [1](#T1){ref-type="table"}). We used MAR Finder (new version) and ChrClass programs for predicting S/MAR binding sites in the sequences (Table [1](#T1){ref-type="table"}). Glazko et alhave classified 5\' flanking regions up to 1500 bp of human tissue specific genes as an out-group, assuming that these regions have no significant association with S/MAR binding \[[@B7]\]. On the contrary, our study reveals that S/MAR binding sequences are enriched in 5\' regulatory regions of Tsg in comparison to the Hkg. The common predictions of both the programs were taken for the analysis. This data indicates a significant enrichment of S/MAR binding sequences in the 5\' flanking regions of Tsg and depletion of S/MARs in the 3\' Hkg regions as compared to Tsg. Chi-square test was applied for both 5\' and 3\' region S/MAR predictions of Hkg and Tsg, to ascertain whether the distributions are significantly different. The chi-square value of 11.37 (df = 1) and P-value ≤ 0.001 obtained for the distribution of S/MARs in 5\' regions of Hkg and Tsg indicate a significant difference in the distribution of S/MAR elements between the two sets. Similarly, for the distribution of S/MARs in 3\' regions of Hkg and Tsg the chi-square value of 5.033 (df = 1) and P-value of ≤ 0.025 show that the Hkg 3\' regions are significantly depleted of S/MARs as compared to Tsg. The observation that the 5\' regulatory regions of Hkg are less enriched in S/MARs in comparison with Tsg might be related to the distribution of housekeeping genes in the genome. Housekeeping genes cluster in chromosomes and therefore, they often would be present in distinct chromatin domains along with housekeeping genes that have a co-ordinated expression \[[@B9],[@B10]\]. The data showing preferential absence of S/MARs in the 3\' regions in Hkg further lend support to this hypothesis. On the other hand, tissue specific genes are known to be dispersed in gene dense as well as heterochromatic regions \[[@B9],[@B11]\]. It may be necessary for them to shield themselves against the effects of positive and negative cis-acting elements of adjacent regions in order to maintain tissue specific expression profile. In this context, the boundary elements or the insulator model has been proposed earlier \[[@B11]\]. S/MARs function as boundary elements and their co-localisation with insulators such as the Drosophila gypsy element is also reported \[[@B12],[@B13]\]. They also function as boundary elements in in vitrosystems by shielding away the position effect \[[@B14]\]. Some earlier reports have suggested a role for S/MARs in maintaining tissue specific gene expression \[[@B15]\]. More recently, the 5\'-HS4 chicken-globin insulator is known to have a CTCF protein binding dependent matrix association \[[@B16]\]. Hence, the over representation of S/MARs seen in Tsg set might possibly be associated with a boundary element function. Our results on the prediction performance of the programs have been quite different from the previous reports \[[@B7]\]. We find that MAR Finder (an under predictor) predicts more number of S/MAR regions in our dataset in comparison to ChrClass program (an over predictor) \[[@B7]\]. This may be attributed to the use of the advanced version of MAR Finder in our study wherein, new parameters/features have been added in the form of the \"New MAR Rules\" option. Analysis of nucleosomal organisation ------------------------------------ The primary template for local and global changes in the chromatin structure of a chromosome is the nucleosomal unit \[[@B4]\]. Chromatin structure and nucleosomal organisation over the promoter regions play a major role in regulation of expression of downstream gene(s) \[[@B6],[@B17]\]. The nucleosome distribution would depend upon the occurrence of nucleosome destabilising elements as well as nucleosome forming sequences. We have analysed both these parameters in our study. Nucleosome destabilising elements --------------------------------- Nucleosome destabilising/excluding elements such as poly (dA.dT) and (CCGNN)~n~in promoter regions have been implicated in maintaining constitutive gene expression \[[@B18]-[@B21]\]. At the functional level, it is known that poly (dA.dT) elements increase the accessibility of promoters of HIS3, URA3 and Ilv1 in yeast to the cognate transcription factor \[[@B18]\]. With the increasing length of poly (dA.dT) repeat, the availability of the sequences to transcription factors improves and similarly, with increasing lengths, the propensity to exclude nucleosomes increases for (CCGNN)~n~sequence motif as demonstrated in yeast and mammalian systems \[[@B19]-[@B21]\]. It has been demonstrated that (CCGNN)~n~sequences promote meiotic recombination and activated HIS4 expression by generating open chromatin \[[@B22]\]. We hypothesised that the differential distribution of nucleosome exclusion elements might be one of the mechanisms involved in maintaining distinct nucleosomal organisation of the housekeeping and tissue specific genes. The frequency of pure poly (dA.dT) stretches \>10 bp and (CCGNN)~2--5~in the 2000 bp 5\' cis-regulatory regions of human Hkg and Tsg(s) were analysed. A significant enrichment of poly (dA.dT) elements in the upstream regions of Hkg is seen in comparison to Tsg (Table [2](#T2){ref-type="table"}). The t-test for the difference in distribution of poly (dA.dT) stretches (\>10 bp) between Hkg and Tsg show significant P-values in the different lengths of the stretches examined. In Hkg, 670 repeats of (CCGNN)~2--5~were detected as against 430 in Tsg. (CCGNN)~2~was the most prevalent repeat unit and uninterrupted repeat units (\>5 mers) were not found in the sequence sets. Although shorter repeat units (2--5 mers) have not been studied for nucleosome exclusion, they might play a role in destabilising the histone octamer \[[@B20]\]. Further, many of them form a part of longer interrupted stretches. The t-test for difference in distribution of (CCGNN)~2--5~between Hkg and Tsg shows a significant P-value of 1.71E-06. Nucleosome formation potential scores and expression level of genes ------------------------------------------------------------------- Using Recon, Levitsky et al(2001) have examined the nucleosome formation potential of 3 classes of human genes namely, Hkg, Tsg and widely expressed genes that differ in their spatial expression status \[[@B2]\]. Their report, based on a small sample size of around 200 genes shows the difference in the nucleosome formation potential between these 3 classes of genes in the upstream 50 bp from the transcription start site. In this study, we examined the nucleosome formation potential values in upstream 2000 bp of 5\' regions of Hkg and Tsg and their correlation with gene expression levels with the complete set of 1083 genes. The Tsg and Hkg sequences show a considerable difference in their nucleosome formation potential scores over an extended upstream region of 2000 bp (Figures [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}). The Tsg region is enriched in nucleosome formation potential scores (peak at 1) in all upstream positions analysed (till 2000 bp). For Hkg, the distribution seems to be shifted towards the negative scores at 400 bp region and this shift diminishes gradually as we move further upstream to finally peak at 1 in 2000 bp upstream region (Figure [1](#F1){ref-type="fig"}). t-test was applied to ascertain the difference in distribution of Recon scores between Hkg and Tsg (Table [3](#T3){ref-type="table"}). The resultant P-values in various intervals of relevance (0.8 to 1, 1 to 1.2, -0.8 to -1 and -1 to -1.2) reflect that the scores in the upstream 400 bp from the gene start site show the maximum difference in all the intervals and at 2000 bp, the difference gradually fades away in intervals 0.8 to 1 and 1 to 1.2 (Table [3](#T3){ref-type="table"}). A correlation analysis between nucleosome formation potential and expression levels was carried out considering the Recon scores at upstream 400bp region, where the P-values reflect the largest difference and the log~10~values of expression levels were taken as inputs \[\"see [Additional file 1](#S1){ref-type="supplementary-material"}\"\]. Initially, we analysed the gross dependence of total expression levels on nucleosome potential in the upstream regions of the two sets of genes (Table [4](#T4){ref-type="table"}). In all the four intervals, no correlation is seen, indicating that chromatin plays an insignificant role in global modulation of levels of expression in these two sets of genes. These results are similar to that observed in case of Saccharomyces cerevisaewhole genome analysis (unpublished results). Further, we refined the analysis to examine the correlation, if any, between nucleosome formation potential in upstream regions and extreme expression levels of genes. The Hkg and Tsg groups were further categorised separately into high and low expression level groups as described under \"Methods\" section and their correlation with the nucleosome formation potential was analysed (Table [5](#T5){ref-type="table"}). The high and low expression genes of Hkg show a low negative correlation with scores in intervals 0.8 to 1.0 and 1 to 1.2 and a low positive correlation with scores in intervals -1.2 to -1 and -1.0 to -0.8. In Tsg, except in one interval, there was no valid correlation seen. This solitary value was not considered since the correlation coefficients in other intervals didn\'t reflect this trend. Our data restates that chromatin in 5\' region plays a major role in determining the ubiquitous or restricted tissue expression of a gene as shown by Levitsky et al(2001) \[[@B2]\]. The abundance of nucleosome exclusion elements in Hkg 5\' regions and the low Recon scores reflect their poor preference for nucleosome assembly. The expression analysis suggests that although chromatin plays a role in bringing about extreme variations of gene expression levels in certain classes of genes such as the housekeeping genes, the relation is not linearly correlated with the total, wider range of expression levels. It is possible that nucleosomes might be involved in fine-tuning of expression levels that may escape our attention, since the difference in the range of expression considered is fairly large. The difference detected in nucleosome formation potential between the two sets might reflect the accessibility to basal transcription factors for Hkg and gene/tissue specific transcription factors for Tsg, considering the difference in spatial and temporal expression patterns of the two groups. Analysis of repetitive sequences -------------------------------- Repetitive sequences are implicated in chromatin organisation and heterochromatinisation \[[@B23]-[@B25]\]. They are differentially enriched in various functional categories of genes and are predicted to play an important role in gene regulation \[[@B24],[@B26]\]. We analysed the distribution of various repeat classes in the 5\' regions of Hkg and Tsg using RepeatMasker software. The total repeat content in Hkg regions is seen to be more than in Tsg regions. As reported earlier, our data shows enrichment of SINES (Alu) in comparison to other classes of repetitive sequences in both the sets \[[@B24]\]. Further, the 5\' sequences of Hkg are more enriched in Alu sequences in comparison to those of Tsg regions (Table [6](#T6){ref-type="table"}). The difference in the distribution of Alu repeats in the two classes of sequences was determined by applying t-test for the number of repeats and the repeat content in terms of length in base pairs in each sequence set (Table [7](#T7){ref-type="table"}). The low total repeat content seen in Tsg upstream regions lends support to the hypothesis that condensed chromatin disfavours transposable element insertions in comparison to open chromatin (Hkg promoters)\[[@B27]\]. Genes with high expression levels are clustered in genomic regions known as ridges. These gene rich regions also have high (G+C) content, SINES and genes with short introns \[[@B9]\]. Eisenberg and Levanon \[[@B28]\] have reported the presence of significantly shorter introns and an overall compact gene structure in Hkg as compared to non-Hkg \[[@B28]\]. We have used the gene list provided by Eisenberg and Levanon \[[@B28]\] for our analysis. The enrichment of SINES in the 5\' regions of Hkg suggests that Hkg might be localised in the ridge regions of the genome. More recently, it has been suggested that the contrasting attributes of gene compactness, GC content and the length of the intronic and intergenic sequences in Hkg and Tsg might be involved in chromatin mediated regulation for maintaining distinct expression patterns in the gene sets \[[@B29]\]. Recently, Alu elements have been shown to house transcription factor binding sites and the presence of such regulatory elements might influence the chromatin structure and gene expression \[[@B30]\]. The paradigm for regulation of gene expression in human tissues has shifted the focus from involvement of a battery of transcription regulators to global regulatory mechanisms \[[@B31]\]. These mechanisms have also gained significance in the context of the low estimates of gene numbers in the human genome \[[@B32]\]. It is in this framework that we have analysed the chromatin characteristics of two groups of genes, one that needs almost a continuous and ubiquitous expression and another demanding tissue specific regulation. It had been predicted that the nucleosomal density in a chromatin domain and the buffering of supercoiling waves by repetitive DNA will play a major role in establishing coordinated gene regulation in a domain in the context of the relevance of maintenance of repetitive sequences during evolution \[\[[@B25],[@B33]\], and \[[@B34]\]\]. A recent report also infers the role of chromatin-mediated mechanisms in the differential gene expression patterns seen in housekeeping and tissue specific genes \[[@B29]\]. Our data and analyses lend support to these hypotheses (Figure [3](#F3){ref-type="fig"}). Another recent report, which addresses the chromatin architecture of the human genome, provides experimental evidence that open chromatin correlates with high gene density regions but not with gene expression \[[@B35]\]. This data further supports our in-silicoobservations and strengthens the domain concept for concerted expression of clustered genes. The role of nucleosome formation potential is apparent from the present analysis in both the housekeeping genes as well as tissue specific genes but with an opposing correlation. Housekeeping genes apparently discourage nucleosome formation to match their expression profile in space and time by ensuring accessibility to transcription machinery. In addition, they also show a significant enrichment in poly (dA.dT) stretches, which are known to destabilise nucleosomes. On the other hand, the tissue specific genes show higher scores for nucleosome formation potential through which they perhaps provide selective accessibility to the transcriptional machinery. Further, our analysis suggests that tissue specific genes resort to additional global regulatory features such as matrix association, which would facilitate maintenance of functionally distinct domains to insulate themselves from both silencing and activating regulatory influence of adjacent domains. The differential distribution of repetitive sequences in housekeeping and tissue specific genes might also play an important role in maintaining distinct chromatin landscape over these regions. Conclusion ========== We have demonstrated that the regulatory regions of housekeeping and tissue specific genes have differential chromatin architecture with respect to S/MAR binding, nucleosome positioning potential and repetitive sequences. This has potential implications for regulation of gene expression in eukaryotic genomes. Methods ======= In this study, the 5\' and 3\' flanking regions of genes were analysed for various attributes of chromatin organisation. The list of human housekeeping genes (Hkg) was retrieved from <http://www.compugen.co.il/supp_info/Housekeeping_genes.html>\[[@B28],[@B36]\]. 532 genes have been categorised as housekeeping because of their ubiquitous and high expression levels in 47 tissues. The list and expression levels of the human tissue specific genes were obtained from Eli Eisenberg (personal communication). 566 genes expressed in only a single tissue were taken as tissue specific genes (Tsg) and analysed. We could unambiguously retrieve sequences of 525 Hkg and 558 Tsg from human genome build 33 (NCBI). Approximately, 2000 bp of the 5\' and 3\' regions from each of these genes were taken for analysis. Scaffold/matrix associated regions (S/MAR) analysis --------------------------------------------------- MAR Finder was used for prediction of S/MAR regions \[[@B37],[@B38]\]. All the default options and the \"New MAR Rules\" were selected for predicting S/MARs. ChrClass program was used for S/MAR prediction \[[@B39],[@B40]\]. Nucleosome organisation and gene expression correlation analysis ---------------------------------------------------------------- The upstream regions (2000 bp) were scanned for nucleosome exclusion elements \[[@B18],[@B20]\] -- poly (dA.dT) pure stretches of \>10 bp length and \[5\' (CCGNN) 3\'\]~2--5~using in-house programs. Recon was used for evaluating nucleosome formation potential in the sequences \[[@B2],[@B41]\]. The score outputs of the 5\' regions were categorised in frequency intervals of 0.2 with a range from -3.2 to +3.2. The Recon scores around +1 and -1 imply strong nucleosome formation and exclusion potentials respectively. The scores in the four intervals of relevance (0.8 to 1, 1 to 1.2, -0.8 to -1 and -1 to -1.2) were taken for all the analyses. Since the promoter region information was not retrieved for these genes, the 2000 bp upstream region from the gene start site was split into 400, 800, 1200 & 1600 bp and analysed. The Recon scores at 400 bp were used to draw correlation between the nucleosome formation potential and expression levels in the two sets of genes. In each sequence set, genes with expression levels \<500 and \>5000 affymetrix expression units were classified as low and high expression genes respectively. We considered a minimum ten fold difference in the expression levels of genes as a relevant criterion for classifying them as high and low expression genes. In Hkg, this criterion yielded 33 low expression and 35 high expression genes. In Tsg, we categorised 416 low expression genes and 24 high expression genes. Repetitive sequence analysis ---------------------------- RepeatMasker version: 20040306-web was used to calculate the repeat content in 2000 bp upstream sequences of the two groups of genes \[[@B42]\]. List of abbreviations ===================== S/MAR: scaffold/matrix attachment regions Hkg: housekeeping genes Tsg: tissue specific genes df: degree of freedom Authors\' contributions ======================= MG, VB and SKB contributed to the study design, analyses and in drafting the manuscript. MG, PS, SKDS, DD, KK and GP were involved in the data retrieval and analyses. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 gene list and expression levels of housekeeping genes, gene list and expression levels of tissue specific genes. \'supplementaryfile1.xls\' contains the list of housekeeping and tissue specific genes and their expression levels used for the analysis. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ MG acknowledges the financial support provided by Council for Scientific and Industrial Research (CSIR), India. SKB thanks Council for Scientific and Industrial Research, India and VB thanks Indian Council for Medical Research (ICMR), for financial assistance through a grant. The authors wish to acknowledge Dr. Beena Pillai, Dr. Rakesh Sharma, Dr. Neeraj Pandey and Dr. Mitali Mukerji for their valuable discussions and suggestions. We would also like to acknowledge Samira for careful checking and helping with the manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Nucleosome formation potential score distributions for 5\' regions of housekeeping and tissue specific genes. The 5\' sequences of human housekeeping and tissue specific genes were analysed by Recon for distribution of nucleosome formation potential scores. Frequency distribution histograms were plotted for scores in various intervals (range -3.2 to +3.2). (A) and (B) show the distribution of nucleosome formation potential scores at 400 and 2000 bp upstream from the gene start site respectively. Nucleosomal density is significantly lower for housekeeping genes as compared to tissue specific ones, in regions close to the gene start site. ::: ![](1471-2105-6-126-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Nucleosome formation potential score distributions for 5\' regions at different positions from the gene start site in housekeeping and tissue specific genes. The 5\' regions of 800, 1200 and 1600 bp from the gene start site of housekeeping and tissue specific genes were taken for the analysis. Frequency distribution histograms were plotted for Recon scores in various intervals (range -3.2 to +3.2). (A), (B) and (C) show the distribution of nucleosome formation potential scores at 800, 1200 and 1600 bp upstream from the gene start site respectively. As we move upstream from the gene start site, the difference in the nucleosome formation potentials between housekeeping and tissue specific genes gradually fades away. ::: ![](1471-2105-6-126-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A model for chromatin landscape in 5\' regions of tissue specific and housekeeping genes. (A) depicts the repressive role of chromatin in maintaining tissue specific gene expression profiles in a chromosome. The chromatin organisation in the 5\' regions of Tsg1 and Tsg2, two different tissue specific genes dispersed in the chromosome is shown. Nucleosome formation potentials and S/MARs -- the boundary elements, are enriched in their upstream regions and might play a major role in facilitating tissue specific expression. This is likely to be a local effect since neighbouring genes might have a different expression pattern. (B) depicts the chromatin organisation in the 5\' regions of Hkg1, Hkg2 and Hkg3, three housekeeping genes clustered in the chromosome. The presence of low nucleosome formation potential regions and enrichment of nucleosome destabilising elements ensure an open chromatin configuration in this domain. As Hkg generally cluster together, they are depleted in S/MARs relative to tissue specific genes as shown in the present analysis by the significant absence of predicted S/MARs in both 5\' and 3\' regions of housekeeping genes as compared to tissue specific genes. ::: ![](1471-2105-6-126-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Distribution of putative S/MARs in housekeeping and tissue specific genes. ::: Putative S/MARs in 5\' regions (%)\* Putative S/MARs in 3\' regions (%)\* ------------------- ------------------------------------------ ------------------------------------------ ------ ------ presence of S/MAR 26.1 34.1 19.1 20.6 absence of S/MAR 26.1 19.2 34.5 25.1 ^\#^Housekeeping genes, ^§^Tissue specific genes, \* ChrClass and MAR Finder programs were used for prediction of S/MARs in housekeeping and tissue specific genes regulatory regions (5\' & 3\' regions -- 2000 bp each). The common predictions of both the programs were used for the analysis. The data is represented as percentage of genes with predicted S/MARs in 5\' and 3\' regions of 525 housekeeping and 532 tissue specific genes. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distribution of poly (dA.dT) repeats of various lengths in the 5\' upstream regions of housekeeping and tissue specific genes. ::: Poly (dA.dT) stretch (bp) No. of repeat stretches in the two classes No. of genes with repeats in 5\' region (%) ------------------------------- ------------------------------------------------ ------------------------------------------------- ------------ ------------ ---------- \>10 443 345 268 (51.0) 240 (43.0) 1.31E-04 \>11 381 297 243 (46.3) 214 (38.4) 3.25E-04 \>12 339 248 226 (43.1) 184 (33.0) 6.10E-05 \>13 295 207 209 (39.8) 156 (28.0) 4.29E-05 \>14 251 168 188 (35.8) 128 (22.9) 2.77E-05 \>15 209 140 164 (31.2) 111 (19.9) 8.83E-05 \>16 180 116 146 (27.8) 99 (17.7) 7.58E-05 \>17 155 103 134 (25.5) 88 (15.8) 2.42E-04 \>18 138 79 120 (22.9) 71 (12.7) 2.23E-05 \>19 112 66 101 (19.2) 59 (10.6) 2.61E-04 \>20 100 58 92 (17.5) 53 (9.5) 5.32E-04 ^\#^Housekeeping genes, \Tissue specific genes. A total of 525 housekeeping and 558 tissue specific genes were analysed. The numbers in parentheses (4^th^& 5^th^columns) represent the percentage of genes containing the repeat stretch. ^§^Difference in the distribution of poly (dA.dT) stretches in Hkg and Tsg analysed by applying t-test (for normalizing the difference in sample size). The repeat lengths from \>12 to \>18 bp are showing very significantly different distributions between Hkg and Tsg. The distributions were examined in 2000 bp upstream region from the gene start site. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### t-test P-values for the difference in the distribution of nucleosome formation potential scores between housekeeping and tissue specific genes. ::: Length (bp)\ P-value in intervals of scores** ------------------- -------------------------------------- ---------- ---------- ---------- 400 3.53E-13 8.73E-17 1.28E-17 4.45E-23 800 1.16E-24 6.27E-24 6.64E-12 6.65E-22 1200 6.91E-26 1.44E-24 2.10E-09 1.64E-18 1600 1.72E-24 6.99E-24 2.72E-07 5.63E-15 2000 2.55E-25 1.84E-25 3.22E-05 6.71E-13 \denotes the length of 5\' upstream region from the gene start site taken for the analysis. The scores were compared in the four Recon score intervals of relevance -1.2 to -1, -1 to -0.8, 0.8 to 1 and 1 to 1.2. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Correlation coefficients of total expression levels (log~10~) with nucleosome formation potential scores in housekeeping (Hkg) and tissue specific genes (Tsg). ::: Category* Correlation coefficient -------------- ----------------------------- ------- ------ ------- Hkg^\#^ 0.10 0.14 0.04 -0.01 Tsg\* -0.10 -0.12 0.15 0.17 ^\#^Housekeeping genes, \Tissue specific genes. The correlation was drawn in the four Recon score intervals of relevance -1.2 to -1, -1 to -0.8, 0.8 to 1 and 1 to 1.2. ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Comparison of the level of correlation between nucleosome formation potential scores and contrasting expression levels of genes. ::: \Category Correlation coefficient** ---------------- ----------------------------- ------- ------- ------- Hkg ↑↑ 0.17 0.30 -0.10 -0.26 Hkg ↓↓ 0.36 0.35 -0.28 -0.39 Tsg ↑↑ 0.03 0.02 0.26 0.11 Tsg ↓↓ -0.12 -0.14 0.15 0.16 \High and low expression level genes were categorised in Hkg (housekeeping genes) and Tsg (tissue specific genes) groups separately. ^\#^Recon score intervals. The genes classified as high and low expression genes in both Hkg and Tsg had atleast a 10-fold difference in their expression levels. The up (↑↑) and down (↓↓) arrows denote high expression and low expression respectively. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### The distribution of Alu repeats in 5\' upstream regions of housekeeping (Hkg) and tissue specific genes (Tsg) is represented in terms of the number of copies and basepairs covered by Alu repeats. ::: Repeat category* No. of copies % of the total sequences covered by the repeat --------------------- ------------------- ---------------------------------------------------- ------ ------ Alu 866 575 20.1 12.3 ^\#^Housekeeping genes, \Tissue specific genes. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### t-test P-values for the difference in the distribution of Alu repeats in 5\' upstream regions of housekeeping and tissue specific genes. ::: Repeat Category* **P-value** --------------------- --------------- ---------- Alu 3.02E-08 7.63E-11 :::	PubMed Central	2024-06-05T03:55:58.312808	2005-5-26	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173084/", "journal": "BMC Bioinformatics. 2005 May 26; 6:126", "authors": [ { "first": "Mythily", "last": "Ganapathi" }, { "first": "Pragya", "last": "Srivastava" }, { "first": "Sushanta Kumar Das", "last": "Sutar" }, { "first": "Kaushal", "last": "Kumar" }, { "first": "Dipayan", "last": "Dasgupta" }, { "first": "Gajinder", "last": "Pal Singh" }, { "first": "Vani", "last": "Brahmachari" }, { "first": "Samir K", "last": "Brahmachari" } ] }
PMC1173085	Background ========== The introduction of DNA microarray technology has enabled investigators to screen thousands of genes simultaneously. One of the main goals of these studies is to identify differentially expressed genes between two biological conditions. For example, many studies \[[@B1]-[@B4]\] have been performed in prostate cancer research to find candidate markers. Since laboratory protocols, microarray platforms and analysis techniques used in these studies were not identical, it is difficult to make a comparison among the results obtained from them. However, systematic integration of gene expression data from different sources increases statistical power of detecting differentially expressed genes and allows assessment of heterogeneity. Meta-analysis is a classical statistical methodology for combining results from different studies addressing the same scientific questions, and it is becoming particularly popular in the area of medical and epidemiological research \[[@B5]\]. Meta-analysis methods have recently been applied to the analysis of microarray data \[[@B6]-[@B11]\]. Rhodes et al. \[[@B6]\] focused on combining p-values for each gene from the individual studies to estimate an overall p-value for each gene across all studies. Their method has been applied to four prostate cancer microarray datasets, two of which are cDNA microarray data and the remainder Affymetrix microarray data. Samples in each data set were taken from prostate cancer cases, but were analyzed with different platforms. Differential expression was assessed independently for each gene in each dataset. Since the method chosen to combine results across studies was based on the statistical confidence measure, the p-value, not on the expression level, this strategy avoids direct comparisons of data sets and related cross-platform normalization issues. Choi et al. \[[@B7]\] focused on integrating effect size estimates to obtain an overall estimate of the average effect size. Effect size is used to measure the magnitude of treatment effect in a given study. Using the same datasets as those used by Rhodes et al. \[[@B6]\], they demonstrated that their method can lead to the discovery of small but consistent expression changes with increased sensitivity and reliability. Parmigiani et al. \[[@B9]\] developed a correlation-based method for assessing reproducibility of gene expression studies with application to lung cancer. They demonstrated that their method can improve correlation across the various studies. Jiang et al. \[[@B10]\] used a distribution transformation method to integrate two lung cancer studies and proposed a gene shaving-based classification approach to identify a small list of differentially expressed genes between lung cancer and normal patients. They noted that many of the selected genes have been experimentally validated. Although some of the above studies (for example Rhodes et al. \[[@B6]\], Choi et al. \[[@B7]\], Parmigiani et al. \[[@B9]\]) demonstrated the utility of integrating cDNA and Affymetrix microarray data, other investigators argued against this approach. Kuo et al. \[[@B12]\] compared Affymetrix and spotted cDNA gene expression measurements based on 60 cell lines from the National Cancer Institute. They found low correlation between the actual gene measurements from the two technologies, and concluded that \"data from spotted cDNA microarrays could not be directly combined with data from synthesized oligonucleotide arrays.\" Moreover, they concluded that it was unlikely that the two types of data could be transformed or normalized into a common standardized index. Jarvinen et al. \[[@B13]\] determined the level of concordance between microarray platforms by analyzing breast cancer cell lines with in situ synthesized oligonucleotide arrays, commercial cDNA microarrays and custom-made cDNA microarrays. Their results demonstrated that data from different microarray platforms are variable to the extent that direct integration of data from different platforms may be complicated and unreliable. In classical meta-analysis, quality measures have often been used when combining results across studies. It has been argued that studies of a higher quality will give more accurate estimates of the true parameter of interest, and therefore studies of high quality should receive a higher weight in the analysis summarizing across studies \[[@B14]\]. The Affymetrix microarray technology has been used worldwide. Its success has been demonstrated by numerous publications in scientific journals. However, it is well-known that only part (approximately 40--50%) of the whole genome is expressed in any specific tissue type, so there are many genes showing low expression and random variability across samples. Furthermore, some genes will be measured less precisely by some technologies, or studies, than others. Therefore, our ability to develop powerful statistical methods for efficiently integrating and weighting information from related genomic experiments will be critical in the success of the massive investment made on genomic studies. The focus of this paper is to design and implement a quality measure appropriate for Affymetrix microarray data. Using our quality measure, we weighted the importance of each gene in each experiment and incorporated our quality measure into the effect size model proposed by Choi et al. \[[@B7]\] to model inter-study variation of gene expression profiles. We believe that applying this approach can lead to a more accurate description of expression patterns than Choi et al\'s method \[[@B7]\]. Results ======= We used two data sets consisting of gene expression profiles in lung cancer and normal subjects. These datasets were collected using different chip types of the Affymetrix oligonucleotide microarrays and were conducted by two research groups, one from Harvard and the other from Michigan (see Methods section for details). A list of 6124 common probe sets found in the two datasets was used for data analysis in this study \[[@B10]\]. We developed a quality weight for each gene in each study by modeling the log of the detection p-values with an exponential distribution, and then summarizing across arrays and groups within each study (see Methods). In order to visualize the effect of the proposed quality weighting, we calculated the mean expression value of each probe set across all normal samples, where the expression variation presumably is less heterogeneous than among the cancer samples. The mean expression value of a probe set in a study is the estimated average of the probe set\'s intensity values across all normal samples in the study. Figure [1](#F1){ref-type="fig"} shows the scatter plot of the average expression values of the probe sets in the Harvard dataset plotted against that of the Michigan dataset: (a) weighted by the quality score (see Methods section for our definition of the quality score), and (b) un-weighted. This plot is intended to be illustrative only -- our algorithm weights the test statistics, rather than the gene expression measures. In Figure [1(a)](#F1){ref-type="fig"}, it can be seen, as expected, that many of the genes with low levels of expression are associated with low quality weights. We show examples of quality scores for selected probe sets in Table [1](#T1){ref-type="table"}. The two datasets may have very different detection p-value distributions, which are reflected in the quality scores. Figure [2](#F2){ref-type="fig"} shows a box plot of the detection p-values for one probe set from Table [1](#T1){ref-type="table"}. When the two datasets give small p-values (e.g., last line in Table [1](#T1){ref-type="table"}), the minimum p-value may be much smaller in one dataset than another. Both, however, will give high quality scores with an appropriate choice of the sensitivity parameter sthat adjusts how the quality measure interprets the detection p-values. Figure [3](#F3){ref-type="fig"} shows the adjusted and unadjusted quantile -- quantile (Q-Q) plots of the observed vs. expected Q values. Q is the test statistic we used for assessing heterogeneity, and is described in detail later in the Methods section. In the adjusted Q-Q plot, the quality score was used as a weight in the computation of Q while it was not considered in the unadjusted Q-Q plot. From these graphs, we can see that the quantiles of the observed Q values are far from the expected quantiles of a ![](1471-2105-6-128-i2.gif) distribution, suggesting that these two datasets generated heterogeneous results beyond random sampling errors. Therefore, we applied the random effect model in this study. The quantiles of the Q statistic were closer to the quantiles of the expected chi-square distribution when quality-adjustment was considered (Figure [3(a)](#F3){ref-type="fig"}) than when it is was not (Figure [3(b)](#F3){ref-type="fig"}). The variance for the unadjusted Qvalues was 9.45, but it was reduced to 3.31 when quality adjustment was used. This result suggests that the incorporation of the adjusted quality measure into effect size estimation can increase consistency and decrease heterogeneity between these two datasets. To identify a list of potentially \"significant\" genes, we adapted the false discovery rate (FDR) algorithm implemented in \[[@B30]\]. We first calculated the adjusted z statistics for all genes based on random-effects model (REM). Genes were then ranked by the magnitude of their z statistic values. A permutation-based approach was used to obtain the corresponding expected ordered z statistic. The potentially \"significant\" genes are genes with a distance between the ordered z statistic from the observed data and that of the permuted data exceeding a given threshold (delta). Figure [4](#F4){ref-type="fig"} shows the relationship between the number of significantly differentially expressed genes and different delta levels. As we see in this figure, the quality-adjusted REM can identify many more significant genes than the quality-unadjusted REM model at any fixed level of delta. We calculated the estimated FDR for each given delta. As expected, the number of genes called significant increased as the value of delta decreased, but at the cost of an increasing FDR. The estimated FDR was zero down to a delta of 0.6, where 228 genes were called significant in the quality-adjusted model and 153 genes in the quality-unadjusted model. In order to get a manageable gene list, we arbitrarily set delta at 1.1. At this delta level, we selected 29 differentially expressed genes (representing 32 probe sets) at a FDR of 0% when the quality weight was incorporated into the estimation of the effect size. However, when the quality measure was not used (Choi et al\'s method \[[@B7]\]), we only selected 20 differentially expressed genes (representing 21 probe sets) at a FDR of 0%. All the 20 genes were also in the top of the list of the 29 genes identified with the quality-weights. Tables 2 and 3 show the selected differentially expressed genes between normal and lung adenocarcinoma patient samples using the quality-adjusted and quality-unadjusted models, respectively, with genes ordered based on their z statistic values \[see [Additional files 1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}\]. As can be seen in Tables 2 and 3, 4 of the 9 genes that were selected by our method, but not by Choi et al\'s method, have also been identified by several other groups including Jiang et al. \[[@B10]\], Beer et al. \[[@B15]\] and Bhattacharjee et al. \[[@B16]\]. In particular, some of these 4 genes, such as TEK and TGFBR2, have been experimentally validated (shown in Table 3 of Jiang et al.\[[@B10]\]). For a cutoff of an adjusted z value of 1.96 (corresponding to a 5% level of significance), the quality-adjusted model identified 9 significantly expressed genes while the quality-unadjusted model identified only 2 of the 9 significantly differentially expressed genes. All these results suggest that our proposed method may have increased sensitivity to detect more differentially expressed and biologically relevant genes than Choi et al\'s method \[[@B7]\]. We compared genes identified with our method with genes identified by Jiang et al. \[[@B10]\], Beer et al. \[[@B15]\] and Bhattacharjee et al. \[[@B16]\], sixteen of the 29 genes identified by our proposed model were also detected in at least one of these studies. In particular, we observed that 6 of the 29 genes were consistently identified in the other three studies. There were 13 of the 29 genes that were uniquely identified in our study. Some of these are plausible candidates for lung adenocarcinoma. For example, Walker et al. \[[@B17]\] found that G protein-coupled receptor kinase 5 (GRK5) is a key gene regulating airway response, that may have implications in obstructive airway diseases. Discussion ========== In this study we proposed a measure to quantify Affymetrix gene chip data quality for each gene in each study. The quality index measures the performance of each probe set to detect its intended target. Furthermore, we extended a traditional effect size model by using the quality index as a weight for combining information from different chip types of Affymetrix microarrays, and incorporating this weight into a random-effects meta-analysis model. We illustrated the advantages of our proposed methods using the Harvard and Michigan gene expression datasets used in \[[@B18]\]. This approach of using the detection p-values to weight the gene expression estimates can be applied in a more general context and to other microarray meta-analysis methods, such as that of Rhodes et al \[[@B6]\]. The assumption of a group-specific exponential distribution for the negative log detection p-values is a very rough approximation. The true distribution is probably closer to the two-parameter log-beta distribution. However, due to the fact that there are only 16--20 probe pairs per probe set used in the Wilcoxon test statistic for the detection p-values, the p-values follow a somewhat discrete distribution. In particular for highly-expressed genes, the p-values for all samples in a group may have the same near-zero value. Estimation of two-parameters is therefore impossible. It would be worth investigating the sensitivity of the performance of this quality-weighting approach to the distributional assumption. The performance of the weighting also depends on the sensitivity parameter s. In this study, a high swas needed since a lower cutoff gave almost all genes extremely low quality scores. This is a consequence of the observed detection p-value distributions. These data were created in 2001, and since that time the quality of the oligonucleotide array technology has improved with more specific probe selections and optimized experimental conditions. For expression data obtained today, smaller values of s would be more appropriate. For example, in an analysis of a dataset generated more recently using improved oligonucleotide technology (not shown), using s= 0.05 gave good performance. The discrepancy between the calculated and expected values for the Q test of heterogeneity was smaller when our quality measure was incorporated than when it was not, as suggested by the quantile-quantile plots, but overall there still remains some variability that has not been fully captured by the weight function, emphasizing the need for a more elaborate weighting strategy and sensitivity analyses. From a biological point of view, lung adenocarcinoma may be heterogeneous originating from different causes \[[@B19]\] and methods like cluster analysis in which subtypes of relatively homogeneous groups of disorders can be identified might be useful. The focus of our paper was to introduce a methodology that can be used to integrate data, and as such the grouping of the samples as \"diseased\" versus \"normal\" was primarily used to identify genes that discriminate the two groups in a broader sense than at a higher resolution in which sub types could be identified. Choi et al. \[[@B7]\] have also provided a brief Bayesian interpretation of their effect size model to integrate information from multiple microarray studies. They argued that a Bayesian approach could offer a more flexible and robust modeling strategy. We did not consider this issue in this study. It will be interesting to investigate how our proposed quality measure could be incorporated into their Bayesian framework in the future. Methods ======= Data source and preprocessing ----------------------------- We selected two Affymetrix microarray data sets from the 4 lung cancer data sets provided by the organizers of the Critical Assessment of Microarray Data Analysis (CAMDA) conference \[[@B18]\]. These datasets were collected using different versions of the Affymetrix oligonucleotide microarrays and were conducted by two research groups, one from Harvard and the other from Michigan. The Michigan study \[[@B15]\] used the HuGeneFL Affymetrix chip, containing 7,129 probe sets, each with 20 probe pairs. This study included 86 lung adenocarcinoma patient samples and 10 normal samples. The Harvard study \[[@B16]\] used the HG\_U95Av2 chip with 12,625 probe sets, each with 16 probe pairs. This study included 17 normal and 127 lung adenocarcinoma patient samples. We use the word \"sample\" interchangeably with \"array\". The main objectives addressed in these two studies were to identify differentially expressed genes related to lung adenocarcinoma, and genes whose expression was related to patient survival. Our interest is in the former. We developed a new method for identifying genes that are differentially expressed between the cancer and normal samples, by modelling the effect size and integrating information from the Harvard and Michigan studies. We converted the probe level data to a single expression measure for each probe set using the robust multi-array average (RMA) algorithm \[[@B20]\], which provides higher specificity and sensitivity in detecting differentially expressed genes. We used 6124 common probe sets in these two studies for the data analysis in our study. These probe sets were selected based on a sequence-based probe matching method, which is believed to produce more consistent results when comparing similar biological data sets obtained by different microarray platforms \[[@B10],[@B21]\]. The detailed procedure used to select these probe sets can be found in the Data processing Section of \[[@B10]\]. Quality measure for Affymetrix Genechip data -------------------------------------------- Affymetrix genechip microarrays are used to monitor gene expression for thousands of transcripts. Each transcript is represented as a probe set, and a probe set is made up of probe pairs comprised of Perfect Match (PM) and Mismatch (MM) probe cells. The level of expression of a gene product is estimated using the intensities of each probe pair in a probe set. Therefore, the probe-specific variability in a probe set can be used as a measure of the performance of that probe set. The detection algorithm proposed by \[[@B22]\] generates a detection p-value, which represents the probability that the probe set (gene) expression is above zero (i.e., turned on), and measured reliably and consistently. A lower p-value is considered as a useful indicator that the measured gene expression is valid and reliable \[[@B22]\]. Specifically, the p-value is based on testing whether the probe-specific differences (PM-MM) are almost always positive. We used the detection p-values to define quality measures for probe sets, summarizing across the arrays and experiments. We realize that some genes may be truly \"off\" under some experimental conditions and hence a large detection p-value may be providing useful information. However, if a gene is \"off\" under all experimental conditions, we argue that analysis of this gene contributes little to understanding of the experiment. For any gene, let pvaluedenote its detection p-value and x~lg~denote - log(pvalue~lg~) for sample lin group g. We assume that each study compares G groups, where there are n~g~samples in group g, and g = 1, 2, \..., G. For example, in the lung cancer data, G = 2, since adenocarcinoma samples are compared to controls. It is well known that p-values follow a uniform distribution when there is no signal, and therefore, we expect x~lg~to follow an exponential distribution with mean λ = 1 if the gene is not expressed. In order to develop a single quality measure for each gene across all samples in one study, we use this relationship with the exponential distribution to motivate a quality measure. We assumed that the detection p-value for a single gene of sample lin group gfollows the distribution x~lg~= - log (pvalue~lg~) \~ Exponential(λ~g~), where different distributions of expression can be expected in each group g. It should be noted that we are modeling the p-value of one gene, across the samples in a group. This is different from the approaches of Allison et al. \[[@B23]\] and Pounds et al. \[[@B24]\] who modeled the distribution of p-values across genes. Although the true distribution of the x~lg~may not be exponential, this assumption leads to a simple model where the one parameter can be estimated by a closed-form expression. Hence, the parameter λ~g~for each gene, study and group gcan be estimated by: ![](1471-2105-6-128-i3.gif) where ![](1471-2105-6-128-i4.gif) is the usual sample mean. This is a maximum likelihood estimator (MLE) with well-known asymptotic optimality properties \[[@B25]\]. To combine across the groups, we assumed a sensitivity parameter s. It is defined as the probability that a representative probe set in a particular treatment group shows a detectable signal, assuming that the relevant distribution (exponential or beta) holds. It can be thought of as the equivalent to a detection p-value defined for a whole group rather than for one array, and an appropriate value for sshould be chosen with this interpretation in mind. For example, the default settings by Affymetrix software for the detection calls are p = 0.06 for a \"Marginal\" call, and p = 0.04 for a \"Present\" call, although these can be altered by the user. In practice, the appropriate choice of smay depend on the signal-detection capability of particular technologies. We recommend plotting the distribution of quality scores for different choices of s, and choosing a value that clearly distinguishes genes of low quality (scores near zero) from high quality genes (scores near 1). The sensitivity parameter sis a chosen cutoff, so that genes that are \"off\" or poorly measured across all experimental conditions will have pvalue≥ s, or in other words, ![](1471-2105-6-128-i5.gif). Therefore, we can define a quality measure across the groups, for each gene and each study as: ![](1471-2105-6-128-i6.gif) The choice of the maximum gives more weight to genes measured with high quality in at least one group, thereby allowing a gene that is \"off\" in one condition and \"on\" under another condition to provide useful information in the analysis. We treat this quality score as a weight for each gene in the subsequent analysis. Modelling effect size with quality-adjusted weights --------------------------------------------------- In order to simplify the discussion, we consider two groups, treatment (t) and control (c) groups, in study i= 1,2,\...,k. Let n~it~and n~ic~denote the number of treatment and control samples in study i, respectively. For each gene, let μ denote its overall mean effect size, a measure of the average differential expression for that gene, and let y~i~denote the observed effect size for study i. We modeled effect size using the hierarchical model: ![](1471-2105-6-128-i7.gif) Where τ^2^is the between-study variability and ![](1471-2105-6-128-i8.gif) is the within-study variance, measuring the sampling error for the i^th^study. Choi et al. \[[@B7]\] used the standardized mean difference as a measure of the observed treatment difference y~i~. This well-known estimator of treatment difference found in Hedges and Olkin\'s \[[@B26]\] work is ![](1471-2105-6-128-i9.gif) where ![](1471-2105-6-128-i10.gif) and ![](1471-2105-6-128-i11.gif) are the average gene expression values in the treatment and control groups of study i, respectively, and ![](1471-2105-6-128-i12.gif) is the pooled standard deviation. For a study with nsamples, an approximately unbiased estimator of ![](1471-2105-6-128-i13.gif) is given by ![](1471-2105-6-128-i14.gif)\[[@B26]\]. The estimated variance ![](1471-2105-6-128-i8.gif) of the unbiased effect size is given by \[[@B27]\] ![](1471-2105-6-128-i15.gif) In a fixed-effects model (FEM), the error of the observed effect sizes is fully assigned to sampling error only, ignoring the between study variance, so τ^2^= 0 and y~i~\~ N(μ, ![](1471-2105-6-128-i8.gif)). On the other hand, a random-effects model (REM) considers that each study estimates a different treatment effect θ~i~. These parameters are drawn from a normal distributionθ~i~\~ N(μ, τ^2^). To assess whether FEM or REM is most appropriate, we tested the hypothesis τ = 0 using the following test statistic, which is a modification of Cochran\'s test statistic \[[@B28]\] that incorporates our quality measure for each study ![](1471-2105-6-128-i16.gif) where ![](1471-2105-6-128-i17.gif) and ![](1471-2105-6-128-i18.gif) ![](1471-2105-6-128-i19.gif) is the weighted least squares estimator that ignores between study variations. Under the null hypothesis of τ = 0, this statistic follows a ![](1471-2105-6-128-i20.gif) distribution. We followed Choi\'s approach \[[@B7]\] to draw quantile-quantile plots of Qto assess whether a FEM or REM model is appropriate. If the null hypothesis of τ = 0 is rejected, we estimate τ based on the method developed by DerSimonian and Laird \[[@B29]\] ![](1471-2105-6-128-i21.gif) Therefore, we can estimate μ that corresponds to a random effects model by ![](1471-2105-6-128-i22.gif) where ![](1471-2105-6-128-i23.gif). Under the REM, ![](1471-2105-6-128-i24.gif) The zstatistics to test for treatment effect under REM is ![](1471-2105-6-128-i25.gif) However, when there are only a small number of arrays in each group, the estimates of standard error sfor each gene can be very variable. Some genes might by chance have very small standard errors, and therefore appear highly significant. To address this problem, different approaches have been developed for \"smoothing\" the variance estimates by borrowing information from the ensemble of genes. This can assist in inference about each gene individually. For example, Tusher et al. \[[@B30]\], Efron et al. \[[@B31]\] and Broberg \[[@B32]\] used t-statistics where an offset was added to standard deviation while Smyth \[[@B33]\] proposed a t-statistic with a Bayesian adjustment to the denominator. We took the offset s~0~as the quantile of the gene-wise standard errors that minimizes the coefficient of variation of the z statistics \[[@B30]\]. Therefore, we can calculate the adjusted zstatistics (used in this study) to test for treatment effect under REM as ![](1471-2105-6-128-i26.gif) The adjusted zstatistics for FEM is the same as that for REM except that τ^2^= 0. Note that all these expressions refer to a single gene. The adjustment for computing z statistic was also used by Garrett-Mayer et al. \[[@B34]\]. Assessment of differentially expressed genes -------------------------------------------- We performed a multiple testing procedure, as described by Dudoit et al. \[[@B35]\], to evaluate statistical significance for differentially expressed genes in the combined studies. The false discovery rate (FDR) \[[@B36]\] has become a popular error measure for this purpose. Tusher et al. \[[@B30]\] developed a permutation-based method to calculate FDR for evaluating differentially expressed genes. We adapt their approach to our meta-analytic framework as follows: 1\. For each gene j, j= 1,2,\..., J, in the original data, compute the adjusted z statistic Z~1~,\...,Z~J~based on the meta-analysis procedure described in the previous section 2\. Order these z statistic values to obtain Z~(1)~≤,\..., ≤ Z~(J)~ 3\. Create B random permutations within both studies. For each permutation b= 1,2,\... B, produce the adjusted z statistic Z~1,b~,\..., Z~J,b~for gene j= 1,2,\..., J 4\. Compute expected order z statistics: ![](1471-2105-6-128-i27.gif) 5\. The potentially \"significant\" genes are those that have a distance between Z~(j)~and ![](1471-2105-6-128-i28.gif), greater than a given threshold delta (Δ). Therefore, we can find the smallest positive Z~(j)~such that ![](1471-2105-6-128-i29.gif), say t~1~. Similarly, we can find the largest negative Z~(j)~, say t~2~ 6\. The estimated FDR for the selected significant genes at the given delta (Δ) is given by: ![](1471-2105-6-128-i30.gif) Authors\' contributions ======================= JB initiated, designed and managed the study. CG proposed the quality measure method and participated in designing and managing the study. PH conducted data analysis and drafted the manuscript. JB and CG revised the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Selected differentially expressed genes between normal and lung adenocarcinoma patient samples (Genes selected based on quality-adjusted modelwith delta = 1.1) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Selected differentially expressed genes between normal and lung adenocarcinoma patient samples (Genes selected based on quality-unadjusted modelwith delta = 1.1) ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We would like to thank Dr. Hongying Jiang who provided us with the list of common probe set IDs for the two data sets used in our study. We also acknowledge helpful suggestions from two anonymous reviewers that greatly improved the quality of the manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Scatter plots of (a) quality-weighted and (b) quality-unweighted mean expression values for the 6124 common probe sets in the Harvard and Michigan datasets based on normal (healthy) samples. ::: ![](1471-2105-6-128-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Detection p-values for a sample probe set (38249\_at). H1 and H2 denote the detection p-values in normal and lung cancer groups, respectively, for the Harvard study; whereas M1 and M2 denote the detection p-values in normal and lung cancer groups, respectively, for the Michigan study. ::: ![](1471-2105-6-128-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Quantile -- Quantile plots of the observed versus the expected Q statistic: (a) with quality adjustment, and (b) without quality adjustment. ::: ![](1471-2105-6-128-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Relationship between number of significantly expressed genes and different delta levels, obtained from fitting the random effects model. ::: ![](1471-2105-6-128-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Quality scores for selected probe sets at a sensitivity parameter s = 0.2 ::: Quality Score -------------- --------------- ------- ------- ------- 38249\_at 0.663 0.001 0.255 4.050 32180\_s\_at 0.263 0.732 0.829 0.194 37174\_at 0.001 0.495 4.050 0.437 32318\_s\_at 0.795 0.795 0.142 0.142 :::	PubMed Central	2024-06-05T03:55:58.316211	2005-5-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173085/", "journal": "BMC Bioinformatics. 2005 May 27; 6:128", "authors": [ { "first": "Pingzhao", "last": "Hu" }, { "first": "Celia MT", "last": "Greenwood" }, { "first": "Joseph", "last": "Beyene" } ] }
PMC1173086	Background ========== Response thresholds and exclusion criteria are applied when presenting a summary of a microarray study; otherwise the data set can be unmanageable. The precise effect of such criteria is, however, rarely discussed or investigated. We believe that many researchers are typically not aware of the effect that their chosen thresholds or \'qualification\' criteria have had on their final set of \"significant genes\". A common example would be data exclusion. In studies using the Affymetrix platform the selection criteria typically uses the absent, moderate or present \'call\' system, calculated with algorithms implemented in MAS5 (microarray suite 5 from Affymetrix) or D-chip \[[@B1]\]. The decision of whether the detection threshold is set to 10% or 20% (for example) will change the number of significant genes when the analysis is followed by the use of a parametric test with multiple testing corrections (for example). How predictable the imposition of such thresholds, on the composition of the output, is not clear. A recent method to identify significant genes is \"Significance Analysis of Microarrays\" (SAM)\[[@B2]\]. An ISI search indicates that SAM is the most popular method employed for microarray analysis (635 citations of the original publication as of October 2004). In SAM, the relative difference (d(i)) is compared to the distribution of d(i) following random permutation of the sample categories. For each d(i), a certain proportion of all genes in the permutation set (control set) will be found to be \'significant\' by chance and this parameter is then used to calculate a \"False Discovery Rate\" (FDR). This is presented as a q value for each gene in the final list of significant genes. The q-values are influenced by the variability in the data set. This implies that changes in the entire data set composition will affect the d(i) distribution in the permuted control set and thus the q-value assigned to a given gene. To what extent this alters the final output when using SAM or subsequent post-hoc analysis has so far not been discussed. As SAM is arguably the most widely utilized method in the microarray analysis field, we felt it was critical to evaluate these considerations. The number of reported significant genes is influenced by the FC setting within SAM and the use of \"Present/Absent\" calls for data inclusion ---------------------------------------------------------------------------------------------------------------------------------------------- The first point to emphasise is that we are examining the effect of a variable FC setting, implemented in the Excel SAM, on a final gene list with a fixed 1.5 FC criteria. Our principle findings can be extended to any relevant final FC criteria but in the example we provide we have focused on a 1.5 FC criteria, for simplicity. When using the FC setting, implemented in the Excel addin, the researcher selects a proportion of the all genes on the chip, and this selection is also utilised for the permutated control data set, however the settings from the full data set SAM analysis are maintained (s\_0, and pi\_0). The FC hurdle setting can therefore change the resulting q-values as genes that pass a certain FC when the original sample categories are used, are less likely to generate high d(i) when the categories are permutated. To test the extent of this effect, we performed SAM at several different FC hurdles within Excel and scored the number of genes reported at the final 1.5 FC criteria. A difference in the number of genes reported as significant above the final FC would indicate that by using the Excel SAM FC restriction setting, during SAM, we change the outcome by altering the d(i) distribution of the permutation data set. We used three different biological data sets to assess how wide-spread any effects were. The first data set was a paired data set from a human skeletal muscle study which examined subjects before and after endurance training using the U95A-E platform (Affymetrix) \[[@B3]\]. The data was RMA normalized \[[@B4]-[@B6]\] and SAM was performed using different biological sample subgroups (groups were formed on the basis of a variety of physiological parameters) and chipset identity (A, B or D). Changing the FC setting in SAM (Excel) altered the final list of significant genes at the 1.5 FC criteria (Figure [1A--D](#F1){ref-type="fig"}). Importantly, the effect was not uniform across all conditions. For some samples a sequential increase in the FC hurdle in SAM correlated with an increased number of reported genes (Figure [1A--B](#F1){ref-type="fig"}) while under other conditions a very small change in FC had an apparently random impact on the composition of the significant gene list (e.g. Figure [1C--D](#F1){ref-type="fig"}). (The number of genes that passed each fold change criteria can be found in \[[Additional file 1](#S1){ref-type="supplementary-material"}\]). The second data set was derived from an in-vitromouse senescence study performed on U74Av2 chips \[[@B7]\] and normalized using RMA \[[@B4]-[@B6]\]. When comparing two of the time points during the induction of senescence we were unable to observe any effect of FC selection within SAM on the yield of significant genes (Figure [2](#F2){ref-type="fig"}). We believe this reflected the very low q-values obtained when originally using SAM, which in turn most likely reflects the low experimental variation that one can achieve using in-vitromodels. The third data set is derived from a study of the aging human brain, containing 21 samples split into two categories, young and old (unpaired data) \[[@B8]\], and normalised using D-chip \[[@B1]\]. When varying the FC hurdle within SAM we again saw a large impact on the number of genes reported as being significant, above the [1.5 FC criteria]{.underline} (a q-value threshold of 1% was considered statistically significant similarly to the original report) (Figure [3A--B](#F3){ref-type="fig"}). The \'significant\' gene list increased from 283 genes when using no FC hurdle (during SAM in Excel) to a maximum of 465 significant genes (67% more) when using a *1.51FC hurdle within Excel SAM (Figure [3A](#F3){ref-type="fig"}). A similar effect was observed when we examined the methods utilised by the original authors (20% \"Present\" call by D-chip). No FC hurdle generates 314 significant genes and a 1.51 FC generates 538 genes (Figure [3B](#F3){ref-type="fig"}). This is a 71% increase in \"significant\" genes associated with aging in human brain, despite the fact that the FC hurdle utilised in SAM was actually greater than the final FC of 1.5! Similar effects arise when other final FC criterions are used during SAM. For example if we only consider the significant genes that pass a final FC \>2 criteria in the aging brain study, then 5% more \>2 FC genes can be found if the FC hurdle within SAM is set to 2.0 compared to 1.5. Clearly, genes that are modulated to a large extent may be of high biological interest and thus our observation is important. Also, there are 73 more significant genes at the FC hurdle of 1.51 if the reduced data set is used compared the full data set (less genes in the data set gave more significant genes). This indicates that Present/Absent filtering also influences the outcome of the analysis, in this case increasing the number of genes defined as being modulated. Changing the FC setting in SAM changes the reported q-value ----------------------------------------------------------- The analysis presented above demonstrates that the q-value obtained for a specific gene depends on the FC hurdle applied during SAM in Excel. To monitor the q-values generated, for individual genes, we obtained the q-values during all SAM calculation using various FC hurdles for all genes that were reported as significant at the 1.51 FC setting (538 genes) in the brain aging study (Figure [4A](#F4){ref-type="fig"}). The highest q-value for a subset of the genes that passed the final fold change criteria was 3.6% when SAM was performed with a 1.0 FC hurdle. The same genes appear as being significant when the \"optimal\" (in the sense that these setting produced the largest significant gene list) Excel SAM FC hurdle setting 1.51 is used while the highest q-value reported was now only 0.97%. Running SAM using different FC settings can change data interpretation ---------------------------------------------------------------------- Often the main reason for carrying out a microarray experiment is to gain greater insight into the molecular processes that contribute to a complex biological phenotype. One standard method for carrying out such analysis is the use of gene-grouping, such as Gene ontology classification \[[@B9]\]. To address whether the biological interpretation may be influenced, we compared overrepresentation of classifications using EASE \[[@B10]\]. We selected the 314 genes and the 538 genes obtained in the brain aging study using the 1.0 and a 1.51 FC hurdles (described above) and calculated the overrepresented classifications. There was a considerable difference in the number of significantly over-represented classifications identified by EASE (figure [4B](#F4){ref-type="fig"}). 35 classifications were overrepresented in both gene lists, 8 were found in the GO analysis of the 1.0 FC list and 29 were unique to the 1.51 FC list. We feel it is unlikely that greater inclusion of randomly determined genes (i.e. false positives) would give rise to a significant increase in statistically enriched functional groups. This indicates that both the number of genes and the interpretation may be substantially altered by arbitrary data filtering, as exemplified by the use of the FC hurdle during the operation of Excel SAM or through the application of present-absent calls using e.g. D-chip. Conclusion ========== The average researcher is highly dependent on the use of \'standard procedures\' for their microarray analysis. An arbitrary filtering option in the Microsoft Excel Addin (we have called this the FC hurdle to distinguish from FC criteria -- which is the finalFC value selected by the researcher, to define a functionally significant change in gene expression) or data exclusion (e.g. present or absent call thresholds) can impact on a study, in a less than predictable manner. A clear problem also arises when SAM is utilised on different software platforms. In the R package Siggenes, no FC hurdle criteria can be made unless an additional programming function is implemented by the individual researcher. This is in contrast to the more widely used Microsoft Excel SAM Addinwhere the researcher can introduce a FC hurdle prior to the q-value calculations (while the SAM algorithm maintains the initial parameters from the analysis of the full data set). It is often unclear in the literature if SAM was performed in Excel or R; and if an FC hurdle was applied within SAM or if the fold change criteria was introduced after completing the operation of SAM, in EXCEL. Indeed, one might question how robust the SAM methodology is, if it is heavily dependent on both pre SAM data selection and within SAM (Excel) data filtering. However, one of the appreciated strengths of SAM is that the real data set is used to estimate experimental variation. One could also question whether it is valid to reduce the data set prior to using SAM. It would seem intuitive that much of the data being removed using a FC hurdle during SAM operation would be below an acceptable response level to be considered as being biologically relevant. However, the FC hurdle may also remove data that is essential for an accurate estimation of the experimental variation. The challenge would then be to remove \'genuine\' noise from non-expressing genes without removing genes that are genuinely expressed and necessary to approximate the data set variation. The effects will be pronounced in data sets demonstrating a large range of gene expression FC\'s or where significant inter experimental variation exists. It is clear, that investigators must be made aware that the impact of \'qualification criteria for inclusion\' and \'out-put filtering\' is less than predictable, when using SAM. Methods ======= Data sets --------- Three data sets were used. The first is a human skeletal muscle study, comparing eight subjects across the U95A-E Affymetrix chips before and after endurance training \[[@B3]\]. A subset of this data set can be derived by creating a low responder group based on their improvement upon training. This subset is referred to as the low responder group in figure [1](#F1){ref-type="fig"}. The data can be obtained from the authors (JAT). The second data set is a part of an mouse in-vitro senescence study performed on U74Av2 (Affymetrix) \[[@B7]\]. Two conditions were selected for comparison: EpiA1-ts58 cells at the basal condition and EpiA1-ts58 cells after 72 hours of senescence induction. This data set can be downloaded from the journal web site in a MAS5 (Affymetrix microarray suite 5) normalised format or will be distributed in RMA normalised format by the authors (OL). The third data set is a brain aging study performed on the U95A chip (Affymetrix) and normalised using D-chip \[[@B8]\]. The authors separated their data set into two groups: young (\<43 years old) and old (\>72 years old) and we used the same grouping in our analysis. The authors further used a data discrimination based on \"present\" calls given by D-chip (20% present in all samples). We used both the full data set and the 20% data set which is referred to as the reduced data set. The data set can be access at GEO, accession number GSE1572. SAM analysis ------------ SAM analysis was performed using the Microsoft Excel addin v1.21. The analysis was performed using different settings of the \"Fold change\". We also performed SAM analysis using the R (R-project.org) package \"siggenes\" as a comparison to the Excel addin without using the fold change setting with consistent results. Classification analysis ----------------------- To look for overrepresented classifications we used EASE \[[@B10]\]. We used all possible classifications and considered a classification as positively overrepresented if the EASE score was lower then 0.05. Abbreviations ============= SAM -- significance analysis of microarrays FDR -- false discovery rate FC -- fold change EF -- extraction factor MAS5 -- microarray suite 5 RMA -- robust multi array average GEO -- gene expression omnibus Authors\' contributions ======================= OL -- Hypothesis generation, practical work with all data sets and drafting of the manuscript CW -- Study design and revising the manuscript. JAT -- Hypothesis generation, practical work on the human muscle data section and drafting of the manuscript Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Genes that pass various FCs The additional file describes how many genes that passed various FCs from all three data sets ::: ::: {.caption} ###### Click here for file ::: Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### FC effects on the endurance training data set:SAM analysis was used at various fold changes studying the exercise data set while scoring genes with a q-value of \<0.05 and FC\>1.5. This was done to asses the effect of the fold change option in the SAM Excel addin on genes reported as significant at a higher fold change. The figure shows the number of scored genes using 4 different chip and sample combinations: (A)All eight subjects before and after training (totally 16 arrays) were compared in a paired analysis using the U95A chips. (B)All eight subjects before and after training (totally 16 arrays) were compared in a paired analysis using the U95B chips. (C)The reduced group consisting of low four low responders (totally 8 arrays) were compared in a paired analysis using the U95B chips. (D)The reduced group consisting of low four low responders (totally 8 arrays) were compared in a paired analysis using the U95D chips. ::: ![](1471-2105-6-129-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### FC effects on the senescence data set:SAM analysis was used at various fold changes using the senescence data set while scoring genes with a q-value of \<0.01 and FC\>1.5. A comparison between non-senescent cells and senescent cells was used (two replicates of the senescent cells and four replicates of the non-senescent cells). ::: ![](1471-2105-6-129-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### FC effects on the brain aging data set:SAM analysis at various fold changes using the brain aging data set (11 old samples and 10 young samples) while scoring genes with a q-value of \<0.01 and FC\>1.5. We used both the full data set and the reduced data set suggested by the authors. (A)The full data set. (B)The reduced data set (20% \"present\" calls by D-chip\"). ::: ![](1471-2105-6-129-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### FC effects individual q-values:q-values of all genes scored as significant in the brain aging study (reduced data set) at FC 1.51 at other FC settings. Genes only acquires discrete q-values and all 538 genes are shown, but overlap. (B) Running SAM with different FC settings changes the biological interpretation:*Venn diagram comparing the number of significantly overrepresented classifications (EASE score \<0.05) using the reduced brain aging data set analysed either with a 1.0 FC setting (314 genes) or a 1.5 FC setting (538 genes). ::: ![](1471-2105-6-129-4) :::	PubMed Central	2024-06-05T03:55:58.318758	2005-5-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173086/", "journal": "BMC Bioinformatics. 2005 May 29; 6:129", "authors": [ { "first": "Ola", "last": "Larsson" }, { "first": "Claes", "last": "Wahlestedt" }, { "first": "James A", "last": "Timmons" } ] }
PMC1173087	Background ========== Whenever experimental data is gathered to examine a process, researchers will try -- implicitly or explicitly -- to define a model describing the process. The purpose of building a model is to generalize the experimental data, allowing the prediction of new experimental outcomes, or to gain insights into the experimental process. In the biological sciences, models are often implicitly built and verbally formulated, such as \"the proteasome preferably cleaves after hydrophobic amino acids\". Such a model is easy to understand and often reflects all the knowledge that can be gathered from a few, difficult experiments. However, with the advent of high-throughput experiments to the biosciences, it has become feasible to generate more quantitative, mathematical models, based on large volumes of data. For the purpose of building a model, conducting experiments can be formally described as collecting pairs of experimental parameters x and experimental outcomes (measurements: y~meas~). Building a model is then equivalent to finding a function f(x) = y~pred~≈ y~meas~. Herein, we call the model function f a prediction tool, and the experimental observations used to generate it its training set T = (x, y~meas~). Developing a prediction tool is a two step process. First, a general prediction method capable of generating specific tools has to be chosen, such as a certain type of neural network, classification tree, hidden markov model or regression function. This choice is to some degree determined by the experimental data available. However, only a few prediction methods are clearly unsuited for a certain type or amount of data, leaving several potentially appropriate methods to choose from. In practice, the personal experience of a scientist often determines which prediction method he applies, as learning to apply a new method is often more costly than the benefits of a slightly better model. After a method is chosen, it is applied to the experimental data to generate a specific prediction tool. Several different terms are commonly used for this second step, such as \'supervised learning\', \'fitting the model\' or \'regression\'. Each method has its own formalism describing how this is done, but essentially a method is capable of generating a certain class of functions, of which one is chosen, that minimizes the difference between measured and predicted experimental outcomes for the training set T. Here we describe a computer program implementing the SMM prediction method, which can be applied to model the sequence specificity of quantifiable biological processes. In such experiments, the input parameter × corresponds to a sequences of amino-or nucleic acids, and the experimental outcome y~meas~is a real numbers measuring the process efficiency. The method utilizes training sets in which the examined sequences all have the same length. The SMM method was previously applied successfully to predictions of MHC binding \[[@B1]\], TAP transport \[[@B2]\] and proteasomal cleavage \[[@B3]\]. However, its software implementation was not made publicly available, because it relied on a commercial library with restrictive licensing terms. Also, applying the method to a new problem required manual changes in the source code, which is impracticable for an external user. Both of these issues are addressed in the software implementation made available here. In addition, this manuscript describes two specific features that we have found to be effective in generating high quality models and which can easily be utilized in other prediction methods, namely handling of bounded data and combinatorial peptide libraries. The SMM software is not meant for \'classical\' sequence analysis, which can be roughly defined as aligning related sequences in order to identify conserved residues or in order to generate classifiers which can identify additional related sequences. Rather, the typical application is the characterization of a sequence recognition event, such as the sequence specific cleavage efficiency of a protease. This characterization does not assume any evolutionary relationship between different recognized sequences. In terms of scientific fields, that makes the SMM software aimed at applications in biochemistry or molecular biology. Implementation ============== Code ---- The SMM algorithm is implemented in C++ code. Only standard libraries or freely available external libraries were used. The two external libraries are Tinyxml \[[@B4]\] to handle the XML input and output and the Gnu Scientific Library \[[@B5]\] for efficient vector and matrix operations. The source code has been compiled and tested using Visual C++ on a windows system and using g++ (1.5 or above) on a Debian and Suse Linux system. A windows executable is also available. The source code, documentation and examples are available as [additional files 1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"} of this manuscript, and on the project homepage at <http://www.mhc-pathway.net/smm>. On a standard 2.6 GHz Pentium 4 PC running windows XP, the creation of a prediction tool from a typical set of training data containing 300 8-mer peptides takes about 10 minutes. Input and output ---------------- The program expects an XML document as its standard input. A more sophisticated (graphical) user interface is not likely to be of great interest for the projected user community, most of which will probably call the executable file from a script or integrate the code into an existing program. The XML input document contains either training data to generate a new tool (Figure [1](#F1){ref-type="fig"}) or sequences for which a prediction should be made with a previously generated tool. The output of the program is again in XML format. For all types of input and output, XML schemas defining their exact structure are supplied. The schemas contain annotation for each input element, documenting their intended use. A simplified alternative input format exists for the most common application, namely the generation of a scoring matrix from a set of standard amino acid sequences. Several example input files are supplied with the program, which should make it easy for users to create similar input files with their own data. Core algorithm -------------- An amino acid can be encoded as a binary vector of length 20, with zeros at all positions except the one coding for the specific amino acid. Extending this notation, a peptide of length N can be encoded as an N\20 binary vector. The same conversion can be made for nucleic residues, resulting in N\4 length vectors. Any fixed length sequence can be converted into a fixed length binary vector following similar rules. The SMM algorithm expects such a vector as the experimental input parameter \'x\'. For a set of experiments, the vectors ^t^x (where ^t^x indicates the transposed vector x) can be stacked up resulting in a matrix H. An example for a set of nucleic sequences is shown in Figure [2](#F2){ref-type="fig"}. The matrix H is multiplied by a vector w which assigns a weight to each possible residue at each position in the sequence. This generates a vector of predictions y~pred~ H w = y~pred~ (1) From a given training set of sequences encoded in H with measured values y~meas~, the \'correct\' values for w can be found by minimizing the difference between the predicted values y~pred~and the measured values y~meas~according to a norm \\|\\| \\|\\|. To suppress the effect of noise in the experimental data, a second term is added to the minimization: \\|\\|H w - y~meas~\\|\\| + ^t^w Λ w → minimum, (2) in which Λ is a positive scalar or a diagonal matrix with positive entries. To better understand the effect of the Λ term, consider first minimizing (2) with Λ set to zero. In this case, the optimal entries for the weight vector w minimize the difference between predicted (y~pred~) and measured (y~meas~) values. Minimizing (2) with a non-zero value for Λ results in a shift of the optimal entries in w towards values closer to zero, especially for entries in w that do not significantly decrease the distance between predicted and measured values. This technique, generally called regularization, suppresses the effect of noise in the measured data y~meas~on the entries in the weight vector w. Refer to \[[@B6]\] or \[[@B7]\] for a general introduction to regularization. If the \\|\\| \\|\\| norm is simply the sum squared error (or L2 norm), equation (2) can be solved analytically for any given value of Λ to: w = (^t^HH + Λ)^-1t^H y~meas~ (3) The optimal value for Λ can then be determined by cross validation: The experimental data points corresponding to rows of (H, y~meas~) are randomly separated into training sets T~i~= (H, y~meas~)~train\ i~and blind sets (H, y~meas~)~blind\ i~. For a given Λ, equation (3) is used to calculate w~i~for each training set T~i~. These w~i~can then be used to make predictions for their corresponding blind sets. Summing the distances for all blind predictions gives the cross validated distance Φ: Φ(Λ) = Σ~i~\\|\\| H~blind\ i~w~i~(Λ) -y~blind\ i~\\|\\| (4) Minimizing Φ as a function of Λ therefore corresponds to minimizing the cross validated distance by \'damping\' the influence of coefficients in w which are susceptible to noise. As the resulting optimal value for Λ and the corresponding w~i~are somewhat influenced by the split into training and blind sets, the same procedure is repeated several times with different splits, which is called bagging \[[@B8]\]. The final w is generated by averaging over all optimal w~i~generated in the cross validation and bagging repeats. Results and discussion ====================== The following sections describe specific properties of the SMM method, and are meant to serve as a guideline when and how to apply it. Additional data validating the SMM algorithm and comparing it with other prediction methods can be found in previous publications \[[@B1]-[@B3]\]. Linear model ------------ If no pair coefficients are incorporated, the output vector w of the SMM method is a standard \'scoring matrix\', which quantifies the contribution of each residue at each position in the input sequence to the prediction. Such a matrix is easy to interpret and analyze without requiring any additional software or expert knowledge of how the matrix was generated, which is especially important when communicating results to experimentalists. Several methods predicting peptide binding to MHC molecules take this approach, e.g. \[[@B9]-[@B12]\], and a comparative study showed that simple statistical methods to generate matrices can perform better than more complex artificial neural networks if the amount of data is limited \[[@B13]\]. Using such a linear model implicitly assumes that the influence of residues at each position in the sequence on the measured value can be considered independent and additive. This has to be a reasonable first approximation in order to successfully apply the SMM method, even if pair coefficients are incorporated. This is the main difference to general learning algorithms such as neural networks, which can in principle model any functional relationship between sequences and measurements. Quantitative data ----------------- The experimental measurements that serve as input to the SMM method and the predicted output are quantitative, not binary. For example, in the case of peptide binding to MHC molecules, IC50 values quantifying binding affinities are used, and not a classification into binding and non-binding peptides. If different representations of the quantitative data are possible, such as either IC50 or log(IC50) values, a representation should be chosen in which the y~meas~values approximately follow a normal distribution. Otherwise the SMM predictions, which are sums of independent contributions and therefore roughly normally distributed themselves, will not be able to fit the experimental data well. In the case of binding affinities, this means log(IC50) values should be used, as IC50 values themselves are usually log normal distributed. Noisy data ---------- Experimental measurements inevitably contain noise. This will cause problems when building models that take the measured values to be exact. Accordingly, the SMM method incorporates a regularization parameter Λ (equation 2), which corresponds to preferring a simpler solution with \'smooth\' values for w to one that exactly reproduces observations. In the first SMM applications \[[@B1],[@B2]\], Λ was a scalar, in which case this approach is called Ridge regression or zero-order regularization. Choosing a scalar value implicitly assumes that the level of signal to noise is roughly the same at each position in the input sequence. In the current version, Λ can also be chosen as sequence position dependent, which is sometimes called local Ridge regression. As shown for an example in Figure [3](#F3){ref-type="fig"}, this makes Λ into a diagonal matrix in which all Λ~i~values belonging to residues at the same sequence position are set to the same value. For a sequence of length N, there are N different Λ~i~values. For a number of training sets containing peptide binding data to MHC molecules, we compared the prediction performance achieved using a position specific matrix Λ~i~to a scalar Λ. The position specific regularization nearly always resulted in better predictions. The difference was especially large if the influence of different sequence positions varied greatly (data not shown). Bounded data ------------ Any experimental technique generates measured values contained within a finite range. For example, in many biological experiments a \"zero\" measurement usually means that the actual value is below the experimental resolution, not that the actual value is 0. Similarly, very large values beyond the expected sensitivity limit are no longer quantitatively accurate. These data points at the upper or lower boundaries of the sensitivity range do not convey the same information as quantified values, but they still do contain information. In the case of MHC binding data available to us, approximately 20% of peptides fall in this category. The SMM method is, to the best of our knowledge, the only method designed to extract information from such boundary values. This is done by means of the novel L2~\<\>~norm, illustrated in Table [1](#T1){ref-type="table"}. For example, if an experimental measurement corresponds to an upper boundary y~meas~\> z, and the predicted value is greater than z, then the distance between y~meas~and y~pred~is zero. This norm has the useful property that any analytical solution according to the L2 norm can be converted into a solution according to the L2~\<\>~norm through an iterative process: First, all measurements including boundary values are treated as normal values, and the solution using the L2 norm is found. In a second step, for each y~pred~, y~meas~value pair for which the L2~\<\>~norm would be zero, the y~meas~value is set to its corresponding y~pred~value. For these y~meas~\* values, the distance L2(y~pred~, y~meas~\) = L2~\<\>~(y~pred~, y~meas~). These y~meas~\ values are then used to solve again according to the L2 norm. This process is repeated until the y~meas~\* values no longer change, as illustrated in Figure [4](#F4){ref-type="fig"}. Randomized peptide library data ------------------------------- As stated before, the experimental data used as input for the SMM method consists of same length amino-or nucleic acid sequences associated with a quantitative measurement. When designing an experiment, the selection of sequences to test can introduce bias into the training data, for example by over-or under-representing residues at specific sequence positions. One way to avoid this is the use of randomized peptide libraries, also known as positional scanning combinatorial peptide libraries, which are mixtures of peptides of the same length. In a given library, all peptides have one residue in common at a fixed position and a random mixture of amino acids at all other positions. For example, the library XAXXXXXX contains 8-mer peptides with a common Alanine at position 2. Such libraries can be used to directly measure the influence of their common residue, by comparing their measured process efficiency to that of the completely randomized library XXXXXXXX. In the case of 8-mer peptides, 160 library experiments are sufficient to characterize the influence of each residue at each position. The results of such a complete scan can be summarized in a scoring matrix. This approach has been used successfully in many different experimental systems \[[@B14]-[@B17]\] A novel feature of the SMM method is that it can combine data from these two sources. When the SMM algorithm is given experimental data from individual peptides and from a randomized library summarized in a scoring matrix mat~lib~, it simply subtracts the values predicted by mat~lib~from each individual peptide measurement. These y\'~meas~= y~meas~- y~pred,lib~values are then used to generate a second scoring matrix mat\'. The final SMM scoring matrix is simply the sum of the two: mat~combined~= mat\' + mat~lib~. Figure [5](#F5){ref-type="fig"} compares the performance of this combined approach to that of a prediction based on peptide or library experiments alone. If enough peptide data is present (roughly the same number as matrix parameters), the combined prediction is better than that of the library matrix alone. At all data points, the combined prediction is better than that using the peptides alone. Importantly, this simple strategy of subtracting library predictions can be used in combination with any prediction method, and is likely to generate similar results, as it effectively increases the training set size. To visualize the prediction quality associated with the distances reported in Figure [5](#F5){ref-type="fig"}, Figure [6](#F6){ref-type="fig"} depicts a scatter plot of the predicted and measured binding affinity for individual peptides corresponding to the data point with the lowest distance in Figure [5](#F5){ref-type="fig"}. Introducing pair coefficients ----------------------------- Pair coefficients quantify the contribution of a pair of residues to the measured value that deviates from the sum of their individual contributions. The form of equation (1) remains unchanged if pair coefficients are introduced in the same binary notations as the individual coefficients. Figure [2](#F2){ref-type="fig"} gives an example how a set of nucleic sequences is transformed into a matrix H, if two such pair coefficients are taken into account. Note that the number of possible pair coefficients is very large. For a sequence of three nucleic acids, there are already (3\4) \ (2\4) / 2 = 48 pair coefficients. For a 9-mer peptide, (9\20) \* (8 \* 20) / 2 = 14400 pair coefficients exist. To the best of our knowledge, only the SMM method and the additive method \[[@B18]\] explicitly quantify the influence of pair coefficients. Since most training sets contain only a few hundred measurements, determination of the exact values of all pair coefficients is not feasible. To overcome this difficulty, the SMM algorithm limits the number of pair coefficients to be determined. First, only coefficients for which sufficient training data exists are taken into account. As a rule of thumb, 5 sequences containing the same pair of residues at the same positions have to be present in the training set for a pair coefficient to be considered. In a second filtering step, only pair coefficients for which the information in the training set is reasonably consistent are retained. In the previous SMM version, this used to be determined by multiple fitting of pair coefficients and discarding those for which a sign change was observed. In the current version, a much faster approach is used. First, a scoring matrix is calculated for the training set without any pair coefficients. Then, for each pair coefficient the predicted and measured values for the sequences containing it are compared. Only if a large enough majority (\>60%) of measured values are above or below the matrix based predictions is the pair coefficient retained. The remaining pair coefficients are determined in complete analogy to the scoring matrix itself, but with a scalar Λ value. We tested the effect of incorporating pair coefficients on prediction quality compared to using a scoring matrix alone for a number of training sets containing peptide to MHC binding data. The pair coefficients showed a consistent positive contribution for large training sets, which comprise more measurements than 1.5 times the number of scoring matrix coefficients. However, the improvement is rather small, as reported before in \[[@B1]\]. This makes it reasonable to ignore pair coefficients if the simplicity of a scoring matrix is more valuable than a small improvement in prediction quality. If higher order sequence interactions such as those described by pair coefficients are expected to be the dominant influence on experimental outcomes, other prediction methods may be better suited than the SMM method. For example, by choosing a different sequence representation than the binary vectors, the information in the training set can be generalized, thereby effectively reducing the degrees of freedom in the input parameters \[[@B19]\]. This allows applying general higher order learning algorithms such as artificial neural networks even with limited input data. Conclusion ========== The SMM method generates quantitative models of the sequence specificity of biological processes, which in turn can be used to understand and predict these processes. It has previously been shown to perform very well compared to other prediction methods and tools for three specific types of experimental data \[[@B1]-[@B3]\]. However, it is difficult to generalize a comparison between different methods, due to two main problems. First, the training data sets utilized in different studies are often not available, so that when comparing tools generated by different methods it is often unclear when good performance is due to a superior method or a better (larger) set of training data. Second, generating tools from the same training set can be difficult, because publications that make the tools available, often only describe the basic principle of the method used. To overcome this second obstacle, we herein presented a computer program implementing the SMM method. Significant effort was devoted to ensuring that the program is robust, documented, cross platform compatible and generates reasonable output without requiring additional parameters. Also, any commercial libraries previously utilized were removed to allow free distribution of the code. This will permit any interested user to apply the SMM method with reasonable effort, allowing for the most important validation: application to scientific practice. Finally, we believe that two strategies demonstrated in this manuscript will be valuable in combination with other prediction methods as well. First, our strategy for the inclusion of experimental data gathered with randomized peptide libraries can be directly transferred to any prediction method. When other experimental data is limited and data from a combinatorial library is available, this should always have a positive effect on prediction quality. Secondly, the L2~\<\>~norm can be applied as an error function for other prediction methods. This will increase the amount of training data effectively available to prediction methods requiring quantitative input, by enabling them to handle experimental boundary values. Availability and requirements ============================= Project name: smm Project home page: <http://www.mhc-pathway.net/smm> Operating system(s): Platform independent Programming language: C++ Other requirements: The Gnu Scientific Library (GSL) has to be installed License: None for the smm code, but GSL requires the GNU GPL license Any restrictions to use by non-academics: None Authors\' contributions ======================= BP conceived of this study and implemented the software, AS participated in the design of the software and helped to draft the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A .zip file containing the SMM source code, documentation and examples for windows systems. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A .tar.gz file containing the SMM source code, documentation and examples for linux systems. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The authors want to acknowledge the help of Emrah Kostem in adopting the source code to g++ and John Sidney for critically reading the manuscript. This work was supported by the National Institutes of Health Contract HHSN26620040006C Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Input training data.The \<TrainingData\> element consists of a series of \<DataPoints\> (2). Each contains a sequence and a measurement value. The characters allowed in \<Sequence\> are specified in \<Alphabet\> (1), and the number of characters has to correspond to \<SequenceLength\> (1). In this example, \<Alphabet\> and \<SequenceLength\> specify 8-mer peptides in single letter amino acid code. Each measurement can optionally be associated with a threshold (3) that can either be \<Greater\> or \<Lesser\>, signaling that the measurement corresponds to an upper or lower boundary of measurable values. ::: ![](1471-2105-6-132-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Converting sequences into matrices.Input sequences of three nucleic acids each are converted to rows of a matrix H. The first column of each row is set to 1, which serves as a constant offset added to each prediction. Columns A1 to T1 contain a binary representation of the first residue in the sequence, in which all columns are set to zero except the one corresponding to the residue. The same is repeated for the second and third residue in the sequence in columns A2 to T2 and A3 to T3. The two last columns G1A3 and A2A3 contain pair coefficients explained at the end of the results section. They are set to one if the two specified residues are present in the input sequence at the two specified positions and zero otherwise. Multiplying matrix H with the weight vector w results in a vector y~pred~of predicted values for the sequences. Rows A1 to T3 of vector w are commonly written as a \'scoring matrix\' which quantifies the contribution of each possible residue at each position to the prediction. Rows G1A3 and A2A3 of vector w quantify the impact of the pair coefficients. ::: ![](1471-2105-6-132-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Sequence position dependent regularization.Example for the regularization term ^t^w Λ w in equation (2). The weight vector w corresponds to a scoring matrix for three nucleic acids as in Figure 2, but without pair coefficients. The diagonal matrix Λ has three different values Λ1, Λ2 and Λ3 effecting values in vector w corresponding to sequence positions 1, 2 and 3. There is no regularization penalty on the \'Offset\' value. ::: ![](1471-2105-6-132-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Iterative model fitting using the L2~\<\>~norm.In this example, the model is a linear function which is fitted to a set of paired values (x, y~meas~). For two of the x values (x = 3 and x = 5), the measured values are thresholds (Greater 3). Fitting a linear function to paired values according to the L2 norm corresponds to the standard linear regression. Adepicts the model fit (straight line) to the measured values (black boxes), ignoring any thresholds. For x = 5, the model value y~pred~taken from the regression curve is 3.4, above the measured threshold value 3. Therefore, in the next iteration the y~meas~\* value is set to the model value 3.4. Bshows the new linear regression with the adjusted y~meas~\* values. This procedure is repeated until the y~meas~\* values no longer change (8 iterations, panel C). ::: ![](1471-2105-6-132-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Combining peptide and library data improves prediction quality.A set of 449 9-mer peptides with measured affinities for TAP taken from \[20\] was split into 5 blind sets. For each of these blind sets, predictions where made from different size subsets of the remaining peptides. The x-axis depicts the number of peptides in these subsets used for generating predictions using either the peptides alone (circles) or in combination with data from a combinatorial peptide library (squares). The dashed line displays the prediction of the library alone, which was taken from \[15\]. The y-axis depicts the L2~\<\>~distance of the predictions for the combined 5 blind sets. ::: ![](1471-2105-6-132-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Visualization of prediction quality.Scatter plot of predicted vs. measured affinity for peptide binding to TAP. The depicted prediction corresponds to the data point in figure 5 with the lowest cross-validated distance, in which 350 peptides and the peptide library were used for training. ::: ![](1471-2105-6-132-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The L2~\<\>~norm ::: Measurement L2~\<\>~distance ------------------------------------- ---------------------- ---------------------- Quantitative (no threshold) (y~pred~-y~meas~)^2^ (y~pred~-y~meas~)^2^ Upper boundary (threshold: greater) 0 (y~pred~-y~meas~)^2^ Lower boundary (threshold: lesser) (y~pred~-y~meas~)^2^ 0 :::	PubMed Central	2024-06-05T03:55:58.320608	2005-5-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173087/", "journal": "BMC Bioinformatics. 2005 May 31; 6:132", "authors": [ { "first": "Bjoern", "last": "Peters" }, { "first": "Alessandro", "last": "Sette" } ] }
PMC1173088	Background ========== Models for codon substitutions are used in computational biology for a wide range of applications such as reconstructing ancestral DNA sequences, determining Ka/Ks ratios to identify periods of adaptive evolution and aligning coding DNA. Methods for estimating mutation matrices from observed substitutions in sequence alignments of proteins were established by Dayhoff \[[@B1]\]. These matrices contain the probabilities of amino acid mutations for a given period of evolution and have long been used for scoring protein sequence alignments, evolutionary studies and homology searches. More than a decade ago, when large-scale protein databases became established, several amino acid substitution matrices based on observed mutation counts in protein alignments were constructed \[[@B2]-[@B4]\], replacing the original Dayhoff matrices that were based on relatively few alignments. However, to describe substitutions at the codon level, parameterized models have been developed \[[@B5],[@B6]\] and are widely used in the study of molecular evolution. In the same way that the growth of protein databases allowed refined construction of amino acid substitution matrices, the recent increase of nucleotide sequence data made it possible to apply these methods at the codon level. The matrices presented here were constructed using the approach of Gonnet \[[@B4]\]. This implies that pairwise alignments using full dynamic programming \[[@B7],[@B8]\] were used in order to count the observed transitions between codons. The sequence data was taken from the complete vertebrate genome databases of ENSEMBL\[[@B9]\]. Results ======= The additional files contain the 64 × 64 matrices presented here. Various aspects of the matrices will be discussed in this section. We present the matrix with the exact counts of the observed substitutions ([Additional file 1](#S1){ref-type="supplementary-material"}), the matrix containing the substitution probabilities derived therefrom ([Additional file 2](#S2){ref-type="supplementary-material"}) and a matrix containing similarity scores for all possible substitutions ([Additional file 3](#S3){ref-type="supplementary-material"}). Substitution counts ------------------- The 17,502 alignments that have been used to construct the matrix presented here contained 8.3 million aligned codon pairs. From each of these aligned pairs, the number of each of the 3730 (61 × 61 + 3 × 3) possible substitutions were counted. The difference between the numbers of frequent and rare substitution is high with the most frequent substitution (the GAG identity) being observed 153,040 times, and the rarest substitution, between TGG and GAG, being counted only 45 times, about 3400 times less often. To estimate the precision of the count of the rarest substitution, a binomial distribution of the counts can be assumed. The substitution with minimal count c~min~occurs with probability p= 45/(8.3·10^6^) = 5.4·10^-6^. The variance of a binomial distribution is σ^2^= N(1 - p)pand thus for very small p, the variance of c~min~is almost equal to c~min~= 45 and the standard deviation σis 6.7. This means that although a very large amount of data is used to construct the matrix, it is just enough to produce codon counts that are of a tolerable accuracy for rare transitions. Only a further increase of high-quality genomic data will allow the clustering of the data into specific subsets. These possibilities will be discussed below. Substitution probabilities -------------------------- The mutation matrix Mconstructed from the counts contains the substitution probabilities for the individual codons. Entry M~i,j~gives the probability that codon jmutates to codon i. (As a consequence, each column of Msums to 1). A convenient measure to express the amount of mutations in a matrix is the percentage identity ![](1471-2105-6-134-i1.gif) with f~i~being the natural frequencies of the codons. For the matrix reported here, pis .35, meaning that in the alignments used, 65% of the codons have undergone substitution (to any other codon, thereby involving up to three nucleotide changes). It is also possible to calculate the percentage of identical amino acids resulting from this matrix: ![](1471-2105-6-134-i2.gif) In the second sum, igoes over all codons that code for the same amino acid as jdoes. The result for p~AA~is .69, therefore 31% of the amino acids are expected to mutate. This allows the determination of the relationship between the codon substitution matrices and amino acid substitution matrices, because the amino acid PAM distance can be derived from the percentage of amino acid identity. Analogously to the definition of 1 PAM, 1 Codon-PAM can be defined as the distance at which 1% of the codons undergo substitution. Again, a codon substitution can involve up to three nucleotide base changes. The substitution matrix for any distance dis approximated by raising the 1 CodonPAM matrix to the power of d. The relationship between CodonPAM, PAM and f2 is shown in Figure [2](#F2){ref-type="fig"}. It shows that amino acid PAM increases almost linearly with CodonPAM. The curve is slightly steeper for the low distances and flattens with increasing distance. The amount of synonymous substitution decays from 1 to .51 (f2 being a measure of synonymous mutation described in the Methods section). Mutation scores --------------- Since the substitution probabilities are influenced by the codon frequencies, it is not possible to see directly which substitutions occur more than expected and which occur less. This issue is corrected in the scoring matrix D, where entry D~i,j~expresses how much more likely it is that codons iand jwere derived from a common ancestral codon compared to a random pairing of them. A higher score for a transition means that this transition is indeed more likely than one with a lower score. The scoring matrix is symmetric, i.e. the transition from codon ito jhas the same score as the transition from jto i. Table [1](#T1){ref-type="table"} displays average scores for different categories of substitutions. It confirms the fact that synonymous substitution scores are generally higher than non-synonymous scores. But it can also be seen that as more nucleotides change, the scores become lower. Synonymous substitutions in which all three bases change, have lower scores than non-synonymous substitutions with only one base change. Discussion ========== Synonymous mutations -------------------- It has been observed that different genes have different Ka/Ks ratios and therefore the fraction of synonymous substitutions will differ between different gene pairs having a certain PAM distance. This is because there are no strong selective constraints on synonymous substitutions and therefore the number of these substitutions accumulates in a clock-like manner \[[@B10]\] while the number of nonsynonymous substitutions is governed by functional constraints. Increasing amounts of genomic data would allow the construction and comparison of matrices from alignments with differing amounts of synonymous and non-synonymous substitutions, representing a two-dimensional array of matrices, where one dimension is the evolutionary distance and the other corresponds to the amount of synonymous change. Unfortunately, the current size of the nucleotide databases does not yet allow such a clustering of the available data. Instead, the alignments selected to construct the matrices were filtered to fall within a window of synonymous mutations, thereby excluding the most extreme values. (see the Methods section for details). Figure [1](#F1){ref-type="fig"} shows the distribution of the alignments\' f2 values. Range of applicability ---------------------- One possible application of scoring matrices is protein and coding DNA alignment. In order to compare alignments based on amino acid substitution matrices and the codon matrices presented here, the likelihood scores are compared. Since these scores express the probability ratios of the two sequences having evolved from a common ancestor to them being aligned by random chance, they serve as a confidence measure of an alignment. The higher the score, the higher the likelihood that the alignment is by reasons of ancestry than by random chance. As the likelihood scores serve as an indicator of alignment quality, orthologous sequences for species pairs of various distance and classes were used to determine when codon matrix based alignments produced higher scores than amino-acid PAM matrices. Table [2](#T2){ref-type="table"} displays for several pairs of species, the number of orthologs used to perform the alignment analysis, the average PAM distance between these orthologs (found by selecting the highest-scoring PAM matrix) and the average ratio of codon based scores to amino acid based scores. A number greater than 1 means that on an average, the codon based scores were higher. The result is that for closely related species, the codon based scores are always higher, but the more distant two species are, the better the performance of the amino acid based alignments. An interesting point is that although codon mutations in different sets of species were found to be significantly different (χ^2^tests, data not shown here), the above finding holds not only for the vertebrates, from which the matrices were constructed, but also for the invertebrates, yeasts and even bacteria. From the results in Table [2](#T2){ref-type="table"}, a PAM distance smaller than 50 would favor the use of codon substitution matrices instead of amino acid based matrices. Conclusion ========== Because codon substitution matrices are substantially bigger than amino acid matrices and also because some of the substitutions are extremely rare compared to the most frequent ones, large amounts of genomic data are necessary to model the transitions accurately. The 17,502 alignments used here produce enough aligned codons to fulfill this criterium, but do not allow further clustering of the data set in order to create more specific matrices. The codon substitution matrix presented here is to our knowledge the first based entirely on empirical data and can serve in many fields of computational biology. We have found that at long distances, when the synonymous mutations have reached saturation, amino acid matrices are better suited for alignments and long-distance homology searching. Codon mutation matrices have advantages for alignment up to medium evolutionary distances and for usages that require DNA such as ancestral reconstruction of DNA sequences and the calculation of Ka/Ks ratios. Methods ======= The basic methods to create scoring matrices are well established. The construction of codon substitution matrices is analogous to that of amino acid transition matrices. The main difference lies in the fact that codon matrices are much larger (4096 elements (64 × 64) instead of 400 elements (20 × 20)). In addition, the stop codons need special consideration. Substitutions between stop codons and sense codons are assumed to be very rare because the effect of such substitutions on the function of the protein would probably be very serious thus the chance of acceptance is very small and usually limited to the 3\' end of the nucleotide sequence. This makes it almost impossible to observe such events. Therefore, these substitutions are not included in the matrices presented here. Substitutions between stop codons, however, are counted and thus also contained in the matrices. This means that the 64 × 64 matrices are block diagonal composed of a 61 × 61 matrix for the coding codons and a 3 × 3 matrix describing substitutions between stop codons. Using orthologs --------------- The matrices are constructed from pairwise alignments of orthologous sequences from five vertebrates -- human (Homo sapiens), mouse (Mus musculus), chicken (Gallus gallus), frog (Xenopus tropicalis) and zebrafish (Brachydanio rerio). The complete genome databases from ENSEMBL\[[@B9]\] were used for this purpose. Using only orthologs has the advantage that no gene is overrepresented in the data set. This is because a particular gene can have many paralogous genes in a genome, but we allowed at most one ortholog per other genome. Circular tours -------------- When counting the substitutions in alignments from all pairs of species, substitutions that occurred early in the tree are counted more often than those that happened later, because paths between two species include the branches near the root more often than those near the leaves. This bias can be prevented by using only species pairs along a circular tour. This way every branch of the tree (and therefore every substitution that ever happened in the history of the genes) examined, is counted at most twice. Concretely, this means that only the orthologs between human and mouse (3107 pairs), mouse and chicken (3691 pairs), chicken and frog (3671 pairs), frog and fish (3441 pairs) and fish and human (3592 pairs) are counted, resulting in 17,502 alignments. Counting substitutions ---------------------- These alignments must fulfill several criteria: 1) they should all be of similar evolutionary distance because substitution probability depends on evolutionary distance. A trade-off exists for the acceptable range of distances as including a broad range of distances, increases the amount of data but at the same time blurs the distance specific information. 2) The alignments must be of a distance that is high enough to allow the observation of rarer substitutions. 3) There must be enough alignments to have statistically significant data for the rare substitutions. After some observations performed on a subset of the data, a distance range of 25 to 60 PAM (57% to 78% identity) of the protein alignments was found to best satisfy these criteria. Another selection criterium was based on the amount of synonymous substitutions between the sequences to eliminate saturation effects from the observed synonymous substitutions. One way to estimate the amount of synonymous substitution is f2, the percentage of conserved synonymous codons at two-fold redundant amino acid sites. Two identical sequences have an f2 value of 1 and it will decay to a value near .5 for increasing amounts of substitutions. An f2 range between .50 and .95 was found to exclude the most extreme cases of synonymous substitutions while leaving enough alignments to fill the matrices. Figure [1](#F1){ref-type="fig"} shows the distribution of f2 values for all alignments within the PAM range of 25 to 60. The sequence pairs with f2 values between .50 and .95 are shown in green and were used to construct the matrices, while the alignments corresponding to the red bins were discarded. Full dynamic programming \[[@B4],[@B7],[@B8]\] was employed in order to construct the alignments. The DNA alignment was obtained by mapping the coding DNA to the aligned proteins. Directly aligning the DNA was not yet possible, because no a priori knowledge about codon similarities was assumed. However, in the refinement steps (see further below), the DNA itself was aligned using the codon substitution matrices from the previous refinement round. Once the sequences were aligned, the actual substitution matrices could be computed. This included counting the observed codon substitutions in the collected alignments and storing them in a 64 × 64 count matrix C. Since the direction of a substitution is not known, for each observed substitution between codons iand j, C~i,j~as well as C~j,i~is increased by 1/2. Insertion or deletion sites were ignored since they provide no information about actual substitutions. From the count matrix, the mutation matrix was derived according to equation 3: ![](1471-2105-6-134-i3.gif) ### Calculating similarity scores The transition scores for a given substitution matrix express the relative probabilities of two codons originating from a common ancestor compared to the probability of them being paired by random chance. The logarithm is taken to make the scores additive, thereby speeding up the computation of alignment scores. Dis calculated as ![](1471-2105-6-134-i4.gif) where f~i~is the frequency of codon iin the observed data set. The factor 10 is used for purely historical reasons. In an iterative process, the sequence pairs were aligned again with dynamic programming \[[@B7],[@B8]\] but this time directly on the codon sequences using the substitution matrices obtained before. Exponentiation of the mutation matrix was used to approximate matrices for different evolutionary distances \[[@B12]\], allowing a maximum-likelihood estimation of the best-fitting matrix to align the sequences. From these new alignments new mutation matrices and finally scoring matrices were constructed in the way described above, until after six iterations a sufficient convergence of the matrix was reached. Authors\' contributions ======================= GMC initiated and guided this project and contributed to the programming. AS did the majority of the programming and analysis. Some of the work was based on previous research by GHG. All work was supervised by GHG. AS and GMC drafted the manuscript which was approved by GHG. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 The 64 × 64 matrix containing the observed substitution counts as described in the results section. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 The 64 × 64 matrix containing the substitution probabilities. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 The 64 × 64 matrix of the transition scores. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ GMC wishes to acknowledge Michael T. Hallett, Jan Norberg, and Xianghong Zhou for starting this project. The authors wish to thank Peter von Rohr, Markus Friberg, Daniel Margadant, Christophe Dessimoz and the anonymous reviewers for careful reading and helpful suggestions. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### f2 histogram.Histogram of the f2 values from the 17,502 alignments used to construct the matrix. ::: ![](1471-2105-6-134-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### CodonPAM vs PAM and f2 ::: ![](1471-2105-6-134-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Analysis of the scores. Average scores for different categories of substitutions. The stop codons are excluded from this analysis. ::: n Substitutions Avg. Score ---------------------------- ----------------- ------------ Identity 61 12.9 Synonymous: all 87 8.7 Synonymous: 1 base change 67 10.0 Synonymous: 2 base changes 14 6.6 Synonymous: 3 base changes 6 -1.7 Non-syn.: all 1743 -7.3 Non-syn.: 1 base change 196 -1.3 Non-syn.: 2 base changes 770 -5.9 Non-syn.: 3 base changes 777 -10.3 All Substitutions 1891 -5.9 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Range of applicability. Ratios of likelihood scores for amino acid and codon based alignments for orthologs between several species pairs, where Nis the number of orthologs used. ::: N Avg. PAM Scores ratio ---------------------------- ------- ---------- -------------- Homo sapiens vs. Mus musculus 14655 17.4 1.150 vs. Gallus gallus 9272 29.3 1.060 vs. X. tropicalis 9953 39.1 1.026 vs. B. rerio 7507 43.7 1.013 Drosophila melanogaster vs. A. gambiae 5059 57.3 .995 vs. H. sapiens 3371 77.5 .959 vs. C. elegans 2156 88.8 .945 Saccharomyces cerevisiae vs. C. glabrata 3467 52.7 1.002 vs. A. gossypii 2909 61.4 .978 vs. H. sapiens 1187 94.1 .931 Escherichia coli vs. E. colistrain O6 3156 2.0 1.323 vs. Salmonella typhi 2557 14.2 1.067 vs. P. aeruginosa 1234 71.6 .980 vs. B. japonicum 765 90.2 .959 :::	PubMed Central	2024-06-05T03:55:58.323706	2005-6-1	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173088/", "journal": "BMC Bioinformatics. 2005 Jun 1; 6:134", "authors": [ { "first": "Adrian", "last": "Schneider" }, { "first": "Gina M", "last": "Cannarozzi" }, { "first": "Gaston H", "last": "Gonnet" } ] }
PMC1173089	Text mining? \...\...\...\...\...\....Why bury it first and then mine it again? =============================================================================== Recently, Sir Tim Berners-Lee, the inventor of the Web, said: \'Life sciences are particularly suitable for pioneering the Semantic Web. For example, within drug discovery, many databases and information systems used by drug researchers are already in, or are ready to be transformed to, machine-readable formats\' \[[@B1]\]. Well, we need our breed of optimists to drive things, and he is obviously one of us. Too much to read ================ Bioinformatics increasingly consists of computer-aided meta-analysis of dispersed articles and database records to assist researchers in the interpretation of massive datasets. Epidemiological studies and high throughput technologies such as microarrays nowadays often lead to sets of potentially relevant papers being identified that surpass human capability for reading, interpretation and synthesis. Aggregating information from many records, followed by logical association of the concepts represented in the full dataset is what I generally refer to here as meta-analysis. Unfortunately, most of the information in the current information sources is not easily accessible, nor, due to ambiguity of the description of concepts in textual format, is it easily readable by computer programs. Semantics: a crucial addition to text ===================================== If all words had only one possible meaning, computers would be perfectly able to analyse texts. In reality however, words, terms and phrases in text are highly ambiguous. Knowledgeable people have few problems with these ambiguities when they read, because they use context to disambiguate \'on the fly\'. Even when fed a lot of semantically sound background information, however, computers currently lag far behind humans in their ability to interpret natural language. Therefore, proper semantic tagging of concepts in texts is crucial to make Computational Biology truly viable. Electronic Publishing has so far only scratched the surface of what is needed. Open Access publication shows great potential, andis essential for effective information mining, but it will not achieve its full potential if information continues to be buried in plain text. Having true semantic mark up combined with open access for mining is an urgent need to make possible a computational approach to life sciences. Databases and pathway tools are not enough ========================================== Of course, we have our curated databases, increasingly equipped with fancy analysis and visualisation tools. These databases hold established knowledge on molecules, their interactions and pathways. These databases are definitely useful tools to rapidly get a rough view on \'what may be switched on or involved\'. However, the inherent restrictions and downsides of such databases include the great efforts that are needed to keep the error level acceptably low and to keep them up-to-date. Furthermore, biological complexity is more than a composition of multiple, man-imagined pathways. Thus free text continues to be a crucial source of cutting edge information for scientists A further reason for the continued value of free text is that pre-digested information in databases only contains explicit knowledgeand will therefore at best cough up \'what somebody already knew\', although I as an individual may not be aware of it. Real serendipitous finding of new things and the association between concepts beyond direct co-occurrence is not supported by most existing tools. However, text is a nightmare for computers ========================================== Unfortunately, free text records are a nightmare of ambiguity. Synonyms and homonyms riddle the records and, in particular, gene and protein names/symbols appear to be impossible to resolve with complete accuracy, based on their textual expression. In an ideal world, scientists would mention formal identifiers from recognized data bases such as EntrezGene or SwissProt in the text, rather than their favourite synonym of the molecule. With traditional search-based text retrieval tools, this problem was little more than a recurrent nuisance, and current traditional search engine providers probably couldn\'t care less, as their users are satisfied when they \'find what they want on page one\'. But now that computational meta-analysis of textual records is increasingly needed, and the underlying datasets may often consist of tens of thousands of papers, it becomes impossible to manually weed out problems relating to synonyms and homonyms. So, like it or not, we have to face the challenge of semantic enrichment ======================================================================== Before any meta-analysis algorithm can be meaningfully applied, semantic analysis and tagging of the underlying texts with unique identifiers for individual concepts is needed. Ideally, this process should be a one-off effort. In a perfect scenario, communication between scientists would take place entirely at the unique concept level. However, this is not reality. What was clear and straightforward in the mind of the researcher after completing the experiments gets lost in a variety of ambiguous expressions during the writing process and ambiguities are produced every minute. Therefore scientific writing can somewhat cynically be called information burying.Authors are actually inclined and stimulated to use variable expressions and synonyms, aphorisms and the like to make their article \'readable\'. This will not change in the foreseeable future and this means we have to face the challenge of semantic tagging of these texts. To keep the challenge manageable, I will restrain this argument to term identification and not include full analysis of the language structure of sentences as attempted by Natural Language Processing approaches. Will authors do it for us? ========================== Scientific text is dry enough as it stands. Should we therefore even contemplate attempts to force authors to make it semantically correct and uniform, according to strict nomenclature or worse, using structured data entry? I would argue that not only would this approach be completely unfeasible, given the creative and individualistic minds of the producers of scientific knowledge, but in addition, the result of hypothetical success would lead to a totally dull form of scientific literature. In the end, no matter what computational aids we offer, researchers will always keep reading for final evidence. So, we need to make things computer-digestible *in the background.* Will computers do it for us? ============================ With the emerging, improved named entity tagging and semantic tagging technologies, a far more elegant and practical solution is available or just around the corner. Correct (computer) recognition of the concepts denoted by words and phrases in free-text, and the semantic enrichment that comes with it, greatly facilitates direct meta-analysis of subsets of the literature. This pre-processing step also facilitates much more accurate cross-linking of information between articles and databases. If yes, why not? ================ It is surprising that publishers have so far taken almost no steps in this direction, even though it holds almost limitless potential, and it is clear that adding value to plain text is the only perspective publishers may have to charge for textual records in the future. Simple, technically feasible additions to current ontology-based semantic tagging software would allow the development of \'tag as you type\' tools that normalise and map ambiguous terms in text to the unique concept they denote and to the corresponding ontologies. The semantic tagging of new and existing text, notably the well over 14 million papers represented in Medline, seems a daunting perspective at first glance. Researchers are neither educated nor encouraged to take a semantic network approach when publishing their data. As a consequence, even decades after electronic publishing became commonplace, most papers are still ending up as dead, non-interactive pages rusting silently away in electronic archives until a search engine happens to be fed with the correct combination of keywords to retrieve them. Or\...will computers and authors do it together? ================================================ In my opinion, author-interactive publication tools should *notforce the writing scientist to use specific terms from pre-structured lists and nomenclatures. Trying to do exactly that, is what made most efforts of nomenclature standardization ineffective. In contrast, the very valuable efforts of the HGNC\'s, Entrezgenes and SwissProts of this world should be used to disambiguate terms on the flyand only ask the author for assistance in the rare cases the ontology driven system can not make an informed decision about the meaning of a term, for example in case a homonym is used without sufficiently distinguishing context surrounding it. If a title such as: \'Epidemiological considerations of BSE\' is typed, the system is unlikely to be able to decide whether the author meant Breast Self Examination or Bovine Spongiform Encephalopathy. If no further context follows, the author could be prompted to resolve the ambiguity. However, if someone decides that he likes EBV more than Human Herpesvirus 4, or she prefers CD154 over TNFSF5, the author should not be forced to change that in the text, as long as the semantic tag added to that term in the background is linked to the unique concept identifier of the virus or the gene in the leading ontologies and nomenclature data bases (see figure [1](#F1){ref-type="fig"}). Obviously, some extra work will be asked from the author, but not anywhere near as much as most critics of the semantic web idea seem to expect. In general, the expectation would be that the vast majority of relevant concepts in biomedical text could be correctly tagged on the fly and in the background without bothering the author at all, other than to resolve the occasional ambiguity, and to review the overall markup once complete. Yes, but.. ========== Obviously, the critics of the Semantic Web approach will bring in their arguments about the difficulties with tagging the legacy documents and the issue of the rapidly developing knowledge, reflecting in ontology changes. Obviously they have a potential point: If only recently published documents are being tagged, computational tools that rely on the tagging will be very restricted in their resources. The technological approach proposed here will enable us to tag existing texts on the fly with high accuracy, by taking care of the \'easy tagging\' and interact with the author or the reader only to resolve the few remaining ambiguities. Moreover, as the tagging is not necessarily static, updated ontologies will lead to (optional) updated proposed tags each time the text is retrieved. Tagging would thus effectively be limited to texts that are still used. If an article is never retrieved by anyone it is probably of limited urgency to tag it. So, let\'s do it! ================= Given the enormous impact a collective semantic enrichment effort would make, particularly on many aspects of bioinformatics research and computational biology, this process should have started many years ago. The Open Access publishers, but BMC Bioinformatics*in particular, should take the lead and promote the use of semantic tagging, probably starting at one of the time points at which stubborn scientists like me are most likely to accept any guidance, and that is when they submit a manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The part of text of this editorial that was marked in italics was fed to an existing semantic tagging tool and it can be seen in the figure how different expressions are mapped to the same concept number. ::: ![](1471-2105-6-142-1) :::	PubMed Central	2024-06-05T03:55:58.325923	2005-6-7	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173089/", "journal": "BMC Bioinformatics. 2005 Jun 7; 6:142", "authors": [ { "first": "Barend", "last": "Mons" } ] }
PMC1173090	Background ========== Dehydrogenases dedicated to L-tyrosine (TYR) biosynthesis comprise a family of TyrA homologs that have different specificities for the cyclohexadienyl substrate: ones specific for L-arogenate (AGN), ones specific for prephenate (PPA), and those that are able to use both \[[@B1],[@B2]\]. Figure [1](#F1){ref-type="fig"} illustrates the biochemical relationship of these specificities to divergent transformations beginning with chorismate (CHA) utilization and converging on TYR formation. Compounding this complexity, a given TyrA enzyme having any of the aforementioned cyclohexadienyl specificities may be specific for NAD^+^or NADP^+^, or may use both. This is consistent with a growing appreciation \[[@B3],[@B4]\] that different substrate specificities are often accommodated across a given protein family that nevertheless maintains a common scaffold of fundamental reaction chemistry. Even within the single category of broad TyrA specificity, there is a continuum ranging from examples where alternative substrates are accepted equally well to other cases where one substrate may be preferred by an order of magnitude or more. Table [1](#T1){ref-type="table"} provides a key to the nomenclature used to identify the various possible substrate-utilization combinations (both cyclohexadienyl and pyridine nucleotide) exhibited by TyrA proteins. The TyrA family is typical of many protein families in that its members have a relatively small core domain that is not highly conserved. As such, substantial challenges for bioinformatic analysis are posed. Here we have not only carried out a labor-intensive manual analysis, but we have also developed tools intended to facilitate and refine follow-on studies of this protein family in the genome era. The approaches implemented in this study with the TYR segment of aromatic biosynthesis hopefully can serve as a template for forthcoming integrant analyses of other pathway segments of aromatic biosynthesis, and indeed for metabolic subsystems in general. This manuscript contains three broad sections. First, the biochemical and enzymological complexity of the TyrA protein family is presented in terms of the diversity that exists in nature with respect to substrate specificity and the association of the core domain with other catalytic or regulatory domains. Secondly, the genomic colinear organization of tyrAgenes with other genes is evaluated, i.e., tyrAis considered in its syntenic context. Thirdly, tyrAis evaluated in its context of regulation. These three sections are tied together in a framework of evolutionary perspective. Results and discussion ====================== Background of TyrA diversity ---------------------------- Our evolutionary analysis is limited by the amount of information that can be managed in a single study, with the focus fixed upon the domain Bacteria(due to the relative density of genome representation for Bacteriain the public databases). However, in order to show where future expansion of the analysis might lead, the selection of TyrA proteins in Fig. [2](#F2){ref-type="fig"} are from all three domains of life, i.e., Bacteria, Archaea, and Eukarya(lower eukaryotes and higher plants). For practicality of presentation, numerous orphan (i.e., without close relatives) TyrA sequences are not shown, and not all members of a given group are necessarily included. The main purposes of the radial tree shown in Fig. [2](#F2){ref-type="fig"} are: (i) to illustrate that TyrA proteins of major phylogenetic groupings are generally congruent with 16S rRNA groupings and (ii) to convey a snapshot visualization of the overall complexity of the TyrA protein family from the vantage point of its varied substrate specificities as well as its multiple fusion partners. As an illustration of the detailed information that follows, note that the TyrA sequences from the beta Proteobacteria at five o\'clock in Fig. [2](#F2){ref-type="fig"} form a cohesive cluster (termed a \'congruency group\'). In this clade there exists a proposed ancestral background of broad specificity where either AGN or PPA in combination with either NAD^+^or NADP^+^could be used. This profile of broad substrate use (which can be denoted as ~NAD(P)~TyrA~c~; see Table [1](#T1){ref-type="table"}) generally persists in the beta Proteobacteria. From this background, narrowed specificities for the AGN/NADP^+^couple emerged once in the lineage represented by Nitrosomonas europaea(Fig. [2](#F2){ref-type="fig"}; dark blue line), narrowed specificity for NAD^+^emerged once in species of Neisseria(orange line), and fusion of tyrA~c~with aroF(which encodes enolpyruvylshikimate-3-P synthase, the sixth enzyme in the common pathway of aromatic biosynthesis; see \[[@B5],[@B6]\] for nomenclature used) occurred recently within the Burkholderialineage. These character-state transformations appear to occur with relative ease, and independent emergence of the same character states can be seen elsewhere in the tree. Phylogenetically congruent TyrA groupings ----------------------------------------- ### Multiple alignments of catalytic-core domains A phylogenetic tree is only as good as the input alignment. An optimal multiple alignment of TyrA homologs requires a trimmed set of sequences that corresponds to the catalytic-core domain. Alignment of sequences with non-homologous N-terminal fusions (such as with chorismate mutase• (AroQ•), HisH~b~•, or plant transit peptides•; note the convention of using a bullet to indicate the fusion point of one domain with another domain) will make them appear to be more closely related than they actually are because residues in the non-homologous N-terminal regions find matches at random. Likewise, those TyrA sequences with C-terminal fusions (such as with •AroF, •ACT, or •REG) will appear to be anomalously close to one another. Even enzyme proteins that have much greater sequence conservation and amino-acid lengths than TyrA proteins cannot reasonably be expected to yield a protein tree that would be congruent over an extensive phylogenetic range with the overall 16S rRNA tree. However, if genome representation is sufficiently dense within a range of closely related organisms, 16S rRNA congruency with a given protein can be expected within that range of organisms provided that (i) the particular functional role has been retained and (ii) lateral gene transfer has not occurred to obscure the relationship. This expectation follows from the outcome of a detailed analysis of tryptophan-pathway proteins in Bacteria\[[@B7],[@B8]\]. ### Congruency within major clades TyrA sequences from higher-plant and yeast Eukaryaform cohesive clusters. Genome representation among Archaeais still relatively limited. (Fig. [2](#F2){ref-type="fig"} does reveal, however, that genes encoding TyrA proteins in Archaeahave experienced various catalytic- and regulatory-domain fusions at least as frequently as those in Bacteria). Eventual expansion of both the tryptophan-pathway and tyrosine-pathway analyses to Archaeashould be quite interesting. The great majority of TyrA sequences available are from Bacteria, and one can see (by inspection of the major clades supported by high bootstrap values in Fig. [2](#F2){ref-type="fig"}) a qualitatively apparent congruence of TyrA-tree sub-sections with 16S rRNA expectations of vertical genealogy. Thus, all cyanobacteria possess a ~NADP~TyrA~a~type of TyrA enzyme, and this is a very cohesive grouping. A few of the larger cyanobacterial genomes have a co-existing second enzyme of the TyrA~c\_Δ~type (discussed in detail later). The low-GC gram-positive bacteria (Bacillus/Staphylococcus/Enterococcus/Listeria) exhibit the ~NAD~TyrA~p~pattern of specificity and also possess a C-terminal domain (ACT) of allosteric regulation. It is interesting that the TyrA~p~•ACT proteins of the Streptococcuslineage (at eight o\'clock in Fig. [2](#F2){ref-type="fig"}) differ from the main low-GC clade in possessing broad specificity for pyridine nucleotides (as indicated with black line color). The most parsimonious evolutionary conclusion would be that in the low-GC gram-positive grouping, acquisition of the ACT domain and narrowed specificity for prephenate preceded narrowed specificity for NAD^+^. Thus, the latter event occurred after divergence of the Streptococcuslineage from the remainder of the low-GC clade. Members of the subclass taxon Actinobacteridae(mostly actinomycetes) possess AGN-specific TyrA enzymes (light blue fill color in Fig. [2](#F2){ref-type="fig"}), but they separate into two distinct groups that correlate either with broad specificity for pyridine nucleotides (Actinobacteridae\_1) or a NAD^+^-specific pattern (Actinobacteridae\_2). The Proteobacteria are discussed immediately below. ### Proteobacteria By far the greatest genomic density available is for Proteobacteria, the group of Bacteriathat includes purple bacteria and their relatives. The various divisions of Proteobacteria, as currently named, lack hierarchical equivalence. For example, the epsilon and delta divisions branch from much deeper positions on the phylogenetic tree than do the alpha Proteobacteria. As genome representation expands for epsilon and delta Proteobacteria, it is probable that these will subdivide to newly named groupings of approximate hierarchical equivalence with alpha Proteobacteria. The most recently diverged Proteobacteria are the beta and gamma divisions. From the combination of our previous analysis of tryptophan biosynthesis \[[@B7],[@B8]\], TYR biosynthesis (this paper), and other segments of aromatic biosynthesis (unpublished data), we find it useful to separate \"upper-gamma\" Proteobacteria from \"lower-gamma\" Proteobacteria (an \"enteric lineage\" with Shewanella oneidensisas approximately the most divergent member). This separation is because the beta Proteobacteria and the upper-gamma Proteobacteria exhibit a smooth continuity of relatively few evolutionary events with respect to aromatic biosynthesis, in striking contrast to extraordinarily dynamic evolutionary events in the lower-gamma Proteobacteria. As a consequence, the lower-gamma Proteobacteria are much more distinct (in terms of aromatic biosynthesis) from the upper-gamma Proteobacteria than the upper-gamma are from the beta Proteobacteria. Figure [2](#F2){ref-type="fig"} shows that alpha, beta and epsilon divisions of Proteobacteria form phylogenetically coherent clusters with respect to their TyrA proteins. Although delta Proteobacteria fall into two well-separated groupings denoted as Delta\_1 and Delta\_2, this should not be surprising since these groupings diverge at a deep level on the 16S rRNA tree where genome representation is poor. In addition, the Myxococcus xanthusTyrA sequence, currently an orphan (three o\'clock in Fig. [2](#F2){ref-type="fig"}), represents a third divergent lineage in delta Proetobacteria. In contrast to delta Proteobacteria, genomic representation for the gamma Proteobacteria is relatively good. Nevertheless their TyrA sequences separate into several well-spaced groupings, albeit for entirely different reasons. In this case, the split seen between two clades of these fairly close relatives (upper-gamma and lower-gamma) is attributed to particularly dynamic evolutionary events compressed into a relatively short time span in the lower-gamma Proteobacteria. (We refer to such a dynamic divergence as an evolutionary jump; see the next section.) Note that the allocation of upper-gamma and lower-gamma Proteobacteria to separate TyrA congruency groups is not the same as being incongruent. It is quite possible that as new genomes come on line, new and intermediate TyrA sequences may result in the merging of the foregoing two congruency groups (currently tyrosine congruency group 1 (TyrCG-1) and tyrosine congruency group 2 (TyrCG-2)). ### Comparison of tryptophan and tyrosine congruency groups Although the true extent of lateral gene transfer (LGT) at present must be described as intensely controversial, there is little doubt that any given organism is mosaic with respect to some unknown fraction of its gene repertoire. Our \"accounting\" system for keeping track of proteins that are faithful to the vertical genealogy is to formulate congruency groupings that are defined by congruence of given protein-tree clusters to a section of the 16S rRNA tree. Ultimately this information will reveal which organisms are \"pure\" with respect to the vertical inheritance of a given pathway or pathway segment. Our congruency groups are intended to be fluid, in that with the continued availability of new sequences, a previous orphan sequence may very well become the seed for a new congruency group. On the other hand, previously separate congruency groups have the potential to merge. (See Methods for more information.) The present tyrosine congruency groups are listed on the AroPath website \[[@B9]\]. Seven tryptophan congruency groups in Bacteriawere previously formulated \[[@B8]\] based upon the correspondence of cohesive clusters in trees of Trp-protein concatenates with sections of 16S rRNA trees. The information input for formulation of tryptophan congruency groups is of greater quality than for tyrosine congruency groups because seven-protein concatenates could be used for the former. On the other hand, the broad information input supporting tyrosine congruency groups in this study is more comprehensive because of greater genome availability. Tryptophan congruency group 1 (TrpCG-1) corresponds perfectly with the organisms represented in TyrCG-1, these being the lower-gamma Proteobacteria (enteric lineage). The upper-gamma Proteobacteria (TyrCG-2) and the beta Proteobacteria (tyrosine congruency group 3; TyrCG-3) are represented by different tyrosine congruency groups. In contrast, the membership of tryptophan congruency group 2 (TrpCG-2) includes both the upper-gamma Proteobacteria and the beta Proteobacteria. The latter merging probably reflects the advantage conferred by the greater information content of the concatenated sequences used to define tryptophan congruency groups. Species of Xylellaand Xanthomonasare usually referred to as gamma Proteobacteria. They probably represent an outlying deeply branching lineage, although trees based on concatenated strings of proteins \[[@B10]\] or 16S rRNA \[[@B11]\] position them with beta Proteobacteria. In any event, Trp-protein concatenate trees placed Xylellaand Xanthomonaswithin TrpCG-2, which contains both upper-gamma and beta Proteobacteria. In contrast, the TyrA domains from Xylellaand Xanthomonaswere well separated (at about two o\'clock in Fig. [2](#F2){ref-type="fig"}) from those of any other organism. This might simply be due to the limited resolving power of a single protein in combination with too few close relatives. (Note that single Trp-protein trees sometimes failed to achieve the congruency-group placements that were resolved by seven-protein Trp concatenates \[[@B8]\]). An additional clue may be relevant. The TyrA proteins from the Xylella/Xanthomonasgenera possess an ACT domain, which has not been observed in any other proteobacterial TyrA proteins thus far. In view of this, origin by LGT seems to be a distinct possibility, but with the important caveat that no likely genome donors are yet obvious on the criterion of sequence similarity. Perhaps more likely is the following possible explanation that postulates a basis for accelerated divergence. The TyrA domains of Xanthomonas/Xylellaproteins have an indel structuring (insertions and/or deletions) that places them within the TyrA~c\_Δ~specificity subclass (see below). We suggest (see below) that such indel structuring reflects interaction of the core TyrA domain with an extra-domain extension. Thus, selection for amino acid changes accomplishing a new domain-domain interaction could account for accelerated divergence of the Xanthomonas/Xylellasequences on the TyrA tree (Fig. [2](#F2){ref-type="fig"}). Cohesive tryptophan congruency groups of the alpha Proteobacteria (tryptophan congruency group 3; TrpCG-3) and the cyanobacteria (tryptophan congruency group 4; TrpCG-4) match up well with the corresponding tyrosine congruency groups (tyrosine congruency group 4 (TyrCG-4) and tyrosine congruency group 8 (TyrCG-8), respectively). The TyrA proteins of epsilon Proteobacteria define a cohesive tyrosine congruency group (tyrosine congruency group 5; TyrCG-5), whereas the Trp-protein concatenates of epsilon Proteobacteria did not exhibit a coherent congruency group, due at least in part to LGT \[[@B8]\]. The delta Proteobacteria separate into two distinct tyrosine congruency groups: Delta\_1 (tyrosine congruency group 6; TyrCG-6) and Delta\_2 (tyrosine congruency group 7; TyrCG-7), as shown in Fig. [2](#F2){ref-type="fig"}. It is likely that corresponding tryptophan congruency groups exist (work in progress), but at the time of the Xie et al. study \[[@B8]\] only Trp-pathway protein concatenates for Desulfovibrio vulgaris(Delta\_2) and Geobacter sulfurreducens(Delta\_1) were available, and they were provisionally listed as \"orphans\". In the present work TyrA sequences from Deinococcus radioduransand Thermus thermophilusare the sole members of tyrosine congruency group 12 (TyrCG-12). At the time of the Trp-pathway work, the genome of Thermuswas unavailable and the Deinococcusconcatenate was listed as an orphan. It is expected that the Deinococcusand Thermusconcatenates will now seed a new tryptophan congruency group. Whereas tryptophan congruency group 5 (TrpCG-5) is defined by cohesive concatenates from actinomycete bacteria, the TyrA proteins from the same organisms separated into two distinct congruency groups. It is intriguing that this partitioning into two congruency groups correlates with narrowed specificity for NAD^+^(indicating an evolutionary jump) in one of the groups. The latter group (tyrosine congruency group 11; TyrCG-11) is denoted Actinobacteridae\_2 in Fig. [2](#F2){ref-type="fig"}, whereas tyrosine congruency group 10 (TyrCG-10) is displayed as Actinobacteridae\_1. The opposite scenario whereby a single tyrosine congruency group corresponds to split tryptophan congruency groups applies in the case of low-GC gram-positive bacteria. Whereas TyrA proteins form a single congruency group in these organisms (tyrosine congruency group 9; TyrCG-9), a small cluster of Trp-pathway concatenates from Bacillus subtilis, B. stearothermophilus, and B. halodurans(tryptophan congruency group 6; TrpCG-6) separate distinctly from the remaining organisms (tryptophan congruency group 7; TrpCG-7). The latter evolutionary jump reflects a dynamic scenario of tryptophan-pathway evolutionary events that include loss of one gene from the trpoperon, insertion of the trpoperon into a 6-gene arooperon to produce a supraoperon, and acquisition of the TRAP (tryptophan-activated RNA-binding protein) mechanism of regulation by an RNA-binding protein \[[@B7]\]. Tyrosine congruency groups and tryptophan congruency groups are maintained and updated at the AroPath website \[[@B12]\]. Distribution in nature of TyrA specificity subclasses for the cyclohexadienyl substrate --------------------------------------------------------------------------------------- Four qualitative classes of specificity for the cyclohexadienyl substrate populate the TyrA superfamily of homologs (Fig. [1](#F1){ref-type="fig"}). These include PPA-specific (TyrA~p~), AGN-specific (TyrA~a~), the broad-specificity cyclohexadienyl (TyrA~c~) dehydrogenases and a fourth class represented by an enzyme of antibiotic biosynthesis (PapC) that converts 4-amino-4-deoxy-prephenate to 4-amino-phenylpyruvate \[[@B13]\]. Representatives of each specificity class have been studied at molecular and genetic levels. TyrA family members sharing a given substrate specificity do not necessarily cluster tightly together, and assignment of substrate specificity to experimentally uncharacterized TyrA homologs is uncertain unless they exhibit very high amino acid identities with experimentally characterized TyrA proteins. In some cases we do not accept older literature reports without more recent verification. For example, the yeast Saccharomyces cerevisiaeTyrA~x~was characterized as a TyrA~p~protein \[[@B14]\] long before it was recognized \[[@B15]\] that PPA preparations were often contaminated with AGN (an unknown compound at that time). Our collection of curated TyrA sequences at AroPath (see Table [3](#T3){ref-type="table"}) contains trimmed sequences that comprise catalytic-core domains. This collection was divided into two groups based on whether the sequences contained the relatively short N-terminal pyridine-nucleotide discriminator segment or the longer C-terminal cyclohexadienyl-substrate core segment. The sequences in the latter group were assembled into subgroups representing established substrate specificities (TyrA~a~, TyrA~p~and TyrA~c~) and were aligned separately to obtain overall consensus sequences for cyclohexadienyl-substrate core segments. The TyrA~c~group members from the lower-gamma assemblage of Proteobacteria (as well as from a few other lineages) were so distinctive that a fourth group (TyrA~c\_Δ~) was defined. This latter group is, in fact, the most divergent of the four. Figure [3](#F3){ref-type="fig"} shows a comparison of the four consensus sequences, with invariant anchor residues shaded yellow and residues conserved across all groups shaded in gray. Residues within each group that are \>50% conserved are shown in capital letters. In pairwise BLAST (Basic Local Alignment Tool) \[[@B16]\]comparisons, TyrA~a~and TyrA~c~consensus sequences are most similar (47% identity), followed by the TyrA~c~/TyrA~p~pair (40% identity), with TyrA~a~and TyrA~p~exhibiting 34% identity. TyrA~c\_Δ~is quite distinct from the other three groupings, exhibiting only 27% identity with TyrA~c~, 23% identity with TyrA~c~, and 18% identity with TyrA~p~. ### Cyclohexadienyl dehydrogenases Many TyrA proteins (at least in the domain Bacteria) are of the TyrA~c~subclass. The cyclohexadienyl dehydrogenases commonly accept PPA or AGN about equally well, but various degrees of preference for one of the alternative substrates are also observed. Detailed molecular and genetic studies of TyrA~c~proteins from Pseudomonas aeruginosa, \[[@B17]\], P. stutzeri\[[@B1]\], and Zymomonas mobilis\[[@B18]\] have been carried out. The distinct variety of TyrA~c~mentioned above, which has been denoted TyrA~c\_Δ~exhibits a number of indels (mostly deletions) within the catalytic-core region when its consensus sequence is aligned with those of the other TyrA classes (Fig. [3](#F3){ref-type="fig"}). It is intriguing that the indel structuring of TyrA~c\_Δ~correlates with the presence of an extra-core extension. This extension is often AroQ, but not always. For example, in the genera Nostocand Anabaenait appears to be a degraded, catalytically inactive AroQ, whereas in Xanthomonasor Xylellait is an ACT domain. Since the one large clade of TyrA~c\_Δ~proteins that has so far been studied prefers PPA over AGN by well over an order of magnitude, an evolutionary relationship of indel insertions to the narrowing of substrate preference for PPA might exist. If so, however, this cannot be the only molecular change to accomplish favored utilization of PPA over AGN since a number of TyrA~c~proteins, (e.g., TyrA~c~from Neisseria gonorrhoeae), also exhibits an overwhelming preference for PPA, even though this class lacks the indel structuring. ### Arogenate dehydrogenases The TyrA~a~class of specificity is currently represented by higher plants and at least three widely spaced bacterial lineages: cyanobacteria, actinomycetes and Nitrosomonas europaea. This discontinuity of phylogenetic spacing is consistent with a fundamental evolutionary scenario \[[@B19]\] whereby the ancestral dehydrogenase was a broad-specificity TyrA~c~that evolved narrowed substrate specificity (to yield either TyrA~p~or TyrA~a~) independently on multiple occasions in modern lineages. The ubiquitous presence of TyrA~a~in cyanobacteria has been heavily documented \[[@B20]\]. Nitrosomonas europaeacurrently (as of March, 2005) has no sufficiently close genome relatives that have been sequenced. The first BLAST hit returned from a ~NADP~TyrA~a~query from N. europaea(March,2005) is the protein from Ralstonia solanacearum(48% identity), which is known to possess broad specificity for both of its substrates (i.e., ~NAD(P)~TyrA~c~) \[[@B21],[@B22]\]. The TyrA sequences of Actinobacteriaseparate into two distinct groupings on the protein tree (Fig. [2](#F2){ref-type="fig"}). Coryneform bacteria in one sub-cluster have been rigorously characterized as the ~NAD(P)~TyrA~a~substrate specificity type. On the other hand, a variety of Streptomycesspecies have been shown \[[@B23],[@B24]\] to possess ~NAD~TyrA~a~, and TyrA proteins of these organisms populate the second Actinobacteriasub-cluster of Fig. [2](#F2){ref-type="fig"}. Figure [4](#F4){ref-type="fig"} shows sequence alignments of the N-terminal pyridine-nucleotide discriminator regions of currently available actinomycetes. The conserved \'D\' residue (highlighted in yellow) in the upper group is a reliable indicator of NAD^+^specificity, in part because NADP^+^is repelled by the negative charge at this position. The asparagine residue (highlighted in blue) in the corresponding position in members of the lower group indicates NAD(P)^+^specificity as discussed by Bonner et al. \[[@B25]\]. Rubrobacter xylanophilusis the most distant representative of the Actinobacteria, being the sole member of the subclass taxon Rubrobacteridae, and its protein (denoted Rxyl) appears as an orphan in Fig. [2](#F2){ref-type="fig"}. A similar relationship of phylogenetic separation associated with narrowed specificity for pyridine-nucleotide substrate exists for the low-GC gram-positive bacteria (eight o\'clock in Fig. [2](#F2){ref-type="fig"}). Here the major clade is NAD^+^-specific, whereas species of Streptococcushave retained the ancestral breadth of specificity for NAD^+^/NADP^+^. Alignments of the pyridine-nucleotide discriminator regions of these latter two groups match up extremely well with the upper alignment of Fig. [4](#F4){ref-type="fig"} where residue 32 of the Wierenga fingerprint \[[@B26]\] is \'D\' and with the lower alignment where residue 32 is \'N\' (data not shown). Recently, a plant tyrA~a~from Arabidopsis thalianahas been reported to consist of two near-identical domains that are fused \[[@B27]\]. The gene encoding this 68-kDa protein co-exists in the genome with a single-domain paralog \[[@B28]\] that encodes a predicted 37-kDa protein, somewhat larger than the catalytic-core domain of TyrA~a~from Synechocystis. TyrA~a~(known to be located in higher-plant chloroplasts \[[@B2]\]) may have originated from cyanobacteria via endosymbiosis. If so, however, the plant TyrA~a~sequences have diverged sufficiently that they no longer share a specific phylogenetic grouping with the cyanobacterial TyrA sequences. This is in marked contrast with the phylogenetic coherence of the tryptophan synthase subunit proteins (TrpEa and TrpEb\_1) from cyanobacteria and higher plants \[[@B29]\]. ### Prephenate dehydrogenases TyrA~p~is conspicuously represented by a large clade of low-GC gram-positive organisms, of which Bacillus subtilisTyrA~p~is the best studied \[[@B30]\]. Thus far, all TyrA~p~proteins are fused to a C-terminal ACT domain, and therefore no \"minimal\" TyrA~p~proteins that consist only of a catalytic core are available as yet. At the level of physiological function, it should be added that those cyclohexadienyl dehydrogenases that exhibit a very substantial preference for prephenate are for all practical purposes prephenate dehydrogenases, even though they carry a formal designation of TyrA~c~or TyrA~c\_Δ~. These include most, if not all, of the AroQ•TyrA~c\_Δ~enzymes of the enteric lineage (lower-gamma in Fig. [2](#F2){ref-type="fig"}). The TyrA~c~protein from Neisseria gonorrhoeae(and by inference, the closely related N. meningitides) is also a well-studied example of overwhelming preference for prephenate \[[@B21]\]. ### PapC dehydrogenases PapC participates in the formation of p-aminophenylalanine as a step in the synthesis of at least two antibiotics (see Fig. [1](#F1){ref-type="fig"}). It is so far represented by only a few sequences. The PapC specificity is strongly indicated by absence of the otherwise invariant residue H197 (E. colinumbering) that is associated with recognition of a 4-hydroxy moiety in the cyclohexadienyl substrates of the aforementioned dehydrogenases. This moiety, of course, differs in being a 4-amino substituent in the substrate used by the PapC dehydrogenase (Fig. [1](#F1){ref-type="fig"}). See Bonner et al. \[[@B25]\] for a more detailed overview. The \"redundant\" trp/arosupraoperon of Nostoc/Anabaena ----------------------------------------------------------- All cyanobacteria possess a highly conserved tyrA~a~gene, as well as a complete suite of tryptophan-pathway genes that are dispersed (unlinked) in the genome. The large-genome cyanobacterial lineage consisting of the Nostocand Anabaenagenera possess in addition a unique and seemingly redundant trp/arosupraoperon consisting of most of the aforementioned genes \[[@B31]\]. These include a second tyrAgene (curated as tyrA~c\_Δ~), six trp-pathway genes (all except trpC), and genes encoding the first two common-pathway steps of aromatic amino acid biosynthesis. All of these supraoperonic genes appear to be redundant in that they are represented by homologs (paralogs or xenologs) elsewhere in the Nostocand Anabaenagenomes at scattered loci. The closest BLAST hits for the Nostoc/AnabaenaTyrA~c\_Δ~proteins are not the co-existing TyrA~a~homologs present in their own genomes (and universally present in cyanobacteria). Rather the closest BLAST hits are to the TyrA~c\_Δ~domains of the AroQ•TyrA~c\_Δ~fusions in the enteric lineage. Since the enteric proteins are NAD^+^-specific and strongly prefer prephenate, it is likely that the \"extra\" cyanobacterial proteins are also ~NAD~TyrA~c\_Δ~proteins. Indeed, this would be consistent with enzymological evidence provided in the literature for both Nostocand Anabaena\[[@B20]\]. Concerning the evolutionary origin of the redundant block of linked genes found in the Nostocand Anabaenagenomes, at least two possibilities await further illumination. (i) These genes might have been acquired by a common ancestor of Nostocand Anabaenavia lateral gene transfer. This is consistent with the observation that biosynthetic-pathway operons are generally absent in the cyanobacteria, and all of the linked genes could have been recruited in a single event. However, at present no candidate donor genomes are known that possess this supraoperon combination of genes. If the TyrA~c\_Δ~proteins of Nostoc/Anabaenaand the enteric lineage are possibly related by LGT, it is of interest that the N-terminal extension of TyrA~c\_Δ~from Nostoc/Anabaenaresembles a degraded AroQ domain of AroQ•TyrA~c\_Δ~from enterics. In both cases the N-terminal residues may compensate for indel deletions within the catalytic core region of TyrA~c\_Δ~. Subsequently, AroQ function may have evolved in one lineage (or have been lost in the other). This possibility of domain-domain interaction is consistent with the established interdependence of the AroQ• and •TyrA~c\_Δ~domains from E. coli\[[@B32]\]. Alternatively, tyrA~a~and tyrA~c\_Δ~(and the duplicated trpand arogenes present in the supraoperon) might be ancient paralogs within the cyanobacterial lineage. If so, at a time following divergence of heterocystous cyanobacteria from the unicellular cyanobacteria, the latter may have lost the clustered block of aromatic-pathway genes in a single event of reductive evolution. The supraoperonic genes might be related to a specialized function associated with \"developmental\" physiological processes that typify the filamentous, heterocyst-forming cyanobacteria. This might be reminiscent of the nature of the phenazine-pigment operon of Pseudomonas aeruginosa. Here unique phenazine-pathway genes are combined with a redundant gene of common-pathway aromatic biosynthesis and two redundant (and fused) genes of tryptophan biosynthesis. This accomplishes the linkage of specific phenazine biosynthesis with a supply of 2-amino-2-deoxy-isochorismate, the branchpoint of divergence toward phenazine and tryptophan \[[@B33],[@B34]\]. This complexity in which multiple paralogs are differentially deployed is consistent with the large genome sizes of Anabaena(7.2 MB) and Nostoc(9.2 MB), compared with the much smaller unicellular genomes of Prochlorococcus marinus(1.7 MB), Synechococcussp. WH8102 (2.4 MB), and Synechocystissp. PCC6803 (3.6 MB). Profile hidden Markov models (HMMs) to distinguish specificity subfamilies for cyclohexadienyl substrate -------------------------------------------------------------------------------------------------------- The limited information thus far available about specific molecular roles of particular TyrA amino acid residues has been summarized recently \[[@B25]\]. The catalytic-core domains of known TyrA~a~, TyrA~p~, TyrA~c~, and TyrA~c\_Δ~proteins were selected from our files of TyrA catalytic-core domains \[[@B35]\], and a new subset of sequences was prepared that lacked the pyridine nucleotide discriminator segment, a glycine-rich βαβ region at the N terminus. Although the glycine-rich βαβ region is not the only segment that contacts pyridine nucleotide substrate, it is the sole region that discriminates between NAD^+^and NADP^+^. The resulting trimmed sequence is defined as the \"cyclohexadienyl-substrate core segment\". No distinctive motifs were found that, in isolation, would be a clear predictive indicator of specificity for cyclohexadienyl substrate. Similar substrate specificity profiles probably can be dictated by alternative patterns of interplay between different residue combinations. Because of the rapid accumulation of incorrectly annotated TyrA entries in GenBank and other databases, partly due to the complications of misnaming that are associated with gene fusions and partly to a failure to assimilate published substrate specificities, the use of BLAST does not return reliable annotations with respect to substrate specificity. Even the HMMs used in Pfam \[[@B36]\] and Interpro \[[@B37]\] were not helpful in this case because the HMM deployed in those databases was broadly but incorrectly defined as \'prephenate dehydrogenase (NADP^+^) activity\' for all TyrA dehydrogenases (accession number PF02153 in Pfam and entry IPR003099 in Interpro). However, Profile HMM is known to be well suited for modeling a particular sequence family of interest and for finding additional remote homologs \[[@B38]\]. It is reputed to outperform methods that rely only upon pair-wise alignment of homologous residues in predicting protein function \[[@B39]\]. Therefore, profile HMMs were constructed using our multiple sequence alignments of each curated TyrA specificity subfamily, using the HMMER package \[[@B38]\]. The profile HMMs obtained are only tentatively reliable for prediction of substrate specificity. To facilitate ongoing and future functional annotations, we have made our profile HMMs available as a working resource for \"specificity prediction\" at AroPath \[[@B40]\]. Users can match query sequences against the four profile HMMs to predict the subfamily to which a query sequence belongs. It is anticipated that future experimental data relevant to substrate specificity will facilitate refinement of the prediction program. For example, at present the program predicts that the TyrA sequences from organisms such as Helicobacter pyloriand Saccharomyces cerevisiaebelong to the TyrA~a~grouping, and it will be interesting to see whether this holds up to experimental confirmation. It is additionally fascinating that (i) the dehydrogenase from Archaeoglobus fulgidusis predicted to belong to the indel-containing TyrA~c\_Δ~grouping and (ii) that it possesses a possible cooperatively interacting extra-core domain extension (an AroQ fusion), just as occurs for the large clade of enteric bacteria. If this is relevant, it is even more fascinating that the Archaeoglobus aroQis fused at the C-terminal side of tyrA~c\_Δ~, rather than at the N terminus as is the case with enteric bacteria. Users at AroPath \[[@B41]\] can enter query sequences into interactive multiple sequence alignments with any of the four sets of \"cyclohexadienyl-substrate core segments\" sequences that were used to train the profile HMMs. An ongoing effort is in process to extend the predictor capability to include the pyridine nucleotide substrate as well. One can also align query sequences of interest with either an assemblage of the complete set of curator-approved TyrA catalytic-core TyrA sequences or with any desired subset of seed sequences. The catalytic-core domain of TyrA proteins ------------------------------------------ The simplest set of fully functional TyrA proteins consists only of the catalytic-core domain (about 180 amino acids) \[[@B1]\] and includes the well-characterized TyrA~c~enzymes from Neisseria gonorrhoeae\[[@B21]\] and Zymomonas mobilis\[[@B18]\], as well as TyrA~a~from a cyanobacterium \[[@B25]\]. In addition the catalytic-core domain from Pseudomonas stutzerihas been engineered for study from a tyrA~c~•aroFfusion \[[@B1]\]. These model core proteins are roughly as divergent from one another on the TyrA protein tree as are the organisms that contain them (Fig. [2](#F2){ref-type="fig"}). In view of the possibility raised in this paper about inter-domain interactions, the single-domain TyrA proteins are undoubtedly the simplest sources for study of the fundamental properties of the catalytic-core domain. Xie et al. \[[@B1]\] suggested that in the set of catalytic-core TyrA proteins, inhibitors bind at the catalytic site and exhibit classical competitive inhibition with respect to the particular cyclohexadienyl substrates that can be accepted by a given organism. This model predicts that the specificity for the sidechains of substrates used would parallel the specificity for inhibitor sidechains. The information summarized in Table [4](#T4){ref-type="table"} supports this expectation. Thus, the TyrA~c~proteins of P. stutzeriand P. aeruginosawill accept either a pyruvyl (as with PPA) or an alanyl (as with AGN) sidechain in the alternative substrates used, and this is paralleled by recognition of either a pyruvyl (4-hydroxyphenylpyruvate) or an alanyl (TYR) sidechain in the competent inhibitor structures. In another case, the N. gonorrhoeaeTyrA~c~exhibits an overwhelming substrate preference for PPA, and consistent with the foregoing, is subject to inhibition by 4-hydroxyphenylpyruvate but not by TYR. A variety of analog inhibitor structures were used by Xie et al. \[[@B1]\] to show that the minimal structure for binding at the substrate-binding site of P. stutzeriTyrA~c~is a six-membered ring with a 4-hydroxy substituent. In contrast to the TyrA~c~proteins just described, the Z. mobilisTyrA~c~is totally insensitive to inhibition by either 4-hydroxyphenylpyruvate or TYR. Since both of these compounds lack a 1-carboxy moiety, it is reasonable to assume that the 1-carboxy substituent present in the two substrates accepted may be required for binding at the catalytic center. Thus, although TyrA~c~from Z. mobiliswill accept the same two substrates as does the TyrA~c~from P. stutzeri, the greatly different inhibition results suggest that Z. mobilisobeys more stringent rules for binding at the catalytic site (i.e., a ring carboxylate must be present). Synechocystissp. and Arabidopsis thalianaTyrA~a~proteins accept as a substrate only AGN, which has an alanyl sidechain. The ring-carboxylate moiety is evidently not absolutely required for binding since these TyrA~a~proteins can recognize TYR (alanyl sidechain) as a competitive inhibitor. In contrast, since N. europaeaTyrA~a~is not inhibited by TYR, it resembles the Z. mobilisTyrA~c~in the putative requirement for a 1-carboxy substituent to secure successful binding at the catalytic site. In summary, some TyrA proteins probably exercise greater discrimination in their requirement for a 1-carboxy moiety for binding at the catalytic site, and these are insensitive to competitive inhibition by the aromatic reaction products (which lack the 1-carboxy substituent). Other TyrA proteins that require the 1-carboxy moiety for the fundamental catalytic process, but presumably do not require it for binding, will recognize product inhibitors that have the same sidechain as any substrate recognized. Specificity for the pyridine nucleotide co-substrate within the TyrA superfamily -------------------------------------------------------------------------------- NAD^+^differs from NADP^+^only in that NADP^+^has a phosphate group esterified at the 2\'-position of adenosine ribose. Therefore, the ability of a dehydrogenase to discriminate between those two lies in the particular enzyme region that contacts the ribose moiety. The glycine-rich region known to constitute the ADP-binding βαβ fold is well known to be this point of contact \[[@B26]\]. This Rossmann β α β fold is inevitably positioned at the extreme N terminus of TyrA proteins, and the typical GXGXXG motif is almost always observed, as illustrated in Fig. [4](#F4){ref-type="fig"}. This region is helpful for assessment of probable specificities for pyridine nucleotide. One can be fairly sure that TyrA proteins possessing D-32 (E. colinumbering, reference \[[@B26]\]) are NAD^+^-specific. A negatively charged residue (D or E) at position 32 is critical for hydrogen binding to the diol group of the ribose near the adenine moiety in NAD^+^-specific enzymes. NADP^+^-specific dehydrogenases cannot tolerate a negatively charged residue at position 32. TyrA proteins that possess an asparagine residue in the corresponding position appear to be broadly specific for both NAD^+^and NADP^+^as discussed above. No clearcut motif has been identified for NADP^+^-specific TyrA proteins, although at least one positively charged residue is expected in the region just beyond residue 32. By elimination, those sequences lacking D-32 or N-32 are strong candidates for NADP^+^specificity. As with the cyclohexadienyl co-substrate, narrowed specificity for NAD^+^(or NADP^+^) also seems to have occurred independently on many occasions (some examples given earlier). The absolute specificity of TyrA~p~proteins for PPA tends to be accompanied by absolute specificity for NAD^+^, as illustrated by the large Bacillus/Staphylococcus/Listeria/Enterococcusclade at eight o\'clock in Fig. [2](#F2){ref-type="fig"}. However, it is interesting that species of Streptococcushave retained the presumed ancestral breadth of specificity for the pyridine nucleotide substrate. The opposite relationship, whereby absolute specificity for AGN tends to be accompanied by absolute specificity for NADP^+^, is also observed. Here three of the four TyrA~a~lineages described earlier exhibit this pattern. Exceptions, though, are the aforementioned TyrA~a~proteins of Actinobacteridae\_1 which accept either NAD^+^or NADP^+^, as well as the TyrA~a~proteins of the sister Actinobacteridae\_2 which are specialized for NAD^+^\[[@B42],[@B43]\]. The TyrA~c~proteins of most complete-genome organisms thus far have happened to be NAD^+^-specific, and this has been the property of the most rigorously characterized ones (from Z. mobilis, P. stutzeri, and P. aeruginosa). However, it is clear from extensive enzymological surveys \[[@B22]\] that TyrA~c~proteins having broad specificity for NAD^+^/NADP^+^are common, examples including species of Ralstoniaand Burkholderia. The spectrum of variation that can exist, even within a clade of organisms that are of fairly close relationship, is illustrated by one striking example. In the pseudomonad clade marked by a common tyrA•aroFfusion, the Acinetobactersp. TyrA~c~is NADP^+^-specific \[[@B44]\], whereas the sister subclade Pseudomonas/Azotobacterexhibits NAD^+^specificity (Fig. [2](#F2){ref-type="fig"}). Here the entire clade marked by a common ancestral fusion shares approximately the same profile of cyclohexadienyl substrate preference, but cofactor specificity has been narrowed in opposite directions. We had previously suggested that there might be a general structural relationship of substrate pairing that tends to favor interaction between PPA and NAD^+^, on the one hand, and, on the other hand, between the greater positive charge of AGN and the greater negative charge of NADP^+^. These relationships may indeed be favored, but it increasingly appears that any combination can occur. Beyond the catalytic core: allosteric domains --------------------------------------------- Various lineages have acquired an amino acid binding domain known as the ACT domain (pfam01842), which is known to bind a variety of amino acids, thus functioning as an allosteric domain for many proteins including phosphoglycerate dehydrogenase, aspartokinase, acetolactate synthase, phenylalanine hydroxylase, prephenate dehydratase and formyltetrahydrofolate deformylase. Recruitment of this domain by fusion with tyrA~p~appears to have occurred in a common ancestor of the large Bacillus/Staphylococcus/Listeria/Enterococcus/Streptococcusassemblage (Fig. [2](#F2){ref-type="fig"}). It is interesting that B. subtilisalso possesses a gene encoding a free-standing ACT domain in its genome (incorrectly annotated as pheB). An additional fusion of genes encoding an ACT domain and tyrA(that arose independently, judging from the widely spaced tree positions) occurred in the common ancestor of Xanthomonasand Xylella. Actinobacteriausually possess a C-terminal extension that probably functions as an allosteric domain. The extension possessed by the Actinobacteridae\_2 assemblage, which includes Streptomyces coelicolorand its relatives, appears to be an ACT domain. On the other hand, it is not all all clear that the C-terminal extension of the Actinobacteridae\_2 assemblage is an ACT domain. This difference, in addition to the differing specificities for pyridine nucleotide substrate, may have contributed to the overall TyrA~a~divergence observed between the two Actinobacteridaegroups. There is no correlation between presence of the ACT domain and specificity for cyclohexadienyl substrate since TyrA~p~from the Bacillusclade is PPA-specific, Xanthomonas/XylellaTyrA~c~is broadly specific, and StreptomycesTyrA~a~is AGN-specific. B. subtilis, which belongs to the large clade having an ACT domain as a carboxy extension, has been extensively characterized \[[@B30]\]. 4-Hydroxyphenylpyruvate is an effective competitive inhibitor, as would be consistent with our proposed effects at the catalytic core for a PPA-specific enzyme. However, TYR, phenylalanine (PHE) and tryptophan were also inhibitors. The violation of the rule that the latter three amino acid inhibitors would not be expected to bind the catalytic core region (because they have alanyl sidechains even though the substrate-binding site only recognizes the pyruvyl sidechain of prephenate) and the finding that some of these were not competitive inhibitors can now be accounted for by the presence of the allosteric ACT domain. A carboxy extension shared by a number of Archaea(denoted \'REG\' in Fig. [2](#F2){ref-type="fig"}) is presumably a regulatory domain as well. This is consistent with the recent result of Porat et al. \[[@B45]\] that not only 4-hydroxyphenylpyruvate, but also TYR, inhibited prephenate dehydrogenase activity of Methanococcus maripaludis. The tyrA gene is a popular fusion partner ----------------------------------------- ### Fusion with aroQ tyrAmay be fused with a number of other catalytic domains, each of them relevant to aromatic biosynthesis (Fig. [2](#F2){ref-type="fig"}). aroQ(encoding chorismate mutase) is frequently fused with a number of other aromatic-pathway genes \[[@B46]\]. The lower-gamma Proteobacteria (enteric lineage) located at twelve o\'clock in Fig. [2](#F2){ref-type="fig"} possess an aroQ•tyrA~c\_Δ~fusion. The fusion physically links chorismate mutase (which forms PPA) with TyrA~c\_Δ~(which utilizes PPA). The two protein domains of AroQ•TyrA~c\_Δ~may have co-evolved to produce cooperative protein-protein interactions since physical separation of the domains evoked relatively low activities of both activities in E. coli\[[@B32]\]. Substantial comparative work shows that the aroQ•tyrA~c\_Δ~fusion has been stably maintained throughout the entire enteric lineage \[[@B47]\]. Exceptions in some genomes lacking this fusion altogether can be attributed to reductive evolutionary loss in pathogens (e.g., Haemophilus ducreyi) or endosymbionts (e.g., Buchnera aphidicola). An independent aroQ•tyrAfusion was generated in the common ancestor of Sulfolobus solfataricusand S. tokodaii(Fig. [2](#F2){ref-type="fig"}). Since the TyrA domain of Sulfolobusspecies lacks the indel structure of the TyrA~c\_Δ~class, it would be interesting to see whether physical separation of the two domains would yield evidence of independent function, in contrast to the results mentioned just above for E. coli. ### Fusion with aroF Secondly, tyrA~c~has been fused with aroFon at least two separate occasions in Bacteria. (The aroFgene encodes enolpyruvylshikimate-3-P synthase, the sixth enzyme in the common pathway of aromatic biosynthesis; see \[[@B5],[@B6]\] for nomenclature used.) One clade includes members of the upper-gamma Proteobacteria: P. aeruginosa, P. syringae, P. putida, P. stutzeri, P. fluorescensand Azotobacter vinelandii. It is interesting that P. syringaehas experienced a deletion of about 200 residues at the N-terminal region of the AroF domain. This has been coupled with the acquisition of a stand-alone aroFgene that is absent in other members of the clade. Interestingly, the latter AroF shows high identity only with AroF from Agrobacterium tumefaciens, an alpha proteobacterium. The A. tumefaciens aroF, in turn, is unique compared to its α-subdivision relatives, both in having divergent sequence and in being unlinked to cmkand rpsA. Thus, it seems likely that the incongruence of AroF belonging to both P. syringaeand A. tumefaciensreflects acquisition via LGT from some as yet unknown source. The disruption of the fused aroFdomain in P. syringaeis an unusual instance where the catalytic function of one fusion domain has become discarded while the function of the second domain has been retained. It is interesting to consider the possibility that the truncated remnant of the aroFfusion domain might be exploitable for use as an innovative source of a new regulatory domain. An additional fusion of tyrAwith aroFhas occurred independently within the beta Proteobacteria in the common ancestor of Burkholderia pseudomalleiand B. mallei. This has been very recent since the closely related B. fungorumand B. cepaciaorganisms lack the fusion. It has been suggested that presence of a given fusion may be useful for sorting out clades that diverged from a common ancestor, independent of other methods \[[@B48]\]. Different fusions offer the power of discriminating clades at various hierarchical levels, i.e., nested clades discriminated by nested gene fusions. The tyrA•aroFfusion occurred in the common ancestor of the clade that includes the upper-gamma Proteobacteria shown in Fig. [2](#F2){ref-type="fig"}. One can reasonably assume that relatively close upper-gamma organisms lacking the tyrA•aroFfusion diverged from the common ancestor of the fusion clade prior to the fusion event. Such would appear to be the case, for example, with Acidithiobacillus ferrooxidans, an outlying member of the upper-gamma Proteobacteria that lacks the fusion. It is reasonable to conclude that the fusion event must have pre-dated the differential specialization for the pyridine nucleotide cosubstrate that distinguishes Acinetobactersp. (NADP^+^-specific) from the large grouping of pseudomonads that are NAD^+^-specific. ### Fusion with hisH~b~ Thirdly, a single organism, Rhodobacter sphaeroides, possesses a hisH~b~•tyrAfusion that must have occurred very recently. hisH~b~encodes an aromatic aminotransferase that is closely related to (or sometimes even synonymous with) imidazole acetol phosphate aminotransferase \[[@B49]\]. The hisH~b~/tyrA/aroFlinkage group is part of a supraoperon in some gram-negative bacteria in which a relatively conserved, yet frequently shuffled gene order is observed \[[@B5],[@B6]\]. Hence, it is reasonable to assume that at the time just prior to fusion, hisH~b~, tyrAand aroFwere adjacent. Note that among the fusions currently known, hisH~b~and aroFare fused to the N-terminal and C-terminal ends of tyrA, respectively. It would be interesting to know the substrate specificity of the R. sphaeroidesTyrA domain. If it is AGN-specific the significance of hisH~b~presumably would be to transaminate PPA to form AGN, the substrate used by TyrA~a~(see Fig. [1](#F1){ref-type="fig"}). On the other hand, if the dehydrogenase is PPA-specific, the significance of the HisH~b~domain would be to transaminate the product of the TyrA~p~reaction. If the enzyme is a TyrA~c~enzyme (as is probable), then HisH~b~likely is competent to catalyze either of the foregoing reactions. ### Fusion with ACT The widespread ACT regulatory domain appears to have been acquired by independent fusions at least three separate times judging from the widely separated lineages that possess a TyrA•ACT fusion (Fig. [2](#F2){ref-type="fig"}). Xie et al. \[[@B5]\] initially noted homologous domains positioned at the N terminus of mammalian phenylalanine hydroxylase and at the C terminus of most microbial prephenate dehydratases. This domain is responsible for phenylalanine-mediated activation and phenylalanine-mediated inhibition of the hydroxylase and dehydratase enzymes, respectively. This domain was later named the ACT domain \[[@B50]\] and shown to be a widely distributed domain family that shares a conserved overall fold. Members of the ACT-domain family possess a wide variety of different ligand-binding capabilities. For example, the ACT domain of 3-phosphoglycerate dehydrogenase binds L-serine as a allosteric inhibitor. ### Fusion with REG Another putative regulatory domain fused to tyrA (denoted tyrA•REG) is thus far restricted to some of the Archaea. This domain is a predicted regulatory domain, as described in COG4937. ### A novel 4-domain fusion Archaeoglobus fulgidusexhibits a striking four-domain fusion consisting of three catalytic domains and a regulatory ACT domain (TyrA•AroQ•PheA•ACT). The TyrA domain is predicted to belong to the TyrA~c\_Δ~class when used as a query input into the AroPath Specificity Predictor Tool \[[@B40]\]. We speculated earlier that the •AroQ fusion domain of Archaeoglobusmay exercise cooperative interactions with TyrA~c\_Δ~, as appears to occur between the AroQ•TyrA~c\_Δ~domains of E. coliand its relatives. tyrAin its syntenic context ----------------------------- Although the genes of prokaryotes have clearly been subject to frequent scrambling, some gene-gene associations persist more tenaciously than others. Xie et al. \[[@B5],[@B6]\] asserted that one such ancestral gene string that has resisted scrambling forces is hisH~b~\> tyrA\> aroF. As suggested above, contemporary gene fusions can serve as frozen-in-time indicators of ancient gene organizations that were later obscured by gene-scrambling events. Another gene string that is often within the syntenic region of hisH~b~, tyrA, and aroFis cmk\> rpsA. Gene synteny in prokaryotes has not been easily recognized because substantial manual scrutiny in combination with a sufficient density of genomic representation on a given portion of the phylogenetic tree is necessary to detect patterns of synteny that are camouflaged by frequent scrambling events (inversion, deletion and transposition). The domain Bacteriais now represented by a collection of sequenced genomes that is progressively approaching the genomic densities needed for meaningful analysis. Figure [5](#F5){ref-type="fig"} provides a visual sense of the frequency with which tyrAis closely positioned with other genes of aromatic biosynthesis, as well as the underlying patterns of overall synteny. These patterns are unstable, and yet persistent traces of synteny can be seen where genomic representation is sufficiently dense. The four genes of particular emphasis in this paper are color coded. Other genes that are engaged in aromatic biosynthesis are colored grey, and any other genes are white. At a very deep level of phylogenetic branching, Thermotogaexhibits a tyrAgene flanked by seven genes encoding all of the common steps of aromatic biosynthesis (two of them being fused). Since closely related genomes are not yet available here, we cannot judge whether these genes came together recently or whether an ancient pattern of synteny has been retained. Although tyrAis not linked to any functionally relevant genes in Aquifex, representing another point of deep phylogenetic branching, this does not necessarily mean that tyrAwas not already generally associated with other aromatic-pathway genes at an early time. For reasons that are totally mysterious, certain scattered lineages exhibit a total lack of operon organization for aromatic-pathway genes (and indeed for most other biosynthetic pathways, such as that for histidine biosynthesis). These lineages (Fig. [5](#F5){ref-type="fig"}) include, besides Aquifex, those of Deinococcus, the actinomycetes, the cyanobacteria, and Chlorobium. Except for the actinomycetes, this phenomenon of total gene dispersal also applies to genes of tryptophan biosynthesis \[[@B7],[@B8]\]. When the various examples of hisH~b~\> tyrA\> aroFlinkage are mapped on a 16S rRNA tree, they first appear in gram-positive bacteria. In Bacillusand related organisms (such as Listeria), the hisH~b~\> tyrA\> aroFunit is associated with a large ancestral operon consisting of aroG\> aroB\> aroH\> hisH~b~\> tyrA~p~\> aroF. Bacillusadditionally possesses the cmk\> rpsAunit, albeit in a separate location. Interestingly, in one narrow subclade (B. subtilis, B. haloduransand B. stearothermophilus) the trpoperon has been inserted between aroHand hisH~b~to yield a supraoperon that has been fully characterized as a complex functional unit \[[@B51]\]. See Xie et al. \[[@B7]\] for a full presentation of evolutionary interpretation relevant to the latter. Though highly scrambled, a pattern of association of pheAwith hisH~b~\> tyrA\>aroFis suggested by linkage patterns seen at the hierarchical level of Cytophagaand Bacteroides(Fig. [5](#F5){ref-type="fig"}). aroQbecame associated with pheAthrough gene fusion as early as the divergence of the Spirochaetesto yield an aroQ•pheA\>tyrA\>aroF\>cmk\>rpsAlinkage unit (Leptospira interrogansin Fig. [5](#F5){ref-type="fig"}). The aroQ•pheAgene associated with tyrAand aroFin Clostridium difficileappears to have arisen from a distinctly different fusion event than that present in delta, epsilon, beta and upper-gamma Proteobacteria or from that present in lower-gamma Proteobacteria (based upon analysis of inter-domain linker regions; unpublished data). Consensus ancestral gene organizations for the most densely represented divisions of Proteobacteria have been deduced as shown at the bottom of Fig. [5](#F5){ref-type="fig"}. Detailed information that supports a deduced consensus for ancestral gene organizations with respect to beta Proteobacteria, upper-gamma Proteobacteria, and lower-gamma Proteobacteria are shown later (Figs. [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). We suggest that the last common ancestor of all Proteobacteria possessed the gene organization aroQ•pheA\>hisH~b~\>tyrA\>aroF\>cmk\>rpsA. This is similar to the synteny that has been retained in general by the beta Proteobacteria and the upper-gamma Proteobacteria. The aroQ•pheA\>hisH~b~\>tyrAportion likely specified all the catalytic requirements for conversion of chorismate to PHE and conversion of chorismate to TYR. Chorismate mutase activity specified by the aroQdomain could supply PPA for both PHE and TYR biosynthesis. Likewise, HisH~b~, widely utilized as an aromatic aminotransferase \[[@B49]\], could also function for both PHE and TYR biosynthesis. Though currently available members of delta and epsilon Proteobacteria exhibit substantial gene scrambling, the various fragmentary linkage patterns seen provide support for the ancestor proposed. Geobacter(and other Delta\_1 members) has the aroQ•pheA\> tyrA\> aroF\> cmk\> rpsAlinkage group (with lytBinserted between cmkand rpsA). Desulfovibrio vulgaris, another delta Proteobacterium (Delta\_2) that is highly divergent from Geobacter, has a very interesting pattern of conservation and scrambling. aroQ•pheA\> aroF\> tyrAhas been attached to a complete 7-gene trpoperon. hisH~b~\> cmk(not shown in Fig. [5](#F5){ref-type="fig"}) is completely separated from rpsA. The supraoperonic gene organization shown for D. vulgarisbegins with two recently discovered genes, herein denoted aroA\' and aroB\', that encode enzymes specifying an alternative biochemical route to dehydroquinate \[[@B52]\]. The epsilon Proteobacteria all display significant gene scrambling, but piecemeal evidence for the unscrambled ancestor proposed is present. For example, Campylobacter jejunipossesses an aroQ•pheA\> hisH~b~unit, as well as aroF \> lytB \> rpsA(Fig. [5](#F5){ref-type="fig"}). Wollinella succinogenesand Helicobacter hepaticusboth possesses an aroF\> lytB\> rpsAunit. The ancestor of alpha Proteobacteria has lost the aroQ•pheAfusion, and a stand-alone pheAis consistently observed. Members of this group are quite uniform in the stable possession of hisH~b~\> tyrAand aroF\> cmk\> rpsAas two separated linkage groups. The beta Proteobacteria are represented by members that have the gene organization: serC\> aroQ•pheA\> hisH~b~\> tyrA\> aroF\> cmk\> rpsA. This is also seen in the members of the upper-gamma Proteobacteria. Figure [5](#F5){ref-type="fig"} includes organisms that illustrate the traces of synteny that can be detected in Bacteriawhere overall genome representation is just barely adequate. The following two figures illustrate how syntenic patterns of more resolution and refinement become evident with denser genome representation. Zooming in on syntenic contexts of proteobacteria ------------------------------------------------- ### Beta proteobacteria and upper-gamma proteobacteria The beta Proteobacteria exhibit a dynamic but still interpretable pattern of altered synteny (Fig. [6](#F6){ref-type="fig"} and Table [5](#T5){ref-type="table"}). Species of Ralstoniahave retained the proposed ancestral synteny that is marked with yellow highlighting in Fig. [6](#F6){ref-type="fig"}. This syntenic organization is such that the aromatic-gene unit aroQ•pheA\> hisH~b~\> tyrA\> aroFis nested between gyrA\> serCat the leftward flank and cmk\> rpsA\> himDat the rightward flank. Species of Burkholderia(the next closest lineage) are almost identical, but exhibit individual evolutionary events (marked by circled numbers on the left, which correspond to a description of the proposed evolutionary events given in companion Table [5](#T5){ref-type="table"}). These events include gene insertion, loss of hisH~b~, translocation of genes away from the ancestral supraoperon, and fusion of tyrAand aroF(in the common ancestor of B. malleiand B. pseudomallei). At a deeper level in the beta Proteobacteria section of the tree, Nitrosomonas europaeaexhibits a separation of the ancestral supraoperon between tyrAand aroF. Either a very large insertion was made between tyrAand aroF, or one of the two gene clusters shown was transposed as part of a sufficiently large segment to include all of the conserved flanking genes. In Chromobacterium violaceum tyrAhas become completely isolated from other gene members of the ancestral supraoperon, and aroFhas assumed an inverted orientation with respect to cmk. Species of Neisseriaexhibit no remnants of supraoperon synteny at all, and wholesale dispersal of all the supraoperon genes has occurred. (It is interesting that among the beta Proteobacteria, Neisseriaspecies are also unique in that all of the trp-pathway genes are dispersed \[[@B7]\]). The gamma Proteobacteria have separated into two distinctly different synteny patterns. The lower-gamma Proteobacteria have undergone marked syntenic change (see below). The assemblage portrayed between Acidithiobacillusand Microbulbiferin the lower part of Fig. [6](#F6){ref-type="fig"} (termed the upper-gamma Proteobacteria) exhibit a strong overall syntenic resemblance of supraoperon genes to that of the beta Proteobacteria. Acidithiobacilluspossesses a near-intact ancestral supraoperon, differing only in having two insertions: one gene encoding 3-deoxy-D-arabino-heptulosonate 7-phosphate (DAHP) synthase between hisH~b~and tyrA, and the other being the insertion of serAbetween serCand aroQ•pheA. Pseudomonas aeruginosaand P. stutzerihave also retained nearly intact ancestral supraoperons, differing only in the fusion of tyrAand aroF. The serC\> aroQ•pheA\> hisH~b~\> tyrA•aroF\> cmk\> rpsAsupraoperon has been studied in P. stutzeri\[[@B5],[@B6]\]. The tyrA•aroFfusion occurred in the common ancestor of the clade shown between Azotobacterand Microbulbiferin Fig. [6](#F6){ref-type="fig"}. The supraoperons of P. syringae, P. fluorescensand P. putidalack hisH~b~. P. syringaeexhibits a recent C-terminal truncation of the aroFdomain, coupled with acquisition elsewhere in the genome of a free-standing •aroFthat is not phylogenetically congruent (probably of LGT origin). Acinetobactersp. and Microbulbifer degradanspossess an aroQ•pheA\> tyrA•aroFunit that has become dissociated from serCat one end and from cmkon the other end. In Xylellaand Xanthomonas, hisH~b~has been deleted from the genome and tyrAhas been transposed away from serC\> aroQ•pheA\> aroF. The latter unit has been transposed away from gyrA, the ancestral flanking gene. On the other hand, cmk\> rpsAhas remained next to himD, the gene usually flanking rpsA. ### The enteric lineage The lower-gamma Proteobacteria differ sharply from upper-gamma Proteobacteria in their possession of the tyrA~c\_Δ~class of tyrAand its fusion with aroQ. In Fig. [2](#F2){ref-type="fig"} this clade of AroQ•TyrA~c\_Δ~fusions was presented as one exhibiting absolute specificity for NAD^+^, combined with an overwhelming but not complete specificity for PPA. In Fig. [7](#F7){ref-type="fig"} the gene synteny associated with tyrA~c\_Δ~is profiled against the 16S rRNA phylogenetic trees of the lower-gamma Proteobacteria possessing these genes, and the proposed evolutionary events are summarized in the companion Table [6](#T6){ref-type="table"}. Figure [5](#F5){ref-type="fig"} has indicated a synteny consensus for the common ancestor at this hierarchical level whereby gyrA\> serC\> hisH~b~\> aroF\> cmk\> rpsAparallels the ancestral synteny of β-Proteobacteria, but without aroQ•pheAor tyrAin the middle of the linkage group. Many dynamic evolutionary events of altered aromatic biosynthesis have occurred within the lower-gamma Proteobacteria since their divergence from the upper-gamma Proteobacteria. This includes the emergence of three allosterically distinct DAHP synthases, one of which now comprises the two-gene, three-domain tyroperon (aroA~Iα\_Y~\> aroQ•tyrA~c\_Δ~). The upper-gamma Proteobacteria characteristically possess the aroA~Iα~paralogs encoding AroA~Iα\_H~(TRP-inhibited DAHP synthase) and AroA~Iα\_Y~(TYR-inhibited DAHP synthase). It has been asserted that AroA~Iα\_F~(PHE-inhibited DAHP synthase) was the most recent paralog, acquired just after divergence of the lower-gamma Proteobacteria \[[@B53]\]. It is bizarre that Shewanella oneidensispossesses a pseudogene of aroA~Iβ~fused to the C terminus of aroQ•pheA. The aroA~Iβ~subclass of Family-I DAHP synthases is not usually observed in gram-negative bacteria \[[@B54]\]. The dissociation of tyrA~c\_Δ~from the serC/rpsAlinkage group correlates with the fusion of aroQwith tyrA~c\_Δ~. The aroQ•pheAfusion has also escaped from the serC/rpsAlinkage grouping and has become linked with the newly emerged tyroperon. Some sort of duplication and recombinational event between aroQ•pheAand tyrA~c\_Δ~may have led to the creation of aroQ•tyrA~c\_Δ~since the AroQ•PheA proteins of lower-gamma Proteobacteria are distinct from AroQ•PheA proteins of other Proteobacteria with respect to the inter-domain linker length and the indel content (data not shown). Although it usually is absent from the lower-gamma Proteobacteria, HisH~b~has persisted as the broad-specificity aromatic aminotransferase in the Pasteurella/Haemophilusgrouping where two hisHparalogs are generally present, one of narrow specificity (denoted hisH~n~) being within the histidine operon. The aspCgene next to aroFin Shewanellais a paralog that probably functions as an aromatic aminotransferase, suggestive of the situation in the E. coligrouping where tyrBis a close paralog relative of aspC, tyrB having become specialized for aromatic biosynthesis \[[@B49]\]. Gene reduction associated with both endosymbiotic and pathogenic lifestyles are evident. Thus, Buchneralacks tyrA, cmk, hisH, tyrB, and possesses only a single aroA~Iα~species (aroA~Iα\_H~). Haemophilus ducreyialso lacks tyrA, as well as aroA~Iα\_H~and the entire trpoperon \[[@B5]\]. TyrA in its context of regulation --------------------------------- ### TyrR regulon Knowledge of the gene regulation impacting TyrA in prokaryotes is sparse, being limited to the lower-gamma Proteobacteria. Here, extensive information gathered from E. colihas revealed that aroQ•tyrA~c\_Δ~belongs to a large regulon controlled by the TyrR repressor. The limited phylogenetic distribution of TyrR, being present only in the lower-gamma Proteobacteria (Fig. [8](#F8){ref-type="fig"}), indicates that it is a recent evolutionary acquisition. In E. colithe regulon members that are under the control of tyrRare the aroA~Iα\_Y~\> tyrAoperon, the aroLMoperon, tyrP, tyrB, aroP, mtr, aroA~Iα\_F~and tyrRitself \[[@B55]\]. Thus, tyrRnot only regulates the tyrosine branch of the pathway, but heavily impacts the common pathway and the transport of all three aromatic amino acids as well. Although outside the scope of this study, a logical expansion of it would be to examine the individual evolutionary histories of all the members of the contemporary E. coliTyrR regulon, i.e., asking when and in what order did these genes come under the influence of tyrR? Clearly, the recruitment of structural genes by tyrRhas been recent, quite dynamic and even now, exhibits evidence of further ongoing change. For example, tyrosine phenol-lyase (a catabolic enzyme that is only sparsely present in gamma Proteobacteria) has been recruited to the TyrR regulons of Erwinia herbicola\[[@B56]\] and Citrobacter freundii\[[@B57]\]. In these cases, not only does TyrR perform as a transcriptional activator, but it requires cyclic AMP receptor protein and integration host factor to do so. As exemplified by E. coli, TyrR is generally a repressor. However, the transcriptional expression of mtris activated by TyrR in the presence of TYR, and tyrPis activated in the presence of PHE (although it is repressed in the presence of TYR). The N-terminal domain of TyrR has been associated with the ability of TyrR to activate transcription in the case of mtrand tyrP\[[@B55]\]. Members of the Haemophilus/Pasteurellalineage have all lost the N-terminal domain and presumably all lack the ability to accomplish transcriptional activation, as has been demonstrated experimentally with H. influenzaeTyrR \[[@B58]\]. In view of the interesting complexity that two operons (mtrand aroLM) in E. coliare regulated by both tyrRand trpR\[[@B55]\], it may be more than coincidental that tyrRand trpRseem to have emerged at about the same evolutionary time, i.e., coincident with the divergence of the upper-gamma Proteobacteria from the lower-gamma Proteobacteria (Fig. [7](#F7){ref-type="fig"}). A possible interaction between the TyrR and TrpR proteins has been noted \[[@B55]\]. ### PhhR in relationship to aromatic catabolism Arias-Barrau et al. \[[@B59]\] have recently characterized a central catabolic pathway (Hmg) that degrades homogentisate in three steps to fumarate and acetoacetate as a source of carbon and energy. One of several peripheral pathways feeding into the central pathway begins with PHE and produces homogentisate via the reactions of phenylalanine hydroxylase (Phh), aromatic aminotransferase, and 4-hydroxyphenylpyruvate dioxygenase (Hpd). In the absence of Phh, a shorter version of the peripheral pathway is one that can use TYR, but not PHE, as a source of carbon and energy. In Fig. [8](#F8){ref-type="fig"} the presence of Phh, Hpd, and Hmg segments of catabolism are mapped on a 16S rRNA tree. (The aromatic aminotransferase distribution is not shown since a multiplicity of aromatic aminotransferases having overlapping substrate specificities makes it particularly challenging to identify the functional role \[[@B49]\].) The cyanobacteria are unique among Bacteriain the use of Hpd for a completely different metabolic role unrelated to aromatic catabolism, i.e., the synthesis of vitamin E derivatives \[[@B60]\]. PhhR is a homolog of TyrR that has been shown in P. aeruginosato be a divergently transcribed activator of a 3-gene operon needed for PHE and TYR catabolism \[[@B61]\]. The structural genes encode phenylalanine hydroxylase (phhA), carbinolamine dehydratase (phhB) and 4-hydroxyphenylpyruvate aminotransferase (phhC), and are powered by a σ^54^promoter \[[@B61],[@B62]\]. PhhR evolved relatively recently since it is only present in some gamma Proteobacteria (Fig. [8](#F8){ref-type="fig"}). The ancestral regulatory gene for the Phh peripheral pathway may have been a member of the leucine-responsive regulatory protein/asparagine synthase C (Lrp/AsnC) family judging from the adjacent and divergently oriented position of asnCgenes to phhAin organisms such as Xanthomonas axonopodisand Mesorhizobium loti. A recent overview of the many different regulator families involved in the control of aromatic catabolism conveys an emerging sense of the variety and dynamic evolutionary processes that underlie aromatic catabolism \[[@B63]\]. Occasional distant homologs of phhRthat appear in erratic fashion (see Fig. [9](#F9){ref-type="fig"}) may have some other regulatory function. For example, Clostridium tetanimay use its PhhR homolog as a transcriptional activator of the gene encoding tyrosine phenol-lyase, as occurs in species of Erwinia\[[@B56]\] and Citrobacter \[[@B57]\]. ### Relationship of TyrR and PhhR What might be of origin of TyrR? TyrR is an anomalous member of the large prokaryote family of σ^54^enhancer-binding proteins that activate promoters dependent upon σ^54^. TyrR is unique within its homology grouping in that it targets σ^70^promoters for regulation, usually (but not always) being a repressor. Its closest homolog relative is PhhR, a canonical member of σ^54^enhancer-binding proteins. σ^54^-dependent enhancer proteins possess a highly conserved σ^54^-contact motif, GAFTGA, that is intimately involved in formation of the ternary complex of enhancer and σ^54^-RNA polymerase holoenzyme \[[@B64]\]. This is perfectly or nearly perfectly retained in the upper clades shown in Fig. [9](#F9){ref-type="fig"}, but is disrupted or completely absent in the clades between Shewanella oneidensisand Pasteurella multocida. The deeper phylogenetic distribution of PhhR (Fig. [8](#F8){ref-type="fig"}) suggests that TyrR evolved as a variant of PhhR. If correct, a regulatory gene that is oriented to catabolism (phhR), and itself of relatively recent origin, was conscripted even more recently for a completely new role in the regulation of primary biosynthesis (tyrR). Consistent with the latter supposition, the gain of TyrR generally correlates with the loss of competence for aromatic catabolism (Fig. [8](#F8){ref-type="fig"}). In contrast to the Citrobacter/Salmonella/Escherichia/Shigellaand the Pasteurella/Haemophilusclades (whose TyrR homologs completely lack the GAFTGA motif), the remaining enteric clades have retained some residues in this region. These residues appear to be more than random remnants. It would be interesting to know if these residues have any functional significance. Indeed, the Photobacterium/Vibrioclade has retained the ancestral catabolic capabilities (Fig. [8](#F8){ref-type="fig"}) that would appear to demand retention of regulation via PhhR; yet the parallelism of the overall features of biosynthesis that are shared with other lower-gamma Proteobacteria would seem, on the other hand, to demand TyrR-mediated regulation. Perhaps this \"TyrR\" species participates in the regulation of both catabolic and biosynthetic genes. In this connection, it is interesting that Chaney et al. \[[@B64]\] found that a change in the GAFTGA motif of NifA could be partially \"suppressed\" by mutational changes in the N-terminal region of σ^54^. Even more striking as a possible evolutionary intermediate is the most outlying member of the lower- gamma Proteobacteria, Shewanella oneidensis. The position of its TyrR on the protein tree parallels expectations based on the 16S rRNA tree. This, plus the conservation of the TyrR regulon features and the overall gene synteny suggest E. coli-like function as TyrR, i.e. acting as a general repressor of regulon-member σ^70^promoters engaged in aromatic biosynthesis. However, the location of \"tyrR\" in S. oneidensisbetween phhAand phhBon one side, and hmgBand hmgCon the other side, strongly implies some kind of regulatory relationship with the catabolic genes. It would be quite interesting to determine experimentally whether \"TyrR\" in S. oneidensis(and maybe Vibrio, as well) can function as a repressor of the usual suite of σ^70^promoters, as well as an activator of σ^54^promoters for phhA/phhBand/or hmgB/hmgC. We suggest that TyrR evolved as a modified version of PhhR as follows. In view of the distribution of genes encoding PhhR and TyrR, as well as the aforementioned catabolic enzymes, the most parsimonious evolutionary scenario may be that central and peripheral catabolic pathways depicted in Fig. [8](#F8){ref-type="fig"} are quite ancient, but acquisition of PhhR as a σ^54^-dependent activator of phenylalanine hydroxylase was quite recent, originating about the time of divergence of gamma Proteobacteria. The clade defined by Shewanella/Vibrio/Photobacteriumretained the catabolic pathway, whereas the other enteric lineages discarded the catabolic pathway, but retained PhhR, which was then recruited as a σ^70^-dependent regulator of aromatic biosynthesis (TyrR). ### Regulation by attenuation A widespread mechanism of regulation is via an attenuation mechanism whereby transcripts initiated at given promoters can be terminated prior to reaching the structural genes of an operon. Whether termination occurs usually depends on the balance (dictated by a variety of mechanisms) between mutually exclusive terminator and anti-terminator structures \[[@B65]\]. Merino has developed a website \[[@B66]\] to provide a database of putative attenuators ahead of various operons in Bacteria. We screened this database for likely attenuators relevant to the regulation of tyrA. Table [7](#T7){ref-type="table"} shows intriguing results that point to significant experimental work that would be desirable. tyrAis frequently a member of apparent supraoperons, as alluded to elsewhere in this paper, and some of these appear to be large gene clusters controlled by attenuation. Substantial work is needed to establish the depth of clades possessing a given attenuator. For example, the hisH~b~\> tyrAoperon is reliably present throughout all alpha Proteobacteria. Since Agrobacterium tumefacienshas been found to possess an attenuator ahead of the hisH~b~\> tyrAoperon, one might reasonably expect that most of the alpha Proteobacteria would possess the attenuator as well. If not, this attenuator would have been a very recent evolutionary innovation. Likewise, since the aroA~Iα\_Y~\> tyrAoperon is widely present throughout the lower-gamma Proteobacteria, it would be interesting to confirm whether only the several species of Vibrioidentified on the Merino website have an attenuator ahead of this operon (or whether other attenuators present are too weak to exceed the threshold imposed for preliminary detection). Some of the supraoperons that appear to be controlled by attenuation are interesting in that they contain the majority of genes needed for both PHE and TYR biosynthesis, e.g., the supraoperons in Enterococcus faecalisand Streptococcus pneumoniae. The latter organism displays two attenuator units. The supraoperon of Desulfovibrio vulgarisis novel in that it begins with two relatively rare genes encoding alternative enzyme steps for aromatic biosynthesis \[[@B52]\], denoted here as aroA\' and aroB\'. The leading five genes are adjacent to the seven-gene trpoperon. Conclusion ========== Protein divergence within a vertical genealogy is not necessarily smooth and progressive. Qualitative biochemical innovations can result in a barrage of new selective pressures that result in evolutionary jumps. The consequent incongruence might easily be mistaken for LGT. The basis for evolutionary jumps will usually only be recognized by detailed and comprehensive analyses of any given subsystem. Examples in this study are as follows. (i) The tyrA~c\_Δ~gene of the lower-gamma Proteobacteria has diverged markedly from tyrA~c~of the upper-gamma Proteobacteria. Here the milestone event was fusion of aroQto a putative tyrA~c~in the ancestor of lower-gamma Proteobacteria to produce aroQ•tyrA~c\_Δ~. Indels within the •tyrA~c\_Δ~domain presumably reflect a multiplicity of selections for functional interactions known to exist between the two fused domains as discussed earlier. (ii) Members of the subclass taxon Actinobacteridaepossess TyrA~a~proteins that separate into two distinct groupings. The presumed ancestral ~NAD(p)~TyrA~a~that is still present in the Actinobacteridae\_1 clade very likely spawned the divergent NAD^+^-specific variety of TyrA~a~to yield the contemporary Actinobacteridae\_2 clade. The previous evolutionary analysis of trp-pathway genes \[[@B7],[@B8]\] can be viewed as a model for comparable studies with other gene systems. Expansion to the greater aromatic pathway is a logical extension. The dynamics of evolutionary change for tyrAcan be matched to the dynamics exhibited by the trpsystem. For example, the lower-gamma Proteobacteria separate as a distinct phylogenetic unit from beta Proteobacteria and upper-gamma Proteobacteria on criteria defined by milestone evolutionary events that altered many character states of both tryptophan and tyrosine biosynthesis in the lower-gamma Proteobacteria. In the future one can anticipate that comprehensive and systematic phylogenetic analysis of each protein member of the TYR, PHE and TRP branches, the common aromatic-pathway trunk, and minor vitamin-like branches (such as the 4-aminobenzoate/folate branch) will accommodate a progressively integrated picture of the entire aromatic network, including catabolic pathways and many other specialized pathways. Methods ======= TyrA sequences -------------- Most TyrA sequences were obtained from the National Center for Biotechnology Information (NCBI) \[[@B16]\]. TyrA sequences from incomplete genomes were retrieved from the PEDANT database \[[@B67]\]. Several sequences in our curated TyrA collection have been corrected for incorrect translation start sites. Various curated TyrA sequence files can be downloaded from our website. These files include complete sequences, trimmed catalytic-core domains, and amino-acid sequence segments that are relevant to specificity for pyridine nucleotide or to specificity for the cyclohexadienyl substrate. The sequence files are summarized in Table [3](#T3){ref-type="table"}. Congruency groupings -------------------- TyrA proteins that cluster together on the TyrA protein tree in congruence with the 16S rRNA tree are called congruency groups. Exact correspondence of branching orders is not necessarily observed. So far, congruency groupings have been assembled for tryptophan-protein concatenates \[[@B8]\] and for TyrA proteins. Completion of equivalent work with the remaining aromatic-pathway segments will identify the repertoire of bacterial organisms in possession of a \"pure\" vertical genealogy with respect to aromatic biosynthesis. Congruency groups for TyrA can be accessed at our AroPath website \[[@B9]\], where a listing of the membership of congruency groups is maintained and updated. Any members of congruency-group clusters, whose position there is incongruent with 16S rRNA expectations, probably (but not necessarily) originated by LGT. The donor lineage may not be obvious, but as more genomes come on line, many cases where donor identities are currently unknown may become revealed. A listing of \"orphan\" TyrA proteins that belong to no current congruency group is given. Such orphans reflect the lack of sufficient genome representation in particular phylogenetic regions and undoubtedly will become the nucleus for additional congruency groups in due course. Alignments ---------- Multiple alignments were obtained by use of the ClustalW or ClustalX programs (Version 1.83) \[[@B68]\]. Manual adjustments were needed in the region of the GxGxxG motif for binding of pyridine nucleotide cofactor in the N-terminal region of TyrA proteins. Guidance for alignment was assisted by maximizing conformation with the Wierenga fingerprint, making allowance for a variable loop of 2--5 residues \[[@B26]\]. This was done with the assistance of the BioEdit multiple alignment tool of Hall (5.0.9 Edition) \[[@B69]\]. The refined multiple alignment was used as input for generation of a phylogenetic tree using the phylogeny inference package (Version 3.2), PHYLIP \[[@B70]\]. The neighbor-joining program was used to obtain a distance-based tree. The distance matrix was obtained by use of Protdist with a Dayhoff Pam matrix. The Seqboot and Consense programs were then applied to assess the statistical support of the tree using bootstrap resampling (1,000 replications). We also used the ANCESCON package \[[@B71]\], which produced similar results as shown in Fig. [2](#F2){ref-type="fig"} (albeit with even wider separation of many groups). The presence of regulatory domains (ACT and REG) was accepted when indicated by the Domain Architecture Retrieval Tool (DART) on the BLAST menu at NCBI \[[@B16]\]. Profile HMMs ------------ Profile hidden Markov models for each of the four TyrA subfamilies, TyrA~a~, TyrA~c~, TyrA~p~and tyrA~c\_Δ~, were built using Sean Eddy\'s HMMER package \[[@B72]\]. The HMMs were generated from our file of curated cyclohexadienyl-substrate core segments (see Table [3](#T3){ref-type="table"}). The seed sequences for each subfamily were first aligned using ClustalW \[[@B68]\]. The resulting multiple sequence alignments were then manually edited to produce more accurate alignment of the seed sequences. Finally, the edited multiple sequence alignments were used to generate the profile HMMs for each TyrA subfamily. Appraisal of gene fusions as one-time or multiple events -------------------------------------------------------- Whether any given contemporary gene fusions tracked back to a fusion event in a common ancestor or whether they occurred independently was evaluated by phylogenetic analysis of the individual protein domains and by inspection of the inter-domain linker region. Linker regions were determined by multiple alignments of fusion sequences with corresponding free-standing domains present in the closest relatives to organisms that lack the gene fusions. Authors\' contributions ======================= JS and MW integrated this specific effort with the broader and ongoing objective of implementing a dynamic and progressively updateable website (AroPath). JS also made substantial contributions to the bioinformatic work. CB did all of the art work and a majority of the bioinformatic analyses. RJ provided initial guiding concepts, a general organizational overview, and assembled the initial manuscript draft. CB, RJ, and JS contributed to the formulation of conclusions made, and all of the authors read and approved the final version of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table S1, entitled \"Key to organism acronyms and sequence identifiers\", is provided as supplementary material in an html document. This table contains the full collection of sequence data and annotations contained in this paper, and gene identification (gi) numbers are included and hyperlinked to facilitate access to the corresponding GenBank records. For future reference to a progressively updated table, refer to the AroPath website \[[@B73]\]. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ R. Jensen thanks the National Library of Medicine (Grant G13 LM008297) for partial support. This research is partially supported by the U. S. Army Research Institute of Infectious Diseases (USAMRIID). This analysis would not have been possible were it not for the yeoman efforts in comparative enzymology carried out over a period of more than 25 years by many students and postdoctoral fellows, most notably Graham S. Byng, Robert Whitaker, Alan X. Berry and Suhail Ahmad. This has produced an invaluable resource of comprehensive data, some of it unpublished. This paper is dedicated to our colleague and collaborator, John E. Gander, on the occasion of his 80^th^birthday. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Composite of alternative biochemical routes from chorismate (CHA) to L-tyrosine (TYR) in nature. An antibiotic synthesis branch from CHA is also shown (dimmed). Here the intermediates shown to intervene between chorismate and pristinamycin or chloramphenicol are p-aminochorismate (ADC), p-aminoprephenate (ADP), p-aminophenylpyruvate (APP), and p-aminophenylalanine (APA). PPA may be transaminated by prephenate aminotransferase (PAT) to yield L-arogenate (AGN). The four TyrA homologs and the reactions they catalyze are colored differently. Arogenate dehydrogenase (TyrA~a~) converts AGN to TYR. Alternatively, prephenate dehydrogenase (TyrA~p~) converts PPA to 4-hydroxyphenylpyruvate (HPP) which is then transaminated to TYR via an homolog of TyrB, AspC, HisH, or Tat \[49\]. A broad-specificity cyclohexadienyl dehydrogenase (TyrA~c~) is competent to catalyze either the TyrA~a~or the TyrA~p~reaction. PapC converts the 4-amino analog of PPA to the 4-amino analog of HPP. AroQ, AroH, and AroR are distinct homologs known to exist in nature for performance of the chorismate mutase reaction. Other abbreviations: AA, amino acid donor, KA, keto-acid accepter. ::: ![](1741-7007-3-13-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Phylogenetic tree for trimmed core domains of selected members of the TyrA Superfamily. Acronyms used for the various organisms are given in alphabetical order in Table 2. (A more extensive listing that includes organisms not shown in Fig. 2 and which also is hyperlinked to all of the individual GenBank records is given in Table S1. A similar table that also includes compilation of known and predicted substrate specificities is maintained at AroPath \[73\]. Lineages possessing experimentally established TyrA~a~, TyrA~p~, TyrA~c~or TyrA~c\_Δ~proteins are indicated by fill colors specified in the legend. Three specificity patterns for the pyridine nucleotide substrate are shown by line colors (see figure box). Although the cyanobacteria are depicted as having NADP^+^-specific TyrA proteins, some of them can also accept NAD, albeit to a lesser degree. All proteins having an aspartate residue homologous to D-32 of the E. coli~NAD~TyrA~c\_Δ~domain are presumed specific for NAD^+^. Fusion of TyrA domains with other catalytic domains is indicated within grey boxes (AroQ•TyrA, TyrA•AroF, HisH~b~•TyrA, and TyrA•AroQ•PheA•ACT) using the convention of a bullet to represent the interdomain area. The boxes overlap any relevant lineages. TyrA proteins having carboxy-terminal fusions with regulatory domains (TyrA•ACT and TyrA•REG) are also shown. The distance scale bar at the bottom left represents substitutions per site. ::: ![](1741-7007-3-13-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Multiple alignment of the HMM consensus sequences obtained for different substrate-specificity groupings within cyclohexadienyl-substrate core segments (see Table 3). Invariant anchor residues are highlighted in yellow, conserved residues in grey. These consensus sequences will change continuously as corrections and refinements are made. The version shown was current as of April, 2005. ::: ![](1741-7007-3-13-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Alignment of the N-terminal glycine-rich P-loop of TyrA•ACT proteins from the Class Actinobacteria. These are specific for L-arogenate as substrate, but fall into two groups with respect to the pyridine nucleotide co-substrate. The top NAD^+^-specific group possesses an aspartate (D) at position 32 (E. colinumbering), whereas the bottom NAD^+^/NADP^+^group possesses an asparagine at the homologous position. Residue numbers are shown at the left. The species in the middle are color coded to match the hierarchical taxon positions obtained from NCBI. The variable loop of the Wierenga fingerprint \[26\], which in E. colicontains five residues (22--26), contains the minimal two residues in all of the Actinobacteriashown. The organisms on the right are color coded according to the taxonomic position indicated on the left (NCBI). The Rubrobacter xylanophilusTyrA~a~sequence is an orphan in the tree displayed in Fig. 2, as consistent with its outlying position in the taxonomy scheme. ::: ![](1741-7007-3-13-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Context of gene organization for tyrA, profiled against the 16S rRNA tree of the domain Bacteria. pheA, hisH~b~, tyrA, and aroFare color coded. Lineages typified by complete dispersal of aromatic-pathway genes are indicated by \"GENE DISPERSAL\". Gmet refers to Geobacter metallireducens; Dace refers to Desulfuromonas acetoxidans; and Ddes refers to Desulfovibrio desulfuricans. Consensus gene organizations are shown for the alpha and beta divisions of the Proteobacteria. The gamma division is subdivided to yield consensus gene organizations for the upper- and lower-gamma (enteric lineage) organisms. Genes that are adjacent and share a common transcriptional direction appear to reside in operons (or supraoperons). Any white spacing indicates substantial separation of the gene clusters shown in the genome. Genes of special interest are color coded, other genes of aromatic biosynthesis are shown in gray and all other genes are shown in white. ::: ![](1741-7007-3-13-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Zoom-in from Fig. 5 showing tyrAsynteny for the beta Proteobacteria and the upper-gamma Proteobacteria. The tree shown, based upon 16S rRNA sequences of the indicated organisms, indicates correct branching orders, but (to facilitate presentation) is not strictly correct in proportion. Circled numbers (in violet) indicate deduced evolutionary events for the beta Proteobacteria (see top of Table 5), whereas circled numbers (in pink; see bottom of Table 5) correspond to deduced evolutionary events for the upper-gamma Proteobacteria. Gene organizations of organisms indicated are shown on the right. The dotted outlining of some gene boxes in Coxiella burnetiiand in P. syringaeindicates pseudogene status. ::: ![](1741-7007-3-13-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Zoom-in from Fig. 5 showing tyrAsynteny for the lower-gamma Proteobacteria (enteric lineage). Deduced phylogenetic events numbered on the left are described in Table 6. The branching position for Buchnerais as suggested in ref. \[7\]. Dotted horizontal lines near the top of the tree indicate branch lengths that were shortened for convenience of presentation. Dotted outlining of boxes around some genes indicates their pseudogene status. It is unknown if the various open reading frame (ORF) insertions are functional. ::: ![](1741-7007-3-13-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Distribution of modules of aromatic catabolism mapped on a 16S rRNA tree. In this figure, only presence or absence (not gene order) is indicated. The Phh module (orange) consists of phenylalanine hydroxylase (PhhA), carbinolamine dehydratase (PhhB), and tyrosine aminotransferase (not shown, see Text), and accomplishes the overall conversion of PHE to 4-hydroxyphenylpyruvate. The Hpd module (yellow) is 4-hydroxyphenylpyruvate dioxygenase, which converts 4-hydroxyphenylpyruvate to homogentisate. The Hmg module (blue) catalyzes the 3-step conversion of homogentisate to acetoacetate and fumarate. The distribution of PhhR and TyrR is shown in boxes. In some cases the HmgC member is shaded light blue to indicate that the gene encoding this isomerase could not be found and is probably encoded by an as yet unknown analog. Some long branches are drawn with gaps that represent 25% of the length of the scale bar. ::: ![](1741-7007-3-13-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Protein tree of TyrR homologs. Nodes supported by bootstrap values of 998 or more are marked with solid circles, and the bootstrap values for nodes internal to these are shown. Generic names relevant to the organism abbreviations can be viewed in Fig. 8. A conserved region containing the σ^54^contact motif GAFTGAis highlighted as an orange band. \"Imperfect\" residues in this region are shown in lower-case fonts. Residue numbers are shown at the right. TyrR and PhhR are regulators of σ^70^and σ^54^promoters, respectively. Four σ^54^proteins of unknown function have very long branches, and to facilitate the visual presentation, the gaps in branch continuity shown represent a scale-bar distance of 0.1. Clades possessing σ^54^regulators are indicated with blue stripes, and σ^70^regulators are indicated with green stripes. ::: ![](1741-7007-3-13-9) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Abbreviations used to designate substrate specificities of tyrA/TyrA homologs ::: Abbreviation^a^ ------------------------- ----------------- -------------------------------------------------------------------------------------------------------------------------- Gene Gene Product Description of specificity^b^ tyrA~x~ TyrA~x~ Specificity for cyclohexadienyl substrate is unknown tyrA~c~ TyrA~c~ Broad-specificity cyclohexadienyl dehydrogenase (CDH) tyrA~p~ TyrA~p~ Narrow-specificity prephenate dehydrogenase (PDH) TyrA~c\_Δ~ TyrA~c\_Δ~ Broad-specificity cyclohexadienyl dehydrogenase having catalytic-core indels in correlation with an extra-core extension tyrA~a~ TyrA~a~ Narrow-specificity arogenate dehydrogenase (ADH) ~NAD~tyrA~a~ ~NAD~TyrA~a~ TyrA homolog is AGN-specific and NAD^+^-specific ~NADP~tyrA~a~ ~NADP~TyrA~a~ TyrA homolog is AGN-specific and NADP^+^-specific ~NAD(P)~tyrA~a~ ~NAD(P)~TyrA~a~ TyrA homolog is AGN-specific but utilizes either NAD^+^or NADP^+^ ~x~tyrA~x~ ~x~TyrA~x~ Specificity for both the cyclohexadienyl and pyridine nucleotide substrates is unknown ^a^Abbreviations in the upper-table (upper 5 rows) indicate the specificities for the cyclohexadienyl substrate. Abbreviations in the lower-table (lower 4 rows) indicate specificities for both cyclohexadienyl (right subscripts) and pyridine nucleotide substrates (left subscripts). Combinations not shown can be deduced from the examples given, e.g., a TyrA homolog specific for prephenate and NAD^+^would be designated ~NAD~TyrA~p~. ^b^The abbreviations CDH, PDH, and ADH (shown parenthetically) have been used frequently in the literature. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Key to organism acronyms ::: ------------------------------------------------------------------------------------------------------- Organism Abbreviation\ Abbreviation\ in Paper on website^b^ --------------------------------------------------------------- ------------------- ------------------- Acidithiobacillus ferrooxidansATCC 23270 Aferr Acinetobactersp. ADP1 ACIN Actinobacillus actinomycetemcomitansHK1651 Aact Actinomyces naeslundiiMG1 Anae Actinoplanes teichomyceticus Atei Agrobacterium tumefaciensstrain C58 Atum Amycolatopsis balhimycina Abal Amycolatopsis orientalis Aori Anabaenasp. PCC 7120 ANAB Arabidopsis thaliana Atha Archaeoglobus fulgidusDSM 4304 Aful Aful\_1 Azotobacter vinelandii Avin Avin\_1 Bacillis anthracisstr. A2012 Bant Bacillus cereusATCC 14579 Bcer Bacillus haloduransC-125 Bhal Bhal\_2 Bacillus stearothermophilus Bste Bacillis subtilis Bsub Bacillus thuringiensisisraelensis Bthu Bifidobacterium longumNCC2705 Blon Blon\_1 Blochmannia floridanus Bflo Bordetella bronchisepticus Bbro Burkholderia cepaciaJ2315 Bcep Burkholderia fungorumLB400 Bfun Burkholderia malleiATCC 23344 Bmal Burkholderia pseudomalleiK96243 Bpse Bpse\_6 Campylobacter jejuni Cjej Chromobacterium violaceumATCC 12472 Cvio Corynebacterium diphtheriaeNCTC 13129 Cdip Corynebacterium efficiensYS-314 Ceff Corynebacterium glutamicumATCC 13032 Cglu Cglu\_1 Desulfovibrio desulfuricansG20 Ddes Desulfovibrio vulgarissubsp. vulgarisstrain Hildenborough Dvul Desulfuromonas acetoxidans Dace Dace\_5 Enterococcus faecalisV583 Efae\_2 Enterococcus faecium Efae\_1 Efae\_1 Erwinia carotovoasubsp.atrosepticaSCRI1043 Ecar Escherichia coliK12 Ecol Geobacter metallireducensGS-15 Gmet Geobacter sulfurreducensPCA Gsul Gloeobacter violaceusPCC 7421 Gvio Haemophilus influenzaeRd KW20 Hinf Helicobacter hepaticusATCC 51449 Hhep Helicobacter pylori26695 Hpyl Klebsiella pneumoniaesubsp. pneumoniaeMGH 78578 Kpne Leifsonia xylisubsp. Xylistrain CTCB07 Lxyl Listeria innocuaClip 11262 Linn Listeria monocytogenesEGD-e Lmon Lotus corniculatusvar. japonicus Lcor Lcor\_3 Lycopersicon esculentum Lesc Methanococcus jannaschii Mjan Methanopyrus kandleriAV19 Mkan Mkan\_1 Methanosarcina barkeristrain Fusaro Mbar Methanothermobacter thermoautotrophicusstrain Delta H Mthe Mthe\_7 Microbulbifer degradans2--40 Mdeg Mycobacterium aviumsubsp. paratuberculosisstrain k10 Mavi Mycobacterium bovisTrEMBL Mbov Mbov\_2 Mycobacterium lepraeTN Mlep Mycobacterium tuberculosisCDC1551 Mtub Myxococcus xanthusDK 1622 Mxan Neisseria gonorrhoeaeFA 1090 Ngon Nitrosomonas europaeaATCC 19718 Neur Nocardia farcinicaIFM 10152 Nfar Nonomuraeasp. NONO Nostoc punctiformePCC73102 Npun Npun\_1 Novosphingomonas aromaticivoransDSM 12444 Naro Oceanobacillus iheyensisTHE831 Oihe Oryza sativassp. japonica Osat Pantoea agglomerans Pagg Pasteurella multocidasubsp. multocidastrain Pm70 Pmul Photorhabdus luminescenssubsp. laumondiiTT01 Plum Prochlorococcus marinussubsp. pastorisstrain CCMP1378 Pmar\_1 Pmar\_3 Prochlorococcus marinusMIT9313 Pmar\_2 Pmar\_10 Propionibacterium acnesKPA171202 Pacn Pseudomonas aeruginosaPAO1 Paer Paer\_1 Pseudomonas fluorescensPfO-1 Pflu Pseudomonas putidaKT2440 Pput Pseudomonas stutzeri Pstu Ralstonia eutrophaJMP134 Reut Ralstonia solanacearumGMI1000 Rsol Rhodobacter capsulatus Rcap Rhodobacter sphaeroides2.4.1 Rsph Rhodopseudomonas palustrisCGA009 Rpal Rhodospirillum rubrum Rrub Rrub\_1 Rubrobacter xylanophilusDSM 9941 Rxyl Saccharomyces cerevisiae Scer Salmonella typhimuriumLT2 Styp Styp\_1 Schizosaccharomyces pombe Spom Shewanella oneidensisMR-1 Sone Shewanella putrifacians Spu Staphylococcus aureussubsp. Aureus MW2 Saur Saur\_2 Streptococcus gordoniistr. Challis Sgor Streptococcus pneumoniaeR6 Spne Streptomyces avermitilisMA-4680 Save Streptomyces caeruleus Scae Scae\_2 Streptomyces coelicolorA3(2) Scoe Scoe\_1 Streptomyces lavendulae Slav Streptomyces pristinaespiralis Spri Streptomyces roseochromogenessubsp. Oscitans Sros Sros\_1 Streptomyces toyocaensisstrain 7 Stoy Sulfolobus solfataricusP2 Ssol Sulfolobus tokodaiistrain 7 Stok Synechococcussp. WH8102 SYNE\_1 SYNE\_1 Synechococcussp. PCC7002 SYNE\_2 Synechocystissp. PCC6803 SYNE\_3 SYNE\_3 Thermobifida fusca Tfus Thermosynechococcus elongatesBP-1 Telo Trichodesmium erythraeumIMS101 Tery Tropheryma whippleiTW08/27 Twhi Vibrio choleraeO1 biovar eltor strain N16961 Vcho Vibrio parahaemolyticusRIMD 2210633 Vpar Wolinella succinogenesDSM 1740 Wsu Xanthomonas campestrispv. campestrisstrain ATCC 33913 Xcam Xylella fastidiosa9a5c Xfas Yersinia enterocolitica(type 0:8) Yent Zymomonas mobilissubsp. mobilisZM4 Zmob ------------------------------------------------------------------------------------------------------- ^a^The system of acronym usage is: the first letter (capital) is the first letter of the genus followed by the first three letters (lower-case) of the species. If there is no species designation the first four letters of the genusare used (all in capitals). Redundant 4-letter acronyms are distinguished by unique following numbers. See \[74\] for a comprehensive listing with hyperlinks to the Taxonomy database records and the GenBank records at NCBI. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Curated TyrA amino-acid sequence files at AroPath \[35\] ::: ------------------------------------------------------ Complete TyrA sequences Catalytic-core domains^a^ Pyridine-nucleotide discriminator segments^b^ NAD^+^-specific NADP^+^-specific Broad specificity Cyclohexadienyl-substrate core segments Arogenate-specific (TyrA~a~) Prephenate-specific (TyrA~p~) Broad specificity TyrA~c~ TyrA~c\_Δ~ Pseudogene TyrA sequences ------------------------------------------------------ ^a^Trimmed free of N-terminal or C-terminal extensions, including any fusions with regulatory domains or other catalytic domains. ^b^High-glycine βαβ Rossmann fold at the N-terminus. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Cyclohexadienyl substrates and inhibitors of TyrA proteins possess identical sidechains ::: Organism Co-substrate Substrate(s) Inhibitor(s)^a^ Reference ------------------------------ ------------------ ------------------ --------------------- --------------- Synechocystissp. NADP^+^ AGN TYR \[25\] Arabidopsis thaliana NADP^+^ AGN TYR \[27, 28\] Nitrosomonas europaea NADP^+^ AGN None \[21\] Corynebacterium glutamicum NAD(P)^+^ AGN None \[42, 43\] Neisseria gonorrhoeae NAD^+^ PPA^b^ HPP \[21\] Pseudomonas stutzeri NAD^+^ PPA/AGN HPP/TYR \[1\] Pseudomonas aeruginosa NAD^+^ PPA/AGN HPP/TYR \[17\] Zymomonas mobilis NAD^+^ PPA/AGN None \[18\] ^a^Abbreviation: HPP, 4-hydroxyphenylpyruvate. ^b^This TyrA~c~enzyme has an overwhelming preference for PPA, but will use AGN poorly. ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Key to evolutionary events asserted in Figure 6 ::: Group Evolutionary event(s) proposed ------------- ------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Beta 1 Dispersal of aroQ•pheA\> hisH~b~\> tyrAaway from one another and away from gyrA\> serCand from cmk\> rspA\> himD; inversion of aroFwith respect to cmk. 2 Complete dispersal of all nine genes originally in the gyrA/himDlinkage group. 3 Insertion of serAafter serC^a^; separation of tyrAand aroFto yield the separated 6-gene unit and 4-gene unit shown. 4 Expulsion of hisH~b~from the genome; insertion of \'ORF\' after serC. 5 Fusion of tyrAwith aroF. 6 Loss of hisH~b~from genome. Upper-Gamma 1 Insertion of serAafter serC^a^; insertion of aroA~Iα~after hisH~b~. 2 Translocation of hisH~b~and tyrAto other regions, leaving two separated 3-gene units. 3 Fusion of tyrAwith aroF. 4 Loss of hisH~b~. 5 N-terminal deletion of •aroFdomain, and acquisition of new aroFgene (probable LGT). 6 Separation of cmk\> rpsA\> himDfrom aroQ•pheA\> tyrA•aroF. 7 Insertion of 4 unknown genes between gyrAand serCin opposite orientation and separation of gyrA\> ORF \> ORF \> ORF \> serCfrom aroQ•pheA\> tyrA•aroF. 8 Loss of himD; translocation of serCaway from gyrAand aroQ•pheA. ^a^Since both Nitrosomonas(beta Proteobacteria) and Acidothiobacillus(upper-gamma Proteobacteria) emerge at deep positions in the tree of Fig. 5, an almost equally parsimonius possibility is that the ancestral serAwas retained in this syntenic position in these two genera, but was transposed elsewhere shortly after early divergence. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Key to evolutionary events asserted in Figure 7 ::: Number Evolutionary events proposed ------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 1 Escape of aroQ•pheAand tyrAfrom the ancestral gyrA\> serC\> aroQ•pheA\> hisH~b~\> tyrA\> aroF\> cmk\> rpsA\> himDsupraoperon. Origin of an aroQ•tyrAfusion. Origin of the aroA~Iα\_Y~\> aroQ•tyrAoperon. Addition of tyrR. Addition of third aroA~Iα~species: aroA~Iα\_F~. 2 Fusion of aroQ•pheAwith aroA~Iβ~pseudogene of unknown origin. Replacement of hisH~b~by aspCduplicate linked with three ORFs. 3 Dissociation of gyrAand serC. 4 Removal of all genes intervening between aroQ•pheAand aroQ•tyrA. 5 Dissociation of aroFfrom both serCand cmk\> rpsA\> himD. Insertion of trpRwithin the intervening region between aroQ•pheAand aroQ•tyrA. 6 Dissociation of serC\> hisH~b~\> aroFfrom cmk\> rpsA\> himD. 7 Loss of aroA~Iα\_Y~from tyroperon. 8 aroFbecomes dissociated from hisH~b~, and aroA~Iα\_Y~is removed from the tyrAoperon. 9 ORF\> gyrAis inserted after aroF. 10 aroQ•tyrAbecomes a pseudogene. 11 hisH~b~is lost. 12 himDis lost. 13 cmk, himDand aroA~Iα\_Y~\> aroQ•tyrAare lost. 14 aroF, himD, aroQ•pheA, and aroA~Iα\_Y~\> aroQ•tyrAare lost. 15 All intervening genes between aroQ•pheAand aroQ•tyrAare eliminated. 16 pheAdomain of aroQ•pheAbecomes a pseudogene. 17 Insertion of ycaLbetween aroFand cmk. 18 Insertion of ORF between aroFand ycaL. 19 Insertion of ORF between aroQ•pheAand aroQ•tyrA. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Putative attenuators^a^associated with tyrA ::: Organism Gene organization^b^ Fig^c^ ----------------------------------- ------------------------------------------------------------------------------------ ------------ Agrobacterium tumefaciens ^¬^hisH~b~\> *tyrA* 4^d^ Bacillus anthracis ^¬^aroG\> hisH~b~\> *tyrA\> aroF* (11) Bacillus cereus ^¬^aroG\> hisH~b~\> *tyrA\> aroF* (11) Bacillus halodurans ^¬^*tyrA\> aroF* (11) Bacteroides thetaiotaomicron ^¬^pheA\> hisH~b~\> aroA~Iβ~•aroQ \> *tyrA* 4 Bordetella parapertussis ^¬^gyrA\> serC\> aroQ•pheA\> *tyrA\> aroF\> cm k\> rpsA\> himD* 5 Desulfovibrio vulgaris ^¬^aroA\' \> aroB\' \> aroQ•pheA\> aroF\> *tyrA\> \[trpoperon\] 4 Enterococcus faecalis* ^¬^aroD\> aroA~Iβ~\> aroB\> aroG\> *tyrA\> aroF\> aroE\> pheA* 4 Lactococcus lactis ^¬^ysaA\> blrG\> kinG\> *tyrA\> aroF\> aroE\> pheA* 4 Lactobacillus plantarum ^¬^ORF \> aroG\> ORF\> aroF\> *tyrA\> aroE* Listeria innocua ^¬^aroG \> aroB \> aroH \> hisH~b~\> *tyrA\> aroF* 4,(11) Streptococcus pneumoniae ^¬^ORF\>aroC~I~\>aroD\>aroB\>aroG\>*tyrA\> ^¬^ORF\>aroF\>aroE\>pheA* 4 Thermoanaerobacter tencongensis ^¬^pheA \> aroA~Iβ~\> *tyrA\> aroF\> ORF\> ORF* Thermus thermophilus ^¬^aroA~Iβ~\> *tyrA* Vibrio parahaemolyticus ^¬^aroA~Iα\_Y~\> *tyrA* 6 Vibrio vulnificus ^¬^aroA~Iα\_Y~\> *tyrA* 6 ^a^Attenuators were extracted from the website of Merino \[66\]. Links are provided for viewing the complete data, including a visualization of the putative attenuator structures. ^b^The symbol ^¬^is used for attenuators. Genes encoding the alternative biochemical steps that were recently reported for formation of dehydroquinate from aspartate semialdehyde and ketohexose 1-phosphate \[52\] are designated aroA\' and aroB\'. ^c^Refers to figures within this manuscript or, if enclosed within parentheses, to the figure in ref. \[7\]. ^d^See the consensus gene organization for α Proteobacteria. :::	PubMed Central	2024-06-05T03:55:58.327311	2005-5-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173090/", "journal": "BMC Biol. 2005 May 12; 3:13", "authors": [ { "first": "Jian", "last": "Song" }, { "first": "Carol A", "last": "Bonner" }, { "first": "Murray", "last": "Wolinsky" }, { "first": "Roy A", "last": "Jensen" } ] }
PMC1173091	Background ========== Three protein complexes that are conserved from yeast to humans, the 19S proteasome lid, the CSN, and eIF3, contain subunits characterized by two protein motifs: the MPN ([M]{.underline}pr1/ [P]{.underline}ad1 [N]{.underline}-terminal) and the PCI ([p]{.underline}roteasome/[C]{.underline}SN/e[I]{.underline}F3) domains \[[@B1]\]. The proteasome 20S catalytic particle and the 19S regulatory subunit cooperate in degrading polyubiquitylated proteins (reviewed in \[[@B2]\]). The 19S proteasome can be separated into the base complex, which binds and unfolds substrates \[[@B3]\], and the eight subunit lid complex, which cleaves ubiquitin from substrates, thus apparently facilitating the entry of substrates into the catalytic proteasome barrel \[[@B4],[@B5]\]. In higher eukaryotes, the subunits of the 19S lid show pair-wise similarity to the eight subunits of the CSN \[[@B6]-[@B9]\]. In vivo, CSN promotes the activity of cullin-RING ubiquitin ligases \[[@B10]-[@B16]\], multiprotein complexes containing cullins, the RING protein RBX1, and one of several hundred substrate-specific adaptors \[[@B17]-[@B22]\]. The MPN domain containing CSN subunit 5 harbors a protease motif \[[@B23]\] that cleaves the ubiquitin-related peptide NEDD8 from cullins \[[@B24],[@B25]\]. This activity, acting in concert with the CSN-associated deubiquitylation enzyme Ubp12, was proposed to promote cullin function by facilitating the recruitment of labile substrate adaptors \[[@B11],[@B16],[@B26],[@B27]\]. The third PCI/MPN complex, eIF3, is more distantly related to CSN and the 19S lid (reviewed in \[[@B28]\]). Whereas human eIF3 consists of up to 13 subunits, consecutively named eIF3a -- l and GA17 \[[@B29],[@B30]\], budding yeast contains only six to eight subunits (depending on purification conditions). Five of these subunits are orthologs of human eIF3a, b, c, g, and eIF3i \[[@B31],[@B32]\] and appear to constitute a conserved core complex \[[@B31],[@B33]\]. Fission yeast contains the same five core subunits, in addition to the non-core subunits eIF3d/Moe1p, eIF3e/Int6p, and eIF3h \[[@B34]-[@B36]\]. A putative eIF3f ortholog was also identified, but biochemical evidence confirming it as an authentic eIF3 subunit functioning in protein synthesis is still outstanding \[[@B34],[@B35],[@B37],[@B38]\]. eIF3 is the most complex translation initiation factor and plays at least two important roles in protein synthesis. First, eIF3 binds to the 40S ribosome and facilitates loading of the Met-tRNA/eIF2· GTP ternary complex to form the 43S preinitiation complex. Subsequently, eIF3 apparently assists eIF4 in recruiting mRNAs to the 43S complex. A critical in vivo function of eIF3 core subunits in these processes was indicated by the lethality of the respective budding yeast deletion strains \[[@B31]\]. In contrast, in fission yeast, the non-core subunits eIF3d and eIF3e are not essential for viability or global protein synthesis \[[@B34],[@B36],[@B38],[@B39]\]. It was therefore proposed that distinct subclasses of eIF3 complexes, containing different combinations of core and non-core subunits, may regulate specific subsets of mRNAs in fission yeast \[[@B34],[@B36],[@B38],[@B39]\]. Our study provides the first experimental evidence substantiating this hypothesis by demonstrating that biochemically distinct eIF3 complexes defined by the PCI domain proteins eIF3e and eIF3m (a novel eIF3 protein) associate with different sets of mRNAs. Results ======= Both csn6and csn7bare essential genes ----------------------------------------- CSN complexes of higher eukaryotes typically contain eight distinct subunits (two MPN and six PCI proteins). However, in fission yeast only six subunits are known (Csn1p, 2p, 3p, 4p, 5p, 7Ap; Ref. \[[@B13]\]), none of which are essential for viability \[[@B40]-[@B42]\]. We noticed two genes in the Schizosaccharomyces pombegenome database, originally annotated as csn6(SPBC4C3.07) and csn7b(SPAC1751.03), which encode MPN and PCI domain containing proteins with considerable similarity to metazoan CSN6 and CSN7B, respectively (data not shown). In order to determine whether these genes might function in the known biochemical pathways regulated by CSN, cullin deneddylation and Ubp12p-mediated deubiquitylation \[[@B24],[@B26]\], we deleted csn6and csn7bin wild-type diploid cells. Upon sporulation and tetrad analysis, only two viable spores could be recovered in each case, both of which were uracil auxotroph (Fig. [1A](#F1){ref-type="fig"}, and data not shown). Thus, unlike the genes encoding the six known subunits of the fission yeast CSN \[[@B41],[@B42]\], csn6and csn7bare essential. Csn6p and Csn7Bp associate with eIF3 components ----------------------------------------------- To elucidate the essential functions of Csn6p and Csn7Bp, we sought to identify their interacting proteins. To this end, we attempted to modify csn6and csn7bat their endogenous genomic loci with five consecutive protein A epitope tags (proA). The C-terminal tag also contained a cleavage site for Tobacco Etch Virus (TEV) protease upstream of the protein A moieties. Diploid analysis indicated that C-terminal tagging destroyed the essential function of csn6(data not shown). In contrast, a csn7b-proAhaploid strain was viable and did not display any obvious phenotypes (data not shown). This indicated that the tagged allele was functional. Total protein lysate from the csn7b-proAhaploid strain was absorbed to Immunoglobulin G (IgG) resin and specifically retained proteins were eluted with sodium dodecyl sulfate (SDS) or by cleavage with TEV protease. The two eluates were resolved by gel electrophoresis and Csn7Bp-proA interacting proteins were identified by peptide mass fingerprinting using matrix-assisted laser desorption/ionization-time-of-flight (MALDI-TOF) mass spectrometry. Both eluates contained an identical set of proteins. In addition to Csn6p, the Csn7Bp complex contained the eIF3 core subunits eIF3a, b, c, and i, as well as the non-core subunit eIF3h (Fig. [1B](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). The eIF3g core subunit was not detected using MALDI-TOF in this purification, but was detected in an independent purification of the Csn7Bp complex using the more sensitive method of nano-liquid chromatography and tandem mass spectrometry (LC-MS/MS, see Fig. [5A](#F5){ref-type="fig"}). However, eIF3e and eIF3d, two previously described components of S. pombeeIF3 \[[@B34],[@B36],[@B38],[@B39]\] were consistently absent from the Csn7Bp complex when analyzed either by MALDI-TOF or by LC-MS/MS (Fig. [1B](#F1){ref-type="fig"} and [5A](#F5){ref-type="fig"}). Notably, no authentic CSN subunits were found in the Csn7Bp complex, consistent with the previous demonstration that S. pombeCSN contains only six non-essential subunits \[[@B13]\]. Structural analysis indicated that Csn6p is not only related to human CSN6, but shows a slightly greater overall similarity to human eIF3f (data not shown). Based on the functional characterization described below (Fig. [2](#F2){ref-type="fig"}), fission yeast Csn6p therefore appears to be the ortholog of human eIF3f. A Csn7Bp ortholog is not known in the human eIF3, although the recently identified subunit GA17 \[[@B30]\] shows considerable similarity to Csn7Bp (K. Hofmann, personal communication). In addition, interactions of metazoan CSN7 with eIF3 subunits have previously been observed \[[@B43],[@B44]\]. With regards to the unified nomenclature of eIF3 subunits \[[@B29]\], we propose eIF3m as a more appropriate name for the novel fission yeast eIF3 subunit encoded by csn7b(see Table [1](#T1){ref-type="table"}). Throughout this manuscript, we will use the proposed new names for the eif3fand eif3mgenes and their products. The original annotations have recently been revised by the curator of the S. pombegenome database, in order to reflect the findings presented here. The eIF3f and eIF3m proteins are primarily cytoplasmic ------------------------------------------------------ The evidence presented above suggested that eIF3f and eIF3m are essential subunits of S. pombeeIF3. To substantiate this possibility, we determined the subcellular localization of eIF3f and eIF3m. If these proteins were authentic eIF3 subunits involved in protein synthesis, they should assume a subcellular localization similar to known eIF3 subunits. We therefore prepared three strains with green fluorescent protein (GFP) tagged alleles of eIF3b, eIF3e, and eIF3m integrated into their respective genomic loci. In addition, since C-terminal GFP-tagging destroyed the essential function of the small eIF3f protein, we prepared a strain mildly overexpressing N-terminally GFP-tagged eIF3f from a pREP81 plasmid. Fluorescence microscopy of live cells revealed that all four proteins were primarily localized in the cytoplasm (Fig. [2A](#F2){ref-type="fig"}), which would be consistent with a function in protein translation. In contrast, CSN subunits are largely nuclear in fission yeast (\[[@B41]\], and our own unpublished observation). eIF3f and eIF3m are required for global protein synthesis --------------------------------------------------------- To directly assess the potential role of eIF3f and eIF3m in protein synthesis, we prepared haploid eif3fand eif3mdeletion strains maintained viable by plasmids driving the regulated expression of eif3fand eif3m, respectively. Expression from these plasmids could be turned off by the addition of thiamin to the growth media, resulting in a strong reduction in protein levels within 20 to 32 hours (Fig. [2B](#F2){ref-type="fig"}). Coincident with shut-off of eIF3f and eIF3m expression, global cellular protein synthesis was diminished by \~80%, as determined by metabolic labeling with ^35^S-methionine (Fig. [2C](#F2){ref-type="fig"}). Further biochemical analysis by sucrose gradient velocity centrifugation revealed a strong reduction in the formation of polysomes in eif3fand eif3mmutants following promoter shut-off (Fig. [2D](#F2){ref-type="fig"}). In contrast, as described previously \[[@B34],[@B36],[@B38],[@B39]\], eif3emutants showed only a minor reduction in total protein synthesis (data not shown) and essentially normal polysomes (Fig. [2D](#F2){ref-type="fig"}). The conserved cullin deneddylation function of CSN was not impaired either in eif3fand eif3mmutants or in eif3emutants, since cullin 1 (Cul1p) did not accumulate exclusively in the neddylated form as it does in csn5deletion strains (Fig. [2E](#F2){ref-type="fig"}). Similarly, unlike in csn5mutants (Fig. [2E](#F2){ref-type="fig"}; \[[@B26]\]), CSN/Ubp12p-mediated inhibition of Cul1p in vitro ubiquitin ligase activity was not impaired upon shut-off of either eif3for eif3mexpression. Cul1p neddylation and activity were also unaffected in eif3emutants (Fig. [2E](#F2){ref-type="fig"}). These data suggested that eIF3f and eIF3m are bona fide subunits of fission yeast eIF3 but not CSN, performing essential functions in protein synthesis similar to eIF3 core subunits. The eIF3m and eIF3e proteins define distinct eIF3 complexes ----------------------------------------------------------- Although the purification of the eIF3m complex robustly retrieved roughly stoichiometric amounts of known eIF3 subunits, two subunits, eIF3d and eIF3e, were conspicuously absent (Fig. [1B](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). To exclude the possibility that we had overlooked these subunits because they were obscured by the IgG bands, we affinity-purified eIF3e-associated proteins from an eif3e-proAstrain. The retrieved proteins were identified using MALDI-TOF mass spectrometry as before. Like the eIF3m complex, the eIF3e complex contained roughly stoichiometric amounts of the eIF3 core subunits a, b, c, g, and i (Fig. [3A](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}). In addition, eIF3f was present while, unexpectedly, eIF3m and eIF3h were missing. We did not detect the proteasome lid subunit Rpn5p, which previously had been shown to associate with eIF3e and eIF3d \[[@B45]\]. This observation indicated that the described eIF3d/eIF3e/Rpn5p interaction is either unstable under our purification conditions or contains only a minor fraction of the total eIF3e engaged in protein interactions. These results raised the intriguing possibility that fission yeast contains two distinct eIF3 complexes that comprise an overlapping set of core subunits, but are distinguished by the presence of either the essential PCI protein eIF3m and the MPN domain protein eIF3h or the non-essential PCI protein eIF3e and its binding partner eIF3d. To exclude the possibility that our mass spectrometry analysis missed substoichiometric amounts of eIF3d in the eIF3m complex, we analyzed eIF3m complexes by immunoblotting with eIF3d antisera. Whereas eIF3d was readily detectable in the eIF3e complex, it was undetectable in the eIF3m complex (Fig. [3B](#F3){ref-type="fig"}). In the reciprocal experiment, eIF3d antibodies co-immunoprecipitated eIF3e-13Myc, but not eIF3m-13Myc or Csn5p-13Myc (Fig. [3C](#F3){ref-type="fig"}). To exclude the possibility that our cell lysis or affinity purification conditions led to uncontrolled release of eIF3 subunits, thus mimicking the existence of distinct subcomplexes, we performed a purification using the protein A-tagged eIF3b core subunit as bait. As determined by LC-MS/MS performed on excised bands detected in the complex upon PAGE separation, the eIF3b complex contained all known eIF3 core components as well as eIF3f and eIF3m (Fig. [3D](#F3){ref-type="fig"}). The eIF3e and eIF3d proteins were also present, albeit in clearly substoichiometric amounts. These findings indicate that purification of eIF3b resulted in the copurification of the distinct eIF3m- and eIF3e-containing complexes. Taken together, these data strongly suggested the existence of two distinct eIF3 complexes defined by the PCI proteins eIF3m and eIF3e. The eIF3m and eIF3e proteins associate with distinct sets of mRNAs ------------------------------------------------------------------ The finding that eif3mis essential, whereas eif3eis dispensable, suggested the possibility that the different eIF3 complexes they define regulate different subsets of mRNAs. To approach this, we sought to identify mRNAs specifically associated with the eIF3m and eIF3e complexes. The complexes were affinity-purified as described above in the presence of RNAse inhibitors, and the associated RNA was extracted, amplified, and converted to Cy3-labeled cDNA as described in the Methods section. The Cy3-labeled cDNA was hybridized competitively with Cy5-labeled cDNA prepared from total RNA onto microarrays representing all 4988 predicted S. pombeopen reading frames (ORFs) and RNAs. Based on the background subtracted hybridization signals, eIF3-associated mRNAs were ranked according to their factor of enrichment in the eIF3m and eIF3e complexes over a mock purified sample (\[see [additional file 1](#S1){ref-type="supplementary-material"}\]). The microarray analysis suggested a global role for eIF3m and a more restricted function of eIF3e in translation. Using an arbitrary cut-off value of three-fold enrichment over mock, eIF3m associated with 2464 different mRNAs, whereas eIF3e associated with only 520 distinct species (Fig. [4A](#F4){ref-type="fig"}). We observed 414 mRNAs enriched more than three-fold in both samples. (Fig. [4A](#F4){ref-type="fig"}). In addition, our analysis revealed 106 mRNAs uniquely enriched in the eIF3e complex, whereas 2050 transcripts were uniquely enriched in the eIF3m complex (Fig. [4A,C](#F4){ref-type="fig"}). A ranked list of all 106 mRNAs that were uniquely enriched more than three-fold in the eIF3e complex was assembled (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]). Approximately 75% of these mRNAs fell into one of three categories: mRNAs encoding proteins involved in intermediary metabolism (25.5%), those encoding proteins involved in protein metabolism (14.2%), and those encoding proteins with unknown functions (35.8%; Fig. [4B](#F4){ref-type="fig"}). The remainder was distributed roughly equally among four categories: mRNAs encoding proteins involved in nucleic acid metabolism, those encoding proteins involved in transcription, those encoding transporter proteins, and those encoding unclassified proteins. A corresponding list of the 106 most highly enriched mRNAs exclusively present in the eIF3m complex (out of a total of 2050 enriched by a factor greater than three) revealed that \~58% also fell into three categories: mRNAs encoding proteins involved in nucleic acid metabolism (9.4%), those encoding transporter proteins (18.9%), and those encoding proteins of unknown function (20.8%). The remainder was roughly equally distributed among four categories: mRNAs encoding proteins involved in intermediary metabolism, those mRNAs encoding proteins involved in protein metabolism, those encoding proteins involved in transcription, and those encoding unclassified proteins (Fig. [4B](#F4){ref-type="fig"}, \[see [additional file 2](#S2){ref-type="supplementary-material"}\]). To exclude the possibility that our affinity purification protocol unspecifically enriched highly expressed mRNAs, we averaged the data from the three competively hybridized samples of total S. pombecDNA to determine the relative expression rank of each mRNA (rank 1 = most abundant mRNA, rank 5407 = least abundant mRNA; see legend to Fig. [4C](#F4){ref-type="fig"} for detail). A plot of expression rank vs. factor of enrichment of the 106 mRNAs most enriched in the eIF3e and eIF3m complexes revealed that the eIF3m complex contained mRNAs that were distributed equally over the entire range of expression levels. In contrast, the eIF3e complex was enriched for rare mRNAs (\~65 % of eIF3e-associated mRNAs had an expression rank \> 4000; see Fig. [4C](#F4){ref-type="fig"}). The lack of a linear correlation between mRNA expression level and enrichment in the eIF3e and eIF3m complexes argues against unspecific copurification of abundant mRNAs. Specific association of mRNAs with eIF3m and eIF3e complexes ------------------------------------------------------------ To validate the results from the microarray hybridizations, we chose several mRNAs at random from each list of 106 and used RT-PCR to confirm their interactions with the respective eIF3 complexes. eIF3e-proA- and eIF3m-proA-associated complexes were affinity-purified as before and separated by PAGE. Both complexes showed the same differential subunit composition as described above (Fig. [5A](#F5){ref-type="fig"}). The associated RNA was purified, split into equal portions, and employed in RT-PCR reactions with gene-specific primer pairs. The RT-PCR analysis of a total of fourteen mRNAs included eight mRNAs identified in the eIF3e complex, five mRNAs identified in the eIF3m complex, and one mRNA found to be associated with both complexes with similar enrichment factors. These mRNAs varied widely in their factors of enrichment and their enrichment ranks as indicated in Fig. [5B](#F5){ref-type="fig"}. Without exception, all eight mRNAs found to be enriched in the eIF3e complex by microarray analysis were also found preferentially associated with this complex by RT-PCR (Fig. [5B](#F5){ref-type="fig"}). Based on this result, we calculate a false positive rate of less than 0.11 with a 95% confidence interval of 0.00 -- 0.31. Whereas mRNAs with a high enrichment factor (\>10, \[see [additional file 2](#S2){ref-type="supplementary-material"}\]) appeared to be exclusively associated with eIF3e-proA (within the limit of detection), those with a lower enrichment factor (\<10) also displayed a low amount of association with the eIF3m complex (Fig. [5B](#F5){ref-type="fig"}). Conversely, all five mRNAs highly enriched in the eIF3m complex as determined by microarray analysis (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]) were almost exclusively bound to this complex when analyzed by RT-PCR (Fig. [5B](#F5){ref-type="fig"}). Notably, the mRNA encoding the proteasome subunit Rpn5p, which was found in both complexes by microarray analysis with similar enrichment factors (eIF3e = 5.1; eIF3m = 8.1) was also confirmed as being associated with both eIF3 complexes by RT-PCR (Fig. [5B](#F5){ref-type="fig"}). These results strongly suggest that the eIF3m and eIF3e complexes specifically associate with distinct sets of mRNAs. Discussion ========== PCI proteins define distinct eIF3 complexes ------------------------------------------- Our results show that Csn6p and Csn7Bp are subunits of fission yeast eIF3. Csn6p was shown to be the eIF3f subunit conserved in all eukaryotes except budding yeast. Consistent with this finding, eIF3f has previously been found to be associated with eIF3g, although the role of eIF3f in protein synthesis had not been examined \[[@B34]\]. Our study further identified two distinct eIF3 complexes defined by the PCI proteins eIF3e and eIF3m (Fig. [3A](#F3){ref-type="fig"}). No direct equivalent of eIF3m is apparent in eIF3 preparations from other eukaryotes. Considering the essential function of eIF3m in protein synthesis in fission yeast, it is possible that other PCI proteins substitute for eIF3m in other organisms. Consistent with this idea is the finding that the PCI domain containing GA17, a recently identified protein copurifying with human eIF3 \[[@B30]\], shows more similarity to eIF3m than to any other PCI protein (K. Hofmann, personal communication). In addition, the PCI domain containing CSN subunits 1, 3, 7, and 8 were found to interact with eIF3 subunits in Arabidopsis thalianaand in human cells \[[@B43],[@B44],[@B46]\]. In addition, Pci8p/Csn11p, a non-essential PCI subunit of the budding yeast CSN \[[@B47],[@B48]\], copurifies with essential eIF3 core subunits \[[@B49]\]. Although the functional significance of these interactions is still unclear, our findings raise the intriguing possibility that they define multiple subclasses of distinct eIF3 complexes. Whereas fission yeast appears to rely on an essential eIF3 complex classified by eIF3m and a non-essential complex specified by eIF3e, higher eukaryotes may divide the task of mRNA translation among a multitude of eIF3 complexes. Since the individual PCI proteins defining these subcomplexes are expected to be present in substoichiometric amounts relative to core components, as is eIF3e in the S. pombecore complex, they may have escaped detection in bulk eIF3 preparations from higher eukaryotes. Distinct eIF3 complexes and translational specificity ----------------------------------------------------- Although our finding of distinct eIF3 complexes that associate with different sets of mRNAs was surprising, circumstantial evidence pointing to their existence was provided in previous studies. For example, fission yeast eIF3e copurifies with eIF3 core subunits, and yet, unlike core subunits, eIF3e is not required for global protein synthesis or viability. Two competing models were proposed to explain the pleiotropic phenotype of eif3edeletion mutants, which includes slow growth in minimal media and meiotic defects \[[@B34],[@B36],[@B38],[@B39],[@B50],[@B51]\]. In the first model, eIF3e regulates the translation of all transcripts, but rare mRNAs may be more affected by a fractional reduction in translation efficiency and thus give rise to distinct phenotypes. In the second model, eIF3e regulates the translation of a specific subset of mRNAs encoding proteins whose depletion leads to distinct phenotypes but not lethality. The findings presented here strongly favor the second model. Our microarray analysis revealed 106 mRNAs uniquely enriched in the eIF3e complex, many being rare, whereas 2050 transcripts were uniquely enriched in the eIF3m complex. The latter finding is consistent with the severe protein synthesis defect and lethality of eif3mdeletion mutants. Whereas these mRNA lists were based on an arbitrary cut-off value of three-fold enrichment over mock, this criterion proved stringent enough to confirm specific association with their respective eIF3 complexes of all 14 mRNAs retested by RT-PCR. Thus, since we calculate the false positive rate to be less than 0.11 with 95% confidence interval 0.00 -- 0.31, the maximum number of mRNAs listed potentially false positively as enriched in the eIF3e complex would be 33 out of 106 (with 95% confidence). However, we consider it unlikely that all mRNAs found enriched in the eIF3e complex are exclusive translational targets of this complex, because the cellular phenotypes of eif3emutants are rather discrete. Although many of the eIF3e-associated mRNAs encode essential proteins, viability, global protein synthesis, and polysome formation are not affected in these mutants when they are grown in rich media (Fig. [2D](#F2){ref-type="fig"}, and data not shown). It is therefore likely that eIF3m or other PCI domain proteins can deputize for eIF3e in many cases. In fact, our microarray study revealed that a substantial portion of eIF3e-associated mRNAs was also recovered in the eIF3m complex (Fig. [4A](#F4){ref-type="fig"}). Nonetheless, we strongly suspect that inefficient translation of a limited set of eIF3e-associated mRNAs contributes to distinct cellular phenotypes found in eif3emutants (see below). The eIF3e protein and translational control during stress response ------------------------------------------------------------------ The eif3emutants display marked sensitivity to a wide variety of cellular stresses, such as osmotic stress, nutrient starvation, and low temperature \[[@B34],[@B38],[@B39],[@B50],[@B52]\] In addition, eif3emutants show synthetic lethality with proteasome mutants, which are presumed to accumulate a large load of misfolded proteins \[[@B45]\]. Conversely, overexpression of eif3econfers resistance to a broad spectrum of unrelated drugs, whose only commonality appears to be that they induce cellular stress \[[@B38]\]. Overexpression of eIF3e also activates genes involved in stress defense, such as thioredoxin \[[@B38]\]. Interestingly, the mRNA encoding thioredoxin was also identified as a putative translational target of eIF3e (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]). Finally, eIF3e undergoes relocalization into cytoplasmic foci in response to heat and osmotic stress, thus implicating it in stress regulation \[[@B35]\]. Global protein synthesis is actively switched off in response to cellular stress and nutrient deprivation (reviewed in Refs. \[[@B53],[@B54]\]). This shut-off protects cells from proteotoxicity by relieving chaperones of their load of unfolded client proteins, conserving amino acids for other essential functions, and attenuating the metabolic consequences of protein synthesis. However, long-term adaptation to stress conditions, repair and recovery require synthesis of new stress-induced proteins. During the stress response, some mRNAs must be translated in the presence of a repressed general translation machinery, suggesting that select eIFs can escape the global repression. Our studies raise the intriguing possibility that the eIF3e complex has such a specialized function during stress response. Lack of eif3emay prevent stress-induced synthesis of its critical translational targets, thus leading to the known stress sensitivity of the mutant. Substantial additional work will be required to address this possibility. Potential mechanisms underlying mRNA discrimination by eIF3 complexes --------------------------------------------------------------------- Several lines of evidence have begun to implicate other general translation factors, including eIF2 and eIF4, in translational specificity \[[@B55]\]. Importantly, these factors are downstream of important signaling cascades, including the nutrient-sensing TOR pathway and a variety of stress-induced kinases (reviewed in Refs. \[[@B55],[@B56]\]). Although the details of these regulatory mechanisms are beginning to emerge, it is still unclear exactly how eIF2 and eIF4 contribute to mRNA specificity. Since eIF3 cooperates with the cap-binding eIF4 complex in directing the 5\' end of mRNAs to the 40S ribosome, and several eIF3 subunits contain RNA binding motifs, our data is consistent with a model in which eIF3 complexes defined by distinct PCI proteins recognize determining features of mRNAs presented to the 40S subunit in concert with eIF4. In agreement with this idea is the recent finding that the human PCI protein eIF3a specifically facilitates the translation of the mRNA encoding ribonucleotide reductase M2 \[[@B57]\]. In addition, A. thalianaeIF3h, a non-core subunit, was recently shown to be specifically required for efficient translation of the transcription factor ATB2 \[[@B46]\]. While we have so far been unsuccessful in identifying common sequence elements within the upstream regions of mRNAs enriched in the eIF3e complex (data not shown), this analysis was severely hampered by the fact that 5\'UTRs are poorly defined in S. pombe. It is therefore still possible that critical features, including secondary structure determinants, lie within the 5\'UTR or other regions of eIF3e-associated mRNAs. Substantial additional work will be required to delineate the exact mechanism of how eIF3e may contribute to the translation of specific mRNAs and the potential impact of this mechanism on human cancers defective in INT6/eIF3e \[[@B58],[@B59]\]. Conclusion ========== We provide biochemical evidence for the existence of two distinct eIF3 complexes in fission yeast. These complexes contain an overlapping set of subunits, but are distinguished by the PCI proteins eIF3e and eIF3m. Based on the finding that the distinct eIF3 complexes associate with different mRNAs, we propose that they have different translational specificities. Methods ======= Fission yeast techniques ------------------------ Preparation of fission yeast cultures, cDNA cloning, yeast transformation, PCR-based genomic epitope tagging and gene deletion, mating, and tetrad analysis were carried out using standard methods as described \[[@B26]\]. The strains used in this study are listed in Table [2](#T2){ref-type="table"}. Affinity purification and mass spectrometry ------------------------------------------- Epitope-tagged strains (eif3m-proAand eif3e-proAstrains) were grown in 12 L yeast extract and supplements (YES) medium to 1.5 optical density (OD) at 595nm wavelength and harvested by centrifugation. Cell lysates were prepared by bead lysis in a lysis buffer containing 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.5% Triton X-100, 1 mM dithiothreitol, supplemented with protease inhibitors (10 ug/ml leupeptin, 10 ug/ml pepstatin, 5 ug/ml aprotinin and 1mM phenylmethylsulfonylfluoride). Upon centrifugation at 17,000 rpm for 50 min at 4°C, 300 mg total protein in a volume of 50 ml was absorbed to 100 ul Dyna beads (Dynal Biotech, Oslo, Norway) coupled to whole rabbit IgG (Jackson Immunochemicals, West Grove, Pennsylvania, USA). Mock purifications were performed in parallel by absorbing lysate from cells lacking any tagged proteins to IgG beads. Beads were collected by a magnetic device and washed five times in 5 ml lysis buffer. TEV protease (Invitrogen, Carlsbad, California, USA) cleavage or elution in 5% SDS solution were used to collect bound proteins. Eluted proteins were separated on trycine gels or by conventional SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and visualized by Coomassie G250 staining. Protein bands were excised from gels, digested with trypsin (Promega Corporation, Madison, Wisconsin, USA), and subjected to mass spectrometry. The eIF3 complexes were analyzed by MALDI-TOF, and by LC-MS/MS at the Harvard NIEHS Center Proteomics Facility, according to standard procedures. Protein synthesis and polysome analysis --------------------------------------- The eif3fand eif3mconditional shut-off strains were inoculated in 10 ml Edinburgh Minimal Media (EMM) with or without thiamin and cultured at 30°C for 30 hrs. 100 uCi ^35^S-methionine were added to the cells, which were labeled for 30 min and then harvested by centrifugation. Cell lysates were prepared by bead disruption and equal amounts of the lysates were analyzed by SDS-PAGE. ^35^S-methionine incorporation was determined by autoradiography and quantified by PhosphorImager analysis (Bio-Rad, Hercules, California, USA). Polysome analysis was performed exactly as described in Ref. \[[@B60]\]. Briefly, cells were grown in EMM with or without thiamin for 30 hrs. Cycloheximide (100 ug/ml) was added to each culture for 20 min. Cells were harvested and lysed in breaking buffer (20 mM Tris/HCl pH 7.4, 50 mM NaCl, 30 mM MgCl~2~, 1 mM DTT, 50 ug/ml cycloheximide, 0.2 mg/heparin, 50 U/ml SuperRNasein, and protease inhibitors), followed by centrifugation at 13,000 × g for 5 min. 20 A~260\ nm~units of each supernatant in a volume of 500 ul were fractionated on 5-45 % sucrose gradients for 3 hrs at 30,000 rpm using a SW40-Ti rotor in a Beckman L70 centrifuge. Polysome profiles were obtained by monitoring the absorbance at 254 nm along the gradient using a Bio-Rad fractionator, and the output was recorded using a Bio-Rad UV detector. RT-PCR ------ RNAs associated with the eIF3e and eIF3m complexes were purified as described above. Primer design and RT-PCR conditions were according to the manual supplied with the Platinum Quantitative RT-PCR Thermoscript kit (Invitrogen). Primers used for RT-PCR are listed in Table [3](#T3){ref-type="table"}. The primer concentration for actin amplification was 0.25 uM. Other primers were used at 1 uM. Total RNA was obtained as described \[[@B19]\]. Immunological techniques ------------------------ Immunoprecipitation, immunoblotting, and in vitro ubiquitination assays were carried out exactly as described previously \[[@B26]\]. Antibodies against protein A and ubiquitin were purchased from Sigma and Genzyme (Cambridge, Massachusetts, USA), respectively. Anti-eIF3d/Moe1p polyclonal antibody was provided by E. Chang (Baylor College of Medicine, Houston, Texas, USA) Microarray analysis ------------------- The eIF3 complexes were affinity-purified as described above, except that 50 U/ml SuperRNasein (Ambion, Austin, Texas, USA) and 5 mM ribonucleoside vanadyl complex (New England Biolab, Beverly, Massachusetts, USA) were added to the lysis buffer. A mock purification was performed under identical conditions with lysate from cells lacking any tagged proteins. The methods for extraction of RNA, cDNA synthesis, RNA amplification, and aminoallyl labeling were adopted from previous publications \[[@B61],[@B62]\]. Briefly, after binding and washing, beads were treated with 3 U/ul DNase I, followed by elution with 5 % SDS, extraction with phenol/chloroform and then ethanol precipitation. The associated RNA was reversely transcribed using Superscript II reverse transcriptase (Gibco-BRL Division of Invitrogen) with 1 uM oligo-dT(15)-T7 primer (5\'-AAA CGA CGG CCA GTG AAT TGT AAT ACG ACT CAC TAT AGG CGC-3\') and 1 uM template switch primer (5\'-AAG CAG TGG TAT CAA CGC AGA GTA CGC GGG-3\'). Full-length double stranded (ds)-cDNA was synthesized using Advantage Polymerase (Clontech, BD Biosciences, Mountain View, California, USA). Newly synthesized ds-cDNA was passed through Bio-6 columns (Bio-Rad) and dried. RNA amplification was performed with the T7 megascript kit (Ambion). Amplified RNA was purified using the RNeasy kit (Qiagen, Hilden, Germany) and reversely transcribed with Superscript II reverse transcriptase in the presence of aminoallyl-dUTP in reactions primed with oligo-dT (Invitrogen). After RNAse H digestion and purification on QIAquick columns (Qiagen), the aminoallyl-cDNA was labeled with Cy3, and purified on QIAquick columns. Control Cy5-labeled aminoallyl-cDNA was made by oligo dT primed reverse transcription of total S. pombeRNA. Labeled cDNAs were hybridized onto glass slide microarrays. Microarrays were made by spotting purified ds-PCR products 500 to 1200 bp in length onto glass slides coated with aminopropylsilane (Erie Scientific, Portsmouth, New Hampshire, USA). The PCR products used represented 4988 predicted ORFs and transcripts as annotated by the Sanger Centre \[[@B63]\] and are located at the 3\' end of predicted genes to maximize sensitivity in detecting oligo-dT primed reverse transcription products. Descriptions of primer pairs and the features on the microarrays are available at the Longhorn Array Database \[[@B64]\]. Median pixel intensities for each spot obtained from hybridizations of eIF3-associated RNA (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 1\]) were corrected by local background subtraction and divided by the corresponding values obtained from a mock purification. This resulted in the factor of enrichment of mRNAs bound to eIF3 complexes over mock. The mRNAs were ranked according to their factor of enrichment (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 2\]). An enrichment factor of three was chosen as an arbitrary threshold. We found 2464 mRNAs were enriched greater than three-fold over mock in the eIF3m complex, while we observed 520 mRNAs enriched in the eIF3e complex (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 3\]). Comparison of the data sets revealed 414 mRNAs that occurred in both complexes, 2050 mRNAs uniquely enriched in the eIF3m complex and 106 mRNAs that were exclusively enriched in the eIF3e complex (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 4; [additional file 2](#S2){ref-type="supplementary-material"}\]). A corresponding list of the 106 mRNAs most highly and uniquely enriched in the eIF3m complex was assembled (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]) and used to select specific mRNAs for RT-PCR analysis. The competitive Cy5-labeled total cDNA provided clear microarray feature identification and a guide to relative sequence abundance in the pre-purification sample from which the relative expression rank of each mRNA was determined (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 2\]). In this study, data were not adjusted by ratio normalization to total RNA signals or local area normalization functions. The microarray data were deposited with the public ArrayExpress (E-MEXP-208) \[[@B65]\] and GEO \[[@B66]\] databases. List of abbreviations ===================== CSN COP9 signalosome Cul Cullin eIF Eukaryotic initiation factor EMM Edinburgh Minimal Media GFP Green fluorescent protein IgG Immunoglobulin G LC MS/MS Liquid chromatography tandem mass spectrometry MALDI-TOF Matrix-assisted laser desorption ionization time-of-flight MPN Mpr1/Pad1 N-terminal OD Optical density PAGE Polyacrylamide gel electrophoresis PCI Proteasome, COP9 Signalosome, eIF3 ProA Protein A RT-PCR Reverse transcription-polymerase chain reaction TEV Tobacco etch virus YES Yeast extract and supplements (growth media) Authors\' contributions ======================= CZ performed the bulk of the experiments shown in this study. FA and SW prepared various strains used in this study. SK and ARI performed mass spectrometry. AO performed the microarray hybridizations, and JL participated in the design of the microarray studies and in data analysis. DAW contributed to study design and drafting of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 This article is accompanied by a Microsoft Excel file (.xls) containing microarray data (see Methods for description). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A Microsoft Excel file (.xls) is included containing the lists of the 106 mRNAs most highly and uniquely enriched in the eIF3e and eIF3m complexes. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank E. Chang for eIF3d/Moe1p antisera, V. Leautaud, G. Dittmar, F. Bachand, and P. Silver for help with polysome analyses, and M. Schmidt and A. Houseman for assistance with microarray data analysis. This study was funded by National Institute of Health (NIH) grant GM59780 to DAW, NIH grant P40 RR16320 to JL, by the NIEHS Center Grant ES-00002, and the NIEHS Training Grant ES07155. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *Analysis of csn6and csn7bgenes and proteins. (A) The csn6and csn7bgenes were disrupted in diploid S. pombecells by inserting the ura4marker. Diploids were sporulated and spore viability was examined by tetrad analysis. Only two spores were viable, indicating that csn6and csn7bare essential. (B) Cell lysate from a strain carrying csn7bmodified with five C-terminal pro-A tags and a TEV cleavage site at the endogenous genomic locus was absorbed to IgG resin, followed by elution of bound proteins with TEV protease (left panel) or SDS (right panel). Gels were stained with Coomassie Brilliant Blue, and proteins were identified by MALDI-TOF mass spectrometry. The asterisks denote degradation products of elF3c. ::: ![](1741-7007-3-14-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Roles of eif3eand eif3min protein synthesis. (A) Subcellular localization. Live cells expressing GFP-tagged alleles of eif3b, eif3e, and eif3mat the endogenous genomic loci were examined by fluorescence microscopy. N-terminally GFP-tagged eif3fwas expressed at low levels from the pREP81 plasmid. (B) Shut-off strains. Diploid heterozygous eif3fand eif3mdeletion strains were transformed with pREP81 plasmids driving the thiamin-repressible expression of Myc-tagged eIF3f and eIF3m, respectively. Diploids were sporulated and homozygous disruptants carrying the eif3fand eif3mplasmids were selected. Cells were grown in liquid medium to an OD~595~of 0.3 in the absence of thiamin, followed by promoter shut-off by the addition of thiamin. Samples were taken at the indicated time points after promoter shut-off and analyzed for the expression of plasmid borne eIF3f and eIF3m by immunoblotting with anti-Myc antibodies. (C) Effect on total protein synthesis. The eif3fand eif3mshut-off strains were maintained in the absence or presence of thiamin as indicated. Strains transformed with empty pREP81 plasmid were included as control. Cells were metabolically labeled with ^35^S-methionine, and aliquots of total cellular proteins were separated by SDS-PAGE and analyzed by autoradiography. The Coomassie Blue stained gel is shown to document equal protein loading. Data were quantified by PhosphorImager analysis and results are displayed in a bar graph. (D) Effect on polysomes. Polysome profiles were determined for the indicated strains as described in the Methods section. An eif3edeletion strain is shown for reference. (E) Effect on CSN-mediated regulation of cullin-RING ubiquitin ligases. Cul1p complexes were immunopurified from the indicated strains and employed in substrate-independent in vitro ubiquitylation reactions with purified E1, the E2 Cdc34p, ubiquitin, and ATP \[26\]. Polyubiquitin chains formed in the reaction are indicated (top panel). In contrast to csn5mutants, derepression of Cul1p activity is not observed in cells lacking eif3f, eif3m, or eif3e. The neddylation state of Cul1p was determined by immunoblotting (middle panel). Hyperneddylated Cul1p only accumulates in csn5mutants. Shut-off of eif3fand eif3mexpression was verified by immunoblotting with anti-Myc antibodies (lower panel). ::: ![](1741-7007-3-14-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Distinct eIF3 complexes. (A) eIF3 complexes associated with eIF3e and eIF3m were affinity-purified by absorption of the respective pro-A-tagged proteins to IgG beads. Bound proteins were eluted by cleavage with SDS (eIF3e) or TEV protease (eIF3m). (B) The eluates described in (A) were analyzed by immunoblotting with eIF3d antibodies. The asterisk refers to cross-reactivity of the secondary antibody with Ig light chains. (C) Lysates from cells expressing Myc-tagged Csn5p, eIF3e, and eIF3m from their endogenous genomic loci were immunoprecipitated with anti-eIF3d antisera. Coprecipitated proteins were identified by immunoblotting as indicated. Total cell lysates are shown to document expression levels of the endogenously tagged proteins. (D) Protein lysate from cells expressing protein A-tagged eIF3b from the endogenous promoter was absorbed to IgG beads, and specifically retained proteins were identified by LC-MS/MS. ::: ![](1741-7007-3-14-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Association of mRNAs with eIF3 complexes. (A) Messenger RNAs associated with eIF3e and eIF3m complexes were identified by microarray hybridization as described in the Methods section. The graph indicates the numbers of mRNAs enriched more than three-fold over mock in both complexes. (B) Classification into functional groups of the 106 mRNAs most highly and uniquely enriched in complexes with eIF3e or eIF3m. (C) The relative expression ranks of the 106 most highly and uniquely enriched mRNAs in the eIF3e and eIF3m complexeswere determined using the relative hybridization signals obtained with total S. pombecDNA (\[see [additional file 1](#S1){ref-type="supplementary-material"}\]). The expression ranks were blotted against the factor of enrichment of each mRNA in the eIF3 complexes. No correlation between expression rank and enrichment is apparent, indicating that highly expressed mRNAs were not unspecifically enriched in eIF3 complexes. The 5407 individual features on the microarray slides represent the 4988 ORFs and transcripts predicted in the Sanger Centre S. pombegenome database and various controls. To simplify the analysis, expression ranks were also assigned to these control spots, thus resulting in expression ranks for some ORFs higher than the theoretically possible number of 4988. ::: ![](1741-7007-3-14-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Specific association of mRNAs with the eIF3e and eIF3m complexes. (A) The eIF3 complexes shown were affinity-purified as described in the Methods section, and separated by SDS-PAGE followed by staining with Coomassie Brilliant Blue. The labeled bands were identified by LC-MS/MS. eIF3f\ refers to a band in the eIF3m complex that was not identified by mass spectrometry in this experiment, and is therefore assigned based on its comigration with eIF3f identified in the eIF3e complex (right lane). eIF3f was positively identified by MALDI-TOF mass spectrometry as a subunit of the eIF3m complex in Fig. 1B. (B) The indicated eIF3e- and eIF3m-associated mRNAs were extracted from the purified complexes and employed in RT-PCR reactions using primers specific for each mRNA (see Table 3). PCR products obtained with total S. pombeRNA are shown for reference in the left panel. The factor of enrichment of each mRNA in the respective eIF3 complex and the enrichment rank (out of all 106 mRNAs enriched in either complex) are indicated below the gels. ::: ![](1741-7007-3-14-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of eIF3 subunits in various eukaryotes ::: *Unified Nomenclature* *Domain* *Human* *S. cerevisiae (Core subunits)* *S. pombe Csn7Bp complex* *S. pombe Int6p complex* *S. pombe eIF3b complex* ---------------------------- --------------- ------------- ------------------------------------- ------------------------------- ------------------------------ ------------------------------ eIF3a PCI p170 Tif32p p107 p107 p107 eIF3b RRM p116 Prt1p p84 p84 p84 eIF3c PCI p110 Nip1p p104 p104 p104 eIF3d \- p66 \- \- Moe1p Moe1p eIF3e PCI p48 \- \- Int6p Int6p eIF3f MPN p47 \- Csn6p Csn6p Csn6p eIF3g RRM/Zn finger p44 Tif35p Tif35p Tif35p Tif35p eIF3h MPN p40 \- p40 \- p40 eIF3i WD repeat p36 Tif34p Sum1p Sum1p Sum1p eIF3j \- p35 \- \- \- \- eIF3k PCI p28 \- \- \- \- eIF3l PCI p67 \- \- \- \- eIF3m PCI GA17 (?) \- Csn7Bp \- Csn7Bp ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Strains used in this study ::: Name Genotype Source ---------- ----------------------------- ----------------------------------------- ------------ DS448/1 leu1-32 ura4-d18 ade6-704 h+ Lab stock DS448/2 leu1-32 ura4-d18 ade6-704 h- Lab stock C399/3 leu1-32 ura4-d18 ade6-704 csn5.13myc kan h- Lab stock C485/3 leu1-32 ura4-d18 ade6-704 eif3m.13myc kan h- Lab stock C617/1 leu1-32 ura4-d18 ade6-704 eif3m.CBP.tev.5proA kan h- Lab stock C642/1 leu1-32 ura4-d18 diploid, eif3f/eif3f::ura4 This study C642/2 leu1-32 ura4-d18 diploid, eif3f/eif3f::ura4 This study C648 leu1-32 ura4-d18 ade6-704 eif3e.cbp.tev.5proA kan h+ This study C650/1 leu1-32 ura4-d18 ade6-704 eif3f::ura4 pRep81.6xhis.myc.eif3f h- This study C650/2 leu1-32 ura4-d18 ade6-704 eif3m::ura4 pRep81.6xhis.myc.eif3m h- This study C652/1 leu1-32 ura4-d18 ade6-704 eif3e::ura4 h+ This study C663/1 leu1-32 ura4-d18 ade6-704 eif3m.gfp kan h- This study C663/2 leu1-32 ura4-d18 ade6-704 eif3e.gfp kan h- This study C665 leu1-32 ura4-d18 ade6-704 eif3e.13myc kan h- This study C701 leu1-32 ura4-d18 ade6-704 eif3b.gfp kan, h- This study C702 leu1-32 ura4-d18 ade6-704 eif3b.cbp.tev.5proA kan h- This study ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### RT-PCR primers ::: Gene Name Sequence Product Size (bp) --------------- ---------------------- ----------------------- actinF GAGCTTCCTGATGGTCAAGT 200 actinR GGATACATAGTGGTACCACC cdc16F AAGTTAGCGCCCAAATCACC 150 cdc16R TTTTGAATGCCCCCCCACGG gar2F GGTGCCATTGAGAAACCTTC 150 gar2R CCGAACCTTCAAAGAAATCA pof6F TCTGATCGGCCTAAGCTGTC 150 pof6R CAATCTTGCAAATTCAACTC rpn5F TGAGAAGCAAGTTCGTCAGG 160 rpn5R TGAAACAATCGATGACAAGT rps8-1F AAGCGTATTCACGAGGTCCG 150 rps8-1R CAACTCGTTGTTAGAAGGGT scd1F TCAGAGTTGGCTGCTTTCTT 150 scd1R ATCCATTGTGTGCCCTGTTC SPAC1348.05F CTTAGTGAACAGTTTGGAAG 150 SPAC1348.05R TGATAAACCAACGGATCCGA SPAC21E11.04F GAGACATCACCTGCTCCAGA 150 SPAC21E11.04R TTTGCTCCGGTGACTAGGTG SPBC12C2.06F TCTGTTCCCAAACCTCAAGC 150 SPBC12C2.06R ATCGATTTTTGCACCTTTAG SPBC1683.01F GGCCGTAAATTTGTCTACGG 150 SPBC1683.01R ACCACCAATACCAACACCAA SPBC36B7.03F CTCTCAATTAAATTTCACCC 150 SPBC36B7.03R AGGAGTACCGTATAAAGCAT SPCC70.05cF GTACCTGGAAATAACTCTCC 150 SPCC70.05cR CATAGCCTTTTCTAAGAGAT tf2-12F AAGCATGTACCAGAGATAGG 150 tf2-12R TGAATCACCTAGAAGAATTA ung1F ACTTTGGAGAGTTCTTGGTT 150 ung1R TGGAGTATGATGTGACCATG F = Forward; R = Reverse :::	PubMed Central	2024-06-05T03:55:58.339959	2005-5-17	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173091/", "journal": "BMC Biol. 2005 May 17; 3:14", "authors": [ { "first": "Chunshui", "last": "Zhou" }, { "first": "Fatih", "last": "Arslan" }, { "first": "Susan", "last": "Wee" }, { "first": "Srinivasan", "last": "Krishnan" }, { "first": "Alexander R", "last": "Ivanov" }, { "first": "Anna", "last": "Oliva" }, { "first": "Janet", "last": "Leatherwood" }, { "first": "Dieter A", "last": "Wolf" } ] }
PMC1173092	Background ========== Prostate cancer is the most common cancer in men resulting in over 30,000 deaths annually \[[@B1]\]. Early detection and treatment has the potential to markedly reduce the morbidity and mortality associated with the disease. While elevated Prostate Specific Antigen (PSA) \[[@B2]\] is the best available indicator of men with cancer \[[@B3]\], its diagnostic utility is limited due to elevated PSA levels in other non-malignant prostate conditions, varying levels in advanced disease and poor correlation between PSA levels and extent of disease. Furthermore, the variable course of prostate cancer -- many patients will not die of the disease -- means that radical therapy for all early cases would result in over treatment of significant number of patients. High throughput genomic technologies, by simultaneously interrogating the expression levels of thousands of genes, offers the potential to identify new biomarkers for early detection, prognosis, targets for therapy and for reclassification of prostate tumors. Using expression microarrays, a number of studies have characterized expression profiles for prostate cancer and other tumors. In some cases, correlations between tumor expression signatures, clinical parameters and outcome \[[@B4]-[@B12]\] have been identified. While potentially powerful, such studies can be significantly impacted by the choice of baseline or \"normal\" tissue used to detect tumor related expression changes. Most prostate expression studies to date have utilized normal appearing tissue adjacent to tumor as the tissue for comparison. However, a variety of methods such as chromosomal analysis \[[@B13]\], SAGE \[[@B14]\] and ploidy analysis \[[@B15],[@B16]\] have shown molecular abnormalities in normal appearing prostate adjacent to tumor. Even the term \"normal appearing\" prostate tissue adjacent to tumor may be misleading, as morphologic researchers using quantitative imaging analysis \[[@B17]-[@B19]\], have identified morphologic changes in the epithelial nuclei and blood vessels architecture in prostate tissue adjacent to tumor that are not routinely commented upon by pathologists. This suggests that, in some cases, tissues adjacent to cancer, although appearing morphologically normal by traditional microscopic examination, may contain genetic changes associated with the genesis of or reaction to cancer. Therefore, the use of adjacent normal as the baseline tissue for comparative gene expression studies may mask tumor related molecular changes preceding the appearance of histological tumor. More recently, a microarray study from our institution \[[@B12]\], using adjacent normal and tumor samples, describes a potential field effect around prostate cancers and regulation of selected genes in both adjacent normals and tumors. Using the same microarray data set for our analysis, we have compared the gene expression profiles of prostate cancer, normal appearing prostate tissue adjacent to tumor, and normal appearing prostate tissue from cancer free tissue donors with the aim of identifying the optimal baseline tissue for expression studies and the gene expression changes between the three specimen types. Methods ======= Clinical profile of cases ------------------------- The 60 tumor samples used in this study consisted of 2, 13, 27, 6 and 12 cases of primary prostatic adenocarcinoma of Gleason grade 5, 6, 7, 8 and 9 respectively. There were 4, 20, 23 and 13 cases spanning the age groups 40--49, 50--59, 60--69 and 70--79 respectively. Of the cases, 36 were stage T3 or higher with 2, 22, 23, 11 and 2 cases of stage T2a, T2b, T3a, T3b and 4 respectively. The 63 adjacent normal samples consisted of 2, 11, 29, 8, and 13 cases of Gleason grade 5, 6, 7, 8 and 9 respectively. There were 4, 21, 25 and 12 cases spanning the age groups 40--49, 50--59, 60--69 and 70--79 respectively. There were 2, 21, 26, 12 and 2 cases of Stage T2a, T2b, T3a, T3b and T4 respectively. Of the donors, 11 are under and 7 are over the age of 40. Samples and sample procurement ------------------------------ The tumor and adjacent normal tissue samples were acquired from the University of Pittsburgh Medical Center under stringent Institutional Review Board guidelines with appropriate informed consent. Specimens were received directly from the operating room. Samples (\>500 mg) were excised and snap frozen in liquid nitrogen within 30 min of excision and stored at -80°C in the University of Pittsburgh Pathology Tissue Bank until extraction of RNA. All samples were submitted for pathology evaluation. In every case, the tissue was excised from the junction between the ejaculatory duct and the prostatic urethra in the transition zone of the prostate. In particular, adjacent normal tissue was excised away from the cancer lesion macroscopically, and their histological diagnosis was confirmed microscopically. Donor tissue specimens were received through a collaborative arrangement with the Center for Organ Recovery and Education (CORE), the local organ procurement agency. The arrangement allows the University of Pittsburgh Pathology Tissue Bank to acquire normal prostates and associated serum/plasma specimens from healthy individuals who have donated their organs for transplant. There is extensive collaborative support from CORE. The donor prostatectomies harvested from brain dead, perfused donors and are bathed in Ringer\'s Lactate solution and transported on wet ice. These donor prostates are transported and handled with the harvested \"transplant\" organs. This significantly reduces transit time and minimizes the degradation of RNA. The processing methodologies used consist of snap freezing tissues in bulk, freezing in OCT and processing the tissues for routine histology (paraffin embedded tissues. For microarray analysis, the donor samples were excised from the same zone as the tumor and adjacent normal samples. cRNA preparation and Affymetrix chip hybridization -------------------------------------------------- cRNA was prepared and hybridized to Affymetrix oligonucleotide arrays as previously described \[[@B12]\] Statistical analysis -------------------- We analyzed prostate tissue samples from 18 donors and 63 prostate cancer patients. From the prostate cancer patients we took samples from the histologic tumor as well as normal appearing tissue adjacent to the tumor. High quality RNA and chip data were obtained from 60 cancer and 63 adjacent to tumor samples. In total 141 samples were run against the Affymetrix U95A chip and analyzed. The raw scanned array images were first processed through the Affymetrix Microarray Analysis Suite 5.0 (Affymetrix Corporation, Santa Clara, Ca) to generate probe cel intensity (\.cel) files. The \.cel files were then analyzed using both MAS 5.0 and dChip software from Harvard University \[[@B20]\], to generate gene expression signal values for each probe set. Data normalization to remove variation in overall chip intensities was perfumed by global scaling to a chip mean target intensity of 200 (MAS 5.0) or by the rank-invariant method (dCHIP). The MAS 5.0 with global scaling data and dChip with rank-invariant normalization data gave similar results in the subsequent analysis. Therefore, in the interests of clarity, we will focus on the MAS 5.0 results in the remainder this paper. In the next phase of analysis, the donors, adjacent normals and tumors were compared for differences in gene expression by using signal values for all 12625 probe sets for each sample. For statistical analysis, we used the Significance Analysis of Microarray (SAM) software package from Stanford University \[[@B21]\]. This method was chosen over conventional statistical tests because of its acceptance in the microarray community, its general simplicity and its ability to provide an estimate of the false discovery rate (the ratio of false positives to total positives). The false discovery rate is particularly important when comparing the expression of thousands of genes simultaneously. For example, when using the Student\'s t test at a P value of 0.05 to examine a population of 10,000 genes, one would expect 500 false positives. If there were in fact 100 true positives, the false positive rate would be an unacceptably high 0.83 (500/600). Briefly, SAM calculates a value for each probe set on the array. This value represents the observed difference in mean expression levels between the specimen classes being compared (i.e. tumor and donor) divided by the variance in the data and a fudge factor (see the original paper for details \[[@B21]\]. The resulting value is called the \"observed d value\". To determine the significance of this value, SAM estimates the \"expected\" d value if there were no difference between the specimen classes. This is done by permutating (randomly changing) the class labels without changing the data and recalculating the SAM value for each probe set. After thousands of permutations, the result estimates the value that would be obtained if the difference in gene expression were due to chance alone. This is the \"expected d value\". The significance of the observed differential gene expression can be estimated by comparing the observed and expected d values. A user defined threshold or \"delta\" (observed d value -- expected d value) can be adjusted to select only those genes observed d value exceeds (for up regulated genes) or is lower (for down regulated genes) than delta. The greater the \"delta\", the greater the stringency of the result and lower the false discovery rate. For each delta value, the SAM output consists of a gene (probe set) list and an associated false discovery rate. The false discovery rate is estimated from the distribution of expected and observed d values. Probe sets are ordered on the basis of observed d value metric, probe sets with high (or low) values represent genes with relatively high differential expression. The \"SAM Plots\" are also very useful in visualizing differences in overall differential gene expression between specimen classes. Results ======= Expression analysis of tumor, adjacent normal and donor tissue -------------------------------------------------------------- The prostate tumors analyzed in this study consisted of Gleason grades 5, 6, 7, 8, 9 and patients spanned the ages of 40 through 79. The goal of our research was to examine the differential gene expression patterns observed when comparing our three specimen classes: tumor versus adjacent normal, tumor versus donor and adjacent normal versus donor. The comparison was made at three points in the analytical process: 1) after normalization to remove variation in overall chip intensity 2) after statistical analysis of the data and, 3) after examining the differentially expressed gene lists. To examine differences in normalized gene expression between tumors, adjacent normals and donors the mean MAS 5.0 and dChip generated signal of each probe set for each specimen class (60 tumors, 63 adjacent normals and 19 donors), was calculated and plotted on a series of scatter plots (Fig [1](#F1){ref-type="fig"}). Figure [1](#F1){ref-type="fig"} shows the scatter plots and Pearson correlations of the normalized expression data analyzed using both MAS 5.0 and dChip as described above (vide supra, Methods). Data scatter is maximum in the tumor versus donor comparison, intermediate in adjacent normal versus donor and minimal in tumor versus adjacent normal. These findings are suggestive of more differential gene expression in tumor versus donor than tumor versus adjacent normal. In other words, donor normal tissue and adjacent normal tissue do not show the same degree of differential gene expression when paired with tumor tissue. Another striking result apparent in Figure [1](#F1){ref-type="fig"} is the close correlation and limited scatter of the tumor versus adjacent plot, even at low levels of signal. Tumor and adjacent normal specimens came from the same population of patients while donor specimens were received from a different set of individuals. To examine potential patient specific expression effects, the 60 tumors and 63 adjacent normal cases were randomly segmented into two groups, one group provided just tumor data and the other just adjacent normal data and a scatter plot of expression was generated. Since the segmentation of 63 cases can be performed in many different ways (permutations), the scatter analysis was performed 1000 times and the correlation between the sample groups determined by obtaining the mean correlation coefficient of the 1000 permutations. (Figure [2](#F2){ref-type="fig"}). In this analysis, the close correlation in expression between tumor and adjacent normal specimens persisted even when tumor and adjacent normal samples were taken from different patients. To determine the statistical significance of the observed differential expression between the three specimen groups, SAM analysis was performed. From each comparison (tumor v adjacent normal, tumor v donor and adjacent normal v donor), a SAM plot was generated and the plots for the three comparisons were overlaid (Fig [3](#F3){ref-type="fig"}). The diagonal line in Figure [3](#F3){ref-type="fig"} represents no differential expression (identical observed and expected d values, for further details see Materials and Methods) with points displaced from the diagonal representing differential expression. Figure [3](#F3){ref-type="fig"} shows that each of the comparisons yields a distinct expression profile with donor v tumor exhibiting more differential expression than adjacent normal v tumor or donor v adjacent normal. To further characterize the expression profiles from these comparisons, differentially expressed gene lists were created from each comparison by selecting genes whose d values (for details, see Materials and Methods) exceed a given threshold. False discovery rates (false positives/ total number of genes in gene list) is no greater than 2.5% at the deltas chosen for this analysis (Table [1](#T1){ref-type="table"}). At a delta of 2.0, when tumor expression is compared to donor expression (Table [1](#T1){ref-type="table"}), 474 differentially regulated genes can be detected. At the same delta, when tumor expression is compared to adjacent normals, only 92 genes are differentially regulated between these two tissues. Furthermore, at this delta, comparison of tumor expression with adjacent normals does not yield any genes up-regulated in tumors whereas the comparison with donors demonstrates up-regulation of 121 genes. Similarly at other deltas, approximately three times more differentially regulated genes can be detected when tumors are compared to donors than to adjacent normals. As was discussed above, tumors and adjacent normal tissues are obtained from the same patients and donor tissues from a different sample population. Therefore the larger gene expression differences between tumors and donors may represent underlying patient specific (genetic, demographic or handling) differences in patient (tumor and adjacent normal) and donor prostates rather than intrinsic differences between tumor, adjacent normal and donor normal tissues. It is significant however, that SAM analysis indicates that adjacent normals exhibit far less differential regulation than tumors when both are compared to donors. At all deltas (Table [1](#T1){ref-type="table"}), tumors v donors exhibit greater differential expression than adjacent normals v donor implying that tumors and adjacent normals are not identical in gene expression. Therefore, tumor specific, and not patient specific, expression changes can indeed by detected by comparing tumors to donor prostates. Significantly, these results establish the presence of unique gene expression profiles for prostate tissue from donors, adjacent normals and tumors (see Fig [3](#F3){ref-type="fig"}) with tumors differing more from donors than from adjacent normals. A potential limitation of our data is that donors span the ages of 5 to 60 and all tumor patients are older than 40. Therefore, the differential gene expression between donors and patients may be due to age specific differences in their prostates. To examine this, we segmented the donors into different age groups and compared only the 40 to 60 year old donors with tumors of the same age group. Although the number of cases in the study were small, the expression pattern observed in this age matched analysis is identical (data not shown) to that when all donors are included suggesting that potential age related differences in donor prostates do not contribute to the results of the donor v tumor analysis. GO annotation of differential gene expression --------------------------------------------- We examined the gene lists produced by SAM analysis of tumor, adjacent normal and donor tissue with two objectives: 1) to identify and functionally annotate some of the genes that contribute to the unique expression signatures of these tissues and 2) to determine whether adjacent normals or donors are the more appropriate baseline tissue for detecting differentially expressed genes in tumors. Functional annotation and comparison of the gene lists was performed using Gene Ontology terms \[[@B22]\], for biological processes and Affymetrix\'s Gene Ontology Mining Tool <http://www.affymetrix.com>. When donors are used as the baseline for comparison, tumors exhibit up-regulation of proliferation related genes including transcription factors, signal transducers and growth regulators (see [additional file 1](#S1){ref-type="supplementary-material"}). This list includes putative oncogenes, signal transducers and growth regulators. Some of the most up-regulated genes are v-fos, jun B, jun D, c-src tyrosine kinase, FGF receptor activating protein, immediate early proteinand early growth response 1. The most down-regulated genes in tumors include those involved in immune response and signal transduction. Some of the genes in this list are the interferon induced transmembrane proteins, Duffy blood group antigenand tumor necrosis factor. In contrast, when adjacent normal tissue is used as the baseline for comparison, tumor tissue exhibits far fewer differentially expressed genes and the genes themselves are less compelling. The list up regulated genes is dominated by ribosomal proteins and metabolic enzymes, while the down-regulated list includes muscle related genes such as tropomysin, actinand actinin. When expression in adjacent normal is compared to donors, an up-regulation pattern remarkably similar to tumors is seen ([additional file 1](#S1){ref-type="supplementary-material"}). Adjacent normals also exhibit up-regulation of putative oncogenes, signal transducers and growth regulators with an almost 70% overlap of the 50 and 100 most up-regulated genes in tumors and adjacent normals, respectively. Similarly there is almost 60% overlap between the most down-regulated genes in tumors and adjacent normals that includes genes involved in immune response. The biological processes regulated in tumors and adjacent normals were also studied using Affymetrix\'s Gene Ontology Mining tool. The up regulated gene lists obtained at a SAM delta of 2.0 (Table [1](#T1){ref-type="table"}) were uploaded to the tool and the resulting annotations examined. Comparison of tumor gene expression to donor expression reveals up-regulation of genes involved in a number of biological processes (Figure [4a](#F4){ref-type="fig"}). Amongst these are genes involved in apoptosis, cell cycle, cell proliferation, immune response, protein phosphorylation, protein biosynthesis and transcription. A subset of these including genes involved in immune response and transcription are also up-regulated in adjacent normals (Figure [4b](#F4){ref-type="fig"}). In contrast when tumor expression is compared to adjacent normals, up-regulation of majority of these processes, except protein metabolism, is not detected (Figure [4c](#F4){ref-type="fig"}). Two important conclusions can be derived from the gene annotations, 1) though there are large number of genes regulated in tumors, there is a relatively small subset of genes including oncogenes and signal transducers that are highly regulated in both adjacent normal and tumor tissues and 2) regulation of a number of potentially important biological processes in tumors can be detected from using donors as the baseline tissues. The common regulation of oncogenes, signal transducers and immune response genes in adjacent normals is a striking result in that it suggests that adjacent normal tissue although appearing morphologically normal, undergo gene expression changes that may be important in tumorigenesis or as a reaction to tumor. Since these genes are regulated in both tumor and adjacent normal, they are not picked up on a direct comparison of the two tissues. While it is possible that donors are different from both adjacent normals and tumors due to processing artifacts -- the tumor and adjacent tissues were taken at surgery and donors at harvesting -- it is unlikely that the large differences seen in donor v tumor are all due to processing differences. This issue is examined further in the discussion section. The up regulation of proliferation markers in both adjacent normals and tumors coupled with the result that more differential regulation is detected when tumors are compared to donors than to adjacent normals suggests that donor prostates may be the more appropriate tissue for expression studies. Regulation of critical of biological processes and pathways may remain undetected if tissue adjacent to tumors is used for comparison. Discussion ========== There is a growing interest in the use of high throughput microarray analysis for the molecular reclassification of diseases. This interest appears to be well founded, as many groups have reported consistent patterns of gene expression associated with pathologic phenotypes, clinical behaviors and outcomes \[[@B4]-[@B11]\]. In the area of prostate cancer numerous groups \[[@B23]-[@B29]\] have all reported significant differential gene expression between histologic tumor specimens and normal appearing prostate tissue from patients with tumor present elsewhere in the prostate. Recently, a group from our institution reported a 70 gene signature that may predict aggressiveness of prostate cancer \[[@B12]\]. Comparison of the gene lists from published data sets with the results of our tumor versus adjacent normal analysis is complicated by the heterogeneity in samples, analysis platforms and analysis methods. Nevertheless, our study is qualitatively similar to other studies in the expression profile of tumors compared to adjacent normal tissue. A number of genes including hepsin, myc, fatty acid synthase SPARC1and EBNA-2coactivator show similar expression patterns across multiple prostate cancer studies \[[@B30]\], and are also regulated in our study. Our donors did not have prostate cancer or prostatic intraepithelial neoplasia (PIN) identified in their prostate and as such are good candidates for \"true normals\". Differential expression was much greater between tumor and donor tissue than between tumor and adjacent normal. The fact that tumor and adjacent specimens come from the same patients could possibly explain this difference but this was ruled out by our analysis. Another possibility is that tissue handing and processing differences could account for some or all of the differential expression seen when donor tissue is use as a baseline. In fact, data in the literature does suggest that tissue processing could effect the expression of genes such as fos, junand egrin prostate tissue \[[@B29]\]. However, the same literature indicates that the effect warm ischemic time is limited to specific genes and in general, involves less than 1% of the regulated genes \[[@B29]\]. Our studies emphasize the need for documentation and quality of all experimental processing steps, from sample acquisition to sample hybridization, in order to completely characterize gene expression differences between prostate donors, tumors and normal tissue adjacent to tumors. In our experiment, tumor and adjacent normal specimens where taken from the same prostates and handled the same way. If differences in patient and donor tissue handling was the major issue driving differential expression in the tumor v donor and adjacent normal v donor comparisons, one would expect tumor v donor and adjacent normal to result in very similar expression profiles. However, we have shown that tumor v donor exhibits far greater differential expression than adjacent normal v donor (see Results). Furthermore, the differentially expressed genes seen in both tumor and adjacent normal include proto-oncogene and transcription factors that one might rationally expect to see in expectation of or in response to a local tumor. Therefore, while the possibility that some expression changes are due to differences in tissue handling cannot be formally ruled out, it is unlikely that the large and specific differences we observe in tumor v donor, tumor v adjacent normal and adjacent normal v donor are entirely due to processing differences. Clearly additional studies, including examination of patient process specimens that do not host prostate cancer (such as cystoprostatectomy for bladder cancer or prostates removed for benign hypertrophy) to examine this process further. The most important finding from our analysis is the potential importance of the donor specimens and the possibility that a field effect exists around prostate tumors, resulting in significant molecular changes in histologically normal appearing tissue adjacent to prostate cancer. Significantly, evidence for such malignancy associated changes have been presented in other organs such as the cervix, bladder and breast \[[@B31]-[@B33]\] Furthermore, a variety of methods such as chromosomal analysis \[[@B13]\], SAGE \[[@B14]\], ploidy analysis \[[@B15],[@B16]\] have shown molecular abnormalities in normal appearing prostate adjacent to tumor. Image analysis has also been employed to identify consistent changes in \"normal appearing\" prostate tissue adjacent to tumor \[[@B17],[@B18]\]. In one study cases of prostatic adenocarcinoma was consistently detected by examining histologically normal tissue using high-resolution image cytometry \[[@B18]\], and in another, combined highly sensitive and discriminating Fourier transform-infrared spectroscopy with statistical analysis was used to detect damaged DNA in normal appearing prostate tissue adjacent to cancer \[[@B34]\]. In expression analysis, while most published prostate studies have used adjacent normals as the baseline tissue, Dhansekaran \[[@B23]\], used both commercially available pooled donor normal tissue and adjacent normal tissue and noted differences in expression profile between the two specimen types. Genes that were differentially expressed in adjacent normals when compared to the pooled donor normals included signal transducers and transcription factors; and expression of these genes in adjacent normals was attributed to a field effect around tumors. Similarly, Yu \[[@B12]\] have noted dysregulation of selected genes in both adjacent normals and tumors when compared to donors. Prakash \[[@B27]\], found that gene expression in asymptomatic benign prostatic hyperplasia adjacent to tumors was different from asymptomatic BPH or symptomatic BPH not associated with tumors. The unique expression signature of BPH next to tumors included fos, jun, immediate early genes and this list was remarkably similar to the most up-regulated genes in the adjacent normals tissue in our study (see adjacent normal v donor, [additional file 1](#S1){ref-type="supplementary-material"}). Finally, within archives of the University of Pittsburgh Pathology Tissue Bank, there was a donor prostate, which was found to harbor prostate cancer. When run on the Affymetrix arrays, the tumor classified with the tumors samples rather than the donor samples. Although this is clearly no more than an anecdotal event, it is an interesting finding. Though microarray technology represents a major advance and provides a powerful tool for high-throughput expression analysis, the most effective use of this technology requires careful consideration of baseline normal tissue. Our results here emphasize the need for careful examination of what constitutes normal tissue and the importance of future studies to fully characterize normal appearing tissue adjacent to prostate cancers. Conclusions =========== Prostate tumor tissue, histologically normal tissue adjacent to tumors and donor normal prostate tissue exhibit unique gene expression profiles with tumor and adjacent normal profiles more similar to each other than to the donors. These results suggest that normal appearing tissue around prostate tumors may also be undergoing tumor related changes and that careful characterization of these different tissues is necessary to understand molecular changes in leading up to prostate cancer. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= URC was involved in the data analysis and interpretation and manuscript writing and revision. RD, as director of the Tissue Bank was involved in obtaining IRB approval and providing deidentified tissue for this research study. CM was involved in data analysis and interpretation. GM and MB as principal investigator and co-investigators of the grant titled \"the Molecular Reclassification of Cancer\" were responsible for design of the study, providing intellectual input and final approval of the version submitted for review. JG was responsible for providing intellectual direction, guidance for the analysis team, review and final approval of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/45/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 The fifty most up regulated and down regulated genes from the tumor v donor, tumor v adjacent normal and adjacent normal v donor comparison. The Affymetrix probe set id, gene names and assignment of biological process for each gene are shown. The biological process annotation includes information from Affymetrix, gene ontology and literature references. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was funded by 1 U01 CA86735-01 (to MB)\"A Shared Resource for the Molecular Classification of Prostate Cancer Comprehensive Prostate Cancer Tissue Resource (CPCTR) Consortium, National Cancer Institute, 5 U01 CA88110-02 (to GM, MB is Co-Principal Investigator) \"Molecular Reclassification of Prostate Cancer\" Director\'s Challenge for the Molecular Classification of Cancer Consortium National Cancer Institute and 2000-98 (to JG) DOD National Medical Technology Test Bed An Advanced Tissue Banking Information System Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Differential gene expression analysis of donor, tumor and adjacent normal protate cancer samples.Scatter plot of MAS 5.0 derived tumor vs donor, adjacent normal v donor and tumor v adjacent normal samples. For each probe set, the mean MAS 5.0 expression values of all the samples in each specimen group was calculated. Scatter plot were constructed using the mean values for each specimen group. ::: ![](1471-2407-5-45-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Regression analysis of permuted donors, adjacent normal and tumor samples.The 60 tumors and 63 adjacent normal tissues were segmented so that tumors and adjacent normal samples in each comparison were selected from different patients. The resulting tumor and adjacent normal samples were then subjected to regression analysis. Donor v tumor, donor v adjacent normal and adjacent normal comparisons were performed. Since the segmentation can be performed in many different ways, the analysis was performed 1000 times. The mean correlation coeifficient and standard deviation from each of these comparisons were plotted as box plots. ::: ![](1471-2407-5-45-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Overlayed SAM plots (for details, see Materials and Methods) from the donor v tumor, donor v adjacent normal and tumor v adjacent normal analyses.Each of the SAM plots was overlayed to direct comparison of the plots. The diagonal line represents no differential gene expression where the observed d value equals the expected d value after 1000 permutations of the class labels. Genes that are differentially expressed are displaced from the diagonal (greater than 0 for up regulation and less than 0 for down regulation). Genes that are more differentially expressed are more displaced from the diagonal than those that are closer to the diagonal. For each of the comparisons, a plot is generated from the d values of the 12625 probe sets in the two specimen groups. Red = donor v tumor plot; green = adjacent normal v tumor plot; black = adjacent normal v donor plot. ::: ![](1471-2407-5-45-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Gene Ontology annotation of differentially expressed gene lists.The fifty most upregulated genes from the donor v tumor, adjacent normal v tumor and tumor v adjacent normal comparisons were uploaded to Affymetrix\'s Gene Ontology Mining Tool, a, donor v tumor; b, donor v adjacent normal; cadjacent normal v tumor; The annotations is presented as a hierarchy of terms, from general to most specific terms (from left to right). The numbers in parenthesis indicate the number of genes that are annotated with the term. In all of the analysis, annotation of all the submitted probe sets is not achieved. Typically, annotation exists for approximately 60% of the probe sets. ::: ![](1471-2407-5-45-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Differential gene expression in the tumor v donor, tumor v adjacent normal and adjacent normal v donor comparisons. The number of genes identified as differentially regulated at each delta (observed d value -- expected d value; for details, see Materials and Methods) are shown. Also, shown are the number of up and down regulated genes at each delta. For each of these deltas, the false discovery rate was no greater than 2.5%. ::: Delta 3.0 Delta 2.0 Delta 1.56 ----------------------------- --------------- --------------- ---------------- Tumor v Donor 65 474 928 Up Regulated 10 121 305 Down Regulated 55 353 623 Tumor v Adjacent Normal 0 92 382 Up Regulated 0 58 Down Regulated 92 324 Adjacent Normal v Donor 12 86 254 Up Regulated 5 25 63 Down Regulated 7 61 191 :::	PubMed Central	2024-06-05T03:55:58.344404	2005-5-13	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173092/", "journal": "BMC Cancer. 2005 May 13; 5:45", "authors": [ { "first": "Uma R", "last": "Chandran" }, { "first": "Rajiv", "last": "Dhir" }, { "first": "Changqing", "last": "Ma" }, { "first": "George", "last": "Michalopoulos" }, { "first": "Michael", "last": "Becich" }, { "first": "John", "last": "Gilbertson" } ] }
PMC1173093	Background ========== Breast cancer is the most frequently diagnosed cancer among women in the United States \[[@B1]\]. Of 668,470 expected female cancer cases in 2004, 215,991 (32.3%) were breast cancer \[[@B1]\]. The incidence of female breast cancer is most common in developed countries such as the U.S. and Western Europe \[[@B2]\], and is higher among whites than other racial and ethnic groups \[[@B3]\]. Among the registries in the Surveillance, Epidemiology, and End Results (SEER) Program of the U.S. National Cancer Institute, female malignant breast cancer incidence rates are lowest in Utah \[[@B3]\], despite approximately 85% of the population being white, non-Hispanic \[[@B4]\]. About 70% of the 2.3-million population in Utah is affiliated with the Latter-day Saint (LDS or Mormon) religion \[[@B4],[@B5]\]. Hence, female breast cancer incidence rates in Utah are largely representative of LDS women and their lifestyle behaviors. A study based on Utah data from 1971 through 1985 showed that LDS women had significantly lower age-adjusted breast cancer incidence rates than did non-LDS women \[[@B6]\]. A recent cross-sectional study involving non-Hispanic white females in Utah compared breast cancer risk factor behaviors between LDS and non-LDS \[[@B7]\]. This study focused on many factors previously shown to influence breast cancer, including age at first birth \[[@B8],[@B9]\], number of pregnancies and number of births (parity) \[[@B10]\], lifetime duration of breastfeeding \[[@B10],[@B11]\], and alcohol consumption \[[@B12]\]. LDS women compared with non-LDS women had a slightly older age at first birth, higher average number of pregnancies and children (parity), and more total years of breastfeeding. LDS women also displayed lower levels of alcohol consumption. A cross-sectional study conducted in Utah in the late 1970s likewise found LDS women to have a later age at first pregnancy, more pregnancies, and lower prevalence of alcohol consumption \[[@B13]\]. However, research has also shown that reproductive factors which may decrease the risk of breast cancer development may explain poorer breast cancer survival \[[@B14]\]. The current study updates the findings of Lyon et al. \[[@B6]\] to the years 1985--99 and also examines female breast cancer survival in Utah according to LDS status and level of religiosity for LDS. Methods ======= Utah Cancer Registry (UCR) data were linked to LDS Church membership records to estimate breast cancer survival according to LDS (religiously active and less active) and non-LDS populations in Utah during 1985--99. Utah Cancer Registry -------------------- The UCR, established in 1966, has continuously participated in the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute since 1973. UCR staff members and local cancer registrars identify incident cases of cancer among Utah residents through routine and systematic review of pathology reports, medical records, radiation therapy records, hospital discharge lists, and vital records. Tumor characteristics including histology, grade, and primary site are coded according to the International Classification of Disease for Oncology-Second Edition (ICDO-2) \[[@B15]\]. Breast cancer-specific death is coded according to the International Classification of Diseases, 9^th^Revision, as 174.0 -- 174.9 \[[@B16]\], and 10^th^Revision, which uses an alphanumeric system and codes them as C50 \[[@B17]\]. Categories of stage of disease at diagnosis are documented in the Summary Staging Guide of the SEER Program of the National Cancer Institute \[[@B18]\]. Registry records also include information regarding treatment, survival, and patient characteristics such as age at diagnosis, gender, race/ethnicity, and place of residence at diagnosis. Such information is ascertained from specific statements in medical records, reports from private pathology laboratories and radiotherapy units, and death certificates. Cancer surveillance in Utah is conducted in accordance with standards instituted by the SEER Program and the North American Association of Central Cancer Registries \[[@B19],[@B20]\]. In order to provide valid estimates of cancer survival, a high percentage of cancer cases must be routinely followed to ascertain both vital status and date of last contact. UCR records are linked four times each year with records of death certificates from the Office of Vital Records and Health Statistics from the Utah Department of Health. Results from these routine linkages identify cancer patients who have died, regardless of their cause of death. Registry staff members work closely with cancer registrars in local hospitals to document, through systematic review of medical records, the date of last contact for those cancer patients not known to be deceased. Records for these patients are also linked annually with administrative databases, including Medicaid reimbursement claims, Utah driver license records, and Utah voter registration lists; these databases are often a source of updated information regarding vital status and/or date of last contact. According to SEER Program standards, follow-up is considered to be current when the Registry documents that a patient has died, or when the Registry documents that a patient was alive within 18 months of the annual anniversary of the date of original cancer diagnosis. By this definition, 97.9% of the patients in the present study were determined to have complete follow-up at the time data were captured for analysis. Linkage of cancer data with LDS church records ---------------------------------------------- UCR records were linked to LDS church membership records to determine membership in the LDS church. A person was identified as a member of the LDS church if the membership record included a baptismal date. The linkage process took place under direct supervision of the Church and the UCR and was conducted in the Church\'s Member and Statistical Records Department. After the records were linked, all personal identifying information was stripped from the database to ensure confidentiality. Records were linked using the probabilistic linking program LinkPro \[[@B21]\]. The program calculates probabilities to identify whether a pair of records refers to the same person. Ten variables that were common to both sets of records were used to link the records: first, middle, and last names; birth day; birth month; birth year; gender; zip code; vital status; and maiden name. SOUNDEX versions of the names were used in the matching process, while actual names were used in the review process. Records were linked if they matched on at least seven of the ten variables. Ambiguous links and records that matched on six of the ten variables were manually reviewed. There were approximately 120,000 incident cases of cancer identified in the UCR database from 1973 to 1999 and approximately 6.6 million records in the LDS Church database. There were 81,617 (68%) UCR records linked to a Church membership record. Of these linked records, 74,829 (92%) matched on at least seven of the ten variables. In addition to information about baptism, LDS church membership records include information on whether born in the church and whether endowed in the temple. To receive one\'s temple endowment requires conscientious effort toward obedience to the lifestyle prescribed by LDS doctrine \[[@B22]\]. This includes adherence to certain health practices such as not smoking or consuming alcohol, abstention from sexual relationships outside of marriage, and being honest in ones dealings with others \[[@B23]\]. A man or woman aged 19 year or older who is considered worthy in a personal interview with a church authority may receive a temple endowment. Generally LDS men and women receive their endowment prior to serving an LDS mission or at the time of a temple marriage. Previous research has considered the temple endowment as an appropriate surrogate marker for religious commitment among adults in the LDS church \[[@B24]\]. Although the data allowed us to identify LDS church membership and religiosity among LDS women, information about religious preference and religiosity was not available among non-LDS women. Data ---- The present study was based on 9,619 female microscopically confirmed, actively followed malignant breast cancer cases (ICDO-2 site code C50.0:C50.9) diagnosed in Utah from 1985 through 1999. There were 5,234 (54.4%) religiously active LDS, 1,369 (14.2%) less religiously active LDS, and 3,015 (31.4%) non-LDS. In addition to LDS membership and level of religiosity, the following variables were considered: age, marital status, summary stage (pathological), histological grade, number of primary cancers in one individual, and treatment. Because of the high percentage of whites in Utah \[[@B4]\], analyses are limited to whites only. Age was categorized as 0--34, 35--44, 45--54, 55--64, 65--74, and 75 years and older; marital status was categorized into four groups: never married, married/cohabitating, previously married, and unknown; and calendar years were categorized into four groups: 1985--88, 1989--92, 1993--96, and 1997--99. Number of primary cancers in an individual describes the chronology of diagnosis of all primary malignant cancers over the entire lifetime of the person \[[@B19]\]. Localized tumors are confined to breast, regional tumors are spread to contiguous organs or lymph nodes, and distant tumors are spread to remote organs. Histological grade is defined by the SEER Program Coding Manual and by the International Classification of Diseases for Oncology, Second Edition in increasing level of severity as either low grade (well differentiated), medium grade (moderately differentiated), or high grade (poorly differentiated) \[[@B15],[@B19]\]. These categories are approximately equivalent to the more widely used Gleason grading system as Gleason score 2--4 (low), 5--7 (medium), and 8--10 (high) \[[@B19]\]. Radiation therapy was identified if radiation was a first course of cancer-directed therapy. Surgery was classified as conservative surgery \[none; no cancer-directed surgery of primary site; partial mastectomy, not otherwise specified (NOS); less than total mastectomy, NOS\], mastectomy \[total (simple) mastectomy, NOS; modified mastectomy; radical mastectomy, NOS; extended radical mastectomy\], or other surgery (mastectomy, NOS; surgery, NOS) \[[@B19]\]. Cases who had died but had unknown date of death, whose cause of death was determined by death certificate or autopsy only, or who represented a second or later primary cancer were excluded from the survival analysis. After these exclusions, there were 8,731 cases available for this portion of the study. SEER registries were included in the study if they provided data to the National Cancer Institute during all the years from 1985 through 1999. These registries are as follows: San Francisco-Oakland, Connecticut, Metropolitan Detroit, Hawaii, Iowa, New Mexico, Seattle (Puget Sound), Utah, and Metropolitan Atlanta. Statistical analyses -------------------- Age-adjusted rates were derived using the direct method and standardized to the 2000 U.S. population. Counts used in the rate calculations were microscopically confirmed. LDS and non-LDS breast cancer incidence rates were compared with SEER incidence rates. Ninety-five percent confidence intervals for the incidence rate calculations were based on standard error estimates from the formula SE(rate) ≈ rate/(cancer cases)^1/2^\[[@B25]\]. Population attributable-risk percent is used to estimate the impact of having a certain breast cancer risk profile in Utah. Among LDS, population values were not available by level of religiosity. Hence, rates are presented for LDS and non-LDS. However, level of religiosity among LDS could be estimated among the cases such that the survival analysis considered religiously active LDS, religiously less active LDS, and non-LDS. Survival time was calculated as the time interval between diagnosis and date of last information. For deceased cases, the date of last information was the date of death. For cases not known to be deceased, the date of last information was the date that the case was last known to be alive. Cases with death from breast cancer (the outcome of interest), and all other cases were censored at the time of last information. We estimated breast cancer-specific survival for cases where breast cancer is a single cancer primary or the first of multiple cancer primaries. Bivariate comparisons were evaluated using the chi-square test. Survival estimates were calculated by the Cox proportional hazards method with one-month intervals \[[@B26],[@B27]\]. Statistical significance based on Cox proportional hazards was determined using 95% confidence intervals. The appropriateness of the Cox proportional hazards model was assessed graphically. Statistical analyses were derived in Statistical Analysis System (SAS) software, version 9.0 (SAS Institute Inc., Cary, NC, USA, 2003). Results ======= Bivariate analyses of the association between religion and selected variables for female malignant breast cancer patients in Utah, 1985--99 are displayed in Table [1](#T1){ref-type="table"}. Significant differences in the distribution of patients among these groups were observed across age categories, marital status, cancer primaries, and histological grade. Religiously active LDS patients tended to be older and were more likely married or previously married. Religiously less active LDS were more likely to have multiple cancer primaries. The high percentage of unknown histological grade at diagnosis makes this variable impossible to evaluate. The median difference in rates between LDS and non-LDS women across the study years was 20.2 per 100,000. The smallest difference in rates was 8.5 in 1992 and the largest difference in rates was 43 in 1991. Age-adjusted female malignant breast cancer incidence rates among white women in Utah are presented by LDS status and calendar year in Figure [1](#F1){ref-type="fig"}. The comparatively low rates in Utah compared with SEER (without Utah) exist across calendar years. The lower rates are primarily explained by LDS women, although non-LDS women also contributed to the lower rates. For the combined years 1995 through 1999, lower female malignant breast cancer incidence rates among LDS compared with non-LDS women in Utah were observed across the age span (Figure [2](#F2){ref-type="fig"}). Age-adjusted breast cancer incidence rates were 107.6 (95% CI: 103.9 -- 111.3) for LDS women and 130.5 (123.2 -- 137.9) for non-LDS women. In SEER (without Utah) the age-adjusted breast cancer incidence rate was 141.9 (140.9 -- 142.9). Hypothetically speaking, if non-LDS women in Utah had the same breast cancer risk profile as LDS women, 214 (4.8%) fewer malignant breast cancer cases would have occurred during 1995--99. If the 9 SEER regions had the same breast cancer risk profile as women in Utah, 12,261 (16.2%) fewer malignant breast cancer cases would have occurred. If the 9 SEER regions had the same breast cancer risk profile as LDS women in Utah, 16,947 (22.4%) fewer malignant breast cancer cases would have occurred. A proportional hazards model was used to estimate the death hazard for female malignant breast cancer among patients, with religiously active LDS as the referent group. The death hazard ratio for religiously less active LDS was 1.09 (0.94 -- 1.27) and for non-LDS was 0.86 (0.75 -- 0.98). Adjusted hazard ratios for age and marital status were also computed for patients based on their already having survived 0 months, 6 months, 12 months, 18 months, and 24 months (Figure [3](#F3){ref-type="fig"}). Religiously less active LDS compared with religiously active LDS showed no significant difference in the death hazards due to breast cancer across each of the conditioned time periods. On the other hand, non-LDS compared with religiously active LDS had significantly lower death hazard rates for breast cancer for each of the conditioned time periods. Discussion ========== Utah presents a unique area for breast cancer studies because of the relatively low incidence of breast cancer among its residents, and the lower risk of breast cancer among members of the LDS church compared with non-LDS in the state. Lifestyle behaviors common among members of the LDS church, such as higher birthrates, higher prevalence and duration of breastfeeding, and lower alcohol consumption have been associated with lower incidence rates of breast cancer. The significantly lower risk of breast cancer among LDS shows the potential preventive effect certain behaviors may have against breast cancer. Yet this study also showed poorer survivability among a population with a traditionally lower incidence rate of breast cancer. It appears that the same factors that are associated with a lower incidence rate of breast cancer are also associated with poorer prognosis once breast cancer is diagnosed. This paradoxical finding is consistent with the results of a study by Korzeniowski and Dyba who reported that \"reproductive factors known to decrease the risk of breast cancer development have an adverse effect on prognosis\" \[[@B14]\]. Results in the current paper confirm those found by Lyon et al. \[[@B6]\]. Specifically, age-adjusted rates are consistently lower for LDS than non-LDS women. In addition, both LDS and non-LDS women displayed lower breast cancer incidence rates than in SEER. Lower rates among LDS women were observed across the age span. The lower rates among LDS women were likely primarily explained by their having significantly higher average number of pregnancies and children (parity), total years breastfeeding, and lower alcohol consumption \[[@B7],[@B13]\]. Not only did LDS women in Utah experienced lower breast cancer incidence rates than did women in SEER, non-LDS women in Utah also had lower rates. This may be attributed to a conforming effect wherein the non-LDS minority approximates the lifestyle behaviors of the LDS majority to increase their social acceptability \[[@B28]\]. Social and cultural factors influence behaviors such as breastfeeding \[[@B29]\]. Religious preference and the dominant religious preference of one\'s residence are two independent factors associated with differences in desired number of children \[[@B30]\]. A religious group may have certain reproductive standards, but when they come in contact with a religious group with different standards, their reproductive health attitudes may change \[[@B31]\]. When a majority group exerts moderate social pressure, especially if greater social acceptance is achievable, the minority group is more likely to conform to the majority\'s values \[[@B32]\]. Each of these factors fits the relationship between LDS and non-LDS in Utah and may explain non-LDS conformity to selected LDS practices. It should be noted that the linked data did not provide information about religious preference or religiosity among the nearly 30% non-LDS women. However, if the women in the study represent a similar representation as those identified in a recent cross-sectional survey of Utah adult women, approximately 26% of the non-LDS women were religiously active, 41% were less religiously active, and 34% had no religious preference \[[@B7]\]. Daniels et al. showed that the significantly lower average number of pregnancies and children (parity), total years breastfeeding, and higher alcohol consumption in non-LDS women were primarily in less religiously active women and those with no religious preference \[[@B7]\]. The population attributable-risk percent was presented to indicate the amount of breast cancer that could have been avoided under the hypothetical situation that everyone in Utah experienced the same breast cancer risk profile as LDS women. If everyone in SEER had the same breast cancer risk profile as LDS women in Utah, this percentage was 22.4%. In other words, approximately 22.4% of all breast cancer in SEER could have been avoided had women in SEER experienced a similar breast cancer risk profile as LDS women in Utah. Applying this percentage to the U.S. female population in 2004, an estimated 48,382 new cases could have been avoided had women in the U.S. displayed similar breast cancer risk behaviors as LDS women in Utah. There was not a significant difference in death hazard rates between religiously active and less active LDS. However, there was a significant difference in death hazard rates between religiously active LDS and non-LDS. Lower death hazard rates in non-LDS became more pronounced as we conditioned on months already survived from diagnosis. Two possible explanations may help explain this, proximity to last childbirth and diagnosis during the lactation period. New studies have indicated that a more recent birth prior to the diagnosis of breast cancer is associated with poorer survivability \[[@B30]-[@B33]\]. Older studies had failed to show an association, arguing that poorer survivability was coincidental and likely the result of significant delays in diagnosis \[[@B34]-[@B37]\]. Religiously active LDS women generally have more children than their non-LDS counterparts. While a greater number of childbirths is not associated with a worse prognosis in breast cancer \[[@B32]\], this higher parity does raise the likelihood that religiously active LDS women will have a more recent birth than non-LDS at the time of breast cancer diagnosis. A less conclusive, yet possible contributor to poorer survivability among active LDS, is diagnosis during the lactation period. One study noted that breast cancer diagnosed during the lactation period has been associated with poorer survival among women younger than 45 years \[[@B35]\]. This relationship persisted despite adjustment for nodal status, tumor size, and age. A more recent literature review, however, concluded that no epidemiologic evidence exists to indicate that breastfeeding increases the risk of breast cancer recurring or of a second breast cancer developing \[[@B41]\]. It is biologically plausible that breastfeeding may contribute to a reduction in the development and growth risk of breast cancer. Breastfeeding reduces the number of ovulations proportionally to its intensity, and maintaining a lower estrogen level than that observed during the menstrual cycle \[[@B42]\]; it mobilizes endogenous and exogenous carcinogens present in the ductal and lobular epithelial cell environment \[[@B43]\]; and it reduces pH, levels of estrogens, and local carcinogens of the lobules and ducts \[[@B44],[@B45]\]. Due to the limited and conflicting evidence supporting an association between diagnosis during lactation and poorer prognosis of breast cancer, and in light of biological mechanisms during lactation which may actually limit cancer growth, the possibility should not be ruled out that poorer prognosis associated with diagnosis during lactation may actually be attributable to the childbirth which most lactating women would have recently experienced. Further research is needed to clarify this association. Limitations =========== The temple endowment was used as a surrogate marker for religiosity among LDS women. Although receiving one\'s temple endowment requires conscientious effort toward obedience to the lifestyle prescribed by LDS doctrine, which includes certain health practices like not smoking or consuming alcohol and abstention from sexual relationships outside of marriage, the data did not allow us to identify whether a previously endowed individual continued to adhere to the doctrine of the church. Yet we assume that given the high level of religious commitment required to receive one\'s temple endowment, the tendency was for these individuals to continue adherence to the health practices of the LDS religion. A related religiosity measure to the temple endowment was not available for non-LDS women. We were concerned with the degree of accurate assignment of the cause of death, but the influence of inaccurate specification of cancer death appeared to be very small. Bias may result when considering multiple cancer primaries because religiously active LDS are less likely to smoke cigarettes. Hence, they are less likely to get a cancer related to smoking, where these cancers tend to be more lethal (e.g., lung and pancreatic cancers). There would have been a selection bias if we had considered death from any cancer. However, limiting disease-specific death to breast cancer made the groups more comparable with respect to smoking history. It was not possible to explore the association between histological grade and survivability because of the large number of patients in the \"unknown\" category. Nevertheless, the difference in distribution of histological grade among patients with known grade information was very similar and not statistically different (χ^2^(4) = 1.71, p = .789). There is no reason to believe that the \"unknown\" grade would have been distributed differently among the religious groups. Finally, specific data on breast cancer risk behaviors was not available on the patient level. However, information was available on whether or not a person had previously received their temple endowment. This measure was used as a surrogate for a host of lifestyle behaviors previously observed among religiously active LDS in Utah, such as higher parity and total time breast feeding and lower prevalence of alcohol consumption. Conclusion ========== LDS lifestyle is associated with lower incidence of breast cancer. The age distribution among LDS breast cancer patients in this study strengthens the argument that higher parity, lifetime breast feeding, and lower alcohol consumption have a preventive effect against breast cancer. Non-LDS in Utah approximate LDS behavior to increase social acceptability, serendipitously contributing to a relatively low breast cancer incidence. LDS showed poorer survival from breast cancer than non-LDS in Utah. Parity and breastfeeding, associated with lower risk of breast cancer, may contribute to poorer prognosis once breast cancer is diagnosed. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Ray M. Merrill conceptualized the study, analyzed the data, and drafted the manuscript. Jeffrey A. Folsom assisted in interpreting the data and writing the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/49/prepub> Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Malignant breast cancer incidence in white women in Utah and SEER (without Utah) ::: ![](1471-2407-5-49-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Age-specific malignant breast cancer incidence among white women in Utah according to LDS status, 1995--99 ::: ![](1471-2407-5-49-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Hazard ratios (95% confidence intervals) of death from breast cancer for religiously less active LDS and non-LDS compared with religiously active LDS conditioned on having already survived xmonths ::: ![](1471-2407-5-49-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Bivariate analyses of selected factors among breast cancer patients by LDS status and level of religiosity in Utah, 1985--99 ::: Religiously Active LDS Religiously Less Active LDS Non-LDS χ^2^P value ------------------------------ ------------------------ ----------------------------- --------- ------------- ------ ------ --------- No. \% No. \% No. \% Age at Diagnosis 00--34 119 2.3 55 4.1 91 3.1 35--44 463 9.1 165 12.3 465 15.6 45--54 962 18.8 250 18.6 639 21.5 55--64 1167 22.9 312 23.2 674 22.6 65--74 1299 25.5 351 26.1 668 22.4 149.7 75+ 1093 21.4 211 15.7 441 14.8 \<.0001 Marital Status at Diagnosis Never Married 123 2.3 110 8.0 217 7.2 Married (cohabitating) 3044 58.2 626 45.7 1944 64.4 Previously Married 2067 39.5 633 46.2 810 26.9 414.9 Unknown 0 0.0 0 0.0 45 1.5 \<.0001 Breast Cancer Primaries Single 4742 90.6 1200 87.7 2735 90.7 11.8 First of Multiple 492 9.4 169 12.3 281 9.3 .0028 Summary Stage (Pathological) Localized 3208 61.3 839 61.3 1859 61.6 Regional 1756 33.5 443 32.4 982 32.6 Distant 166 3.2 54 3.9 106 3.5 4.3 Unknown 104 2.0 33 2.4 69 2.3 .6340 Histological Grade Low (well-differentiated) 576 11.0 152 11.1 336 11.1 Medium (moderately diff.) 1796 34.3 463 33.8 1076 35.7 High (poorly diff./undiff.) 1272 24.3 357 26.1 796 26.4 13.7 Unknown 1590 30.4 397 29.0 808 26.8 .0337 Radiation Therapy No 3504 67.0 896 65.5 1961 65.0 3.5 Yes 1730 33.0 473 34.5 1055 35.0 .1736 Surgery Conservative Management 1835 35.1 508 37.1 1144 37.9 Mastectomy 3368 64.3 856 62.5 1858 61.6 8.4 Other Surgery 31 0.6 5 0.4 14 0.5 .0775 Year of Diagnosis 1985--88 1139 21.8 306 22.4 693 23.0 1989--92 1282 24.5 374 27.3 732 24.3 1993--96 1525 29.1 371 27.1 855 28.3 7.8 1997--99 1288 24.6 318 23.2 736 24.4 .2498 :::	PubMed Central	2024-06-05T03:55:58.347376	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173093/", "journal": "BMC Cancer. 2005 May 18; 5:49", "authors": [ { "first": "Ray M", "last": "Merrill" }, { "first": "Jeffrey A", "last": "Folsom" } ] }
PMC1173094	Background ========== Since 1994, colorectal adenocarcinoma has been the third most frequent cancer in the Grand-Duchy of Luxembourg (Western Europe), preceded by breast and prostatic cancer \[[@B1]-[@B3]\]. In the United States of America (USA), Japan and France, there have been changes in the site distribution of large bowel cancers over the last decades \[[@B4]-[@B9]\]. Because of the lack of a systematic, organized screening for colo-rectal cancer in Luxembourg, the data presented here may help to define new preventive screening approaches, to increase funding and hasten the introduction of screening for this tumour. The aim of this study was to investigate changes in incidence, stage, anatomical site distribution, and the observed survival rates of colon cancer over two 5-year periods, 1988--1992 and 1994--1998. Analogous to the report on rectal cancer published earlier \[[@B10]\], descriptive epidemiological data on colon cancer were collected to underline the need for the initiation of a nation-wide strategy for the early detection of colorectal cancer. Methods ======= Between January 1^st^, 1988 and December 31, 1998, all consecutive new cases of high-grade adenoma and all new cases of invasive colonic adenocarcinoma (COAC) were registered by the national Morphologic Tumour Registry (MTR) in Luxembourg. These cases had been exclusively diagnosed and \"double-read\" in the central department of pathology by nine senior pathologists. Over the decade the population increased from 374,900 to 429,000, -- an average increase of 1.3% per year. In the elderly age group \>69 years of age, there was a continuous increase by 1.8% (0.5 -- 4.4%) \[[@B11]\]. Patients of all nationalities, race or ethnical origin living in Luxembourg were considered. Disease histology was limited to high-grade adenomas (i.e. adenomas with severe dysplasia) and invasive adenocarcinomas of the colon. Tumours recorded as other carcinomas, metastatic or recurrent disease, metachronous adenomas or invasive adenocarcinomas, malignant carcinoid and mesenchymal malignant tumours were excluded from the current review. For patients with synchronous colon neoplasias, only cases with the worst stage category were taken in account. Data on patients with high-grade adenomas (i.e. stage 0 patients), who by definition could not have metastases in regional lymph nodes or distant metastases, were obtained from 254 biopsy-forceps and polypectomy specimens removed by 26 gastroenterologists and 11 senior-surgeons. Adenomatous proliferations in marginal zones of invasive adenocarcinomas and adenomas with severe dysplasia appearing as recurrent invasive disease within 2 years at the same site were not taken into account. Data on 1158 surgical resection specimens allowed an interpretation in relation to the local and regional spread, the lymph node and the residual tumour status. Hence the stratification of the invasive colonic adenocarcinomas by stage (stage I-IV) led to the exclusion of patients with biopsy diagnosis alone and patients who had only undergone polypectomy without segmental resection (i.e. without regional lymphadenectomy). Patients with known preoperative debulking radiotherapy were not considered. The surgical resection specimens were analysed with respect to tumour-stages (stage I to IV), according to the guidelines of the Union Internationale Contre le Cancer (UICC) and the American Joint Committee on Cancer (AJCC) \[[@B12]-[@B14]\]. We defined stage I and stage II as early cancers and stage III and stage IV as advanced colon cancers. To evaluate the influence of the number of colonoscopic procedures on the incidence rates and the stage distribution of colonic cancers in Luxembourg, the data on colonoscopies (1992--1998), gratefully made available by the National Health Found (NHF), were analysed \[[@B15]\]. The R-classification according to the TNM-system was performed by conventional methods \[[@B16],[@B17]\]. In our series the histological examination of the proximal, distal and lateral (deep) line of resection was mandatory. R0 indicates no residual tumour; R1 corresponds to microscopically and R2 to macroscopically residual tumour, while Rx did not allow a statement on residual tumour. The histopathological diagnoses were made according to the WHO-classification \[[@B18]\]. On this basis, we defined \'high grade\' -- adenomas, the equivalent of stage 0 or former adenocarcinoma in situ, as adenomas of tubular, villous or tubulo-villous histological type with high-grade intraepithelial neoplasia (i.e. severe dysplasia). This category includes cancer cells confined within the glandular basement membrane (intraepithelial) or lamina propria (intramucosal) with no extension through the muscularis mucosae into the submucosa. We considered all colon segments as large bowel, with the exception of the rectum and the recto-sigmoid junction. Anatomically the rectum was defined as a 16 cm long segment between the ano-cutaneous line and the sigmoid colon. Cancers described at the recto-sigmoidal junction without statement on the centimetres from the anal margin were considered as rectal adenocarcinomas and were excluded. Left-sided colon cancers were defined as adenocarcinomas of the descending colon and the sigmoid colon. Right-sided colon cancers included adenocarcinomas situated above the splenic flexure (i.e. splenic flexure, transverse colon, hepatic flexure, ascending colon and caecum with its prolongation, the appendix vermicularis). Beside the evaluation of the frequency of invasive COCA by the 17 \'classic\' 5-year age categories (0--4 years to 80 years and above) the changes of anatomic sub-site distribution in relation to 3 age groups young (\<40 years of age), middle aged (40--69 years) and elderly (70 years and above) patients were analyzed. The observed survival rates of patients with colon cancer were measured from the time of diagnosis and calculated by the actuarial method (life-table) with a 95% confidence interval (c.i.). To evaluate the observed survival rates in relation to the anatomical extent (stage I-IV) and the surgical resection margins (R-status), 164 patients with colon adenocarcinoma diagnosed only by biopsy-forceps without consecutive surgical resection data in our files, who were without final stage and who are part of the 1379 patients with an invasive COCA, were analysed separately. The mortality rates of colon cancer were issued from the anonymous files of the death certificates published by the National Health Direction (Ministry of Health) \[[@B19]\]. The age-standardized incidence rates of the colon cancers diagnosed in the Grand-Duchy of Luxembourg during the period 1983--1997 were compared with the data of other geographical European regions with similar population density and socio-economic characteristics, published by the WHO in \"Cancer in Five Continents, volumes VI, VII, VIII\" \[[@B1]-[@B3],[@B20]-[@B22]\]. The statistical evaluations include the chi-square test (χ^2^) with a level of significance p \< 0.05 and the life-table survival analysis. The age-standardized incidence rates were calculated by the direct method, the standard error of the age-standardized rates by the Poisson approximation \[[@B23]\]. Results ======= The histopathological diagnoses of 1379 consecutive invasive colon adenocarcinomas (COAC) were provided from 164 samples obtained by biopsy only (11.9%), 57 polypectomy specimens (4.1%) and 1158 surgical resection specimens (84.0%). The latter allowed a stratification by stage (UICC/AJCC, 1997) and had reliable follow-up data. The 1379 invasive adenocarcinomas of the colon involved 658 males (47.7%) and 721 females (52.3%); M/F-ratio of 1: 1.1. The mean age was 69.7 years (range: 26--97). During the study period, 254 patients were diagnosed with high-grade adenomas. These patients constitute a separate group from the 1379 invasive COAC. The average, annual, age-specific rates of colon AC in Luxembourg over the period 1988--1998 were 31.1 per 100,000 for all ages and both genders, for males 30.3 per 100,000 and for females 31.9 per 100,000 (Figure [1](#F1){ref-type="fig"}). The comparison of the crude incidence rates of patients with colon AC diagnosed in the 5-year periods 1988--1992 and 1994--1998 revealed an increase by 13.5% for both genders from 28.9 to 32.8, for females by 15.5% from 29.6 to 34.2 and for males by 11.3% from 28.2 to 31.4 per 100,000. The average, annual, age-standardized incidence rates (ASR-world population) of colon cancer (period 1988--1998) were 17.5 per 100,000, for males 20.5 per 100,000 and for females 15.4 per 100,000. Table [1](#T1){ref-type="table"} documents the world age-standardized incidence rates (ASR/W) of the invasive COAC in the Luxembourgish population (females and males) over the period 1983--1997 in comparison to other European countries. The time trend (1988--1998) of the frequency of invasive COAC showed that the total number of invasive COAC rose significantly from 109 cases in the year 1988 to 162 cases in 1998 (in females from 61 to 93, in males from 48 to 69). Comparing the two 5-year periods 1988--1992 and 1994--1998, there was a significant increase in the absolute number of all COAC from 557 to 687 cases (p \< 0.001). The absolute number of COAC in males increased significantly from 266 to 323 cases (p \< 0.02), and in females significantly from 291 to 364 cases (p \< 0.01). Figure [2](#F2){ref-type="fig"} shows the age-distribution of all patients with histologically confirmed invasive colonic adenocarcinoma (n = 1379 cases). Of these patients 1.7% were \<40 years of age; 3.8% between 40 and 49 years; 41.0% between 50 and 69 years; 31.5% between 70 and 79 years and 22.0% of the patients were 80 years and above. Concerning the COAC stage distribution, two peculiarities appeared at the extremes of patient age scales. Below the age of 40, none of the patients (n = 23) had early stage cancer (i.e. stage I or stage II). On the other hand, out of the 303 patients aged 80 years and above, most had stage II (39.0 %) and stage III (38.2 %) disease, whereas stage I (10.6 %) and stage IV (8.3 %) were much less frequent. Time trends in relation to the tumour-stages \[TNM-System (UICC/AJCC\] are represented in Figure [3](#F3){ref-type="fig"}. During the two 5-year periods 1988--1992 and 1994--1998, there was a highly significant increase (p \< 0.001) in the crude incidence rates of the stage 0 cases (i.e. high-grade colonic adenomas) by 190%, an increase in stage I cases by 14.3%, in stage II cases by 12.9%, in stage III cases by 38.5%, and a decrease in stage IV cases by 11.8%. The relation of the 254 new colonic high-grade adenomas to the 1379 new invasive adenocarcinomas stratified by years is shown in Figure [4](#F4){ref-type="fig"}. The comparison of the average of crude incidences of the high-grade adenomas to the invasive adenocarcinomas for the two 5-year periods indicates a highly significant improvement of the ratio by 148% in the second period. There were no significant changes in the number or crude incidences of colorectal endoscopies or colonoscopies with biopsy examinations. The distribution of the 1244 new invasive colon adenocarcinomas by anatomical site and two 5-year periods (1988--1992: n = 557 cases and 1994--1998: n = 687 cases) are given in Figure [5](#F5){ref-type="fig"}. Twelve cases (1.0%), which were not otherwise specified, had to be excluded. In relation to three age groups (\< 40 years of age, 40 -- 69 years and \>69 years) the anatomical sub-site distribution shows a shift to right-sided colon adenocarcinomas by 9.2% in the 40 to 69 years age group (n = 553) and by 76.2% in the group of patients of \>70 years of age (n = 658). The number of the left-sided colon cancers decreased by 14.5% in the 40--69 years age group and increased by 22.9% in those \>70 years. Patients with no precise site and those of the year 1993 were excluded for statistical (two 5-year periods) comparison reasons. In Table [2](#T2){ref-type="table"} the anatomical sub-site distribution in relation to the three age groups and their age-specific rates per 100,000 is represented. The overall observed survival rates of the 1379 patients with an invasive COAC, diagnosed by biopsy or polypectomy or surgical resection specimen were calculated and stratified by years: 1-year: 72 ± 2% (n = 997/1379); 2-year: 60 ± 3% (833/1379); 3-year: 54 ± 3% (n = 740/1379); 4-year: 48 ± 3% (n = 662/1379) and 5-year 44 ± 3% (n = 604/1379). In Table [3](#T3){ref-type="table"}, the observed 5-year survival rates of 931 invasive COAC stratified by stage and operation for cure (R0), and those of 58 patients with COAC with microscopically (R1) or macroscopically (R2) residual tumour after surgical resection are represented. The analysis of the observed survival rates of the 164 patients with colon AC, diagnosed by biopsy-forceps only (i.e. without known final stage), revealed that 70 patients (42.7%) survived 12 months, whereas 20 (12.2%) died in the first month after diagnosis, 46 (28.1%) within the next 6 months and 28 (17.1%) in the period from 6 to 12 months. The age-distribution of these patients showed that 10 (6.1%) patients were \<50 years, 105 (64.0%) were aged between 50 and 79 years and 49 (29.9%) were \>80 years of age. The mortality rates for colon cancer in the Grand-Duchy of Luxembourg over the last 15 years (1984--1998), grouped by 5-year periods, did not change significantly (Table [4](#T4){ref-type="table"}) \[[@B19]\]. There was a slight, insignificant (p \< 0.05) decrease in the crude death rates from 26.3 per 100,000 in the period 1984--1988 to 23.8 per 100,000 in the period 1989--1993 and 23.9 per 100,00 in 1994--1998 \[[@B19]\]. Discussion ========== In contrast to the USA and Canada, the age-standardized incidence rates of colon cancer in Luxembourg steadly increased from 1983--1997 for both genders (Table [1](#T1){ref-type="table"}). Our findings are in concordance with those of many other European countries, where the increase occurs mainly in the male population \[[@B20]-[@B22],[@B24]\]. Some authors believe that the decreasing incidence rates in the USA are due to preventive cancer screening \[[@B25]\]. We found 3.2% cases with synchronous colon adenocarcinomas, which is comparable to the 3.0% and 4.8% of synchronous cancers identified by authors in other countries \[[@B26],[@B27]\]. The effect of age was examined in relation to the classic age groups and in relation to three cohorts of patients, (a) under 40 years of age, (b) between 40 and 69 years, and (c) 70 years and older. The number of colon cancers increases around the age of 45 (Figure [2](#F2){ref-type="fig"}). This is consistent with the findings of Imperiale et al., who found a low incidence of colon adenomas in the age groups 40 to 49 years \[[@B28]\]. Patients under 40 years of age had a higher incidence of advanced stage colon cancers. Similar observations were reported by Mostafa et al. \[[@B29]\]. Considering that high-grade adenomas are obligate precancerous lesions with a slow progression, that polypectomy can mean prevention of cancer, and that the high-grade adenoma/invasive adenocarcinoma ratio (Figure [4](#F4){ref-type="fig"}) improved in the late 1990\'s, we observed evidence of earlier diagnoses \[[@B30]\]. We consider the continous registration of high-grade adenomas and determination of the adenoma-adenocarcinoma ratio as a useful additional quality assurance marker in evaluating clinical and histopathological diagnostic procedures. To explain the unexpected increase of precancerous colonic lesions, we analysed our files in order to identify parameters that might have changed, such as staff composition or other factors. Over the period 1992--1995 the number of physicians (n = 38) practicing endoscopy did not change and the number of high-grade adenomas (n = 109 cases) removed by each endoscopist varied between one and six cases. Considering that, the number and the crude annual incidence rate of the colonoscopies did not rise significantly over the time, that the staff of pathologists remained the same, the histopathological criteria and coding did not change, and that a systematic organized screening programme until the year 2000 was not launched and a spontanous, unorganized screening is not known, we believe that a more selective indication for colonoscopy examinations may be one of the causes of this increase. The abrupt change after 1993 in the number of stage 0-cases may be the consequence of the changes in the statute of the National Health Fund. In 1992 the notions of patients and medical care providers \'profiling\' at the national level were introduced. At the same time, the media campaign for preventive medical care strategies against cancer (breast, malignant melanoma) launched by the governmental institutions in collaboration with the NHF and the Luxembourg Foundation against Cancer may have had a beneficial (secondary) effect. Not only the incidence of stage 0-cases rose, but we found also a positive influence on the crude incidence rates of stage I and stage II patients increasing over the two 5-year period (1988--1992) and (1994--1998) by 14.3 and 12.9% respectively. Unfortunately there has been a highly significant (p \< 0.001) increase in stage III patients by 38.5% too. Colon AC in young patients (40 years of age or less) seems to be more aggressive, since those patients were diagnosed exclusively in stage III and stage IV. Since these patients often present a positive family history of colorectal cancer, there is an obvious need for health-care providers to heighten awareness, and to identify high risk persons in this population group \[[@B25],[@B31],[@B32]\]. In North America, Japan, France and other high-risk areas, there has been a proximal shift in the sub-site distribution of colorectal cancer \[[@B4],[@B29],[@B32]-[@B34]\]. Similar changes have occurred in Luxembourg. This proximal shift was found in both sexes, especially in patients aged 70 and above. In males the increase was 14.0%, in females 73.6%. Mitry et al. \[[@B33]\] found that right colon cancer rates had increased by 21.6% in males and 10.4% in females. This underlines the need to discuss the opportunity for total colonoscopy in patients of higher risk or elderly age groups \[[@B30]\]. In our cohort, the overall 5-year survival rates of the patients with surgery for colon cancer without distinction of the residual tumour-status were 46 ± 3% (95% c.i.). Because the data on age-specific risk of death from other causes than colon cancer during the 1990\'s are not yet available, the relative survival rates could not be calculated. It must be admitted that relative survival rates will give better data for colon cancer in Luxembourg than indicated by the observed survival rates reported here. The relative 5-year survival rates reported in the literature vary between 30.6 and 55.3% \[[@B6],[@B35]\]. Because COAC incidence increases at the age 50 approximately in our series, because a significant shift to right-sided colon cancer has been detected in patients of 70 years of age and above and because survival rates would probably improve if the disease is treated in its early stages, a strategy to perform twice-lifetime colonoscopy (at ages 50 and 60 years or at the transition to the pension age with 65 years) for asymptomatic patients without family history or other risk factors, as discussed by others, needs further investigations \[[@B4],[@B29],[@B34],[@B36],[@B37]\] Conclusion ========== In contrast to the time trends in the USA and Canada, we found increasing age-standardized incidence rates in colonic adenocarcinomas and an increase of the crude incidence rates especially in stage III patients. Despite the 4-fold increase in high-grade adenoma diagnoses (i.e. stage 0 cases), efforts should be reinforced in preventive strategies to detect more precancerous lesions and/or early tumour-stages, to reduce the true incidence of advanced tumour-stages at diagnosis, and by this the mortality rates. The findings of this study should provide baseline information and scientific evidence for the elaboration and implementation of cancer preventive and intervention strategies to those populations at average and high-risk for colorectal cancers. As long as new diagnostic procedures, such as stool-based DNA-test for colorectal tumours, are not yet available for mass-screening purposes, and as long as the effectiveness of a complete colonoscopy screening for colorectal cancer in asymptomatic patients beginning at the age of 50 has not been proven by population-based controlled trials, the data underline the need of at least a systematic faecal occult blood test (FOBT) screening for colorectal cancer beginning at the age of 50 \[[@B30],[@B38],[@B39]\]. Our data support mainly that screening could be considered in preventing deaths from colo(rectal) cancer. The findings can be used by the national health care authorities to define a screening strategy and to help in deciding the target age-groups in Luxembourg. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors (RS, PP, YW, NK, CC) collaborated intensely on all aspects of the manuscript, from research design to data preparation and presentation. RS wrote, and all authors approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/52/prepub> Acknowledgements ================ We thank Mrs Mireille Braun, Mrs Martine Becker and M. Guy Weber for their helpful assistance during the preparation of this manuscript. The study received financial support from the national Morphologic Tumour Registry (RMT a.s.b.l.). The opinions stated in this document are those of the authors and do not necessarily represent the official position of the RMT. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Time trend of the crude incidence rates of the invasive colon adenocarcinomas (n = 1379) by gender; 1988--1998. ::: ![](1471-2407-5-52-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Age-distribution of the invasive colonic adenocarcinomas (n = 1379 cases); 1988--1998. ::: ![](1471-2407-5-52-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Crude incidence rates of the high-grade adenomas (n = 254) and the invasive adenocarcinomas of the colon (n = 1379) stratified by stage (TNM/UICC-AJCC, 1997) \[14\]. ::: ![](1471-2407-5-52-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Crude incidence rates and colonic high-grade adenoma / invasive adenocarcinoma-ratio in comparison to colonoscopies: 1988--1998; both genders; n = 254\/1379 cases (\diagnoses by biopsy only included). ::: ![](1471-2407-5-52-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Number of invasive colonic adenocarcinomas (n = 1244; cases of the year 1993 excluded) by anatomic site; 1988--1992 and 1994--1998. (N.O.S.\* = not otherwise specified) ::: ![](1471-2407-5-52-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Colon cancer in the European Community: world age-standardizedincidence rates \[ASR(W)\\], time trends: 1983--1987; 1988--1992; 1993--1997. ::: Males* 83--87 88--92 93--97 Females 83--87 88--92 93--97 --------------------- ------------ ------------ ------------ --- --------------------- ------------ ------------ ------------ --- F/Bas Rhin 27.2 30.2 31.3 ↑ D/Saarland 19.4 20.4 21.5 ↑ D/Saarland 23.7 25.5 27.3 ↑ NL/Maastricht 17.1 17.9 19.4 ↑ I/Parma 23.0 24.4 27.0 ↑ DK/Denmark 19.3 19.9 18.6 ↓ IRL/Southern 21.8 24.2 25.3 ↑ IRL/Southern 19.7 20.3 18.3 ↓ E/Navarra 13.8 19.2 22.1 ↑ I/Parma 16.1 17.8 17.9 ↑ NL/Maastricht 19.7 21.9 22.1 ↑ UK/South Western 15.9 16.2 17.3 ↑ UK/South Western 18.2 19.2 22.1 ↑ F/Bas Rhin 17.4 18.8 16.9 ↓ L/Luxembourg 15.5 20.2 21.9 ↑ L/Luxembourg 11.3 14.9 15.0 ↑ DK/Denmark 20.3 20.6 20.7 ↑ S/Sweden 15.8 15.9 15.0 ↓ CH/St.Gall 16.4 16.5 19.7 ↑ E/Navarra 9.6 13.0 12.4 ↑ S/Sweden 17.5 17.7 17.7 ↔ CH/St.Gall 11.7 12.0 12.1 ↑ SF/Finland 11.9 12.8 14.5 ↑ SF/Finland 10.6 11.6 12.1 ↑ CND/Canada 27.8 26.9 25.9 ↓ CND/Canada 23.6 21.3 20.0 ↓ USA/Seer\\, white 31.1 28.1 25.5 ↓ USA/Seer\\, white 23.6 20.8 19.4 ↓ \* Cancer incidence in five continents. Vol VI; Vol VII; Vol VIII \[20, 21, 22\] \\ SEER: Surveillance, Epidemiology and End Results Program ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Invasive colon adenocarcinomas (n = 1232\) by anatomic site (left and right colon), by age groups (\<40 years; 40--69 years; \>69 years) and age-specific incidence rates; 1988--1992 and 1994--1998. ::: Age groups* Number of incident cases (ri) Person- years of obervation (ni) Age-specific rates per 100,000 (ri/ni) *p-values \\* ------------------------ ----------------------------------- -------------------------------------- -------------------------------------------- ------------------- \<40 years colon right 1988--1992 5 1 072 099 0.5 p = n.s. colon right 1994--1998 9 1 151 878 0.8 colon left 1988--1992 2 1072 099 0.2 p = n.s. colon left 1994--1998 5 1151 878 0.4 40--69 years colon right 1988--1992 130 676 772 19.2 p = n.s. colon right 1994--1998 142 742 577 19.1 colon left 1988--1992 150 676 772 22.2 p = n.s. colon left 1994--1998 131 742 577 17.6 \>69 years** colon right 1988--1992 122 175 629 69.5 p \< 0.001 colon right 1994--1998 215 196 145 109.6 colon left 1988--1992 144 175 629 82.0 p = n.s. colon left 1994--1998 177 196 145 90.2 \* excluded colon cancers N.O.S.(n = 12) and those of 1993 (n = 135) \\ level of significance p \< 0.05; n.s. = not significant ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Prognoses of invasive colonic adenocarcinomas resected for cure (R0) or with residual tumour (R1+R2) and stratified by stage-TNM/UICC \[12, 13, 14, 23\]. ::: --------------------------------------------------------------------------------------------------------------------------------------------------------- Stage TNM Observed 5-year survival rates (actuarial method, 95% confidence interval) ------------------------ -------------- -------------------------------------------------------------------------------- --------------------- ---------- UICC, 5^th^edition R0 R1+R2 *p-values \\\ I \*** T1 N0 M0\ 71 +/- 8%\ 100% \\\ \<0.5188 T2 N0 M0 (n = 129) (n = 1) II \* T3 N0 M0\ 63 +/- 5%\ 45 +/- 30%\ \<0.2392 T4 N0 M0 (n = 388) (n = 11) III \* anyT N1 M0\ 38 +/- 5%\ 4% \\\ \<0.0007 anyT N2 M0 (n = 331) (n = 25) IV \* anyT anyN M1 17 +/- 8%\ 5% \\\ \<0.1584 (n = 83) (n = 21) I-IV \* 51 +/- 3%\ 14 +/- 9%\ \<0.0001 (n = 931) (n = 58) --------------------------------------------------------------------------------------------------------------------------------------------------------- \* only surgical specimens with available survival data \\ number of cases too low for calculating the 95% confidence interval \\\* level of significance p \< 0.05 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Analysis of the mortality rates from all colon adenocarcinomas of the last 15 years (1984--1998) declared in Luxembourg by death certificate only \[19\]. ::: Males Females both genders p-values\* ----------------- ------------ ------------- ------------------ ---------------- 1984--1988: 26.5/10^5^ 26.2/10^5^ 26.3/10^5^ p = n.s. 1989--1993: 22.5/10^5^ 25.0/10^5^ 23.8/10^5^ p = n.s. 1994--1998: 23.6/10^5^ 24.3/10^5^ 23.9/10^5^ \* level of significance p \< 0.05 ; n.s. = not significant for both genders :::	PubMed Central	2024-06-05T03:55:58.351136	2005-5-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173094/", "journal": "BMC Cancer. 2005 May 24; 5:52", "authors": [ { "first": "René", "last": "Scheiden" }, { "first": "Paul", "last": "Pescatore" }, { "first": "Yolande", "last": "Wagener" }, { "first": "Nelly", "last": "Kieffer" }, { "first": "Catherine", "last": "Capesius" } ] }
PMC1173095	Background ========== The WWOXgene, originally cloned by our laboratory, spans a genomic region greater than 1 Mb in size and is the second most common chromosomal fragile site, FRA16D (16q23) \[[@B1],[@B2]\]. Abnormalities affecting WWOXat the genomic and expression level have been reported in numerous neoplasias and cancer derived cell lines including, breast, ovarian, esophageal, lung, stomach, liver, pancreas and hematological malignancies \[[@B3]-[@B12]\]. We observed that ectopic WWOX expression inhibited anchorage independent growth and in vivotumorigenicity of highly aggressive breast carcinoma lines, suggesting a putative tumor suppressor role for this novel protein \[[@B13],[@B2]\]. WWOXencodes a 46 KD, 414-amino acid protein that contains two WW domains at the NH~2~terminus and a short chain oxidoreductase (SDR) central domain \[[@B1]\]. The first WW domain- is involved in protein-protein interactions by binding the specific proline rich motif PPXY and several potential candidate partner proteins have been postulated \[[@B14],[@B15]\]. Within the SDR domain, the presence of WWOX amino acid residues, serine 281, and 293-[Y]{.underline}NRS[K]{.underline}-297 make up a catalytic signature motif conserved in short-chain steroid dehydrogenases \[[@B16]\]. We originally reported high WWOXmRNA expression levels in ovary, prostate, testis and breast \[[@B1]\]. In this study we analyzed WWOX protein expression pattern in normal ovary and ovarian carcinomas. We correlated WWOX protein expression with ovarian carcinoma histotypes and clinico-pathological parameters. In addition, since we recently observed a strong association between loss of WWOX expression and estrogen and progesterone receptor (ER and PR) status in breast cancer \[[@B12]\], we also investigated any potential association between expression of sex steroid hormone receptors and WWOX in ovarian cancer. Methods ======= Western blot analysis --------------------- Total protein extracts were prepared from snap frozen tissue of 38 human ovarian carcinomas and 5 normal human ovarian tissues. As negative control for WWOX protein expression we used protein extracts from the ovarian cell line PEO1 that does not express WWOX due to a homozygous deletion affecting exons 4--8 of this gene \[[@B4]\], a kind gift of Dr. Hani Gabra at Imperial College London, UK. As positive control we used the same cell line stably transfected with a WWOX expressing vector (PE01-WWOX) \[[@B10],[@B12]\]. Total cell protein lysates were made using RIPA buffer (50 mM Tris pH7.5, 150 mM NaCl, 0.5% sodium deoxycholate, 1% Triton X-100, 0.1% SDS) containing protease inhibitor cocktail (Roche, Mannheim, Germany). For western blotting 50 ug of total protein was separated by 12.5% SDS-PAGE and transferred to PVDF membranes (Millipore, Billerica, MA). Immunodetection was performed using Protein Detector™ (KPL, Gaithersburg, MD) western blotting reagents as described by the manufacturer. WWOX protein was detected using affinity-purified anti-WWOX rabbit polyclonal primary antibodies developed in our laboratory (final concentration 280 ng/ml) \[[@B12]\] and HRP conjugated anti-rabbit secondary antibody (KPL, city, state, 1:2000 dilution) followed by chemiluminescence autoradiography. Actin was used as the protein loading control and it was detected using monoclonal anti-actin antibody (ICN biomedicals, Burlingame, CA, 1:1000 dilution) and HRP conjugated anti-mouse secondary antibody (KPL, 1:5000). Quantitation of X-ray films exposed to western blot chemiluminescence-emitting membranes was performed using a Kodak digital science Image Station 440CF. Protein loading and signal intensity was controlled by normalizing each sample to the aforementioned positive control, PE01-WWOX as previously described \[[@B10],[@B12]\]. Ovarian tissue microarrays -------------------------- The MD Anderson Cancer Center ovarian TMA was prepared by D.R. at the J.L laboratory \[[@B17]\] with samples from 441 patients with primary epithelial ovarian cancer that had undergone surgery at M. D. Anderson Cancer Center between 1990 and 2001. Follow-up information was updated through June 2003. Histopathologic diagnoses assigned at the time of treatment were based on World Health Organization criteria, each sample was reclassified by grade based on the Gynecologic Oncology Group (GOG) criteria \[[@B18]\] and each case was also classified by disease stage according to the International Federation of Gynecology and Obstetrics (FIGO) system \[[@B19]\]. The appropriate institutional committee approved use of human tissue blocks and clinical records reviews. Each core was scored individually and the results are reported as the mean of at least two replicate core sample measurements \[[@B20]\]. We also used a second ovarian TMA set generated at the Fox Chase Cancer Center (FCCC) tumor bank by R.P. at the A.J.P.K laboratory. This TMA set included a total of 86 ovarian tumors and 21 normal ovarian samples. Cores of normal ovaries were used as positive controls. Only primary mucinous carcinomas of the ovary where included on this cohort of patients. Selection criteria was based on histopathological diagnosis made by a trained pathologist on whole sections, histochemical stainings, immuhohistochemical staining, and clinical assessment of the patient. By pooling the TMA sets from both institutions we were able to successfully analyze in duplicate WWOX protein staining from a total of 444 invasive surface epithelial derived primary ovarian carcinomas that included Serous (n = 375), Endometroid (n = 40), Mucinous (n = 10) and Clear Cell Carcinoma (CCC) (n = 19) histological subtypes. A subset of these cases (n = 323) was successfully processed and analyzed for allowing comparative analysis of WWOX protein expression with ER and PR status. Immunostaining method --------------------- Anti-WWOX immunostaining was performed as previously reported \[[@B12]\]. Mouse monoclonal antibodies for ER, NCL-ER-6F11 (Novocastra, UK) and PR, NCL-PGR-312 (Novocastra, UK) were used according to manufacturer recommendations. The anti-PR antibody used in these studies is able to recognize only the A isoform of this receptor \[[@B21]\]. Evaluation of immunohistochemical staining ------------------------------------------ Staining intensity was measured using a Chromavision Automated Cellular Imaging System (Chromavision^®^ACIS^®^, San Juan Capistrano, CA) as previously described \[[@B12]\]. The methods for scoring cytoplasmic or nuclear staining are based on different program functions of the ACIS^®^and are the recommended by the manufacture. For anti-WWOX staining measurements we employed the generic DAB software application provided by ACIS. The area of each individual TMA core was considered for the measurements in its totality. The software determines brown intensity (i.e. positive stained cells) regardless of the area covered by the positive cells. The cutoff to differentiate positive and negative staining was determined to be a mean intensity of 63 (arbitrary staining intensity units, s.i.u.) over a total of 255 (color saturation scale). Therefore, cores with values = 63 s.i.u were \'negative\' for WWOX immunostaining, i.e. no brown staining observable. Values between 63 and = 65 s.i.u. were considered to be in the \'weak/low\' staining category, i.e. barely visible brown staining, 65 to 81 s.i.u were considered \'moderate\' and cores with staining intensity values \> 81 s.i.u were considered to fall in the \'strong\' staining category. For sake of simplicity cases were classified in two groups: a) negative + low = Negative-Low WWOX group, i.e. any core with a value = 65 in staining intensity and b) moderate + strong = High WWOX group, i.e. any core with values above 65 in staining intensity. The Negative/Low WWOX group was composed of 79% (105/133) completely negative cases and 21% (28/133) weak cases. The mean WWOX intensity staining for normal ovarian surface epithelium (OSE) fell in the moderate category (mean value 70 ± 4). Anti-ER and PR antibodies ------------------------- For performing anti-ER and anti-PR determinations we utilized the nuclear antigen software application provided by ACIS. Only areas of tumor in each TMA core were considered for measurements. The free form drawing tool provided by the software was used to select tumoral areas and to exclude other structures. A final score was determined by the percentage of brown stained nuclei (i.e. positive cells) over the total of tumor cell nuclei measured (i.e. positive plus negative cells). Only cores with at least 10% tumor were included in the analysis. For the receptor analyses, positivity of nuclear staining was defined as follows: weak \> 5% stained nuclei, moderate 5--40% stained nuclei and strong \>50% stained nuclei. Statistical methods ------------------- Univariate analysis of clinical-pathological parameters based on WWOX staining was determined by χ^2^test with Yates\' correction. The basic significance level was fixed at p\< 0.05 and all data was analyzed using SPSS statistical software (Version 11.0; SPSS Inc., Chicago, IL). Results ======= Western blot analysis --------------------- Immunoblotting analysis demonstrated that full length WWOX protein was the predominant WWOX isoform expressed in both normal ovary and ovarian carcinomas (Figure [1B](#F1){ref-type="fig"}). The anti-WWOX antibody used was raised against the WW domains \[[@B2]\] and should detect all potential WWOX isoforms. The specificity of this WWOX antibody in the immunoblotting analyses was demonstrated by observing no immunoreactive products in cell extracts from the PEO1 ovarian carcinoma cell line used as negative control (Figure [1A](#F1){ref-type="fig"}). In addition, as a positive control PEO1 cells were stably transfected with a WWOX expression vector (Figure [1A](#F1){ref-type="fig"}). Normal ovarian tissues displayed a consistently strong WWOX specific signal while WWOX protein expression levels were extremely variable among tumor samples. Some tumors displayed barely detectable, if any, WWOX protein when compared to normal tissue, e.g. T108 while other samples had significantly higher WWOX levels e.g. T578 in Figure [1](#F1){ref-type="fig"}. We concluded that 26% (10/38) of tumors over-expressed WWOX, 34% (13/38) expressed normal WWOX levels and 40% (15/38) of tumors had levels lower than 50% that observed in normal ovary. Pattern of WWOX staining in normal ovary ---------------------------------------- Initially we characterized the cellular localization of WWOX protein expression in normal ovaries by IHC using the antibody described in the previous section. The specificity of the WWOX antibody for IHC analysis was validated by observing that the immunoreactivity was abolished by pre-absortion to the recombinant GST-WWOX fusion protein used to raise the antibody (data not shown). The analysis of TMA cores and whole sections of normal human ovary showed that WWOX protein is constitutively expressed at high levels in normal OSE cells (Figure [2A](#F2){ref-type="fig"}), in epithelial inclusion cysts within the cortex (Figure [2B](#F2){ref-type="fig"}) and in scattered granulosa-lutein cells surrounding aged corpora lutea or corpora albicans (not shown). Positive staining was also detected in one observed Walthard nest within the ovarian hilus and in a few stromal cell patches likely related with hormone production. In all cases the inmunostaining had a homogeneous and diffuse pattern that was localized to the cytoplasm. WWOX staining in ovarian carcinomas ----------------------------------- WWOX expression was highly variable among ovarian tumors (n = 444). WWOX staining intensity was determined in all four major surface epithelium derived ovarian carcinoma histotypes, i.e. Serous, Endometroid, Mucinous and Clear Cell (Table [1](#T1){ref-type="table"}). Thirty percent of ovarian carcinomas (133/444) showed loss of WWOX protein expression while 70% (311/444) demonstrated positive staining including cases with very strong staining (Figure [3A--I](#F3){ref-type="fig"}) (Table [1](#T1){ref-type="table"}). Among serous carcinomas, 29% of the cases (109/375) fell in the negative/low group and 71% (266/375) stained moderately or strongly positive. The most typical WWOX staining pattern observed for all histotypes was cytoplasmic and diffuse, no tissues demonstrated any nuclear staining. Interestingly, some of the carcinomas showed a distinctive strong cell membrane staining pattern (Figure [3](#F3){ref-type="fig"}). Other tumors displayed a predominantly apical border staining, in conjunction with staining of what appeared to be luminal secretions (Figure [3B](#F3){ref-type="fig"}). Among endometroid carcinomas 23% of the cases (9/40) were negative/low and 77% (31/40) stained moderately or strongly positive for WWOX expression. This histological subtype showed a diffuse and apical cytoplasmic pattern of protein expression (Figure [3D, E](#F3){ref-type="fig"}). Among the CCC group, 42% of the cases (8/19) were negative/weak while 58% (11/19) stained moderately or strongly for WWOX (Figure [3G--I](#F3){ref-type="fig"}). Within the mucinous carcinoma group, we observed that most tumors (7 out of 10 cases) did not stain for WWOX expression. Furthermore the three positive mucinous carcinomas were clearly below the mean value for normal OSE. The immunoreactivity localized to the cytoplasm as in the other histotypes but in a very specific perinuclear fashion (Figure [3J](#F3){ref-type="fig"}). The high rate of absence of WWOX protein expression among the mucinous carcinoma group was statistically different compared to the other three histotypes (p = 0.0131, Table [1](#T1){ref-type="table"}). WWOX staining analysis according to stage and grade --------------------------------------------------- WWOX expression was next correlated with tumor Stages (FIGO). We observed that 23% of Stage I (7/31), 29% of Stage II (8/28), 31% of Stage III (74/242) and 48% of Stage IV (31/65) tumors showed negative/low WWOX protein expression. We detected a statistically significant trend between the decrease in WWOX expression and more advanced FIGO stages (-b = -0.135; p trend=0.007, Table [2](#T2){ref-type="table"}). No correlation was found between tumor grade (GOG) and WWOX staining (p = 0.707, Table [2](#T2){ref-type="table"}). WWOX staining analysis according to relapse and survival -------------------------------------------------------- An analysis of WWOX protein expression relative to disease relapse was performed in a subset of cases from the MDACC TMA (n = 354). Between relapsing and non-relapsing cases no statistical difference was observed regarding WWOX expression (p = 0.094) (Table [2](#T2){ref-type="table"}). Among cases with progressive disease, 46% (27/59) showed negative/low WWOX levels of expression while 54% (32/59) fell in the positive category. It was possible to evaluate survival in cases from the MDACC group as shown by the Kaplan-Meier plot (Figure [4](#F4){ref-type="fig"}). Importantly, we observed a statistically significant correlation of WWOX expressing cases with longer overall survival and loss of WWOX protein expression correlated with shorter overall survival (p = 0.03). WWOX staining analysis in correlation with ER and PR status ----------------------------------------------------------- We analyzed the association of WWOX with ERα and PR status in ovarian carcinoma cases (n = 323). When ER status was correlated with WWOX expression levels no significant differences were found (Table [2](#T2){ref-type="table"}). On the other hand, a statistically significant correlation was observed between WWOX expression and PR expression (p = 0.008, Table [2](#T2){ref-type="table"}). In the WWOX negative cases 85% (70/82) were PR negative while only 15% (12/83) were PR positive. Discussion ========== Since our original cloning of WWOX\[[@B1]\] abundant evidence has accumulated indicating that this gene likely plays a role in either tumor initiation or progression in various neoplasias including ovarian cancer \[[@B3]-[@B12]\]. WWOX spans the second most common chromosomal fragile site, FRA16D. This genomic region is prone to chromosome breakage, recombination and gross rearrangements. Multiple ovarian carcinoma studies have demonstrated high frequency of allelic losses for specific chromosomal regions on 16q \[[@B22]\] and since the losses have been found both in low and high-grade tumors it has been proposed to be an early event in ovarian carcinoma development \[[@B23]\]. Loss of heterozygosity at 16q23.2-q24.2 has been correlated with ovarian carcinoma metastasis and advanced tumor stages \[[@B24]\]. Comparative Genomic Hybridization (CGH) studies confirmed those observations also indicating high frequency of 16q21-q24 genomic losses in advanced invasive ovarian carcinoma stages \[[@B25],[@B26]\]. Taken together the data indicates that a gene, likely WWOX, located at 16q23-24 is associated with a worse prognosis in ovarian cancer. In this study WWOX protein expression in normal ovary was predominantly observed in the OSE and epithelial inclusions, both sites proposed to be where ovarian epithelial neoplastic transformation originates. Because of this finding together with the observations of Paige (2001) \[[@B4]\], that ovarian carcinoma cell lines have high frequency of genomic losses affecting the WWOXgene including homozygous deletions, we considered it of relevance to analyze WWOX protein expression among the four major surface epithelial derived ovarian carcinoma histotypes. Our analysis by immunoblotting and IHC demonstrated that about one third of ovarian carcinomas displayed extremely reduced or absent WWOX protein expression. It is worth noting that in all instances, normal and tumor, WWOX staining was exclusively cytoplasmic and in some serous carcinomas staining was also localized to the apical cell membrane or luminal border. This is similar to our observations in breast tumors \[[@B12]\], where WWOX was always cytoplasmic without nuclear staining. This is also in agreement with observations from other groups \[[@B11]\]. This is in stark contrast with Watanabe (2003) \[[@B27]\] who observed that some cells of normal and tumor breast and gastric cases had WWOX protein expression localized to the nuclei. Possibly this discrepancy is due to the use of different antibodies. In this ovarian cancer study, we also observed that Serous and Endometroid histotypes express normal or high WWOX but two rare but well described ovarian carcinoma histotypes, mucinous and clear cell (28,29,30), demonstrated a higher frequency of loss of WWOX protein expression. This association was more significant for the mucinous type, for which we observed that 70% (7/10) of cases were devoid of WWOX protein expression. Eventhough a small number of mucinous cases were studied the overall high frequency of decreased or loss of WWOX in these tumors as a group is intriguing. Borderline significance was found for the pure CCC for which 42% of cases demonstrated no or very low WWOX protein expression (Figure [3H](#F3){ref-type="fig"} versus [3G](#F3){ref-type="fig"}). Primary mucinous ovarian tumors have been shown to carry a higher frequency of K-ras mutations as one of the few genetic hallmarks able to distinguish it from other histotypes, while CCC are characterized by a lack of p53 mutations. Mucinous carcinomas and CCC, has been associated with a higher frequency of resistance to chemotherapy (31). Interestingly, in our study, we observed a statistically significant trend between loss of WWOX expression and worse patient outcome. Patients whose tumors expressed WWOX at moderate or high levels fared better than those with low expression of this enzyme (Figure [4](#F4){ref-type="fig"}). We also found a statistically significant correlation with loss of WWOX and stage IV at initial diagnosis (FIGO). The association with overall survival could be due to an increased frequency of loss in higher stage tumors that are associated with a poorer outcome or alternatively due, in part, to an increased frequency of clear cell and mucinous tumors which have a lower likehood of responding to therapy \[[@B31]\]. It was also of interest that we observed a positive correlation between WWOX and PR expression. In contrast to our previous observations in breast cancer in which we observed a strong correlation between ER and PR with WWOX expression \[[@B12]\], in this ovarian study, only PR was associated with WWOX expression (Table [2](#T2){ref-type="table"}). Specifically PR negative ovarian carcinomas also lacked WWOX expression. Progestins acting through the PR have been postulated to be protective for ovarian carcinoma development \[[@B32],[@B33]\] due to a postulated ability of inhibiting cell proliferation and inducing apoptosis (34,35). We noticed that all (100%) of our analyzed mucinous carcinomas were PR negative which agrees with the literature regarding absence of PR expression in mucinous tumors \[[@B36],[@B37]\] while 90% of CCC were PR negative. These correlations raise the question of whether the observed positive association between PR loss and WWOX loss is a consequence of the predominance of WWOX loss in the two aforementioned histotypes, rather than a direct mechanistic association between WWOX and PR. Changes in progesterone signaling and regulation of progesterone-responsive genes can be of critical significance in the ovarian tumorigenic process by itself \[[@B38]\]. The opposite could be true as well i.e. that high levels of WWOX expression is associated with PR expression and as a consequence related to the putative protective effect of progestins that in turn could be associated with more favorable patient outcomes. Conclusion ========== We provide evidence that WWOX protein expression is frequently altered and highly variable in ovarian carcinomas. The loss of WWOX expression shows association with Mucinous and CCC histotypes more than the others and also shows a tendency with PR negative expression. Significantly, reduced WWOX protein expression correlates with less favorable outcome. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= CMA conceived, coordinated, designed the study, interpreted the data and revised the manuscript. MIN coordinated, performed pathological and immunohistochemical study, interpreted the data, drafted and write the report.DR constructed M.D. Anderson Cancer Center Tissue Microarray and interpreted the data. JHLM participated in preparation of the antibody, Western Blot analyses, interpretation of the data and critically reviewed the manuscript. MCA performed statistical analysis, figures and tables and interpreted the data. HK carried out Western Blot analyses. RP constructed Fox Chase Cancer Center Tissue Microarray. AJPKS, AG, JL and GM provided tumor samples, interpretation of the data and critically reviewed the manuscript. All of the authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/64/prepub> Acknowledgements ================ The authors gratefully acknowledge Nancy W Abbey for her outstanding technical assistance. The authors acknowledge NCI RO1 CA102444 (C. M. Aldaz), SPORE in Ovarian Cancer P50 CA83639, tissue pathology core grant NCI PO1 CA64602 and ACS RSG-04-028-CCE (J. Liu), and center grant ES-07784. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### WWOX protein expression in normal ovary and ovarian adenocarcinomas as determined by immunoblot analysis. A) Total protein extracts from PEO1 cell lines transfected with an empty vector or a WWOX expressing vector were analyzed by immunobloting with the anti-WWOX antibody. Note: no immunoreactive bands are observed in the vector transfected cell line. B) WWOX protein expression was determined by immunoblotting of total protein extracts from 38 ovarian carcinoma samples. Five normal ovarian tissue extracts are shown on the first five lanes. Quantitation of WWOX protein expression. Autoradiographs of WWOX and actin were digitized using the Kodak digital science Image Station 440CF. WWOX expression in each sample was normalizated to actin to correct for loading differences. In turn these numbers obtained from each tumor were normalized and expressed as relative to the normal ovarian values (i.e. Relative Expression). ::: ![](1471-2407-5-64-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### WWOX immunohistochemical staining of normal ovary. A) Representative photomicrograph (20X) of normal ovary displaying positive staining in ovarian surface epithelial cells. B) Photomicrograph (20X) showing strong WWOX inmunostaining localizing to the cytoplasm of inclusion cyst epithelial cells. ::: ![](1471-2407-5-64-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### WWOX inmunohistochemical staining in ovarian carcinoma samples by histotypes. A-C. Serous ovarian carcinomas. A, Strong-Moderate. Note heterogeneity in staining intensity pattern in this tumor sample; B, Moderate. Note predominance of apical staining in this papillary serous ovarian carcinoma; C Weak-Negative, WWOX lack of staining observed in approximately 30% of serous carcinoma cases. D-F. Endometroid ovarian carcinomas. D and E positive WWOX cytoplasmic staining and F negative staining. G-I. Clear Cell ovarian Carcinomas. G, representative photomicrograph of one of a moderately WWOX positive CCC case, while H and I, represent typical negative cases. J-L. Mucinous ovarian carcinomas, J, representative mildly positive case and K representative mucinous carcinoma of the endocervicoid subtype with demonstrating no WWOX staining. L, mucinous carcinoma of the intestinal subtype also negative for WWOX staining. ::: ![](1471-2407-5-64-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Kaplan Meier Plot analysis showing the pattern of Overall Survival relative to WWOX protein expression. ::: ![](1471-2407-5-64-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### WWOX intensity staining according to the four major epithelial derived ovarian carcinoma histotypes. ::: *Histology* *WWOX intensity* *Total* --------------------------- ---------------------- ------------------- --------- Serous carcinoma 109/375 (29%) 266/375 (71%) 375 Endometroid carcinoma 9/40 (23%) 31/40 (77%) 40 Clear cell carcinoma 8/19 (42%) 11/19 (58%) 19 Mucinous carcinoma \* 7/10 (70%) 3/10 (30%) 10 Total 133/444 (30%) 311/444 (70%) 444 \ (χ^2^= 8.67; p = 0.0131)* ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Clinical parameters and steroid receptor status analyzed in correlation with WWOX staining. ::: *Clinical parameters* *WWOX intensity* *Total* *p value* --------------------------- ---------------------- ------------- --------------- ------------- Grade (GOG) 1 4 (23%) 13 (77%) 17 p = 0.707 2 9 (33%) 18 (67%) 27 3 107 (33%) 215 (67%) 322 Stage (FIGO) I 7 (23%) 24 (77%) 31 p = 0.007 II 8 (29%) 20 (71%) 28 III 74 (31%) 168 (69%) 242 IV 31(48%) 34 (52%) 65 Relapse No 31(32%) 66 (68%) 97 p = 0.09 Yes 61(31%) 137(69%) 198 Progressive disease 27(46%) 32 (54%) 59 ER Negative 22 (25%) 66 (75%) 88 p = 0.984 Positive 59 (25%) 176 (75%) 235 PR Negative 70 (29%) 170 (71%) 240 p = 0.00 Positive 12 (14%) 71 (86%) 83 :::	PubMed Central	2024-06-05T03:55:58.354966	2005-6-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173095/", "journal": "BMC Cancer. 2005 Jun 27; 5:64", "authors": [ { "first": "María I", "last": "Nunez" }, { "first": "Daniel G", "last": "Rosen" }, { "first": "John H", "last": "Ludes-Meyers" }, { "first": "Martín C", "last": "Abba" }, { "first": "Hyunsuk", "last": "Kil" }, { "first": "Robert", "last": "Page" }, { "first": "Andres JP", "last": "Klein-Szanto" }, { "first": "Andrew K", "last": "Godwin" }, { "first": "Jinsong", "last": "Liu" }, { "first": "Gordon B", "last": "Mills" }, { "first": "C Marcelo", "last": "Aldaz" } ] }
PMC1173096	Background ========== Myocarditis is an inflammatory cardiomyopathy that occurs in all age groups \[[@B1]\]. Most cases are thought to be sub-clinical, but myocarditis can manifest as fulminant or chronic heart failure \[[@B2]\]. The inflammation is presumed to most commonly start as an infectious process, although autoimmune and idiopathic forms also occur \[[@B1]\]. Bacterial or protozoal myocarditis is rare in the developed world, with the vast majority of cases being viral \[[@B3]\]. Although most viruses that are pathogenic for humans can cause myocarditis, common etiologic agents are enteroviruses and adenoviruses \[[@B4]\]. It remains unclear if the primary problem in severe disease is ongoing damage from a virus, a post-infectious inflammatory reaction, or a combination of both factors. If ongoing infection is the primary problem, IVGG could be efficacious if it contains antibodies to the microbe. IVGG also has anti-inflammatory properties that prevent the formation of and downregulate cytokines \[[@B5]\] so could be efficacious even if the primary problem is a post-infectious inflammatory reaction. The objective of this study was to review the evidence evaluating the use of IVGG in the treatment of patients with acute myocarditis. A secondary objective was to determine if there is an identifiable group of acute myocarditis patients (based on age, duration of symptoms, acuity of onset of symptoms, cardiac function at presentation, virologic results, or the presence or absence of histologic evidence of acute myocarditis on cardiac biopsy in patients where a biopsy was performed) who would benefit from IVGG. The review of randomized controlled trials has been registered with the Cochrane Collaboration and regular updates will be available in the Cochrane Library. Methods ======= Searching --------- We systematically searched the following electronic databases: Cochrane Central Register of Controlled Trials (4^th^quarter, 2004), MEDLINE (1966 to January Week 4, 2005), EMBASE (1988 to Week 6, 2005), CINAHL (1982 to January Week 3, 2005), Web of Science (1975 to February 5, 2005), and several trial registries in addition to the Cochrane Heart Group\'s Trial Registry \[[@B6]-[@B11]\]. A comprehensive search strategy was developed by a medical research librarian based on the following terms: immunoglobulin, gammaglobulin, ivig, immunoglobulin, igg, immune, serum, myocarditis, cardiomyopathy, myocardiopathy, carditis, heart, inflammation. The complete search strategy is available from the authors on request. We also contacted the primary author of the RCT, and reviewed the reference lists of all selected articles. In addition, we handsearched proceedings from the following meetings for RCTs: American Heart Association (1999--2004), American College of Cardiology (1998--2005), European Society of Cardiology (1998--2004), and International Heart and Lung Transplantation Society (1998--2004). The search was not limited by language or publication status. Study selection --------------- Two reviewers independently screened the searches for potentially relevant studies. The full manuscript of all potentially relevant studies was retrieved and each study was assessed independently by two reviewers for inclusion using predetermined eligibility criteria. Primary studies of any design were eligible for inclusion. Inclusion criteria included reports of patients given IVGG for possible viral myocarditis within six months of onset of symptoms. Because of the poor sensitivity of cardiac biopsy as a diagnostic tool for acute myocarditis, a histologic diagnosis was not required. Exclusion criteria included reports of patients with evidence of ischemic heart disease, rheumatic heart disease, or any other non-viral etiology, onset of myocarditis less than six months postpartum (as the pathogenesis of postpartum myocarditis may differ from that of presumed viral myocarditis), or receipt of immunosuppressive therapy (unless this was not possible as they were reported as part of a case series). Reports of use of high-dose immunoglobulin directed at specific organisms were excluded as efficacy may be influenced by the formulation available and the accuracy of the etiologic diagnosis. Assessment of methodological quality ------------------------------------ Study quality was assessed independently by two reviewers and discrepancies were resolved by consensus. RCTs were evaluated using the previously validated Jadad 5-point score to assess randomization, double blinding, and losses to follow-up \[[@B12]\]. Allocation concealment was assessed as adequate, inadequate, or unclear \[[@B13]\]. Non-RCTs were graded according to the levels of evidence presented in Table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Levels of evidence for studies of therapeutic interventions from the Oxford Centre for Evidence-based Medicine (website title <http://www.cebm.net/levels_of_evidence.asp>) \[14\] ::: Level Study design ------- -------------- -------------------------------------------------------------------------------------------------------------------- 1 A Systematic review (with homogeneity) of randomized controlled trials B Individual randomized controlled trial (with narrow confidence interval) C All or none 2 A Systematic review (with homogeneity) of cohort studies B Individual cohort study (including low quality randomized controlled trial, e.g., \< 80% follow-up) C \"Outcomes\" research; ecological studies 3 A Systematic review (with homogeneity) of case-control studies B Individual case-control study 4 Case series (and poor quality cohort and case-control studies) 5 Expert opinion without explicit critical appraisal, or based on physiology, bench research or \"first principles\" ::: Data extraction --------------- Data were extracted by one reviewer using a standard form and checked for accuracy and completeness by a second. The primary outcome was rate of survival without a transplant or requirement for placement of a left ventricular assist device. Secondary outcomes included change in echocardiographic measures of cardiac function, duration of hospitalization, and improvement in functional symptoms (determined by increased exercise tolerance as measured by any objective test and New York Heart Association Functional Capacity) when available. Data were collected on complications and adverse events. Data analysis ------------- Data from studies that were not RCTs were reported but not analyzed further. We planned to combine data from RCTs, to do sub-group analyses, and to test for publication bias, however these were not possible as only one RCT was identified. For the RCT, dichotomous data on efficacy (e.g. event-free survival) were expressed as an odds ratio (OR) with 95% confidence interval. Dichotomous data on adverse events were reported as a risk difference and number needed to harm with 95% confidence intervals. Continuous data (change in left ventricular ejection fraction and peak oxygen consumption) were converted to the mean difference with 95% confidence intervals. For peak oxygen consumption, we assumed a correlation of 0.5 and used the methods of Follmann \[[@B15]\] to calculate the standard deviations of the change from baseline estimates. Since the number per group was not given for this variable, an estimate of the sample size in each group had to be estimated by pro-rating the original group sample sizes to the new total sample size. Results ======= Figure [1](#F1){ref-type="fig"} presents a flow diagram of studies considered for inclusion in the review. Only one RCT evaluating IVGG for acute myocarditis has been reported to date \[[@B16]\]. The trial was of moderate methodological quality according to the Jadad scale (3 out of maximum 5 points, where 5 indicates highest quality); allocation concealment was unclear. The study enrolled 62 adults (mean age 43.0 +/- 12.3 years; 37 men), of which only ten had cellular inflammation on endomyocardial biopsy (four fulfilled the Dallas criteria for acute myocarditis, and three for borderline myocarditis). The authors reported that the given sample size would provide 80% power to detect a difference between groups of ≥ 8% in ejection fraction (EF) change scores. Patients were randomized to receive either 2 g/kg IVGG or an equivalent volume of 0.1% albumin in a blinded fashion. The incidence of death or requirement for cardiac transplant or placement of a left ventricular assist device was low in both groups. Event-free survival was not significantly different but favored the control group (OR 0.52, 95% CI 0.12, 2.30). Follow-up at 6 and 12 months showed no significant difference in improvement in left ventricular ejection fraction for cases and controls (mean difference 0.00, 95% CI -0.07, 0.07 at six months; 0.01, 95% CI -0.06, 0.08 at 12 months). Functional capacity as assessed by peak oxygen consumption was not significantly different in the two groups at 12 months (mean difference -0.80, 95% CI -4.57, 2.97). Infusion-related side effects occurred significantly more often in the treated group (RD 0.33, 95% CI 0.17, 0.50; number needed to harm = 3, 95% CI 2, 6), but all appeared to be mild. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow diagram of studies considered for inclusion in the review ::: ![](1471-2261-5-12-1) ::: Sixteen additional studies of various designs met the inclusion criteria \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. These included a retrospective case series \[[@B17]\], an uncontrolled trial \[[@B18]\], two case series \[[@B5],[@B19]\], and twelve case reports \[[@B20]-[@B31]\]. In a retrospective cohort study, Drucker compared children who were treated with IVGG (n = 21) to historic controls (n = 26); 65% of the sample was less than two years of age \[[@B17]\]. The treatment group showed improved cardiac function and a trend (though not statistically significant) towards improved survival. McNamara conducted an uncontrolled trial involving 10 adults (mean age 35.8 +/- 15 years) \[[@B18]\]. One patient with refractory arrhythmias and severe heart failure died during the infusion; the other nine patients showed an impressive improvement in cardiac function following administration of IVGG, with a significant increase in left ventricular ejection fraction at 12-month follow-up (p = 0.008). In ten case reports and both case series, improvement in cardiac function was described after use of IVGG in a total of 21 cases, although improvement was considered to be minimal in one case \[[@B26]\]. In the other two case reports, there was no benefit of IVGG in a patient presenting with hand, foot and mouth disease and acute myocarditis from coxsackievirus A16 \[[@B27]\] and in a patient with fulminant myocarditis of unclear etiology \[[@B28]\]. Hemolytic anemia was attributed to use of IVGG in one case report \[[@B26]\]. We could find no ongoing trials of the use of IVGG in acute myocarditis, although the European Study of Epidemiology and Treatment of Cardiac Inflammatory Diseases (ESETCID) is evaluating hyperimmunoglobulin for patients with cytomegalovirus-positive myocarditis \[[@B32]\]. Discussion ========== The small number of studies evaluating IVGG in the treatment of acute myocarditis reflects the immature state of this body of literature. We identified only one RCT (Level 1 evidence), which involved 62 adult patients with idiopathic cardiomyopathy (of which only 10 had histologic evidence of myocarditis) and showed no apparent benefit \[[@B9]\]. This is in contrast to an uncontrolled trial and numerous observational studies (Level 4 evidence), most of which suggested potential benefit. This indicates a possible bias towards publication of case reports or case series with positive results. The validity of evidence was generally poor, with the majority of included studies ranking low in terms of level of evidence. The most important threat to validity for fifteen of the seventeen included studies was the lack of controls, which could result in an overestimate of the benefit of IVGG. Spontaneous improvement in cardiac function is common with acute myocarditis and can be rapid or gradual, so it is possible that the improvement noted in these cases was part of the natural history of the disease. The retrospective cohort study demonstrated a greater improvement in cardiac function in patients given IVGG compared to historic controls \[[@B17]\]; this type of study is more susceptible to bias because of the inability to fully control for all potential confounders, such as other changes in patient management over the eight-year study period. Acute myocarditis is a relatively non-specific entity, as the diagnosis is often clinical. Laboratory confirmation consists of microbiologic, histologic, and immunohistochemical methods. With regard to microbiologic confirmation, it is rare to isolate the etiologic agent from a myocardial biopsy (probably because the biopsy is usually done too late in the course of the illness), and identification of organisms by molecular techniques is in its infancy \[[@B4]\]. However, with the development of new effective antivirals, there should be increased efforts towards making a virologic diagnosis early in the course of myocarditis as it is possible that antivirals would improve the prognosis. Furthermore, there are case reports of hyperimmunoglobulin showing an apparent effect in cases of varicella \[[@B33]\], cytomegalovirus \[[@B34]\], and parvovirus \[[@B35]\] myocarditis. With regard to histologic confirmation, the Dallas criteria (lymphocyte infiltration with myocyte necrosis) \[[@B36]\] and more recently the World Heart Federation criteria (\>14 leukocytes/mm^2^with necrosis or degeneration) \[[@B37]\] have been used for histologic diagnosis, but the sensitivity and specificity of these criteria are not known. There are no standardized values for immunohistochemical markers \[[@B38]\]. Our aim in this study was to analyze reports of use of IVGG for presumed viral myocarditis, but it is possible that many of the patients in the reports did not have viral myocarditis (only 10 of the 62 patients in the RTC had cellular inflammation on endomyocardial biopsy). Once our ability to accurately diagnose viral myocarditis improves, it may be possible to identify a subset of patients who will respond to IVGG. This might be patients whose disease was precipitated by a specific virus, or patients who are treated with IVGG early in the course of their illness when they have ongoing viral replication in the myocardium. Perhaps pediatric patients are more likely to respond. Children are thought to be more likely to present in the acute inflammatory stage of illness \[[@B16]\], and may have a worse prognosis than do adults when they present with fulminant myocarditis \[[@B39],[@B40]\] In conclusion, the value of IVGG in patients with acute myocarditis is obscured by the poor quality of evidence. We were not able to identify a subgroup of patients who appear to be more likely to respond to IVGG. A large RCT is required to evaluate the efficacy of IVGG for acute myocarditis with emphasis on the etiology of the myocarditis. Until there are RCTs demonstrating benefit, use of IVGG for acute myocarditis should not be part of routine practice. Moreover, there is a great need for further studies of the pathophysiology of acute myocarditis, which would allow for a better understanding of the etiology and the natural history of the disease. This might allow for improved diagnostic criteria, which would make it much easier to design studies of treatment options. This may also assist in identifying sub-groups of patients where IVGG or other therapies have a greater potential to confer clinical benefit. List of abbreviations ===================== IVGG -- intravenous gamma globulin RCT -- Randomized controlled trial Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= The idea for the study originated with JLR. The protocol was written by LH with help from JLR and EC. The literature review was completed by EC, and LH and JLR reviewed the papers and derived the data. Statistical analysis was done by BV. The manuscript was written by JLR and LH with input from TPK, EC and BV. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/12/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Observational studies and nonrandomized trials examining the use of IVGG in acute myocarditis. This table summarizes all studies reporting the use of IVGG for acute myocarditis. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Philip Berry and Marilyn Josefsson, from the Alberta Research Centre for Child Health Evidence, for administrative support and collection of articles.	PubMed Central	2024-06-05T03:55:58.357326	2005-6-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173096/", "journal": "BMC Cardiovasc Disord. 2005 Jun 2; 5:12", "authors": [ { "first": "Joan L", "last": "Robinson" }, { "first": "Lisa", "last": "Hartling" }, { "first": "Ellen", "last": "Crumley" }, { "first": "Ben", "last": "Vandermeer" }, { "first": "Terry P", "last": "Klassen" } ] }
PMC1173097	Background ========== Hypertension is among the most prevalent chronic diseases in the United States \[[@B1]\]. Despite the availability of effective medications and well-published guidelines for the treatment of hypertension \[[@B1]-[@B4]\], the majority (approximately 75%) of hypertension in the United States remains poorly controlled \[[@B5]\]. Hypertension is particularly burdensome among racial/ethnic minority groups and hypertension-related cardiovascular disease has been shown to be the greatest contributor to previously documented racial differences in mortality \[[@B5]-[@B12]\]. Although higher rates of hypertension control and a reduction in racial differences in outcomes from hypertension may be obtained by increasing providers\' aggressiveness in intensifying therapy when indicated \[[@B2],[@B12]\], several studies have demonstrated that providers often allow their patients to remain poorly controlled \[[@B12],[@B13]\]. We previously examined the association of patients\' race/ethnicity with processes of hypertension care \[[@B13]\]. We found that, among a cohort of hypertension patients with repeatedly elevated blood pressures, Hispanics were significantly less likely to have therapy intensified and were more likely to have uncontrolled blood pressure (BP) than were other racial and ethnic groups. In an effort to identify potential targets for interventions to improve hypertension care among our patients, we further examined which patient and provider characteristics may explain racial differences in rates of therapy intensification. Methods ======= Study sample and procedures --------------------------- To determine which patient-centered characteristics are associated with providers intensifying drug therapy for hypertension and whether provider experience is related to differences in intensification of therapy, we studied a random subset of 1,205 patients who had a minimum of two hypertension-related outpatient visits to one of twelve general medicine clinics in community health centers and community practices affiliated with a large urban academic medical center from July 1, 2001 through June 30, 2002 (totaling 3,257 visits). To determine hypertension-related visits, we reviewed the electronic medical record (EMR) for all clinic visits with a primary or secondary diagnostic code of hypertension (HTN) (ICD-9 401- 401.9, 405- 405.99). The study protocol was approved by the Human Studies Committee at the Brigham and Women\'s Hospital. Medical record and administrative data -------------------------------------- We hired three abstractors who were trained by the primary investigator to review the EMR of each patient in our study sample, each abstractor reviewed records for approximately 400 individual patients. From each hypertension-related visit note in the EMR they collected the following data: patient race/ethnicity, name of the patient\'s primary provider, patient age at time of initial study period visit, sex, primary insurer at time of initial study period visit, presence of comorbid disease (diabetes, congestive heart failure (CHF), coronary artery disease (CAD), or renal failure) listed on the patient problem list, BP control (defined using cut point of \<130/85 for diabetic or renal failure patients and \<140/90 for all others), and any changes made to antihypertensive drug therapy during visit (decrease or discontinuation of drug, change to another class of drug, or increase of drug dose or addition of new drug). The EMR of each patient contains racial/ethnic data obtained via self-report the first time each patient registers in his/her clinic that then is classified by the data entry clerk as Hispanic ethnicity or not and race is then classified as Black, White, other, or unknown. Hence, it is impossible to know the country of origin of our Hispanic participants and we could not determine the exact racial/ethnic mix of patients classified as \"other\" or \"unknown.\" For these reasons, we excluded patients who\'s EMR listed their racial/ethnic classification as \"other\" (N = 25) or \"unknown\" (N = 67) and we limited our analyses to White patients (who identify themselves as non-Hispanic), Black patients (who identify themselves as non-Hispanic), and self-identified Hispanic patients. From the medical center\'s administrative database, we abstracted each provider\'s level of experience (intern, resident, or attending). Patient zip code was also obtained and linked to 2000 U.S. Census data to obtain the median annual household income in each patient\'s zip code. Intensity of therapy -------------------- Our methods for identifying intensified cases among our cohort have been previously described \[[@B13]\]. Because we were examining the quality of care delivered to hypertension patients prior to July of 2002, approximately one year before the release of the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation and Treatment of High Blood Pressure (JNC VII) \[[@B14]\], we used JNC VI definitions for blood pressure control in order to examine whether therapy was appropriately intensified. We classified each visit into two categories, intensified visits (an increase in intensity of drug therapy) versus non-intensified visits (a decrease or no change in intensity of drug therapy) according to previously published definitions of changes in medications \[[@B2]\]. We developed an algorithm to determine whether a patient received at least one increase in drug therapy (an increase in drug dose or addition of new medication) in response to a repeatedly elevated BP during the study period (Figure [1](#F1){ref-type="fig"}). Each patient with fewer than two visits with an elevated BP (N = 356) was excluded from the algorithm. Each patient with an uncontrolled BP at more than one visit was identified as either an intensified case (at least one drug increase) or a non-intensified case (no drug increases). Each reviewer examined a subset of 30 records; we then tested for inter-rater reliability and found excellent agreement among reviewers (kappa = 0.90). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow diagram of algorithm for determining an \"intensified case.\" Inter-rater reliability was high (kappa 0.90). We identified a total of 782 cases as either an \"intensified case (N = 600) or a \"non-intensified case\" (N = 157). ::: ![](1471-2261-5-16-1) ::: Our intensification algorithm is potentially limited because it does not account for differences in the number of times medication is increased in comparison to the number of uncontrolled BP visits for each patient. To address this issue, we created a secondary measure of intensification by calculating the proportion of times each patient had their therapy increased when their BP was uncontrolled (0 representing no increases and 1 representing an increase in therapy with every uncontrolled visit). We then determined whether the two intensification measures were correlated in order to further validate our algorithm and found that the two measures were highly correlated (coefficient 0.71, P \<0.001). Because our algorithm-derived method has the additional advantage of using more than a single visit\'s BP in determining whether therapy intensification is indicated, we used this measure of intensification for all subsequent analyses. Data analysis ------------- We compared patients\' demographic, clinical, and provider\'s characteristics by patient race/ethnicity (Table [1](#T1){ref-type="table"}) using chi-squared tests for categorical and Student\'s t-test for continuous variables. We also estimated the association of being an intensified case with patients\' demographic, clinical, and provider\'s characteristics (Table [2](#T2){ref-type="table"}). We report two-tailed P values with statistical significance set at P≤0.05 for all analyses. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic and clinical characteristics of population by race and ethnicity ::: Variable White (N = 304) Black (N = 309) Hispanic (N = 144) P value\* ------------------------------------ --------------------- --------------------- ------------------------ --------------- Men (%) 27.3 24.3 35.4 0.05 Aged \<65 years (%) 54.6 65.1 66.7 0.009 Annual household income^†^: \<0.001 = \$35,500 11.9 37.8 29.4 \$35,501--\$43,140 11.9 40.8 21.0 \$43,141--\$55,365 24.7 15.5 42.7 \> \$55,365 51.6 5.9 7.0 Insurance: \<0.001 Private/Medicare (%) 87.7 69.1 40.3 Medicaid (%) 4.3 17.6 28.5 Free care/Self-pay (%) 8.0 13.4 31.3 Mean number of hypertension visits 2.8 3.0 2.8 0.06 Diabetic (%) 14.5 27.5 27.8 \<0.001 Coronary artery disease (%) 5.3 4.2 2.1 0.02 Provider experience (%)^‡^: \<0.001 Intern 2.1 9.3 10.1 Resident 4.8 23.0 10.1 Attending 93.2 67.8 79.8 \Using chi-square tests for categorical variables and ANOVA for mean number of hypertension visits. ^†^N = 734 for those patients with zip code available. ^‡^N = 691 for those patients with provider information available. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Demographic and clinical characteristics of population by intensification of therapy ::: Variable* Non-Intensified Case (N = 157) Intensified Case (N = 600) P value\* ------------------------------------ ------------------------------------ -------------------------------- --------------- Patient race/ethnicity (%): 0.03 White 19.7 80.3 Black 18.1 81.9 Hispanic 28.5 71.5 Men (%) 29.1 27.7 0.74 Age: 0.10 \<65 years 18.8 81.2 ≥65 years 23.8 76.2 Annual household income^†^: 0.93 ≤\$35,500 20.4 79.6 \$35,501--\$43,140 19.7 80.3 \$43,141--\$55,365 22.4 77.6 \> \$55,365 20.5 79.5 Insurance: 0.15 Private/Medicare (%) 19.5 80.5 Medicaid (%) 27.8 72.2 Free care/Self-pay (%) 20.0 80.0 Mean number of hypertension visits 2.6 3.0 \<0.001 Diabetic (%): 0.01 Yes 27.8 72.2 No 18.7 81.3 Coronary artery disease (%): 0.24 Yes 12.5 87.5 No 21.1 78.9 Provider training level (%)^‡^: 0.91 Intern 22.7 77.3 Resident 21.4 78.6 Attending 20.3 79.7 \Using chi-square tests for categorical variables and Student\'s t-test to compare mean hypertension visits. ^†^N = 734 for those patients with zip code available. ^‡^N = 691 for those patients with provider information available. ::: Using logistic regression, we assessed whether race/ethnicity was associated with intensification of therapy after adjusting for all measured confounders. Data were available on every variable for 685 of the 757 patients (90.5%) for multivariable analyses. We report adjusted odds ratios with 95% confidence intervals for the intensified cases. In secondary analyses, we included interaction terms for patient race/ethnicity and provider experience level to determine whether racial/ethnic differences in intensification differed by provider experience level. All non-significant interaction terms were removed from the final model. All models were estimated using SUDAAN statistical software to adjust for within clinic correlation of visits \[[@B15]\]. Results ======= Patient, clinical, and provider characteristics ----------------------------------------------- Of the 757 patients, 304 (40%) were White, 309 (41%) were Black, and 144 (19%) were Hispanic (Table [1](#T1){ref-type="table"}). The majority of patients in our cohort were women (72%) and most were either privately insured or had Medicare (71%). There were 169 (22%) patients in our cohort with diabetes. We were able to determine the primary provider for 691 patients, of who 44 (6.4%) received care from an intern, 89 (12.9%) received care from a resident, and 558 (80.8%) received care from an attending. Demographic and clinical differences between patients\' racial/ethnic groups are presented in Table [1](#T1){ref-type="table"}. Intensification of therapy -------------------------- Of the 757 patients with multiple visits whose BP was uncontrolled at two or more visits, there were 600 (79.3%) who had their medications intensified. Hispanic patients were significantly less likely to have their medications intensified than White and Black patients (Table [2](#T2){ref-type="table"}). Diabetic patients and those with fewer visits during the study period were also less likely to have their medications intensified. After adjustment for age, insurance status, number of hypertension-related visits, diabetes, and physician experience, there were no remaining racial/ethnic differences in rates of intensification (Table [3](#T3){ref-type="table"}). Compared to patients aged = 65 years, younger patients were more likely to have their therapy intensified, and the odds of having therapy intensified increased with each visit. There were no significant interactions between race/ethnicity and provider experience in the multivariable model. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Adjusted odds ratios\ \[95% C.I.\] of being an intensified case ::: Variable Odds Ratios of being Intensified Case 95 % Confidence interval ------------------------------------- ------------------------------------------- ------------------------------ Black^†^ 1.09 \[0.71--1.67\] Hispanic^†^ 0.78 \[0.58--1.06\] Age \< 65 years^‡^ 1.61 \[1.38--1.88\] Medicaid^§^ 0.76 \[0.57--1.02\] Free care or self pay^§^ 1.22 \[0.73--2.03\] Number of hypertension visits^\\|\\|^ 1.37 \[1.17--1.61\] Diabetes\\ 0.61 \[0.34--1.08\] Intern provider^††^ 0.86 \[0.39--1.88\] Resident provider^††^ 0.89 \[0.55--1.46\] \* Adjusted for patient race/ethnicity, age, insurance status, number of hypertension-related visits, diabetes, and physician experience. ^†^Whites use as reference group ^‡^Aged ≥ 65 years used as reference group ^§^Privately insured or Medicare used as reference group ^\\|\\|^Represents increase in odds with each additional visit \\No diabetes used as reference group ^††^Attending provider used as reference group ::: Discussion ========== In our prior work, we found that Hispanic patients were more likely to receive the JNC recommended antihypertensive drug class then Whites, however, Hispanics were also significantly less likely than Whites to have their therapy appropriately intensified in response to an uncontrolled BP \[[@B13]\]. Furthermore, we found appropriate intensification of anti-hypertensive therapy was associated with subsequent BP control for all racial/ethnic groups, suggesting that the poorer rates of BP control among Hispanics in our prior study may have been due to significantly lower rates of antihypertensive medication intensification \[[@B13]\]. In this study, we found that these previously noted racial/ethnic disparities in rates of antihypertensive therapy intensification may be due to differences in visit patterns among patients and in physicians\' aggressiveness in managing BP in diabetic patients, suggesting that racial/ethnic differences in disease severity are likely determinants of unequal treatment of uncontrolled hypertension. There is substantial literature, that suggests that racial/ethnic minority groups are less likely to have their antihypertensive therapy appropriately intensified \[[@B16]-[@B19]\]. However, these studies were limited because the investigators were not able to assess practice patterns such as the frequency with which individual anti-hypertensive drugs were intensified in response to uncontrolled BP. We found that Hispanic patients in our cohort were also less likely to have their anti-hypertensive medications intensified at least once in response to repeatedly uncontrolled BP than were other racial/ethnic groups and our findings also expand beyond documenting racial/ethnic disparities in aggressiveness of therapy by determining the roles racial/ethnic differences in clinic utilization among patients and racial/ethnic differences in the prevalence of diabetes play in confounding differences in providers\' aggressiveness in treating hypertension. An important national priority in health care is the elimination of racial/ethnic disparities in healthcare; however, a better understanding of the determinants of disparities is needed to address this issue \[[@B20],[@B21]\]. Our finding that higher rates of diabetes among Hispanics in our cohort play a major role in the insufficient management of their uncontrolled BP is of particular concern given that Hispanics suffer a disproportionately larger burden from hypertension and diabetes compared to Whites; and that Hispanics are at a higher risk of having hypertension and diabetes, are less likely to be aware that they are hypertensive, are more likely to have target organ damage, and have significantly higher age-adjusted diabetes and hypertension-related mortality compared to whites in the U.S \[[@B5],[@B6],[@B22],[@B23]\]. Our findings are supported by several studies that have documented lower rates of BP control among diabetic patients \[[@B24],[@B25]\], and others showing that providers often do not attain adequate BP control for patients, even after multiple opportunities to do so \[[@B2],[@B24]\]. Our study has several limitations. We were unable to collect measures of patient adherence to prescribed therapy from the EMR. Providers may not want to intensify therapy at the same rate for patients they know are less compliant with therapy, although it is hard to identify these patients. We were also unable to determine English proficiency of each patient from the medical records. When a language barrier exists providers may be less likely to intensify therapy. We examined disparities in quality of hypertension care, as measured by intensification of therapy, among a cohort of patients and physicians during 2001--2002 using the established guidelines available during that time period, the JNC VI, as the reference for our analyses. In comparison to JNC VI, the JNC VII guidelines recommend much more aggressive management of hypertension both in terms of the accepted level of BP control (130/80 for diabetic and renal failure patients and 140/90 for all others) and the recommendations to providers to intensify antihypertensive therapy more rapidly when BP is uncontrolled \[[@B14]\]. However, since the JNC VII guidelines were not available until 2003, it would have been impossible for providers in our study sample to have been fully aware of them and it would have been misleading to assess the quality of hypertension care among our cohort using quality guidelines that were not in existence during that time. By using the less aggressive guidelines that were in existence during the study period, we may have underestimated the gap in rates of intensification between racial/ethnic groups and between patients with diabetes and those without. Lastly, we examined patients who received their care at primary care practices affiliated with a single large urban teaching hospital, and although there was substantial socioeconomic diversity in our sample our results may not be generalizable to smaller, rural, or non-teaching hospitals. Conclusion ========== We found significant racial/ethnic differences in intensification of drug therapy, and that these differences were largely accounted for by differences in frequency of clinic visits and in the prevalence of diabetes. Future interventions should focus on increasing physician awareness of the need to intensify drug therapy more, particulary among Hispanic patients, and on encouraging providers to treat hypertension more intensively in diabetic patients. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= LH conceived of the study, and participated in its design and coordination and helped to draft the manuscript. SS participated in the design of the study, performed the statistical analysis, and helped to draft the manuscript. DWB and JZA participated in the design of the study, and helped to draft the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/16/prepub> Acknowledgements ================ The authors would like to thank Deborah H. Williams for statistical programming assistance. This study was supported by a grant from the Agency of Healthcare Research and Quality (\#3U18 HS11046). Dr. Hicks was supported by the Robert Wood Johnson Foundation\'s Harold Amos Medical Faculty Development Program (\#043486).	PubMed Central	2024-06-05T03:55:58.358889	2005-6-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173097/", "journal": "BMC Cardiovasc Disord. 2005 Jun 22; 5:16", "authors": [ { "first": "LeRoi S", "last": "Hicks" }, { "first": "Shimon", "last": "Shaykevich" }, { "first": "David W", "last": "Bates" }, { "first": "John Z", "last": "Ayanian" } ] }
PMC1173098	Background ========== Hypertension is a continuous, independent, yet modifiable risk factor for cardiovascular, cerebrovascular and renal disease. It has been estimated that 62% of cerebrovascular disease and 49% of ischemic heart disease can be attributed to suboptimal blood pressure(BP) control \[[@B1]\]. Since 24 hour ambulatory blood pressure monitoring (ABPM) is currently recommended as the measurement of choice in difficult cases (uncertainty of diagnoses, fluctuating office visits, unresponsiveness to treatment, white coat affect(WCE)) \[[@B1]-[@B5]\] it is, in effect, the \'final arbiter\' or \'gold standard\' for the diagnosis of hypertension and the assessment of whether target has been achieved. The generally accepted blood pressure target on ABPM is a daytime average pressure of \< 135/85 mmHG \[[@B2]-[@B4]\]; the target blood pressure when measured in the doctors office is \< 140/90 mmHG. Increasing evidence suggests that ABPM more closely predicts target organ damage than does clinic measurement. The Ohsama study found that cardiovascular mortality, but not all cause mortality, was more closely predicted by the daytime systolic blood pressure than clinic pressures \[[@B6]\]. ABPM also better predicted cardiovascular events such as MI, CHF, stroke and TIA \[[@B4],[@B5]\] as well as other target organ damage such as ventricular hypertrophy, proteinuria, plasma creatinine, and stroke \[[@B5],[@B7]-[@B14]\]. Many patients have clinically significant WCE, making clinic measurements an unreliable indicator of true blood pressure control \[[@B11]\]. Future guidelines may suggest ABPM as routine in the diagnosis and monitoring of hypertension. This would create difficulties as this test is expensive and often difficult to obtain. Recently, automatic blood pressure measuring devices have been used either at home or in the clinic setting \[[@B16]\]. It has been found that the measurements taken at home, or while not in the presence of clinic staff, have a better correlation with daytime ABPM than do nurse or physician measurements \[[@B15]-[@B17]\]. Home or clinic self measurement is also preferred by patients over 24 hour monitoring \[[@B18]\]. The purpose of this study is to determine the possible clinical utility of the BpTRU automated blood pressure monitor in the diagnosis and monitoring of hypertension in the primary care clinic setting. Specifically, we set out i) to determine how BpTRU measurements related to ABPM measurements and ii) to determine the level of BpTRU measurement that best predicted a mean daytime blood pressure of \<135 mmHG systolic and \< 85 mmHG diastolic. The BpTRU monitor has been developed by VSM MedTech Ltd of Vancouver Canada specifically for the clinician\'s office. Methods ======= The data used in this study is a subset of the data collected for two ongoing RCTs being conducted at the Centre for Studies in Primary Care at Queen\'s University, Kingston, Ontario, Canada -- the Home Monitoring of Blood Pressure Study (ISRCTN25105161) and the Intensive Scheduled Management of Hypertension Study (ISRCTN05874865) which are funded by the Heart and Stroke Foundation of Ontario. A group of 481 subjects for which all relevant data were available were used for analyses of correlations. Subjects were recruited from 51 family practices in eastern Ontario. All patients with a diagnosis of hypertension who were being treated with antihypertensive medications were identified in each practice. Each patient\'s chart was reviewed to abstract the blood pressures recorded on the patient\'s chart at the last three office visits where blood pressure was measured. Only one recording was used from any single office visit. If there was more than one recording at a given visit, the last measurement recorded for that visit was used. These visits ranged from several weeks to several months apart, depending on the practice of the physician regarding follow-up of hypertensive patients. If the mean of these three readings, taken at different visits, was ≥140/90, the patient was labeled as \'uncontrolled\' by office measurement, meaning that the patient had not achieved the treatment target (≥130/80 was used for diabetics). The research nurse then contacted these uncontrolled patients and invited them to participate in the study. Subjects were excluded if they were \<18 years of age, pregnant, or had a known secondary cause for their hypertension. BpTRU measurements ------------------ The BpTRU device uses the oscillometric technique used by most ambulatory and home blood pressure measuring devices \[[@B19],[@B20]\]. It is designed to take an initial reading while the clinician is present, and then with the patient alone in the room, proceeds to take 5 more measurements at intervals of 1--5 minutes and then provides an average of these five readings. The specific steps we used were: i) the subjects were seated for at least 5 minutes, ii) the BpTRU cuff was applied to the non-dominant arm by the research nurse, iii) the initial BpTRU blood pressure reading was taken and recorded, iv) the staff then left the room while the BpTRU device took a minimum further 5 readings at intervals of either one minute or two minutes, v) these five readings were averaged by the device and this average was recorded. The BpTRU device has passed the standards of the British Hypertension Society and the Association of Advancement of Medical Instrumentation \[[@B20],[@B21]\]. Ambulatory monitoring --------------------- All patients then had ABPM monitoring using oscillometric A&D Model TM2430 equipment (A&D Medical, Milpitas, California, USA). The ABPM equipment was applied at the same visit as the other research measurements. The cuff was fixed to the non-dominant arm and the device was set to obtain automatic readings every 15 minutes during the day (0600--2200) and every 30 minutes at night (2200--0600). This monitoring took place on working days and subjects were instructed to behave and work as usual. The A&D Model TM2430 has been clinically validated according to the British Hypertension Society protocols \[[@B22]\]. Results ======= Of the 481 subjects, 210 (43.6 %) were male and 271 (56.3 %) female. The average age was 64.9 (SD 11.59) with a range of 33 to 92 and average BMI was 30.6 (SD 5.22). All but 2 subjects were on antihypertensive medications. In essence, this was a population of treated known hypertensive patients. Comparison of group mean blood pressures ---------------------------------------- Table [1](#T1){ref-type="table"} shows the mean and standard deviations(SD) of systolic and diastolic BP as measured in the family physicians\' offices (the average of the blood pressure measurements recorded on the patient chart at the last three visits), the initial BpTRU reading, the average BpTRU reading, and the 24 hour daytime average. There was no statistical difference in group mean between clinic measurement and initial BpTRU readings, and no statistical difference between average BpTRU readings and ABPM daytime average. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Mean systolic and diastolic BPs ::: ------------------------------------------------------------------------- ----------------------------------- Average of the blood pressures measured at the last three office visits 150.8 (SD10.26) / 82.9 (SD 8.44) BpTRU initial reading 150.0 (SD21.33) / 83.3 (SD12.00) BpTRU Average 140.0 (SD17.71) / 79.8 (SD 10.46) 24 hour daytime average 141.5 (SD 13.25) / 79.7 (SD 7.79) ------------------------------------------------------------------------- ----------------------------------- ::: Comparison of BP target achievement ----------------------------------- A large proportion of those subjects who had uncontrolled hypertension according to the mean of the three last clinic BPs had lower blood pressure on either or both of BpTRU or 24 hour monitor. In 393 (81.7%) of the 481 subjects the systolic blood pressure was less with the BpTRU than with the clinic readings and in 312 (64.8%) of the 481 subjects the systolic blood pressure was less with daytime ABPM than with the clinic readings. Diastolic pressure similarly decreased in 332 (69.0%) using the BpTRU average and 292 (60.7%) with daytime ABPM. In all, 250 of the 470 (53.2%) were found by BpTRU monitor to have achieved the accepted clinic measurement target of \<140/90 mmHG. ABPM revealed that 162 subjects (33.4%) were actually normotensive with a daytime average of \<135/85 mmHG. Correlations ------------ Pearson correlations (Table [2](#T2){ref-type="table"}) were calculated comparing the mean daytime systolic and diastolic ABPM results with the average of the blood pressure measurements recorded on the patient\'s office chart, the first BpTRU measurement taken in the presence of the staff, and with the average of five readings on the BpTRU device. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Pearson Correlation coefficients ::: Mean 24 hr daytime [systolic]{.underline} Mean 24 hr daytime [diastolic]{.underline} ------------------------------------------------------------------------- ------------------------------------------- -------------------------------------------- Mean of the blood pressure measurements at the last three office visits r = 0.145 r = 0.316 1st BpTRU measurement r = 0.473 r = 0.554 Average BpTRU(out of 5) r = 0.571 r = 0.610 ::: Using BpTRU to predict achievement of targets --------------------------------------------- ABPM is accepted as the most reliable way to determine if blood pressure targets are achieved in hypertensive patients, but using ABPM after every medication adjustment is impractical and costly. To determine which level of BpTRU blood pressure best predicted achievement of target (i.e. \<135/85) on ABPM, various prediction estimates (sensitivity, specificity, and predictive values) were calculated for different levels of BpTRU readings (Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Two-By-Two Tables And BpTRU Predictive Characteristics For Achievement Of ABPM Systolic Target ::: Systolic Target Achieved on ABPM (\<135 mmHG mean daytime blood pressure) ---------------------- ----- --------------------------------------------------------------------------- ----- --------------------------------------------- Yes NO \< 140 mmHG on BpTRU Yes 131 137 Sens = 80%; Spec = 55% PPV = 49%; NPV = 84% No 32 170 \< 135 mmHG on BpTRU Yes 111 101 Sens = 68%; Spec = 67% PPV = 52%; NPV = 80% No 52 206 \< 130 mmHG on BpTRU Yes 86 59 Sens = 53%; Spec = 81% PPV = 59%; NPV = 76% No 77 248 \< 125 mmHG on BpTRU Yes 62 31 Sens = 39%; Spec = 90% PPV = 67%; NPV = 73% No 101 276 \< 120 mmHG on BpTRU Yes 42 12 Sens = 26%; Spec = 96% PPV = 78%; NPV = 71% No 121 295 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Two-by-Two Tables and BpTRU predictive characteristics for achievement of ABPM diastolic target ::: Diastolic Target Achieved on ABPM (\<85 mmHG mean daytime blood pressure) --------------------- ----- --------------------------------------------------------------------------- ----- --------------------------------------------- Yes NO \< 90 mmHG on BpTRU Yes 341 58 Sens = 92%; Spec = 43% PPV = 86%; NPV = 61% No 28 43 \< 85 mmHG on BpTRU Yes 299 36 Sens = 81%; Spec = 64% PPV = 89%; NPV = 48% No 70 65 \< 80 mmHG on BpTRU Yes 238 13 Sens = 65%; Spec = 87% PPV = 95%; NPV = 40% No 131 88 \< 75 mmHG on BpTRU Yes 161 5 Sens = 44%; Spec = 95% PPV = 97%; NPV = 32% No 208 96 \< 70 mmHG on BpTRU Yes 82 1 Sens = 22%; Spec = 99% PPV = 99%; NPV = 26% No 287 100 ::: Discussion ========== The measurements of blood pressure done with a provider present, that is the mean of the physician office measurements, and the observed first BpTRU measurements were not significantly different from each other and probably are both affected by the white coat phenomenon. This also suggests that the mean of the last three office blood pressure measurements is a good proxy for the patient\'s current blood pressure, taken with a health care professional present (an observed blood pressure measurement). Both of these measures, that is, the mean of the last three office blood pressures and the first, observed reading from the BpTRU, were significantly higher than multiple measurements taken when a provider is not present, that is the mean of the five BpTRU measurements and the daytime mean on ABPM. The mean of the BpTRU and the daytime mean on ABPM were not significantly different. Recommended BpTRU Thresholds ---------------------------- Our data suggest that the degree of agreement between the BpTRU and 24 hour ABPM is not sufficient for clinicians to use BpTRU alone to determine if BP targets have been achieved. However, since the agreement between the BpTRU and ABPM is so much better than the usual sphygmomanometer-based, observed, clinic measures, it seems reasonable to use the BpTRU to make treatment adjustment decisions to a predetermined BpTRU level and then confirm it with a 24 hour ABPM. If one considers Table [3](#T3){ref-type="table"}, it would seem that a systolic BP level of \<135 mmHG provides the best overall agreement with the 24 hour ABPM. Similarly, for the diastolic two-by-two tables in Table [4](#T4){ref-type="table"}, one would likely choose \<85 mmHG diastolic as the most appropriate BpTRU level for diastolic pressure. Clinicians should consider treating patients to a level of \<135/85 mmHG using the BpTRU or similar automated multiple reading device, and at that point conduct a 24 hour ABPM to confirm that a daytime mean pressure of \<135/85 mmHG has been achieved. If it has not been achieved, the clinician should further treat to a BpTRU level of \<130/80 mmHG or \<125/75 mmHG before re-assessing with 24 hour ABPM. Limitations ----------- Our data are the best currently available to provide some sense of what a BP target level should be when using the BpTRU in office practice. However there are some limitations. The BpTRU and ABPM were carried out in a research setting not a true clinical setting. All the patients had uncontrolled blood pressure according to office BP measurements, so it does not represent a full spectrum; it does not include normotensive patients or patients where the chart records suggest target has been achieved. The data was collected in the course of another study. There is a need for a study that sets out, a priori, to compared BpTRU with ABPM in the full spectrum of patients, and which follows those patients for a longer term to assess clinical outcomes. Conclusion ========== The control of hypertension is vital to decrease cardiovascular morbidity and mortality, however the current recommended BP targets are often not being met. There is also growing evidence that office measured sphygmomanometer-based blood pressures are unreliable and do not predict outcomes as well as ABPM, whereas there is increasing evidence from prospective trials that 24 hour monitoring has prognostic significance. We have shown that the BpTRU has potential to be used in the clinic setting to help overcome the difficulties caused by the WCE without the cost of having to conduct frequent 24 hour ABPMs. Although this study demonstrates that the BpTRU has a sensitivity and specificity that are not ideal when compared to the ABPM device we used, it is superior to usual office measurement and can be used by clinicians as part of their strategy for determining whether BP target has been achieved. We suggest that the BpTRU be used to adjust treatment until patient\'s BpTRU pressure is below 135/85 mmHg and then a 24 hour ABPM be conducted to confirm. ABPM measurement is not practical as a means of monitoring targets after each medication adjustment but in conjunction with the BpTRU can form the basis for clinical decisions that will promote more effective control of hypertension. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= LB abstracted the data from the larger baseline dataset from the randomized controlled trials. She analysed the data, did the initial interpretation and wrote the initial draft of the article. MG designed and is conducting the randomized controlled trials from which the data for this study was drawn. He worked with LB on the analysis and the data interpretation and helped with the substantial re-writing of the original draft of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/18/prepub>	PubMed Central	2024-06-05T03:55:58.361194	2005-6-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173098/", "journal": "BMC Cardiovasc Disord. 2005 Jun 28; 5:18", "authors": [ { "first": "Linda", "last": "Beckett" }, { "first": "Marshall", "last": "Godwin" } ] }
PMC1173099	Background ========== The quality of health care is determined by two main factors: the quality of the judgements and decisions that determine what actions are to be taken, and the quality with which those actions are executed \[[@B1]\]. Within health care, there are wide variations in clinicians\' judgements on diagnosis and in the management of patients with the same symptoms and diagnosis \[[@B2]\]. These variations are seen within different disciplines, among both experts and novices. In spite of this evidence, clinicians generally believe that the decisions made by their colleagues would be similar to their own, and hence they assume that there is a broad consensus in medical practice \[[@B2]\]. Patients with symptoms from the upper gastrointestinal tract are regularly seen in clinical practice. The symptoms may be either of organic origin -- e.g. ulcer disease, oesophagitis, and malignancies in the oesophagus and the ventricle -- or of functional origin \[[@B3]\]. Dyspepsia is a collective term and includes conditions in both categories. Dyspepsia is common, and the subjective symptoms in either category varybetween patients. During a 3-month period, about 30% of the adult population suffers from dyspepsia \[[@B4]\]. Among these sufferers, only a minority has ulcer diseases (10%) \[[@B5]\] or reflux (12%) \[[@B6]\]. According to the literature, certain management strategies are recommended for some of these conditions, for example reflux \[[@B7]\]. For other conditions, for example functional dyspepsia, the evidence is more ambiguous \[[@B8]-[@B11]\]. One may expect a wider variation in the latter than in the former treatment strategies. The aim of this study was to describe, using a qualitative approach, the variation in the management strategies and treatments suggested by clinicians in three different disciplines for patients with symptoms from the upper gastrointestinal tract. According to this aim, the hypotheses that guided our design were: 1\. Management strategies and treatments suggested by different clinicians vary considerably, even when optimal treatment is clear cut, as documented by evidence in the literature. 2\. Clinicians believe that the management strategies of their colleagues will be similar to their own. Methods ======= Informants ---------- Informants were selected to represent clinicians who regularly encountered patients with dyspeptic symptoms in their daily practice and who thus were expected to have a treatment policy for these patients. We therefore invited Swedish clinicians who were specialists in medical gastroenterology, surgery, and general practice to participate in this study. All clinicians gave their consent to be interviewed. They also suggested other clinicians who might be interested in participating in the study. Altogether, 27 clinicians participated, nine from each of the three specialities. This number is the number commonly used in studies that use judgement analysis to ensure a variation in answers \[[@B12]\]. The specialists in medical gastroenterology (four women and five men) all worked at Lund University Hospital at the time of the study. The surgeons (one woman and eight men) worked at Lund University Hospital or Helsingborg Hospital. The general practitioners (two women and seven men) worked in the Public Health Service in the southern part of Sweden. The clinicians were 35--56 years (average 47 years) of age and their experience in their speciality ranged from 1 to 22 years (average 10 years). Interviews and case scenarios ----------------------------- A useful approach for studying clinicians\' strategies is judgement analysis \[[@B14]\], which describes the cognitive process involved in making a decision. It focuses on the actual decision made by the clinician and on which information, for example symptoms and clinical findings, that the clinician uses to reach that decision. In our study, each of the clinicians was interviewed separately. The interviews were based on four case histories. The histories described frequent symptoms and findings in 30-year-old men seen in general practice and in specialist out-patient clinics (Table [1](#T1){ref-type="table"}). Besides age and sex, all histories contained information on the localisation of symptoms, but information on subjective symptoms and findings from the examination varied from history to history. The questions, listed in Table [1](#T1){ref-type="table"}, focused on how the clinician viewed whether there was a need for treatment, what management the clinician would suggest, which factors the clinician thought were most important to consider, and what decisions the clinician thought colleagues in their own speciality or in other specialities would make. The same questions were asked for each patient. The case histories were presented in the sequence described in Table [1](#T1){ref-type="table"}. Thus, the questions were answered four times by each informant. One of the authors (KK) with experience in judgement and decision analysis interviewed each informant at her or his office. Each interview began with the interviewer reading the first history. The interviewer then asked the questions aloud and wrote down the informant\'s answers after each question. If the informant wished, the history and questions were reread. At the end of the question-and-answer session after each case history, the interviewer read aloud the participant\'s answers and in cases of ambiguity, adjusted the answers. The interviews thus took the form of an oral questionnaire, but with the possibility to explain the answers more fully. Content analysis ---------------- The analysis was performed after all data had been collected from the informants. Two of the authors (KK, MT) with experience in qualitative studies read the interview notes. In a first step, each of the clinicians\' answers was scrutinised to find content or evaluations in common, which were then coded as concepts. The following is an example of answers coded as a concept: To question A (\"Do you think that it is a problem to decide whether there is a need for treatment in a case like this?\" and if yes: \"What problems?\") informants gave answers like \"there is no demonstrable disease, but the patient has symptoms\" and \"there is no diagnosis but a suffering patient\" (patients 1 and 2). These answers were coded as the concept \"The patient suffers, but there is no distinct diagnosis\". In the second step, related concepts were combined into categories, common to all the case histories. The authors compared their coding and resolved any differences through discussion to reach a consensus. The concepts and categories were compared with the original answers to ensure that the answers were covered by the classifications. Results ======= The varied answers of the clinicians are presented in Table [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}. Different categories of problems existed that could explain the variation in the decision-making process. When answering the question \"Do you think there is a problem to decide whether there is a need for treatment in a case like this?\", five different categories of answers were given. For example, the basis for deciding whether or not to treat and what would be optimal treatment was ambiguous and, further, the evidence for treatment was poor. Clinicians also felt that they lacked competence in evaluating test results when making treatment decisions and that patient expectations and requirements were problems with patients 1--3. In these cases, the patients sometimes requested investigations and treatment, even though the clinician informed them that there was no reported evidence of effect in the literature for the requested treatment or investigation. For the same patient, the management strategies considered by the different clinicians varied widely, including extensive examinations, non-pharmacological treatment, drug treatment, and surgery (Table [3](#T3){ref-type="table"}). For patients 2 and 3, who had tested positive for Helicobacter pylori (H. pylori), some clinicians stated that they would suggest eradication treatment only if the patient had an ulcer, whereas other clinicians said they would suggest this treatment irrespective of the presence of ulcers. For patient 4, who had a duodenal ulcer, some clinicians suggested eradication treatment only if the patient tested positive for H. pylori, whereas others recommended that treatment irrespective of such test results. This variation was also observed among clinicians in the same speciality. Variation was more pronounced if the case history noted symptoms but no organic findings, as in patients 1 and 2, than if the case history noted unambiguous findings and symptoms, as in patients 3 and 4. When asked \"Which factors are most important to consider in your decision?\" (Table [4](#T4){ref-type="table"}), the clinicians defined three categories of information. For patients 1 and 2, whose symptoms were diffuse, several clinicians answered that the medical history was the most important piece of information to consider. They also claimed to need a more comprehensive medical history for patients 1 and 2 than for patients 3 and 4, whose symptoms were more obviously related to an organic diagnosis. A majority of the clinicians believed that their treatment decisions would be similar to those of most of their colleagues in their speciality. Some gastroenterologists believed that surgeons would prefer surgery to medication, given the same test results. Several gastroenterologists and surgeons believed that family clinicians investigated less, informed more, and prescribed acid inhibitors on wider indications than gastroenterologists and surgeons. Discussion ========== Methodological and results approaches ------------------------------------- The four case histories in this study are representative of patients with conditions frequently seen in medical practice. Diseases of the gastrointestinal tract account for about 6% of all reported consultations in Sweden \[[@B13]\]. Gastritis, dyspepsia, and unspecified diseases in the ventricle and duodenum (patients 1 and 2) are diagnosed in 1.6% of the population; oesophagitis and reflux (patient 3) in 0.9%; and different ulcers (patient 4) in 0.6%. The clinicians who participated in this study were not representative of a randomised selection of clinicians in Sweden but rather of a select group with a special interest in patients with symptoms from the gastrointestinal tract. Since their experience with these patients was extensive, they should have been able to develop a treatment policy. One would expect a variation in treatment strategy in this group to be more limited than among a randomised selection of clinicians. The results underpin existing evidence that variations in medical practice exist \[[@B14]-[@B16]\]. This study presents variations both in treatment strategy and in what information is considered important for making treatment decisions. The histories of patients with diffuse symptoms but no objective findings (patients 1 and 2) gave rise to more extensive variations than did the case histories of patients 3 and 4 where the symptoms were more obviously related to a diagnosis (patient 3) or where more obvious organic changes existed (patient 4). Uncertainty in diagnosis could lead to uncertainty regarding which outcome of alternative interventions is optimal \[[@B17],[@B18]\]. A meta-analysis that evaluates the most effective treatment in patients with functional dyspepsia \[[@B10]\] recommended the eradication of H. pylori if the treatment is to be effective from the patient\'s perspective, whereas other randomised, double-blind, controlled studies find that eradication has no beneficial effect \[[@B11]\]. These results were underpinned by the variations observed in this study. Uncertainty due to lack of knowledge or professional competence may lead to qualitative differences. In this study, different management strategies were recommended for patients with reflux disease (patient 3), for example, eradication treatment as one strategy and changes in life-style as another. However, evidence in the literature presenting eradication as an optimal management of reflux is lacking \[[@B17]\]. Instead, this management strategy raises the cost to society and to the patient and causes unnecessary antibiotic pressure. Randomised, double-blinded, prospective studies conclude that the only indication for treating dyspepsia with proton pump inhibitors (PPIs) is the presence of an ulcer or reflux \[[@B7]\]. However, the majority of clinicians prescribed PPIs and H~2~-receptor antagonists for the treatment of dyspepsia, although no evidence of an effect with this treatment compared to placebo has been documented \[[@B9],[@B10]\]. Thus, this regimen has led to high costs for society without any benefits. The drug costs for treatment of ventricular and duodenal ulcers and of reflux, including eradication treatment, were SEK 1.6 billion \[USD 210 million\] in 1998 in Sweden \[[@B13]\], while The Swedish Council on Technology Assessment in Health Care reports that society\'s direct and indirect costs for dyspepsia were between SEK 3.7 and 4.4 billion \[USD 490 and 590 million\] in 2000 \[[@B3]\]. From the informants\' answers to the question \"Which factors are most important to consider in your decision?\", the completeness of some of the answers given in the interviews could be questioned. Most likely, the clinicians failed to answer the question in full as they left out many factors worthy of consideration. Other factors in the patient\'s life such as stress, mental mood, working situation, and diet were not mentioned by the clinicians in the interviews but should often be considered in treatment strategies in an actual clinical situation. Within each of the three specialities, clinicians believed that their colleagues would treat patients in the same way as they themselves did. This aspect of \"professional certainty\" implies that clinicians believe that their practice is correct, irrespective of how much it in fact differs from that of other clinicians \[[@B18],[@B19]\]. This unawareness of variation in the management of frequently seen patients may indicate a lack of communication and discussion about everyday cases. Such discussions are perhaps reserved for more \"complex\" and rare cases. Opportunities to change clinicians\' practice --------------------------------------------- Among the causes of variation in medical practice, the influence of factors like patient characteristics (e.g. age, sex, morbidity, and personal preferences) could be more or less regarded as legitimate to explain variations in practice \[[@B20]\]. Other factors, like resource capacity, could be influenced by, for example, budget restrictions while management policy and practice style are more resistant to change \[[@B20]\]. These latter factors should therefore be the targets of efforts to change. However, many attempts to implement evidence by information alone, for example in the form of clinical practice guidelines, have failed to change management strategies \[[@B20]-[@B23]\]. Instead, a combination of methods is most likely needed if a permanent change is to occur \[[@B23]\]. Furthermore, it is of utmost importance that potential barriers are identified and that the clinicians who will be affected support the clinical practice guidelines to be implemented. In recent years, the individual autonomy of patients and letting patients\' preferences influence the choice of intervention have been emphasised \[[@B24]\]. In this study, such a strategy could explain why some clinicians mentioned that they felt pressured by patients to perform an intervention, even when the optimal strategy was non-intervention. Conclusion ========== In conclusion, our study adds to the present scientific literature by showing that, even in cases with a consensus in the scientific literature on treatment, clinicians can differ in their opinion of which management is optimal. Despite these variations, the clinicians believed that the decisions made by their colleagues would be similar to their own. Overall, we as researchers must make scientific evidence comprehensible and communicate evidence in such a way that clinicians are easily able to interpret and implement it in practice. Of particular significance is that scientific evidence leads to an evidence-based care, which is effective clinical practice, and to the promotion of health from the perspective of the patient, together with cost-effectiveness as a priority. Cost-effectiveness is a vital concern. Targeting high patient satisfaction may lead to ineffectiveness if the economic consequences of a treatment strategy are ignored. Declaration of competing interests ================================== The authors declare that they have no competing interests. Authors\' contributions ======================= Kerstin Knutsson, DDS, Odont dr, has special competence in judgement and decision-making within health care. KK contributed to the overall concept and design of the study and conducted the interviews. Grants from the Swedish Research Council (grant no. 521-2001-6341) supported this study. Bodil Ohlsson, MD, PhD, is a specialist in gastroenterology and contributed with her knowledge in this area to the design of the study and the interpretation of the data. Margareta Troein, MD, PhD, has special competence in qualitative methods in medical care and contributed to the qualitative analysis and interpretation of the data. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-230X/5/15/prepub> Acknowledgements ================ We thank Professor Åke Nilsson, Lund University Hospital, Lund University, Lund and Professor Stefan Lindgren, Malmö University Hospital, Lund University, Lund for their valuable opinions in the initial phase of this study and on the manuscript. Figures and Tables ================== ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Case histories and questions presented to the informants. ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Patient 1. A 30-year-old man with symptoms from the upper part of the abdomen. The patient has had intermittent pain, made worse by stress, for a few years. One year ago an ultrasound examination and an oesophagogastroduodenoscopy were performed. No abnormal findings were registered. H. pylori serology was negative. Patient 2. A 30-year-old man with symptoms from the upper part of the abdomen. The patient has had intermittent pain, made worse by stress, for a few years. One year ago an ultrasound examination and an oesophagogastroduodenoscopy were performed. No abnormal findings were registered. H. pylori serology was positive. Patient 3. A 30-year-old overweight man with symptoms from the upper part of the abdomen and retrosternally. The pain worsens when he leans forward and when he rests in a prone position. He sometimes experiences a sour or bitter taste in his mouth. H. pylori serology was positive. Patient 4. A 30-year-old man with symptoms from the upper part of the abdomen. The ultrasound examination revealed nothing abnormal, but the oesophagogastroduodenoscopy indicated a duodenal ulcer. After each case the following questions were asked:\ A. Do you think that it is a problem to decide whether there is a need for treatment in a case like this? If yes: What problems?\ B. How would you manage a case like this?\ C. Which factors do you think are most important to consider in your decision?\ D. What do you believe your colleagues would decide in a case like this? ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Clinicians\' answers coded as concepts and organised into categories. Answers of 27 clinicians to the question \"Do you think that it is a problem to decide whether there is a need for treatment in a case like this?\" If the physicians admitted that it was a problem, they were then asked \"What problems?\" Answers were coded as concepts and organised into categories for each patient. Numbers of answers are presented within parentheses. Each clinician may give several answers that could be included in different categories. ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Categories Concepts ----------------------------------------------------- ------------------------------------------------------------- ---------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------- --------------------------------------------------------- There is no problem to decide (16) (20) (19) (22) Basis for treatment decision is insufficient There is no diagnosis (5) There is no diagnosis (3) The amount of discomfort is not clear (1)\ No examination of Helicobacter pylori has been made (3) More examinations are needed (2) Evidence for treatment is poor There is hardly any evidence for benefits of treatment (1)\ Patient suffers, but there is no distinct diagnosis (1) Recurrent condition (2) It is difficult to treat dyspepsia (1)\ Patient suffers, but there is no distinct diagnosis (2)\ Psychosomatic problems (2) Clinician is uncertain of her or his own competence Uncertainty about indication for treatment when Helicobacter pylori serology is positive (1) Uncertainty about indication for treatment when Helicobacter pylori serology is positive (1) Patient\'s expectations Patient wishes treatment (1) Patient requires treatment due to positive Helicobacter pylori serology (2) Patient is not motivated to reduce weight (1)\ Patient requires treatment due to positive Helicobacter pylori serology (2) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Clinicians\' answers coded as concepts and organised into categories. Answers of 27 clinicians to the question \"How would you manage a case like this? \" Answers were coded as concepts and organised into categories for each patient. Numbers of answers are presented within parentheses. Each clinician may give several answers that could be included in different categories. ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Categories Concepts ---------------------------------------------------------------------------------- ------------------------------------------------------------------ ----------------------------------------------------------------------------- ---------------------------------------------------------------------------- ---------------------------------------------------------------------------- Extend the examination Exclude coeliac disease, examine colon, liver, gall bladder (6)\ Oesophagogastroduodenoscopy, expiration tests for Helicobacter pylori (10)\ Oesophagogastroduodenoscopy, expiration tests for Helicobacter pylori (18) Diagnostic tests for Helicobacter pylori (8) Evaluate patient\'s psychosocial conditions (4)\ Evaluate patient\'s psychosocial condition (2)\ Evaluate outcome of previous treatment (3) Evaluate outcome of previous treatment (1) Prescribe non-pharmacological treatment Reassuring information (8)\ Reassuring information (7)\ Smoking cessation (2)\ Unspecific life-style changes (1) Expectancy (3)\ Expectancy (4)\ Weight reduction (3)\ Dietary advice, reduction of alcohol intake (11)\ Dietary advice (6)\ Dietary advice (3)\ Reduction of stress (7) Reduction of stress (3) Unspecific life-style changes (6) Prescribe drugs against acidity Antacids (7)\ Antacids (7)\ Antacids (1)\ Proton pump inhibitors (11) H2-receptor antagonists (13)\ H2-receptor antagonists (9)\ H2-receptor antagonists (8)\ Proton pump inhibitors (5)\ Proton pump inhibitors (6)\ Proton pump inhibitors (15) Sucralphate (1) Sucralphate (1) Prescribe triple treatment (antacid and antibiotics against Helicobacter pylori) Only if patient has an ulcer (7)\ Only if patient has oesophagitis or an ulcer (1)\ Only if patient has a positive test for Helicobacter pylori (11)\ Yes, irrespective of whether patient has an ulcer (4) Yes, irrespective of whether patient has oesophagitis or an ulcer (1) Yes, irrespective of whether Helicobacter pylori serology is positive (14) Prescribe other drugs Drugs which increase gut motility (2)\ Drugs which increase gut motility (1) Alginate (6) NSAID (1)\ Bulking agent (1) Recommend surgery If patient has hiatus hernia (1) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Clinicians\' answers coded as concepts and organised into categories. Answers of 27 clinicians to the question \"Which factors are most important to consider in your decision?\" Answers were coded as concepts and organised into categories for each patient. Numbers of answers are presented within parentheses. Each clinician may give several answers that could be included in different categories. ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Categories Concepts --------------------------- ------------------------------------------------- ----------------------------------------------------------------------------- ----------------------------------------------------------------------------- ----------------------------------- Medical history Patient\'s age (4)\ Patient\'s age (5)\ Patient\'s age (2)\ General state of health (1)\ Symptoms including localisation of pain (12)\ Symptoms including relation to meals (12)\ Symptoms generally and in relation to position (21) Symptoms (1) Life-style/stress (12)\ Effects of previous medication (1)\ Effects of previous medication (1)\ Similar problems for a long time (2) Similar problems for a long time (3) Results from examinations Oesophagogastroduodenoscopy and ultrasound (6)\ Oesophagogastroduodenoscopy and ultrasound (1)\ Oesophagogastroduodenoscopy (3)\ Oesophagogastroduodenoscopy (26)\ Test for Helicobacter pylori (5) Test for Helicobacter pylori (7) Test for Helicobacter pylori (1)\ Test for Helicobacter pylori (3) Weight (1) Patient\'s expectations Patient\'s preferences (1) Patient requires treatment due to positive Helicobacter pylori serology (3) Patient requires treatment due to positive Helicobacter pylori serology (3) ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- :::	PubMed Central	2024-06-05T03:55:58.363499	2005-5-15	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173099/", "journal": "BMC Gastroenterol. 2005 May 15; 5:15", "authors": [ { "first": "Kerstin", "last": "Knutsson" }, { "first": "Bodil", "last": "Ohlsson" }, { "first": "Margareta", "last": "Troein" } ] }
PMC1173100	Background ========== With the development of molecular techniques to identify genetic polymorphisms, numerous markers and in particular Single Nucleotide Polymorphisms (SNPs) \[[@B1],[@B2]\] are now available within and between genes for establishing their possible role in disease susceptibility. One recurrent question in the literature is the way these markers could be used to test for association with complex diseases \[[@B3],[@B4]\]. Should one focus on one marker at a time and simply compare allele or genotype distributions for the studied markers in cases and controls or should one consider haplotypes formed by several linked markers? Haplotypic methods have been suggested to be more powerful at detecting the role of a given genomic region in disease susceptibility \[[@B4]-[@B7]\]. Indeed, disease susceptibility may be due to the combined effects of variants at different markers. Different methods have been proposed to test for association between markers located on the same reconstructed haplotypes and disease susceptibility \[[@B4],[@B7]-[@B10]\]. In these tests, the number of haplotypes being large, the degrees of freedom to compare cases and controls are also large, decreasing the power to detect the association. Moreover, as some haplotypes would only be carried by one or two individuals, there could be statistical problems owing to small sample sizes making difficult the evaluation of their possible effect on the susceptibility to the disease. A solution to reduce these problems is to group haplotypes in order to decrease the degrees of freedom and to increase the number of individuals in the different haplotype groups. The cladistic method as described by Templeton et al. \[[@B11]\] is a method to carry out such groupings of haplotypes and to determine which haplotype or group of haplotypes is likely to be responsible for the phenotypic variation observed among the population. It consists in using parsimony methods to build a phylogeny of the haplotypes, and to group haplotypes according to the clades defined by the phylogenetic tree. When applied to a case/control data set, the proportions of cases and controls in different clades are compared. If a clade shows a significantly larger number of cases than the others, one will conclude, first, that there is an association between the disease and one haplotype or a group of haplotypes belonging to the clade, and, second, that the mutations defining the clade containing an excess of cases are good candidates to be functional gene polymorphisms involved in the disease susceptibility. The cladistic analysis relies for a large part on the reconstruction of an accurate phylogenetic tree, but recombination is known to bias phylogenetic reconstruction processes \[[@B12]\]. The recent discovery of the haplotype block structure of the human genome \[[@B13]\] now provides an interesting framework for cladistic methods. Indeed, haplotype blocks define regions in which the recombination rate is quite low. Moreover, once a block is found to be associated with a disease, it will not be suitable to use classical methods based on recombination to fine map susceptibility loci. Haplotype phylogenies being based on the history of mutations may then help to identify disease susceptibility loci. The cladistic method was first described in a series of five articles \[[@B11],[@B14]-[@B17]\] and applied in different case-control studies focusing either on quantitative or qualitative data. For example, Haviland et al. used a cladistic analysis to study the association between the apolipoprotein AI-CIII-AIV and plasma lipid, lipoprotein and apolipoprotein levels \[[@B18]\]. Keavney et al. \[[@B19]\] used cladograms to localize the sites that significantly influence Angiotensin-I Converting Enzyme (ACE). Zhu et al. \[[@B20]\] worked on quantitative data simulated for the twelfth Genetic Analysis Workshop (GAW). Using regression analysis and analysis of variance (ANOVA), they were able to find a short region containing the simulated mutation. Wang et al. \[[@B21]\] used a nested linear model to analyze the association between ApoA4haplotypes and plasma lipid level in the cladistic framework. Among the studies that consider discrete data is the one by Kittles et al. \[[@B22]\] who tested the association between haplotype variations on the Y chromosome and alcohol dependence and related personality traits. Lobos and Todd applied the same cladistic method to four RFLPs in the tyroxine hydroxylase gene \[[@B23]\], and to five SNPs in the dopamine receptor D2 gene \[[@B24]\]. Darlu and Génin \[[@B25]\] applied the method to the GAW 12 simulated data and were able to identify some of the disease susceptibility sites. For a more detailed review on the subject, see \[[@B26]\]. More recently Seltman et al. \[[@B27]\] developed a method that can be applied on family-based data. They proposed a test, the ET-TDT, that combines the transmission disequilibrium test (TDT) and the use of unrooted phylogenetic trees to group haplotypes. They further extended their method to the analysis of qualitative and quantitative data from case-parent trios or population-based studies \[[@B28]\]. Finally, an alternative method where phylogenetic trees are reconstructed by a distance method was proposed by Durrant et al. \[[@B29]\] and implemented in the software CLADHC. From these different examples, cladistic methods appear to be innovative methods to detect an association between a candidate gene and a disease and to localize disease susceptibility sites. However, the validity and the power of these methods to detect association have only been tested in a very limited number of situations (fixed pattern of linkage disequilibrium, fixed phylogenetic tree) \[[@B27],[@B29]\]. In this paper, we develop a new cladistic test (CT), based on the analysis of a rooted tree, which allows to make hypotheses about functional polymorphisms that are involved in the susceptibility to the disease. Through simulations on ten different tree topologies, we evaluate the conditions in which this cladistic method is efficient and compare it to other methods. Finally, we apply this method to a real data set concerning Crohn disease and the CARD15 gene \[[@B30]\]. Results ======= Association test ---------------- As detailed in the method section, four different tests are performed: a single locus allele-based test referred to as SbST (Site by Site Test), an haplotype-based test referred to as HT (Haplotypic Test), a test based on haplotype phylogenies referred to as CT (Cladistic Test) and a test based on the clustering of haplotypes described by Durrant et al. \[[@B29]\] and referred to as CLADHC. To compare their respective power to detect an association, simulations were performed under three disease models involving one (model 1) or two (model 2 and model 3) susceptibility sites for ten different tree topologies. Results for model 1 are presented in Table [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}. When the susceptibility site is included in the studied SNPs (Table [1](#T1){ref-type="table"}), SbST is the most powerful test for all trees but one (tree 8) (between 99.1% and 99.5% depending on the tree, 99.35% on average), while the other tests have a weaker power (92.9%, 93.7%, and 91.2%, on average, for CT, HT, and CLADHC respectively). CT and CLADHC have almost the same average type I error, slightly but significantly lower than the type I error of HT and SbST. In these simulations, as the type I errors cannot be fixed a priori, powers may be difficult to compare. However, if the type I errors of CT and CLADHC were increased, their power would also be increased and, consequently, the difference in power with HT would be even more pronounced and the difference in power with SbST would be more reduced. In the trees showing comparable type one error (Trees 1, 5, 7), the power of SbST is clearly higher than the power of CT. When the susceptibility site is removed (Table [2](#T2){ref-type="table"}), the performances of the four tests are, on average, similar (about 89%), CT being the most powerful test for four trees, SbST for four other trees, HT and CLADHC for one tree each. However, the type I errors being slightly different, by adjusting them one expects that CT and CLADHC could be on average, slightly more powerful than HT and SbST. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Power to detect an association, one susceptibility site simulated and included in the analysis ::: Tree CT CLADHC HT SbST ------------- --------------------------- --------------- --------------- ---------------------------- 1 96.4 (3.3) 88.9 (3.3) 91.1 (5.1) [99.3]{.underline}(3.4) 2 99.0 (4.4) 95.7 (4.1) 91.3 (6.0) [99.5]{.underline}(5.3) 3 86.0 (2.5) 91.1 (4.7) 91.2 (4.6) [99.3]{.underline}(4.4) 4 96.3 (3.8) 95.6 (2.9) 91.5 (4.2) [99.3]{.underline}(4.6) 5 88.8 (4.1) 95.3 (3.3) 90.4 (4.7) [99.1]{.underline}(4.6) 6 84.5 (3.5) 88.8 (2.7) 90.7 (4.4) [99.5]{.underline}(4.4) 7 97.4 (4.0) 96.3 (3.7) 91.9 (4.8) [99.2]{.underline}(4.3) 8 [99.7]{.underline}(3.2) 98.5 (2.4) 91.2 (4.9) 99.4 (4.7) 9 88.1 (3.5) 94.0 (4.4) 91.3 (4.7) [99.4]{.underline}(4.3) 10 92.5 (2.6) 93.0 (2.9) 91.3 (4.7) [99.5]{.underline}(3.6) Average 92.9 (3.5) 93.7 (3.4) 91.2 (4.8) [99.35]{.underline}(4.4) Std dev.^a^ 5.63 (0.62) 3.24 (0.76) 0.41 (0.49) 0.13 (0.54) Sample size: 100 cases and 100 controls, 1000 replicates, penetrance vector: \[0.03;0.06;0.3\], frequency of the susceptibility allele: 0.205 ± 0.005. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test. ^a^Standard deviation ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Power to detect an association, one susceptibility site simulated and removed from the analysis ::: Tree CT CLADHC HT SbST ------------- --------------------------- --------------------------- --------------------------- ---------------------------- 1 [94.8]{.underline}(3.8) 89.1 (3.5) 91.2 (5.6) 89.5 (3.9) 2 94.0 (3.8) 92.7 (4.1) 87.6 (5.6) [98.5]{.underline}(5.3) 3 79.3 (3.2) 88.9 (4.7) 86.2 (4.7) [94.2]{.underline}(4.1) 4 88.7 (2.3) 85.7 (3.1) 87.8 (4.6) [93.6]{.underline}(4.4) 5 85.6 (4.3) [94.4]{.underline}(4.1) 90.5 (4.5) 94.0 (4.1) 6 78.6 (3.0) 81.6 (3.3) [88.0]{.underline}(4.2) 67.0 (4.2) 7 92.4 (3.8) 90.5 (3.7) 87.8 (5.2) [92.7]{.underline}(4.0) 8 [99.7]{.underline}(3.0) 98.6 (2.8) 91.2 (4.7) 99.6 (4.5) 9 [91.9]{.underline}(3.7) 89.2 (4.5) 91.2 (4.8) 89.1 (4.1) 10 [88.8]{.underline}(2.9) 82.8 (3.0) 81.3 (4.1) 79.1 (3.5) Average 89,38 (3.4) 89.35 (3.7) 88,28 (4.8) [89,73]{.underline}(4.2) Std dev.^a^ 6.71 (0.6) 5.2 (0.6) 3.06 (0.5) 9.81 (0.5) Sample size: 100 cases and 100 controls, 1000 replicates, penetrance vector: \[0.03;0.06;0.3\], frequency of the susceptibility allele: 0.205 ± 0.005. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test. ^a^Standard deviation ::: As expected, the power is higher when the susceptibility site is included than when it is removed. In this latter case, the power depends on the degree of linkage disequilibrium (LD) between the susceptibility site and the other sites as expressed by the maximum value of LD, referred to as LD~max~. Moreover, since the pattern of LD can be different from one tree to another, one expects more differences in power between trees when the susceptibility site is removed. Indeed the standard deviation of the power is larger in this case. This standard deviation is also particularly large for SbST (σ= 9.81). This test, being an allele-based test, is more influenced by LD than the three others tests that are haplotype-based tests, as proved by the high positive correlation observed between the power of SbST to detect association and the LD~max~(Spearman rank correlation ρ equals 0.851, p-value \< 0.01). For CT and CLADHC, Spearman rank correlations are not significant (p \> 0.1) confirming that LD has a lesser impact on these two tests. Therefore, the high standard deviations observed for these two tests are probably more related to the phylogeny of haplotypes than to the linkage disequilibrium. The results for model 2 and model 3, differing by the values of the penetrance vector, are presented in Table [3](#T3){ref-type="table"} and in Table [4](#T4){ref-type="table"} respectively. There is no real difference in power between the four tests for model 2 (about 97%). For model 3, the power of the four tests is approximately half the power for model 2 owing to the reduced value of the penetrance for the genotype associated to the highest risk in this model. In this case, SbST and CT are, on average, the most and the less powerful test respectively ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Power to detect an association for two susceptibility sites simulated under model 2 ::: Model 2 ------------- --------------------------- ---------------------------- --------------------------- --------------------------- Tree CT CLADHC HT SbST 1 97.8 (3.7) [99.3]{.underline}(3.8) 97.7 (5.0) 99.1 (6.9) 2 95.1 (3.6) 97.7 (2.9) [97.9]{.underline}(5.6) 95.8 (4.3) 3 94.2 (4.9) [99.4]{.underline}(4.9) 97.9 (5.1) 98.6 (4.9) 4 97.6 (4.6) 97.8 (4.6) 97.8 (5.5) [98.6]{.underline}(4.6) 5 98.5 (5.3) [99.5]{.underline}(4.5) 98.9 (5.7) 98.2 (7.4) 6 [98.9]{.underline}(4.4) 97.9 (3.9) 98.5 (5.7) 96.9 (6.6) 7 93.0 (5.1) 93.5 (4.6) [96.9]{.underline}(5.6) 95.5 (5.2) 8 95.2 (6.1) [99.0]{.underline}(5.3) 98.2 (6.2) 95.4 (4.2) 9 96.7 (3.6) 97.3 (3.1) 97.6 (4.1) [97.7]{.underline}(4.0) 10 [99.9]{.underline}(4.5) 99.5 (4.0) 98.2 (60.) 99.8 (5.8) Average 96.7 (4.6) [98.06]{.underline}(4.2) 97.96 (5.45) 97.56 (5.4) Std dev.^a^ 2.24 (0.8) 1.83 (0.8) 0.54 (0.6) 1.58 (1.2) Sample size: 200 cases and 200 controls, 1000 replicates, penetrance: 0.9 for the homozygotes carrying the two susceptibility alleles, 0.3 for the other genotypes. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test ^a^Standard deviation ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Power to detect an association for two susceptibility sites simulated under model 3 ::: Model 3 ------------- --------------------------- --------------------------- --------------------------- --------------------------- Tree CT CLADHC HT SbST 1 40.7 (4.7) [54.0]{.underline}(4.5) 42.0 (5.9) 47.5 (4.7) 2 31.7 (3.4) 32.7 (3.9) 43.0 (5.9) [45.8]{.underline}(5.3) 3 26.8 (4.2) [47.9]{.underline}(4.4) 40.7 (6.4) 45.2 (3.5) 4 34.8 (4.8) 39.2 (4.7) 38.9 (7.0) [47.0]{.underline}(4.4) 5 41.5 (5.6) [49.7]{.underline}(4.7) 44.2 (6.6) 37.2 (4.6) 6 42.0 (3.6) 34.8 (4.7) [42.4]{.underline}(5.6) 41.3 (5.3) 7 28.6 (5.2) 32.1 (4.7) [41.6]{.underline}(6.4) 39.9 (4.3) 8 33.2 (4.3) [47.2]{.underline}(4.0) 42.9 (6.1) 37.7 (4.0) 9 33.5 (4.4) 39.1 (4.1) 41.2 (7.0) [48.5]{.underline}(5.3) 10 [65.6]{.underline}(6.6) 48.1 (4.1) 42.1 (7.8) 56.9 (4.2) Average 37.84 (4.7) 42.48 (4.4) 41.9 (6.5) [44.7]{.underline}(4.6) Std dev.^a^ 11.07 (0.9) 7.84 (0.3) 1.45 (0.7) 5.94 (0.6) Sample size: 200 cases and 200 controls, 1000 replicates, penetrance: 0.6 for the homozygotes carrying the two susceptibility alleles, 0.3 for the other genotypes. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test ^a^Standard deviation ::: Localization of the susceptibility site(s) ------------------------------------------ The efficiency of SbST, CT and CLADHC to localize the susceptibility site(s) when it is included among the SNPs is presented in Figure [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}. The efficiency of CT is plotted against the efficiency of SbST or CLADHC. To correctly interpret these figures, it has to be noted that the criteria used to compare the methods are not strictly identical. Indeed, the identified SNP is always unambiguously detected by the SbST method, since this is the site giving the most significant chi-square and two chi-squares are never exactly identical. On the other hand, the susceptibility site may not be the only one detected by CT because two or more sites could have the same values of the criteria (V~i~, see the Methods section). However, in our simulations, the susceptibility site is the only one detected in from 80% (for 1 tree) to 100% (for 5 trees) of the replicates. When the susceptibility site is not found alone, it is almost always found with only one other site, rarely with two other sites (\<0.5% of the replicates). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Comparison of the efficiency to localize the susceptibility site between CT, and SbST or CLADHC. The efficiency to localize the susceptibility site is measured by the percentage of replicates in which the simulated susceptibility site is either uniquely detected or detected among other putative sites. black circles: comparison of CT and SbST, crosses: comparison of CT and CLADHC. Each symbol corresponds to one of the ten tree topologies. ::: ![](1471-2156-6-24-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Comparison of the efficiency to localize the two susceptibility sites between CT and SbST. The efficiency to localize the two susceptibility sites is measured by the percentage of replicates in which the two sites are detected (black circles) or at least one of the two sites (white circles) Each symbol corresponds to one of the ten tree topologies. (A): penetrance 0.9 for homozygotes carrying the two susceptibility alleles and 0.3 for all other genotypes; (B): penetrance 0.6 for homozygotes carrying the two susceptibility alleles and 0.3 for all other genotypes. ::: ![](1471-2156-6-24-2) ::: For model 1 (Figure [1](#F1){ref-type="fig"}), CT is slightly less efficient than SbST (for 9 trees out of 10) and more efficient than CLADHC (for all 10 trees). For SbST, we note a high negative correlation between the efficiency to localize the susceptibility site and the LD~max~between the susceptibility site and the other sites (Spearman rank correlation ρequal -0.88 with a p-value of 0.008). It shows that the presence of a site in relatively high LD with the susceptibility site disturbs the localization process, increasing the uncertainty of the susceptibility site\'s identification. Interestingly, there is no such statistically significant correlation with CT and CLADHC (p \> 0.08), due to the fact that these last two tests use both the information on the whole haplotypes and on their history through the tree reconstruction. The presence of sites in high LD with the susceptibility site does not prevent these two methods from localizing correctly the susceptibility site. For model 2 and model 3, (Figure [2A](#F2){ref-type="fig"} and [2B](#F2){ref-type="fig"} respectively), CT is more or at least as efficient as SbST to localize the two sites, whatever the penetrance model. For the detection of at least one of the two sites, we can observe that, whatever the tree, CT is always more efficient than SbST under model 2, whereas this is only true for 6 out of the 10 trees under model 3. Application to CARD15/NOD2 polymorphisms and Crohn disease ---------------------------------------------------------- The reconstruction of the phylogenetic tree of the 33 haplotypes leads to a very high number of equally parsimonious trees. One thousand of them were retained and analyzed. In table [5](#T5){ref-type="table"}, for the two rooting methods mfH and consH, the SNPs are ranked from the least to the most significant (based on the V~i~parameter, see the Methods section). Association is detected with CT (p~mfH~= 2 × 10^-5^, p~consH~= 3 × 10^-5^) and SNP 8 is selected as potential susceptibility site. If haplotypes carrying the susceptibility allele at SNP 8 are removed, an association can still be detected (p~mfH~= 4 × 10^-5^, p~consH~= 3 × 10^-5^) and SNP 12 is identified as potential susceptibility site. When removing haplotypes with the susceptibility allele at SNP 12, there is still a remaining association (p~mfH~= 8 × 10^-5^, p~consH~= 3 × 10^-5^) and depending on the rooting method either SNP 13 (with consH) or SNP 9 (with mfH) are retained. When haplotypes with the susceptibility allele at SNP 13 are removed, the association test is not significant (p~mfH~= 2.9 × 10^-1^, p~consH~= 1.1 × 10^-1^). When haplotypes with the susceptibility allele at SNP 9 are removed, the association test is significant (p~mfH~= 2 × 10^-5^, p~consH~= 5 × 10^-5^) and the identified site is SNP 13 for both rooting method. In the end, the combination of SNP 8, SNP 12 and SNP 13 is the most parsimonious combination that can explain the susceptibility to the disease. Moreover, we can note that these 3 SNPs are those showing the highest V~i~on the whole data set. So, we finally have identified three SNPs involved in the susceptibility to Crohn\'s disease. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Cladistic analysis of Crohn\'s data 1000 equiparsimonious trees are studied. SNPs are ranked from the least to the most significant. In bold, the presumed susceptibility sites previously identified by Hugot et al. \[30\]. ::: least significant \-\-\-\-\-\-\--\> most significant ---------- ------------------- ------------------- ------------------ ------- -- ------------ ------------ ---------- mfH^a^ SNP 5 SNP 9 SNP 6 SNP 7 SNP 13 SNP 12 SNP8 consH^b^ SNP 9 SNP 5 SNP 6 SNP 7 SNP 13 SNP 12 SNP8 ^a^Rooting on the most frequent haplotype ^b^Rooting on a consensus sequence ::: Discussion ========== In this paper, we have presented a method to perform phylogeny-based nested haplotype analysis of case/control data and introduced the cladistic test (CT). We have shown by simulations that the CT test is approximately as powerful to detect an association as a test that compares haplotype distributions in cases and controls with no prior grouping. However, a major advantage of CT towards other haplotypic tests is the possibility of making inferences on potential susceptibility sites. Indeed the method of parsimony used to build the phylogeny allows the reconstruction of character state changes on the tree and thus the identification of the most likely susceptibility sites. Finding the polymorphisms responsible for the susceptibility to a disease is an important issue in genetic epidemiology and in the past few years, different methods have been proposed that use linkage disequilibrium to fine-map candidate genes and to localize disease mutations (\[[@B31],[@B32]\] and \[[@B33]\] for example) These methods are based on the fact that a disease variant arises on a particular haplotype, the haplotype and the disease thus being initially associated. Then, over time, this association is broken by recombination and only a small region of the ancestral haplotype is preserved around the disease variant. In such models, recombination is the event of interest. Other parameters such as mutations can be integrated in the models, but they are considered as nuisance parameters. The cladistic test developed here is different in the sense that it is based on the reconstruction of the evolutionary history of the mutations that occur to form the haplotypes. Contrary to the other fine-mapping methods, the event of interest is mutation, and recombination is a nuisance parameter. Through the simulation study, we have shown that this test is a powerful tool to suggest etiological SNPs. This is also illustrated here on Crohn disease since the cladistic test was able to correctly identify the 3 sites within the CARD15 gene that are already reported to be involved in the disease susceptibility \[[@B30]\]. The cladistic test requires the building of a phylogeny of the different haplotypes present in the sample. Different methods are available to build haplotype phylogenies: distance methods, parsimony methods and maximum likelihood (ML) methods. A parsimony method was chosen here because it allows the reconstruction of apomorphies that can be used to define potential sites involved in the disease susceptibility and because it runs faster than ML methods, thus allowing a simulation study. We have compared CT with another clustering method, CLADHC, a distance-based method that also differs from CT in the statistical analysis (CLADHC performs regression analysis, CT is based on chi-square tests). We have shown that the power of these two tests to detect association is quite similar, CT being slightly more powerful when one site is simulated, and CLADHC when two sites are simulated. For the localization of the susceptibility site, the efficiency of the two methods is difficult to compare because CT directly localizes the site on short sequences while CLADHC localizes a window containing the susceptibility site on longer sequences. The two test should thus be used in different situations. Nevertheless CT appears to be more efficient in our simulation conditions In the phylogeny reconstruction, one is faced to different problems. The first one is the choice of the ancestral sequence to root the phylogeny. With real data, the ancestral sequence is usually not known, unless at least two outgroups can be postulated (in the parsimony context). In the absence of ancestral sequences, hypotheses have to be made, and they are numerous, like taking the most frequent haplotype, but this haplotype could be different from one population to an other or from one sampling to another, and this most frequent haplotype may not have a significantly different frequency from the next most frequents. Or one can take the consensus haplotype, but this sequence may not be observed in the sample, or not be among the most frequent sequences. In this study, our strategy was to consider the uncertainty of the ancestral sequence as a nuisance parameter, since our goal was not primarily to focus on this problem and because we must be in the same situation for all the performed simulations. Extensive further simulations will be needed to assess the properties of the different rooting methods. However, several ancestral hypotheses should be considered, discussed, and tested on real data, as we have done for the Crohn data. Second, the presence of recombinations would restrict the use of phylogenetic reconstruction. Indeed, phylogenetic reconstruction is known to be biased by recombinations \[[@B12]\] in a manner depending on how and when the recombination events occur in the tree. This context led us to simulate the history of mutations along the haplotypes without introducing recombinations between haplotypes. However, because in our method the reconstructed tree has not to represent the exact history of the haplotypes, but only gives a way to cluster them in an appropriate way, we believe that our method can tolerate some low degrees of recombination, as it is probably the case for the Crohn data. Further simulations would however be required to better assess this point. Third, the choice of the way characters are optimized on the tree is another important point in cladistic analysis. Two classical options can be used to optimize the character state changes along the tree: deltran, (delayed transformation) that favors convergences or acctran(accelerated transformation) that favors reversions. In this paper, only the results performed with the deltran option are presented. Simulations (one susceptibility site) have also been performed with the acctran option. For the association detection, the results are very similar to those obtained with deltran showing that this parameter does not influence the power of the CT method to detect association in our simulation conditions. For the localization of the susceptibility site, the results are different. The method is more efficient with the deltran option than with the acctran option for eight trees out of ten, it is less efficient for one tree and the results are equivalent for the last tree. It is partly due to to the uncertainties of the character Sand to the frequency of the disease susceptibility allele. When the frequency of the susceptibility allele is low, as in our simulations, and when Sis coded \"?\" in some nodes of the tree, one expects the change of Sfrom 0 to 1 to occur later in the tree, and the delayed option (deltran) will then be more appropriate. Conversely, when the frequency of the disease is large, the change of Sfrom 0 to 1 has more chances to occur earlier in the tree since the Smutation can then include more cases, and the accelerated option (acctran) will be more appropriate. However, when analyzing a real data set, it is advisable to compare the results obtained with both options. The variability of the results obtained with the cladistic method is due for a minor part to the LD between the susceptibility site(s) and the other sites and for the major part to parameters specific to the tree. Among these parameters, the position of the mutation in the tree (near the root or the leaves, sample size of the clades near the mutations\...) has already been reported to have an influence on the power of cladistic methods to detect association \[[@B27]\]. Apart from the position of the mutation, the presence of sites that co-mutate with the susceptibility site, the degree of multiple mutations of the susceptibility site in the tree, and the existence of one or more reversions for the susceptibility site influence the power of the cladistic test to detect an association, and they have an even greater impact on the efficiency of the cladistic test to localize the susceptibility site. As all these factors are not independent, the evaluation of their own effects remains problematic. The four methods compared here (CT, HT, SbST and CLADHC) suppose that haplotypic data are available, which is not usually the case. The most likely haplotypes for an individual have to be reconstructed using numerical methods (for a review of these methods, see \[[@B34],[@B35]\]). The availability of familial data may help to determine the phase of the markers and then, to obtain the most likely haplotypes by using the genetic information of the two parents. However, even when these familial data are not available, the numerical methods have been shown to be very efficient to reconstruct haplotypes, except for rare haplotypes that may either be missed or falsely created. In this work, since we choose to focus on the comparison between several methods, the problem of haplotype inference is ignored. Moreover, as the four studied methods are all based on already known or inferred haplotypes carried by the individuals, the uncertainty due to the haplotype inference should not play differently from one method to another. Further extension of the CT method will take this uncertainty into account by using, for example, likelihood-based tests. Conclusion ========== In conclusion, as proved by our simulations and the Crohn example, the use of phylogenies to group haplotypes by clades and the nested analysis of case-control data on the basis of these clades turns out to be an interesting and complementary strategy for the genetic study of multifactorial diseases, especially to pinpoint the sites that are likely to be involved in disease susceptibility among the different markers identified within a gene. Methods ======= Principle of the method ----------------------- The material on which the method is based is a sample of haplotypes composed of a combination of SNPs, each haplotype being labeled either as control or as affected depending on the phenotype of the individual. After building the phylogenetic tree of these different haplotypes, a series of nested homogeneity tests are performed to detect differences in the distribution of cases and controls in the different clades. Once an association is detected, a new character is defined according to the proportion of cases carrying an haplotype. Then, it is optimized on the tree and the sites that significantly co-mutate with this new character are putative susceptibility sites for the disease. Simulation study ---------------- To test the power of this method to detect an association and its efficiency to precisely localize the susceptibility site, we performed the following steps : i) simulation of the haplotypes and choice of the susceptibility site(s) (one or two susceptibility sites are considered) ; ii) attribution of a status (case or control) to these haplotypes taking into account the genetic model of the disease; iii) reconstruction of the phylogeny of haplotypes and inference of the character state changes along the branches of the tree; iv) association test: nested analysis of the case/control ratio in the clades; v) localization of the susceptibility site using the equiparsimonious (ie, all the trees that have the minimum number of character state changes) ; vi) estimation, by simulations, of the power to detect the association between haplotypes and the disease and comparison with other methods; vii) estimation of the efficiency to localize susceptibility sites and comparison with other methods. ### Haplotype simulation 24 different haplotypes are simulated using the TREEVOLVE software (version 1.32) \[[@B36]\]. An ancestral haplotype sequence of 20 sites is randomly produced, a guide tree is generated using the coalescent model without recombination and the evolution of this ancestral sequence along the tree is simulated to obtain a sample of 24 different haplotypes. Since TREEVOLVE can only simulate the evolution of DNA sequences (4 nucleotides) and not the evolution of bi-allelic markers, we set the frequencies for bases Aand Tto very low values (A= T= 0.001 and G= C= 0.498) in order to obtain an ancestral sequence composed only of Gand C, as an equivalent of SNPs coded by 0/1 character states. The mutation rate is set to 2 × 10^-6^, and the substitution rates are chosen so that only G-C transversions could occur: A- C= 0.001, A- G= 0.001, A- T= 0.001, C- T= 0.001, G- T= 0.001, G- C= 0.995. No heterogeneity in the substitution rate among sites is specified. The population size was set to 100,000 and its growth was stationary with no subdivision. Once the haplotypes are obtained by simulation, the susceptibility sites are chosen among the 20 sites. For the simulations with only one susceptibility site, the program picks up the first site in the sequence having a minor allele frequency (p) in the sample of haplotypes within the range F\<p\<F+ 0.01 where Fis a predefined allele frequency (in our simulations, F= 0.20). For two simulated sites, the program picks up the first two simulated sites verifying a user-defined constraint. In our simulations, the two sites are chosen such that f(A~1~B~1~) = f(A~1~B~2~) = f(A~2~B~1~) = f(A~2~B~2~) = 0.25, A~1~and A~2~being the two alleles at the first site and B~1~and B~2~, the two alleles at the second site. When no site is found satisfying these conditions, a new set of haplotypes is simulated until the required conditions are obtained. This entire procedure is repeated ten times to generate ten different data sets. The pairwise linkage disequilibrium (r^2^measure, mean values for the 1000 replicates) and the tree topologies corresponding to these ten data sets are provided as Additional file [1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}. ### Attribution of disease status To generate the genotypes of each individual, pairs of haplotypes are sampled at random with replacement and the disease status (case or control) is determined by his genotype at the susceptibility site(s) and by the penetrances. In our simulations, three disease models are studied, involving either one susceptibility site (model 1) or two interacting susceptibility sites (model 2 and model 3). The first model is the disease model described in Bourgain et al. \[[@B37]\] that roughly corresponds to the effect of ApoE4in Alzheimer\'s disease \[[@B38]\]. We set the penetrance of the three genotypes DD, Ddand ddto respectively P~DD~= 0.03, P~Dd~= 0.06, P~dd~= 0.30, dbeing the disease susceptibility allele. For model 2 and model 3, an at-risk haplotype formed by the combination of the alleles A~2~at the first locus and B~2~at the second locus is defined. The penetrance associated to the genotype A~2~B~2~A~2~B~2~is set to 0.9 for model 2 and to 0.6 for model 3. All the other penetrances are set to 0.3 for both models. The sampling process is carried on until Ncases and Ncontrols are obtained (N= 100 or 200 in our simulations, depending on the number of simulated susceptibility sites). This constitutes a replicate. As the haplotypes are sampled randomly from the set of 24 different haplotype, they are almost equi-frequent in the sample of cases and controls, and their frequencies are close to ![](1471-2156-6-24-i1.gif). ### Phylogeny of haplotypes All the different haplotypes are used to build a phylogenetic tree by a parsimony method which consists in choosing the tree that minimizes the number of mutation events \[[@B39]\]. The PAUP software (version 4.0b10 \[[@B40]\]) is used to reconstruct the tree. A heuristic search is performed, using stepwise addition and TBR (tree bisection and reconnection). All sites are equally weighted, and changes from allele 0 to allele 1 and from allele 1 to allele 0 are equally allowed for all sites, including the susceptibility site (unordered option). Among the list of all parsimonious trees, only the first one obtained is used for the association test, for computation time reasons and because we found that identical results are obtained for more than 98% of the equiparsimonious trees (data not shown). For the localization of susceptibility sites, all the equiparsimonious trees (or up to 100 different trees if there are too many equiparsimonious trees) are analyzed. The trees are rooted using the real ancestor obtained with TREEVOLVE. The character states are inferred along the branches by using the classical option deltran (delayed transformation), but acctran (accelerated transformation) was also investigated. ### Association test Starting from the root of the tree, series of nested homogeneity tests comparing the number of cases and controls in different clades are performed. The principle of the method is explained in Figure [3](#F3){ref-type="fig"}. Briefly, at each level of the tree, homogeneity in the distribution of cases and controls is tested among all the nclades defined at this level. If the test is significant, an association is detected and the analysis ends. If the test is not significant, one homogeneity test is performed between all the sub-clades descending from the nclades. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Description of the nested clade analysis. (A) shows the homogeneity test performed at level k (between clades C~1~and C~2~). If it is not significant (B), a test will be performed at the following level (k+1), between all the sub-clades descending from clades C~1~and C~2~, i.e between clades C~1.1~, C~1.2~, C~2.1~and C~2.2~(3 degree of freedom). If it is significant the analysis ends because an association is detected. ::: ![](1471-2156-6-24-3) ::: P-values of the homogeneity tests are calculated by Pearson chi-square tests except when only two clades are compared in which case a Fisher exact test is performed, and except when more than two clades are tested and the sample size is less than four in a category. In this latter case p-values are estimated by permutation tests. The influence of the LD on the power of the different tests is assessed by using the r^2^measure of LD. ### Localization of the susceptibility sites For each replicate, several equiparsimonious trees may exist. We will analyze the Tfirst equiparsimonious given by PAUP, Tbeing limited to 100 for computation time reasons. For a tree t, a new character Sis allocated to each haplotype h. The state of Sis \"0\", \"1\" or \"?\" depending on the proportion (p~h~) of cases carrying the haplotype hcompared to the proportion p~0~of cases in the whole sample. if ![](1471-2156-6-24-i2.gif), Sis coded \"0\" (high number of controls); if ![](1471-2156-6-24-i18.gif), Sis coded \"1\" (high number of cases); else, Sis coded \"?\" (missing data). with n~h~being the number of individuals carrying the haplotype h. This character Sis optimized on the Tequiparsimonious trees using PAUP and the deltran option. For each site i, let ![](1471-2156-6-24-i3.gif) and ![](1471-2156-6-24-i4.gif) be the observed number of times each transition (0→1 for ![](1471-2156-6-24-i5.gif) and 1→0 for ![](1471-2156-6-24-i6.gif)) co-mutates with a 0→1 change of the character Son tree t. Let ![](1471-2156-6-24-i7.gif) and ![](1471-2156-6-24-i8.gif) be the expected number of co-mutations on tree tunder the hypothesis of a random distribution of the mutations on tree tand an equal probability of mutation on each branch. ![](1471-2156-6-24-i9.gif) where ![](1471-2156-6-24-i10.gif) (resp. ![](1471-2156-6-24-i11.gif)) is the number of 0→1 (resp. 1→0) transitions of the site ion tree t; s~t~is the number of 0→1 transition of the character Son tree t; b~t~is the number of branches of tree t. Then, for each site ion tree t, we measure the correlated evolution of the site iand the character S, by defining ![](1471-2156-6-24-i12.gif) (resp. ![](1471-2156-6-24-i13.gif)) as follow: ![](1471-2156-6-24-i14.gif) then, for the Tequiparcimonious trees, we define ![](1471-2156-6-24-i15.gif) (resp. ![](1471-2156-6-24-i16.gif)) as follow: ![](1471-2156-6-24-i17.gif) Finally, V~i~is defined as the max between ![](1471-2156-6-24-i15.gif) and ![](1471-2156-6-24-i16.gif). The site or the two sites corresponding to the highest V~i~are selected as putative susceptibility sites ### Estimation of power and efficiency For each of the ten samples of 24 haplotypes simulated along the ten different tree topologies, the process detailed above (attribution of disease status, phylogenetic reconstruction, association test and localization of the susceptibility site) is repeated 1000 times to obtain 1000 replicates, each of them including Ncases and Ncontrols. The power to detect the association is measured by counting the proportion of replicates in which a significant heterogeneity in the distribution of cases and controls is detected between clades. The efficiency of CT to localize the true susceptibility site is simply evaluated as the number of times, among the 1000 simulations, the true susceptibility site is among the detected sites. When two susceptibility sites are simulated, we evaluate the number of times the two sites are detected and the number of times at least one of the two sites is detected. The test using the cladistic method (CT) is compared to three other association tests: i) a haplotypic test (HT) that compares the haplotype distributions in cases and controls: if hdifferent types of haplotypes are observed, a chi-square test with h- 1 degree of freedom (df) is performed. As with the CT test, when sample sizes are small, a permutation procedure is used to perform the chi-square test. This HT test is only used to test for association and not to localize the susceptibility sites. ii) An allelic test, referred to as Site by Site Test (SbST), that compares the allele distributions in cases and controls at each site by a chi-square: if the number of sites is s, then schi-squares, each with one df, are performed. The site or the two sites that give the most significant results are proposed as susceptibility sites. iii) A test based on CLADHC, the program developed by Durrant et al. \[[@B29]\]. In this test, the data set is analyzed by overlapping windows sliding along the haplotypes. In each window, a tree is reconstructed using a distance method, and a regression analysis is performed at each level of the tree (nested analysis). The level of the tree that maximizes the evidence of disease marker association in the likelihood-ratio test is selected. For each window, the program also provides a significant threshold calculated using the Bonferroni correction. In our simulations, a window size of 6 is selected. We consider the association as detected when at least one of the windows is significant. Concerning the localization of the susceptibility site, CLADHC can only find a window containing the susceptibility site, but not the susceptibility site itself. Therefore, we only use CLADHC in the case of one susceptibility site and we consider the localization as correct when the window with the maximum value of the statistic contains the susceptibility site. For CT, SbST and CLADHC, we are confronted to a problem of multiple testing. When a real data set is analyzed, a permutation procedure can be used to estimate p-values of the association tests that are corrected for multiple testing. This procedure is very time-consuming and cannot be used on the simulations we have performed here. As an example, the 100,000 repetitions for the Crohn data run in about 24 hours on a Pentium III, 930 MHz, 512 Mo of RAM. Therefore, for SbST and CT, we estimate the nominal type I error required to obtain a global type-one error of 5% by a simulation under the null hypothesis of no association between the studied sites and the disease (5000 replicates). We found that for the different simulation conditions studied here, these nominal errors are equal to 1% for CT and varies from 0.2 to 0.4% for SbST. For the analysis with CLADHC, the program provides a significant threshold calculated using Bonferroni correction. As the Bonferroni correction is very conservative, to obtain similar type I errors with CLADHC as with CT, nominal type I errors were set to values varying from 15 to 17% for simulations with one susceptibility site and from 20 to 25% for simulations with two susceptibility sites. Analysis of real data on Crohn disease -------------------------------------- The data set is described in a paper of Hugot et al. (2001) \[[@B30]\] and includes 232 nuclear families with two affected children and their parents genotyped for 13 SNPs in the CARD15/NOD2 gene. Haplotypes were reconstructed in these families using GENEHUNTER 2.0b\[[@B41]\]. In each family, one affected child is selected at random to create the case sample (his two haplotypes are labeled cases). The control sample is formed with the parental haplotypes non transmitted to the selected children. The phylogenetic reconstruction assumes that no recombination has occurred between the SNPs. To determine if this was true for the different SNPs tested here or at least for a subset of them, linkage disequilibrium (LD) is estimated between all the SNPs. From the pairwise LD, a group of 7 SNPs (5, 6, 7, 8, 12, 9 and 13) can be identified as forming a block in strong LD among which no or very few recombinations are likely to have occurred (see [Additional file 3](#S3){ref-type="supplementary-material"}). Interestingly, this same block of 7 SNPs was described by Vermeire et al. \[[@B42]\] in the CARD15/NOD2 gene. Only these 7 SNPs that form 33 different haplotypes are used in the cladistic analysis. The phylogenetic trees are reconstructed using the same parameters as in the simulations. However here, contrary to the simulations, the ancestral sequence being not known, two different analysis are performed using as ancestral sequence either the most frequently observed haplotype in our sample (mfH) or a consensus sequence formed by the combination of the most frequent alleles at each SNP (consH). For the association test, p-values are estimated on the first equiparsimonious tree obtained by PAUP, using an algorithm similar to the one described in a paper of Becker and Knapp (2004) \[[@B43]\]. This permutation process corrects for multiple testing in a lesser conservative way than the Bonferroni correction because it takes into account the dependency between the tests. For the localization, among all the equiparsimonious trees, 1000 of them are retained and analyzed with CT. Authors\' contributions ======================= CB perfomed the simulation analyses, contributed to the conception of the study, to the elaboration of the algorithm and wrote the manuscript. VD contributed to the elaboration of the algorithm. J-PH provided the CARD15 data. PD and EG contributed to the design and to the conception of the study and to the manuscript preparation. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Average pairwise linkage disequilibrium for the 10 simulated date sets. The average pairwise LD (over 1000 replicated) is calculated using the r^2^measure for each of the 10 tree topologies (T1 to T10). The matrices are plotted with GOLD\[[@B44]\]. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Tree topologies of the 10 simulated date sets ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 Pairwise linkage disequilibrium for the Crohn data. (A) shows the pairwise linkage disequilibrium (LD) calculated with the D\' measure. (B) shows the LD calculated with the r^2^measures. The matrices are plotted with GOLD \[[@B44]\]. A high degree of LD can be observed. The differences between the r^2^and the D\' values are explained by the difference in the allelic frequency for the different SNPs. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We wish to thank Marie-Claude Babron and two anonymous reviewers for helpful comments and Arnaud Legrand for letting us use one of his perl function. We are also grateful to Habib Zouali and Suzanne Lepage from the fondation Jean Dausset for providing us with Crohn data and to Caroline Durrant for kindly providing us with the software CLADHC. Many thanks to the runtime team from the LaBRI for letting us run a part of our simulations on their cluster.	PubMed Central	2024-06-05T03:55:58.366282	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173100/", "journal": "BMC Genet. 2005 May 18; 6:24", "authors": [ { "first": "Claire", "last": "Bardel" }, { "first": "Vincent", "last": "Danjean" }, { "first": "Jean-Pierre", "last": "Hugot" }, { "first": "Pierre", "last": "Darlu" }, { "first": "Emmanuelle", "last": "Génin" } ] }
PMC1173101	Background ========== N-acetylation polymorphism is one of the earliest discovered and most intensively studied pharmacogenetic traits that underlie interindividual and interethnic differences in response to xenobiotics. In humans, acetylation is a major route of biotransformation for many arylamine and hydrazine drugs, as well as for a number of toxins and known carcinogens present in the diet, cigarette smoke and the environment \[[@B1]-[@B3]\]. Genetically determined differences in N-acetylation capacity have been proved to be important determinants of both the effectiveness of therapeutic response and the development of adverse drug reactions and toxicity during drug treatment \[[@B4]\]. In the last decades, numerous investigations have been made to elucidate the genetic basis of N-acetylation polymorphism in various ethnic groups in order to develop efficient genotyping tests and to adapt therapies to specific patients and populations in accordance with their genetic makeup. Some of the drugs excreted by acetylation are indeed crucial in the treatment of diseases representing a worldwide concern, such as tuberculosis and AIDS-related complex diseases \[[@B5],[@B6]\]. Moreover, a number of epidemiological studies have suggested possible associations between the N-acetylator phenotype and a variety of complex human diseases, the most consistent findings being those regarding urinary bladder cancer and familial Parkinson\'s disease \[[@B7]-[@B10]\]. The gene coding for the arylamine N-acetyltransferase 2 (NAT2) enzyme has been established as the site of the classic human acetylation polymorphism \[[@B11]-[@B13]\] and the molecular basis of individual and interethnic variation in acetylation capacity is now well documented \[[@B14],[@B15]\]. All mutations reported to date are found within the 870-bp coding region of the NAT2gene. Among the seven single nucleotide polymorphisms (SNPs) that are commonly found in human populations, four result in an amino acid substitution that leads to a significant decrease in acetylation capacity (single base-pair substitutions at positions 191, 341, 590, 857). The other three are either silent mutations (C282T, C481T) or a non-synonymous substitution that does not alter phenotype (A803G). In the consensus gene nomenclature of human NAT2that encompasses all currently recognized alleles \[[@B16],[@B17]\], sets of SNPs located throughout the coding region are linked in terms of haplotypes, that is they are organized as they segregate together on one individual\'s chromosome at the NAT2locus. Each combination of SNPs identified so far constitutes a distinct haplotype that is treated as an allele of the haplotype system. The consideration of multilocus haplotypes seems more desirable since there is growing evidence that for genes containing multiple SNPs in high linkage disequilibrium (LD) such as NAT2\[[@B18]\], haplotype structure rather than individual SNPs can be the principal determinant of phenotypic consequences \[[@B19]-[@B21]\]. A functional polypeptide is indeed the product of a haplotype, covering the entire coding region and coded by a single chromosome. The NAT2alleles described so far contain up to four of the acknowledged mutations in various combinations. Each allele is associated with an acetylator phenotype depending on which mutations they contain: for instance, substitutions at positions 191, 341, 590, and 857 are diagnostic for defective NAT2function and hence for the slow acetylator phenotype (Table [1](#T1){ref-type="table"}). Three NAT2phenotypes have been described: subjects with two low activity alleles are classified as slow acetylators, while those with two functional alleles are considered rapid acetylators. If only one allele is of the slow type, an intermediate phenotype is observed. Many early studies did not distinguish between fast and intermediate acetylators, categorizing both types of subjects as fast acetylators. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The major human NAT2alleles and their associated phenotype^a^. ::: Allele Nucleotide change ^b^ Phenotype ----------------- ----------------------- ----------- ------- --- ------- --- ------- ------- NAT2\4* rapid **NAT2\5A* x x slow NAT2\5B* x x x slow NAT2\5C* x x slow NAT2\6A* x x slow NAT2\6B* x slow NAT2\7A* x slow NAT2\7B* x x** slow NAT2\12A* x rapid NAT2\12B* x x rapid NAT2\13* x rapid **NAT2\14A* x slow NAT2\14B* x** x slow Nucleotide substitutions shown in bold have a functional consequence on enzyme activity. Bold-faced alleles contain functional polymorphisms and are hence associated with the slow acetylator phenotype. Classification of NAT2alleles into different clusters is based on the most functionally significant nucleotide substitution present: the NAT2\5, NAT2\6, NAT2\7and NAT2\14clusters possess signature nucleotide substitutions at positions 341, 590, 857 and 191, respectively and are hence all decreased function alleles (\'slow alleles\'). The others display enzymatic activity comparable to the rapid acetylator allele NAT2\4. There are significant interethnic differences in NAT2allele distribution and frequency \[14, 15\]. ^a^Adapted from Hein et al. \[3\]. NAT2nomenclature is accessible on the internet at website <http://www.louisville.edu/medschool/pharmacology/NAT.html> ^b^Only changes from the reference sequence (NAT2\4) are indicated. ::: Problems may occur when individual multi-site NAT2genotypes have to be assigned correctly to a particular combination of two multilocus haplotypes. Indeed, current routine genotyping and sequencing methods typically do not provide haplotype information in diploid organisms such as humans, and the gametic phase of haplotypes is inherently ambiguous when individuals are heterozygous at more than one locus. As illustrated in Figure [1](#F1){ref-type="fig"}, a subject carrying two inactivating mutations can be either rapid or slow acetylator depending on whether these mutations are located in the same or different chromosome, respectively. It is thus crucial to unequivocally assess mutation linkage patterns, this step being a prerequisite to obtain accurate haplotype frequency estimates in populations and reliable genotype-phenotype predictions. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The ambiguous gametic phase of haplotypes for a given multilocus genotype. To illustrate the relevance of linkage phase ascertainment, let us consider the following case of a four-site heterozygous individual at positions 191, 341, 481 and 803 within the NAT2coding sequence. Eight possible combinations of haplotypes can be inferred from this multilocus genotype, two of whom are shown here. Depending on the location of mutations to either DNA strand, the individual\'s NAT2genotype composed of two multilocus haplotypes will not be the same. Moreover, an incorrect resolution of mutation linkage patterns may entail an error in individual phenotype prediction: the subject will be classified either as a slow or as a rapid acetylator depending on the haplotypic combination chosen. Symbol (\) points at mutations leading to a decrease in NAT2enzyme activity, while symbol (×) indicates those with no impact on the acetylator phenotype. ::: ![](1471-2156-6-30-1) ::: However, in spite of its high relevance, this issue has not been handled properly by most past studies investigating NAT2polymorphisms. Early genotyping studies only screened for the presence of three polymorphisms (C481T, G590A, G857A), and a subject was defined as a slow acetylator if he was homozygous for one, or heterozygous for two (each located on one DNA strand), of this three nucleotide changes. Such a definition assumed that there could be no single allele with two or more of the tested mutations. In other studies that screened a larger number of SNPs within NAT2, patterns of LD between point mutations were often assumed, in reference to the haplotypes previously described and which are commonly found in populations of European origin. For instance, the designation of NAT2alleles is usually based on the assumption that 481T and 803G are strongly linked to 341T, and 590A and 857A are linked to 282T \[[@B22]\]. However, in rare cases the typical allelic linkage pattern of mutations may be disrupted because of genetic recombination and this may result in misclassification of alleles. Indeed, although such assumed linkage patterns are very strong, other allelic variants carrying either unusual combinations of mutations or mutations in isolation have been described in a few cases \[[@B23]\]. Furthermore, the designation of NAT2alleles in such a way that it necessarily conforms to the existing consensus nomenclature of acknowledged haplotypes precludes the disclosure of unexpected combinations of mutations, different from the established allelic variants, and hence, the discovery of new alleles. Such a manner of inferring haplotypes from unphased multilocus genotypes may introduce biases in NAT2allele designation and individual phenotype prediction, and these potential biases are of higher magnitude when non-European populations are concerned. Most studies assumed particular patterns of linkage previously described in populations of European origin, but which may not hold in other ethnic groups. Indeed, recent works have shown that patterns of LD can differ markedly among populations with different ethnic and demographic backgrounds. As an example, Loktionov and colleagues \[[@B24]\] pointed out the high occurence of isolated mutations 803G and 282T (defining alleles NAT2\12Aand NAT2\13, respectively) in Black South Africans, while these nucleotide changes are almost always tightly linked to other mutations in European populations \[[@B25]\]. Likewise, Dandara et al.\[[@B26]\] recently identified a novel mutation linkage pattern (NAT2\6E) that appeared to be common in three African populations and that had not yet been reported in Europeans. As well, Anitha et al.\[[@B27]\] revealed a new combination of acknowledged mutations (NAT2\5G) in the Malapandaram tribe of South India that has not been described so far in any other world population. The genotype-phenotype discordance observed in many ethnic groups where mutation linkages have not been extensively proven experimentally might result from such unexpected compound alleles. It is thus necessary to systematically verify the postulated allelic combinations. To avoid such potential problems, many authors designed special analytical techniques to unequivocally determine the allocation of mutations to either DNA strand. Molecular methods using combinations of mutation-specific polymerase chain reaction (PCR) reamplification coupled to restriction mapping of the PCR products have been developed; these allow the separate analysis of each allele in order to obtain a complete map of both genes in every individual. Some studies applied these procedures to all multiply heterozygous subjects \[[@B24],[@B28]-[@B31]\], while others limited their application to particular cases as those where an alternative linkage pattern of mutations would lead to a change in phenotype \[[@B5],[@B27],[@B32]-[@B36]\]. However, these experimental methods of molecular haplotyping are not entirely satisfying because they entail an additional cost and are currently labour-intensive, time-consuming, prone to experimental errors and difficult to automate. Therefore, they do not appear to be good candidates for routine clinical applications and for a generalization at a large scale. A cheap and relatively straightforward alternative for haplotype reconstruction based on genotype data from unrelated individuals is the use of computational algorithms. The most widely used algorithms developed so far are based either on a parsimony, a maximum-likelihood, or a Bayesian approach (see \[[@B37]\] for a review). In the last decade, numerous investigations based on empirical data and extensive simulation studies have demonstrated that such in silicohaplotype-inference methods could give effective and accurate prediction of haplotype phases, especially in regions with high LD values between polymorphic sites and small probabilities of recombination events \[[@B18],[@B38],[@B39]\]. Therefore, they could be fairly efficient alternatives to molecular-haplotyping methods when applied to NAT2gene data. Surprisingly, to our knowledge, only three studies have used computational methods to reconstruct NAT2haplotypes and estimate allele frequencies in population samples \[[@B40]-[@B42]\]. One explanation for such limited use may be the lack of evidence documenting the performance of in silicoapproaches when applied to actual NAT2data. Indeed, the accuracy of these strategies, compared with molecular methods, needs to be assessed before their applications can be advocated at a large scale. A recent study provided preliminary results on this issue: Xu and colleagues \[[@B18]\] empirically evaluated and compared the accuracy of the Clark\'s algorithm \[[@B43]\], the expectation-maximisation (EM) algorithm and a Bayesian method implemented in the PHASE program \[[@B44]\] in phase inference at NAT2, taken as an example of a locus with pronounced LD over a 850-bp region. In this study, NAT2haplotypes (consisting of five genotyping SNPs at position 282, 341, 481, 590, and 803 nt) were experimentally determined through cloning and sequencing in 81 individuals of European ancestry. They found that all three computational methods provided remarkably accurate and reliable estimates for NAT2haplotype frequencies and individual haplotype phases. The objective of the present study was to extend this investigation to more precisely assess the performance of the computational approach. We conducted an extensive study based on experimental data from a larger number of samples, issued from populations of various ethnic origin, and tested for haplotypes involving the seven major polymorphic loci of NAT2. Furthermore, the larger population samples investigated are of greater significance: as the sample size grows, there is more opportunity to observe rare haplotypes that are the most difficult to infer statistically. This comparative study is designed to evaluate the performance of different haplotyping algorithms and to assess the consistency of their estimates. In addition, it provides information on the impact of various data-set characteristics (sample size, haplotype frequency distribution, haplotype frequencies, deviation from Hardy-Weinberg (HW) equilibrium, \...) on estimation accuracy; we then explore the utility of data-based diagnostics for assessing probable accuracy. Results ======= Molecular haplotyping of the NAT2locus revealed between eight and twelve distinct haplotypes in each of the five population samples investigated. The theoretical maximum number of haplotypes for a set of seven biallelic variable sites is 128 (2^7^) if there is random association between polymorphic sites, whereas it is only 8 in the absence of recombination, recurrent and back mutation. Hence, the small number of haplotypes observed at NAT2suggests strong LD over the short physical distance spanning this gene. Indeed, we observed complete or near complete LD for all pairs of SNPs with sufficiently high frequencies (only alleles with frequencies in the range 0.05--0.95 were included in the analysis because estimates of LD for low-frequency alleles in small samples are not informative): 85% of all pairwise r^2^values were highly significant (Exact p-value \<0.0001). Although LD patterns were rather similar among the different population samples, substantial differences in LD levels were observed (Figure [2](#F2){ref-type="fig"}): the Korean sample, and especially the South African sample, displayed much smaller values of average pairwise r^2^(0.27 and 0.20, respectively) than the two European and the Nicaraguan samples (values between 0.39 and 0.57), for which a strong haplotypic structure was observed. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Linkage disequilibrium (r^2^value) between SNP markers in the NAT2locus. Graphical representation of the disequilibrium matrices obtained through computation of the r^2^coefficient between each pair of markers, for the Spanish, Korean and Black South African samples. The British and Nicaraguan samples provided patterns and levels of LD comparable to those of the Spanish data. For each marker pair, GOLD \[60\] plotted the color-coded pairwise r^2^statistics at the Cartesian coordinates corresponding to marker location, and the plots were completed by interpolation. These graphs point out the strong level of LD between markers at positions 341, 481, and 803, as well as between SNPs located at 282 and 590: these markers are thus strongly predictive of one another. In Black South Africans, LD patterns are less pronounced and more diffuse across marker pairs. ::: ![](1471-2156-6-30-2) ::: Among the 1608 individuals investigated over the five data sets, 45.5% (732/1608) were either homozygous for all SNP sites or heterozygous at only one SNP site; thus, their haplotype pairs could be assigned directly. Besides, 35.7% (574), 10.0% (160), 0.9% (15), and 7.9% (127) individuals were heterozygous at two, three, four and five SNP sites, respectively. We inferred their haplotype phases with four computational haplotyping methods, and compared the results with those obtained through molecular haplotyping. Since the Hapar program often provided several equally parsimonious solutions for a given multilocus genotype, it could not resolve a relatively large fraction of heterozygous individuals in each sample and hence, we could not deduce frequency estimates for the haplotypes observed. Therefore, we evaluated Hapar only on its ability to identify the set of haplotypes present in a sample. Haplotype identification ------------------------ Hapar found for each sample the smallest set of haplotypes that could explain the genotype data, and PL-EM, Haplotyper and PHASE provided the list of all the haplotypes selected to appear in at least one of the subjects in the \"best\" reconstruction, that is when the most likely pair of haplotypes is selected for each individual. The I~H~indices of the four programs are displayed for each population sample in Table [3](#T3){ref-type="table"}. For the British and Korean samples, all computational methods identified exactly the same haplotypes as those determined experimentally. In contrast, in the other three samples, the algorithms sometimes inferred one additional haplotype, that was not actually present, and/or missed one haplotype that was shown to be present by means of molecular haplotyping. Nevertheless, these prediction errors always concerned rare haplotypes of frequency \<0.75% (singletons in most cases). The PHASE algorithm performed the best. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Performance of the four computational methods in haplotype identification, as measured by the I~H~index. ::: Hapar PL-EM Haplotyper PHASE --------------------------------- ------------- ------------- ------------- ------------- 258 Spanish \[45\] 0.933 (1)* 0.933 (1) 0.933 (1) 0.933 (1) 137 Nicaraguans \[30\] 0.952 (1) 0.952 (1) 0.909 (2) 0.952 (1) 112 UK Caucasians \[24\] 1 1 1 1 101 Black South Africans \[24\] 0.917 (2) 0.917 (2) 0.917 (2) 1 1000 Koreans \[31\] \- \* 1 1 1 Numbers in brackets indicate the number of haplotypes for which an error of prediction was made. \* The size of the Korean sample was too large to be correctly handled by the Hapar program. ::: Prediction of individual haplotype phases ----------------------------------------- We also evaluated and compared the effectiveness of the computational methods in reconstructing haplotype pairs for individuals. Table [4](#T4){ref-type="table"} gives, for each data set and for each algorithm, the individual error rate. Whatever the method, the number of incorrectly reconstructed individuals was remarkably low, with error rates always under 4%. The largest number of mistakes were observed for the African sample and, among the three algorithms tested, PHASE yielded the lowest error rates. Furthermore, it is interesting to note that, in all cases of incorrectly predicted phases, there was no impact on phenotype prediction. Thus, despite these errors, the proportions of slow, intermediate and rapid acetylators in each population sample were in 100% agreement with those deduced from molecular haplotyping. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Individual error rate in haplotype reconstruction ::: PL-EM Haplotyper PHASE --------------------------------- ------- ------------ ------- 258 Spanish \[45\] 0.39% 0.39% 0.39% 137 Nicaraguans \[30\] 2.19% 3.65% 2.19% 112 UK Caucasians \[24\] 0.89% 0% 0% 101 Black South Africans \[24\] 3.96% 3.96% 2.97% 1000 Koreans \[31\] 0.30% 0.30% 0.30% The individual error rate is defined as the ratio of erroneous phase calls to the total number of phase calls (see text). ::: Estimation of haplotype frequencies ----------------------------------- A comparison of the haplotype frequencies determined molecularly with those estimated computationally showed very high concordance. Both PL-EM and PHASE methods provided similarity index (I~F~) values very close to the maximal value of 1 in all investigated data sets (Table [5](#T5){ref-type="table"}). Such high values may be explained by the fact that the I~F~index gives more weight to common haplotypes whose frequencies are the most accurately estimated by computational algorithms. To investigate the effect of haplotype frequency on estimation accuracy, we plotted the change coefficient (C) against the larger of the two haplotype frequencies (Max \[![](1471-2156-6-30-i1.gif), p~0i~\]), for all possible haplotypes with nonzero frequency estimates determined by either analysis in any of the five population samples. As shown in Figure [3](#F3){ref-type="fig"}, substantial percentage changes \>30% occur only at the lowest haplotype frequencies (\<0.007). Even the moderate changes (range 10%--30%) occur only when the frequency estimates are \<0.03, and any change in percentage value \>5% concerns only haplotype frequencies \<0.035. Two-thirds (67%) of the haplotype frequency estimates showed either no change or a small change (\< 3%). In Table [6](#T6){ref-type="table"}, we compared the relative estimation accuracy of PL-EM and PHASE programs by computing the average change coefficient for three classes of haplotype frequencies. The two methods perform similarly for haplotypes with frequencies \>0.05, whereas PHASE provides more accurate estimates when rarer haplotypes are concerned. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Index of similarity (I~F~) between haplotype frequencies estimated with and without molecular haplotyping information. ::: PL-EM PHASE --------------------------------- ------- ------- 258 Spanish \[45\] 0.996 0.996 137 Nicaraguans \[30\] 0.986 0.986 112 UK Caucasians \[24\] 0.994 0.998 101 Black South Africans \[24\] 0.981 0.988 1000 Koreans \[31\] 0.997 0.998 Computations of I~F~indices were based on the haplotype frequency estimates obtained by considering all possible haplotype configurations (with a nonzero probability) inferred for each subject, weighted by their estimated probability. Note that such estimates can be rather different from those obtained by gene counting on the basis of the \"best\" reconstruction (that is, when only the most probable pair of haplotype is selected for each sampled individual). Since Haplotyper only provides a summary of the frequency with which each haplotype occurred in the \"best\" reconstruction, it was excluded from the comparison. ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The change coefficient (C) as a function of haplotype frequency. The change coefficient reflects the discrepancy between haplotype frequencies deduced from phase-known data and those estimated computationally (here with the PHASE program). All haplotypes occurring in any of the five population samples are considered. ::: ![](1471-2156-6-30-3) ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Average change coefficients of PL-EM and PHASE programs computed for three classes of haplotype frequency. ::: Haplotype frequency ------- --------------------- ------ ------ PL-EM 30.6% 8.3% 1.2% PHASE 17.4% 6.8% 1.2% ::: Partially resolved data sets ---------------------------- We also performed similar analyses on six other previously published data sets, in which linkage phase patterns were only partially resolved by molecular haplotyping. These data concerned 844 German \[[@B32]\], 248 Polish \[[@B33]\], 303 Turkish \[[@B34]\], 50 non-caste Dogons from Mali, 52 Gabonese and 60 Caucasians \[[@B5]\]. Haplotype phase information was available for 41%--74% individuals in these six population samples (including phase-resolved genotypes as well as non ambiguous homozygous or simply heterozygous genotypes). The PHASE algorithm was applied on the unphased genotype multilocus data of each of these samples, and a 100% concordance was observed between individual haplotype phase reconstruction through the computational method and the empirically determined linkage patterns, for all investigated data sets (data not shown). This means that, despite the efforts invested, in terms of work, time and money, to resolve mutation linkage phase in a part of each sample, no more information was added by molecular haplotyping than what could be extracted from computational algorithms applied to these data. Discussion ========== This empirical study demonstrates how closely the frequencies computationally estimated from phase-unknown data approximate those from gene-counting estimates based on phase-known data. In the particular case of the NAT2gene, where there is near complete LD between SNPs within the coding region, all in silicoapproaches provided highly effective and accurate estimates for haplotype frequencies and individual haplotype phases. Estimated frequencies of common haplotypes were nearly identical to those empirically determined, whereas rare haplotypes were occasionally miscalled when their presence/absence had to be inferred. As already pointed out by Stephens et al.\[[@B44]\] and Lin et al.\[[@B39]\] and confirmed in this study, lower-frequency variants are less easily estimated statistically; indeed, there is less contextual information about phase for singletons versus nonsingletons. Thus, for those research questions for which the NAT2common haplotypes are most important, frequency estimates based on the unphased SNP-typing results from unrelated individuals will be sufficient. However, accurate identification of rare haplotypes may be critical for many researchers, such as population geneticists interested in detecting features of recent demographic history that are population-specific or signatures of selective effects in NAT2sequences; as well as for epidemiologists and clinicians concerned with the possibility that rare haplotypes may be important for disease risk or for predicting drug response. In such cases, molecular haplotyping will be necessary to determine linkage phase unambiguously \[[@B57]\]. For a locus such as NAT2where a strong haplotypic structure is observed, all algorithms provided highly effective and accurate results for haplotype reconstruction. Thus, such \"ideal\" data for statistical inference did not permit to properly discriminate between the different methods investigated. Nevertheless, despite roughly similar performances, slightly better results were observed with the PHASE program. In particular, PHASE outperformed the other programs when frequencies of rare haplotypes have to be inferred. This is consistent with the results of some previous studies which evaluated and compared the performance of several algorithms on both empirical and simulated data \[[@B44],[@B54],[@B61]\]. PHASE provided the most accurate reconstructions, probably because the true haplotypes conformed more closely to the assumptions of the approximate coalescent prior than to those of the Dirichlet prior. Many factors may influence the estimation accuracy of computational approaches. They can be assessed empirically within a dataset, to be further used as \"diagnostics\" for predicting potential inaccuracies in estimation caused by features in the relevant data set \[[@B38]\]. Sample size did not appear to have a large effect on the haplotype frequency estimates comparing phase-known and phase-unknown results for the five data sets included in this study. Perhaps the low error rate observed in Koreans is partly due to the huge size of this sample (1000 individuals): an improvement in accuracy of the estimation procedure with increased sample size is indeed expected since information redundancy in the form of multiple copies of the same haplotype in the data set is required for the statistical algorithms to work properly \[[@B38],[@B48]\]. On the other hand, computational methods may also perform best in small samples, in which there is little chance to observe rare haplotypes that are the most difficult to infer statistically. Nevertheless, since the number of new haplotypes is not expected to increase linearly with sample size, the analysis of sufficiently large samples should guarantee a good reliability in the resulting estimates. Although most of the tested algorithms assume HW equilibrium, significant departures from HW proportions did not seem to have any impact on the accuracy of their predictions. HW equilibrium holds well for the Nicaraguan, UK Caucasian, and Black South African samples (nonsignificant results), whereas the genotype distribution in the Spanish and Korean samples show significant departures from HW proportions (Table [2](#T2){ref-type="table"}). The data sets investigated are not the most suitable to evaluate the effect of a deviation from HW equilibrium: significant results for tests of HW ratios are quite close to the threshold value (5%), and in the case of the Korean sample, an excess homozygosity is observed, which should not compromise the algorithm performance. Indeed, in such a case, there is a balance between loss of accuracy caused by violation of HW equilibrium and gain of accuracy caused by the decrease in missing phase information through an excess of homozygotes \[[@B38],[@B62]\]. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The five phase-resolved NAT2molecular data sets investigated. ::: Population sample Proportion of phase-unknown multiple heterozygotes^a^ **NAT2gene diversity^b^ Deviation from the Hardy-Weinberg equilibrium (exact p-value)^c^ --------------------------------- ----------------------------------------------------------- ----------------------------- ---------------------------------------------------------------------- 258 Spanish \[45\] 66.7% 0.65 0.012 137 Nicaraguans \[30\] 59.1% 0.70 0.072 112 UK Caucasians \[24\] 52.7% 0.69 0.222 101 Black South Africans \[24\] 63.4% 0.86 0.122 1000 Koreans \[31\] 50.0% 0.52 0.016** All population samples were genotyped for the same seven nucleotide changes (G191A, C282T, T341C, C481T, G590A, A803G, G857A), except Koreans where the C190T mutation was investigated instead of G191A. ^a^Proportion of multiply heterozygous individuals with ambiguous genotype whose phase has been resolved molecularly. ^b^Expected heterozygosities for the NAT2haplotyped system were estimated as where n is the number of gene copies in the sample, and pi is the sample frequency of the i-th haplotype. ^c^The significance of deviations from Hardy-Weinberg equilibrium was tested for genotypic data with known gametic phase using the random-permutation procedure implemented in the Arlequin package \[51\]: a Fisher\'s exact test using a Markov chain random walk algorithm was performed for each data set. The resulting p-values were considered significant if inferior to 0.05 (significant p-values are shown in bold). ::: Among the five data sets investigated in this study, Black South Africans displayed the highest error rate in haplotype computational inference. One possible explanation may be the presence in this sample of a large number of different multiple heterozygotes with ambiguous multilocus genotypes, occurring at roughly similar frequencies (this is reflected in the high NAT2gene diversity displayed by this population sample (Table [2](#T2){ref-type="table"})). Indeed, both the number of different ambiguous multiple heterozygous genotypes and their relative frequencies have been shown to be of high importance in the assessment of haplotype estimation accuracy \[[@B57]\]: both would be good indicators of the difficulty level of a given data set for haplotyping algorithms. The existence of many different multilocus genotypes uniformly distributed implies that many different haplotypes occur at low frequency, and that, consequently, greatest error and uncertainty occur in the estimation of haplotype frequencies (since no single haplotype is overwelmingly frequent). In contrast, the presence of a small number of multiple heterozygous genotypes at proportionately high frequencies implies that some individual haplotypes exist at high frequencies, and the estimation of those haplotype frequencies will be easier and accomplished with greater accuracy \[[@B38],[@B57]\]. In such cases, molecular haplotyping may add little information for the resolution of haplotype phases. The amount of LD between SNP markers may be another determining factor for the prediction of estimation reliability since when multiple polymorphic sites display little disequilibrium, as was observed in the African sample compared to the others, a large proportion of the chromosomes may occur as uncommon or rare haplotypes, implying a greater difficulty level in the statistical inference of haplotypes. Therefore, we advocate to examine beforehand the unphased NAT2genotype data for both the frequency distribution of multiply heterozygous genotypes and the level of LD between polymorphic markers; this will allow to assess the difficulty level displayed by the data set for statistical inference, and hence, to predict the ability and accuracy with which computational algorithms would infer haplotype phases from such data. Of course, statistical methods can be used in conjunction with experimental methods to provide more accurate estimates of individual haplotypes. It has been claimed that the ability of certain computational methods to accurately assess the uncertainty associated with each phase call gives them the substantial practical advantage of allowing experimental effort to be directed at sites and/or individuals whose phases are most difficult to reconstruct statistically or that are critical to the conclusions of the study \[[@B20],[@B44],[@B61]\]. However, in our study, we observed that most erroneous phase calls inferred at the individual level were strongly supported, with a probability close to the maximal value of 1. Thus, these errors could not have been avoided since they would not have been selected for the molecular targeting. This stresses why, in the case of the discovery of a novel NAT2allele through computational haplotyping methods, the unusual linkage pattern should always be confirmed by cloning and sequencing the allele under question, as advocated by Cascorbi and Roots \[[@B25]\] for novel allelic combinations detected by molecular techniques. Throughout this study, we assumed that the NAT2linkage patterns molecularly determined were the \"true\" ones, and hence, that there was no error in the haplotype assigments based on experimental methods. However, molecular techniques may have experimental error rates as high as the rate of statistical error associated with the computational haplotype determination algorithms \[[@B19]\]. Indeed, molecular haplotyping bears the risk of false positive or false negative allele-specific amplification (because of the nucleotide-dependent specificity of that technique) as well as uncomplete or non-specific digestions with the enzymes used in restriction analyses \[[@B25]\]. In the present study, we have estimated the computational error rates to be of no more than 3--4% for all investigated algorithms. This is not higher than the corresponding error rate from molecular haplotyping techniques, on the order of 2--3% \[[@B20]\]. Therefore, it is difficult to determine whether the discrepancies observed between experimental and computational estimates are actually due to statistical errors from algorithms; they may be due to technical errors during manipulations and the molecular data used as a reference for comparisons might be wrong. The disadvantage of in silicoapproaches is that algorithmic techniques are statistical and require the analysis of a population rather than a single individual. This is not a limitation in clinical trials and epidemiological surveys, which are always performed on a cohort basis. In clinical pharmacy, however, if a specific individual\'s haplotypes are of interest to predict his response to drug treatment, his unphased multilocus genotype must be combined with a standard reference set of haplotypes to infer the phasing \[[@B20]\]. This implies a thorough knowledge of the NAT2genotypic distribution in the ethnic population from which this individual was drawn. Conclusion ========== This study demonstrates that computational methods can provide an effective and accurate prediction of haplotype phases, in the particular case of the NAT2gene which displays high values of LD between polymorphic sites. The objective of this study is not to advocate the systematic use of computational approaches for NAT2haplotype inference at the expense of molecular haplotyping methods. We are convinced that these last ones remain the most reliable and effective way to resolve linkage phase patterns and that they can produce, for a fixed sample size, much more precise estimates of haplotype frequencies than other approaches \[[@B63]\]. However, the considerable effort required to obtain and analyse individual chromosomes make alternative designs preferable; and the in silicoapproach appears to be the most practical one. Thus, for researchers not willing to invest time and money in the preliminary step of NAT2haplotype reconstruction, the use of computational algorithms constitutes a safe and effective way to get reliable haplotypic data on which further analyses could be carried on. Once haplotypes are constructed, various statistical methods can be applied on NAT2haplotype data to detect allele-disease associations or to classify patients according to their acetylation status. Methods ======= NAT2 molecular data sets ------------------------ To evaluate the performance of in silicoapproaches in NAT2haplotype reconstruction, we based our study on data collected from the literature, for which linkage phase was resolved directly through molecular haplotyping. Molecular data from five previously published data sets were analysed: they concerned 258 Spanish from Central Spain \[[@B45]\], 137 Nicaraguans with a Central American Indian-European mixed origin \[[@B30]\], 112 British from the Cambridge area \[[@B24]\], 101 Black South Africans (mostly Tswana-speaking people) \[[@B24]\], and 1000 Koreans \[[@B31]\]. All subjects included in these studies were randomly selected, unrelated healthy volunteers whose ethnic origin had been clearly defined. In each population sample, seven SNPs were typed at NAT2for all individuals (no missing data), and mutation linkage phase of all multiply heterozygous individuals was resolved molecularly through allele-specific PCR and restriction mapping. A summary description of the data sets is given in Table [2](#T2){ref-type="table"}. These data provide an opportunity to compare haplotype frequencies estimated by direct gene counting on experimentally haplotyped data with haplotype frequencies estimated by haplotyping algorithms when phase information is ignored. Throughout this report, we will use the term «phase-known» to refer to the individual\'s genetic constitution for the NAT2haplotyped system, including the linkage phase of the component SNP alleles. Whereas we will use the term \"phase-unknown\" to refer to an individual\'s multilocus genotype in the absence of phase information. Computational haplotyping methods --------------------------------- We evaluated the ability of four population-based haplotype inference methods to reconstruct NAT2haplotypes from the phase-unknown genotype data. ### Hapar The first method is based on maximum parsimony: it searches for a set of minimum number of haplotypes that explain the observed genotype samples. Clark\'s method, the first developed algorithm for haplotype reconstruction \[[@B43]\] which can be viewed as a sort of parsimony approach, requires homozygote or single-site heterozygote in the sample to start its inferential cascade. Wang and Xu \[[@B46]\] overcame this limitation by designing an algorithm with a global optimization goal. This method, recently implemented in the Hapar program \[[@B46]\], was tested at its default settings on the phase-unknown NAT2data. ### PL-EM We also applied the EM algorithm \[[@B47]\] to obtain the maximum-likelihood estimates of haplotype frequencies in the samples, given the observed data \[[@B48]-[@B50]\]. This algorithm starts with initial arbitrary values of haplotype frequencies and iteratively updates the frequency estimates, to maximize the log-likelihood function, until convergence is reached. Several EM-based algorithms have been developed. We used three different implementations, Arlequin \[[@B51]\], HAPLO \[[@B49]\] and PL-EM \[[@B52]\], that all gave us identical results on comparable analyses (data not shown). Thus, we presented only the results obtained with the PL-EM program: this software implements an algorithm derived from the standard EM but incorporating the computational strategy of partition-ligation \[[@B53]\] to handle a larger number of loci. We performed 50 independent runs with different initial conditions to minimize chances of local convergence so as to ensure finding the global maximum likelihood estimates. For a given unphased genotype pattern, the probability of each possible haplotype configuration was calculated by using the estimated population haplotype frequencies, and all compatible haplotype phases with nontrivial probabilities were generated. The haplotype pair with the greatest probability was considered to be the haplotype phase for each individual, and population haplotype frequencies were estimated as a function of each inferred haplotype pair, weighted by their estimated probability. ### Haplotyper and PHASE Finally, two Bayesian statistical methods based on Gibbs sampling procedure were applied to the phase-unknown NAT2data. Such methods treat the unknown haplotypes as random quantities and combine prior information-beliefs about what sorts of patterns of haplotypes are expected to be observed in population samples- with the likelihood-the information in the observed data \[[@B54]\]. The conceptual difference between the two investigated Bayesian algorithms lies in the prior information incorporated into the statistical model. The algorithm implemented in the Haplotyper program \[[@B53]\] uses a Dirichlet prior distribution, which assumes that the genetic sequence of a mutant offspring does not depend on the progenitor sequence \[[@B54]\]. Instead, the algorithm implemented in the PHASE program \[[@B44]\] uses a prior approximating the coalescent, which is one of the evolutionary models most commonly used in population genetics (see \[[@B55]\] for a review): it assumes that unresolved haplotypes will tend to be the same as, or similar to, known haplotypes. We employed the lastest version of PHASE (PHASE v 2.1 \[[@B54]\]) to evaluate the performance of this method, using the default parameter values in the Markov chain Monte Carlo simulations. For each data set investigated, we applied the algorithm ten times with different seeds for the random number generator, and checked for consistency of the results across the independent runs in order to verify that the algorithm did not converge to a local, rather than global, mode of the posterior distribution. We chose the results from the run displaying the best average goodness-of-fit of the estimated haplotypes to the underlying coalescent model. Besides, to evaluate the Bayesian algorithm implemented in Haplotyper, we performed 20 independent runs of the program on each sample. This software could not be run directly on the 1000-Korean sample as it can only handle 500 individuals at most per data set. To circumvent this limitation, we randomly generated ten pairs of complementary data sets, each composed of 500 individuals, and we ran Haplotyper on each of them. Results were averaged over the ten complete data sets. Both programs Haplotyper and PHASE provide a list of the most likely pairs of haplotypes for each subject. They also quantify the uncertainty associated with each phase call by outputting an estimate of the probability that each call is correct. This prevents inappropriate overconfidence in statistically reconstructed haplotypes. Measures of estimation accuracy ------------------------------- Computational algorithms of haplotype reconstruction may be used for many different purposes. We focus here on three particular tasks: finding the list of all haplotypes present in a sample, inferring the most likely pair of haplotypes for each sampled individual, and estimating haplotype frequencies in the population. Thus, three different measures of accuracy were used to evaluate the performance of the tested algorithms. ### -haplotype identification To assess accuracy in terms of haplotype identification, we used the I~H~index introduced by Excoffier and Slatkin \[[@B48]\]. It compares the number of different haplotypes detected experimentally with the number of different haplotypes inferred by the computer programs. We considered that a given haplotype is identified as being present in the true sample if its estimated frequency is above the threshold value of 1/(2n) in a population sample of n individuals. I~H~is given by: ![](1471-2156-6-30-i3.gif) where k~true~is the number of haplotypes in the true sample, k~est~is the number of estimated haplotypes with frequency above the threshold, and k~missed~is the number of true haplotypes not identified in the sample. Values of I~H~can vary between 1 (when the computational identified haplotypes are exactly the same as those determined experimentally) to 0 (when none of the true haplotypes are identified computationally). ### - reconstruction of the haplotypes of each sampled individual We specified the haplotype pair for an individual by choosing the most probable haplotype pair consistent with the individual\'s multilocus genotype. We measured performance by the individual error rate, which is the proportion of individuals whose haplotype pairs were incorrectly inferred by the program \[[@B53]\]. ### - estimation of sample haplotype frequencies To examine how close the computationally estimated haplotype frequencies are to the observed frequencies in the phase-known data, we used the similarity index I~F~of Renkonen \[[@B56]\], defined as the proportion of haplotype frequencies in common between the estimated and observed frequency distributions \[[@B18],[@B48]\]. ![](1471-2156-6-30-i4.gif) where ![](1471-2156-6-30-i1.gif) and p~0i~denote, respectively, the estimated and observed sample frequency of the i-th haplotype. This measure incorporates all hhaplotype frequencies and thus captures the overall difference between estimated and observed frequencies for a particular data set. It varies between zero, when true haplotypes have estimated frequencies tending to zero, and one, when observed and estimated frequencies are identical. Since this index gives more weight to the high-frequency haplotypes, we used a second criterion to assess the accuracy of computational algorithms in haplotype frequency estimation: the change coefficient C, defined in Tishkoff et al.\[[@B57]\] as ![](1471-2156-6-30-i5.gif) where Max \[![](1471-2156-6-30-i1.gif), p~0i~\] indicates the maximum value of ![](1471-2156-6-30-i1.gif) or p~0i~. This coefficient measures the percentage change in haplotype frequencies across the two information conditions (phase-known versus phase-unknown data). Ccoefficients were computed for each possible haplotype in each population. The value of Cranges from 0 to 1, with 0 indicating that the estimated and observed frequency are identical. The maximal value of 1 indicates that molecular haplotyping showed either the presence of a haplotype that was assigned a zero through computational haplotyping, or vice versa. Measure of pairwise LD between SNP markers ------------------------------------------ We used the phase-known data to quantify the amount of LD between all pairs of polymorphic sites by computing the correlation coefficient r^2^\[[@B58]\] for each population sample separately. These statistics are expected to be 1 (perfect LD) when the variation is segregating in a population as only two distinct haplotypes. Statistical significance of LD between pairs of sites was assessed by Fisher\'s exact tests. Computations were performed with the software PowerMarker v3.21 \[[@B59]\], and a graphical summary of disequilibrium matrices was displayed by the GOLD program \[[@B60]\]. Abbreviations ============= Single nucleotide polymorphism (SNP) Expectation-Maximisation (EM) algorithm Linkage disequilibrium (LD) N-acetyltransferase 2 (NAT2) Authors\' contributions ======================= AS participated in the conception and design of the study, in collection of data, in performing all the computational analyses, and in drafting the manuscript. PD participated in the design of the study, in interpretation of data and in revising the article critically for important intellectual content. All authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:58.370975	2005-6-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173101/", "journal": "BMC Genet. 2005 Jun 2; 6:30", "authors": [ { "first": "Audrey", "last": "Sabbagh" }, { "first": "Pierre", "last": "Darlu" } ] }
PMC1173102	Background ========== The accurate replication and repair of genetic material, which is a prerequisite for normal functioning of the eukaryotic genome and the prevention of cancer, relies on coordinated and faithful DNA synthesis. One important mechanism that ensures a high fidelity of DNA replication is a \"cleansing\" of the DNA precursor pool from deoxyribonucleoside triphosphates containing a modified base \[[@B1]-[@B4]\]. Such modified bases may have ambiguous base-pairing properties that will result in a high mutagenic activity after their incorporation into DNA during replication. A classic example of the detoxification mechanism is the elimination of dUTP and 8-oxo-dGTP from the dNTP pool by the E. colidUTPase and MutT proteins, respectively \[[@B1],[@B5]\]. Purine analogs 6-hydroxyaminopurine (HAP) and 2-amino-HAP (AHA) are powerful mutagens in bacteria, yeast, and higher eukaryotes \[[@B6],[@B7]\]. It has been suggested that HAP-deoxyriboside-triphosphate (dHAPTP) is a possible endogenous contaminant of nucleotide pools under peroxyl radical stress \[[@B8]\]. HAP and AHA closely resemble the natural purines, hypoxanthine and xanthine (Fig. [1](#F1){ref-type="fig"}), and therefore, could be exploited to investigate the mechanism preventing mutations that are caused by non-canonical purine nucleotides \[[@B9]-[@B11]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Chemical structures of HAP and AHA and natural purine bases. ::: ![](1471-2156-6-31-1) ::: It was proposed that purine salvage enzymes convert base analogs to the corresponding deoxyribonucleoside triphosphates, which are misincorporated or misreplicated during DNA synthesis, resulting in induction of mutations \[[@B12],[@B13]\]. HAP-induced mutagenesis in yeast is elevated in strains with defects in proofreading activity of replicative DNA polymerases \[[@B14],[@B15]\] and does not depend on excision, mutagenic recombination, and mismatch repair systems \[[@B14]-[@B16]\]. We have described several systems protecting cells from the mutagenic and inhibitory effects of HAP (see review \[[@B16]\]). One is the novel molybdenum cofactor-dependent system in E. coli\[[@B17]\]. It has yet to be determined if a similar system exists in higher eukaryotes. Another, versatile HAP-detoxification pathway relies upon the action of triphosphatase, Ham1p, which hydrolyze HAP-containing ribo- and deoxyribo-nucleotides to nucleoside monophosphates, and which prevent incorporation of base analog into DNA and RNA. We initially described the elevated sensitivity to HAP in yeast due to mutations in the HAM1gene \[[@B18]\]. When we cloned and sequenced the HAM1gene, we found that it has homologs in many organisms, from bacteria to humans \[[@B13]\], and proposed that the gene might code for new triphosphatase \[[@B16]\]. Then, the crystal structure of the Ham1p homologue from a thermostable bacterium (protein Mj0226) was determined \[[@B19]\]. It was found that the Ham1p ternary structure has common features with MutT. Homologs of the yeast Ham1p from other organisms possessed triphosphatase activity on dITP, ITP, XTP, and dHAPTP substrates (Kozmin and Pavlov, unpublished; Burgis and Cunnigham, personal communication; and \[[@B19]-[@B21]\]). There are additional, less thoroughly studied, factors modulating purine base analogues mutagenesis in yeast (see \[[@B16]\] for review). For example, aah1mutants are sensitive to HAP, suggesting that adenine deaminase Aah1p may deaminate HAP base to hypoxanthine \[[@B16]\]. In the present study, we carried out a genome-wide search for HAP and AHA sensitive mutants. The release of several complete sets of deletion mutants by the Yeast Deletion Project provides a powerful approach for different types of genome screens in yeast \[[@B22]\]. Haploid and diploid strains have already been used to detect new genes controlling sensitivity to different agents such as UV, ionising radiation, iron, and methyl methane sulfonate (MMS) \[[@B23]-[@B27]\], as well as spontaneous mutability \[[@B28]\]. This approach, when combined with other genomics approaches, helps to establish the biological functions of uncharacterized ORFs in yeast, many of which have human orthologs. This approach also allows us to decipher the network responses to endogenous and environmental stress \[[@B29]\]. The present study is the first systematic, genome-wide search for the mutations conferring sensitivity to mutagenic purine base analogs. Results ======= Development of the screening method ----------------------------------- To develop a useful method for searching the yeast mutants sensitive to base analogs, we calibrated the experimental conditions using the wild-type strain, BY4742, and two previously described HAP-sensitive mutants, ham1and aah1(see \[[@B6],[@B16]\]), created on BY4742 background. As shown in Fig. [2](#F2){ref-type="fig"} and described in Materials and Methods, yeast were grown in a 96-well microtiter plate and then transferred, using a multiprong replicator device, to YPD plates containing base analogs. An induction of the Can^r^mutants was monitored by replica-plating to the minimal-medium plates containing L-canavanine. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Scheme of the protocol for screening the yeast deletions library for base analog sensitivity and induced mutagenesis. ::: ![](1471-2156-6-31-2) ::: The results are presented in Fig. [3](#F3){ref-type="fig"}. The left panel of Fig. [3](#F3){ref-type="fig"} shows the induction of canavanine-resistant mutants by HAP and AHA; and the right panel represents the survival of the tested strains on YPD plates in the presence of base analogs. In the wild-type strain, as in the ham1and aah1mutants, 1--3 spontaneous canavanine-resistant colonies per spot arise in the absence of mutagen (Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}, left panel). In our experimental conditions, 1 μg/ml of HAP did not induce Can^r^mutants in the wild-type strain. A moderate induction of Can^r^clones (fewer than ten per spot) was observed at 10 μg/ml of HAP and a very strong HAP-induced mutagenesis was observed in the wild-type strain at 100 μg/ml of HAP (Fig. [3A](#F3){ref-type="fig"}, left panel). For comparison, 100 μg/ml of AHA were only moderately mutagenic (compare Fig. [3A](#F3){ref-type="fig"} and Fig. [3E](#F3){ref-type="fig"}, left panel). Furthermore, both HAP and AHA did not affect the viability of the wild-type strain, even at the maximal concentration of 100 μg/ml (Fig. [3A](#F3){ref-type="fig"} and [3E](#F3){ref-type="fig"}, right panel). In the ham1strain, 1 μg/ml of HAP induced Can^r^mutants with the similar frequency that was observed in the wild-type strain at 2 orders in magnitude higher concentration of analog, 100 μg/ml (compare Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}, left panel, phenotype of hypermutability, HM). The hypersensitivity of the ham1mutant to the toxic action of HAP was clearly detectable at 100 μg/ml of HAP (Fig. [3B](#F3){ref-type="fig"}, right panel). We will refer later to this phenotype as hypersensitivity, HS. Note that a reduction of the number of canavanine-resistant clones at 10 and 100 μg/ml of HAP, in comparison with 1 μg/ml of HAP observed in the ham1mutant (Fig. [3B](#F3){ref-type="fig"}, left panel, another manifestation of hypersensitivity phenotype, HS), is also due to a dramatic decrease of survival. When the aah1mutant was tested (Fig. [3B](#F3){ref-type="fig"}, second row, for HAP and Fig. [3F](#F3){ref-type="fig"}, first row, for AHA), the drop of viability was less severe than that for the ham1mutant (phenotype of elevated sensitivity, S). HAP-induced mutagenesis was detectable at a low dose of 1 μg/ml, but was much less compared to the ham1mutant (phenotype of elevated mutability, M). Mutagenesis was somewhat stronger at the dose 10 μg/ml and was not seen at 100 μg/ml (another manifestation of elevated sensitivity, S). For the aah1mutant, AHA-induced mutagenesis was moderate at the dose of 10 μg/ml and very strong at a concentration of 100 μg/ml. The aah1mutant exemplified what we expect to observe for mutants moderately sensitive to both HAP and AHA. These data suggested that the procedure we devised for micro-titre plate format is effective for the detection of mutants with altered parameters of HAP and AHA sensitivity and mutability. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Results of the screening of the yeast deletion library for elevated mutagenesis and sensitivity in micro-titer plates. Left panel -- Mutagenesis on selective plates with canavanine. Right panel -- The estimation of the number of colony-forming units on YPD medium. ::: ![](1471-2156-6-31-3) ::: Screening of the deletion strains library ----------------------------------------- We screened the yeast deletion strains library as described above. After an initial screening of 4,823 deletion strains, 43 mutants strains appeared to affect base analog-induced killing or mutagenesis. However, in subsequent testing of candidate strains in same type of plate tests, we have confirmed HAP-sensitivity of 16 mutants (Fig. [3](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}, column 4, where we refer to phenotypes of mutants according to abbreviations described in previous section). Next, we examined the mutability and survival of these 16 mutants in quantitative tests with HAP (see Materials and Methods). Based on the results of these two types of tests, we categorize HAP-sensitive strains in three groups, as shown in Table [1](#T1){ref-type="table"}. Group I comprises, in addition to ham1and aah1strains, mutants ade12and ade2. These strains were hypersensitive to the mutagenic and lethal action of HAP in both types of tests (Fig. [3](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}, columns 4--6). Strains of this group were sensitive to the low doses (1 and/or 10 μg/ml) of HAP, in contrast to the wild-type strain. The ade12mutant was almost as mutable as the ham1strain, but the hypermutability could only be demonstrated in a quantitative test, due to poor plating efficiency (compare Fig [3B](#F3){ref-type="fig"}, row 3 and Table [1](#T1){ref-type="table"}, column 5, row 3). The ade2mutant was as mutable as the aah1mutant (Table [1](#T1){ref-type="table"}). Deletion mutants of the first group showed variable degree of sensitivity to HAP-induced killing Table [1](#T1){ref-type="table"}, column 6). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### HAP-sensitivity of the mutants selected in our screening. ::: Class of mutants Gene or ORF name deleted Functional group Response HAP in spot tests^†^ Induced mutant frequency (×10^-7^) by HAP^\#^ Survival in presence of HAP^\#^ -------------------------------------------------------- -------------------------- ------------------ ------------------------------- ----------------------------------------------- --------------------------------- Wild type \- 400 100% Class I: mutants hypersensitive to HAP HAM1 DNA^1^ HM, HS 13000 17% ADE12\* DNA HM, HS 10000 30% AAH1\* DNA M, S 5500 30% ADE2 DNA M, S 4400 70% II class: Mutants sensitive to mutagenic effect of HAP VIP1 Cell^2^ M, S 840 76% VID27 Metabolic^3^ M, S 600^°^ 82% IPK1 Metabolic M, S 1400 100% ADE5,7 DNA M 1900 100% ADE8 DNA M 1500 100% ADE6 DNA M 1100 100% RIM101 meiosis M 1100 100% ADE3 DNA M 940 100% ADE1 DNA M 860 100% YGR035c unknown M 500^°^ 100% Yjl055w\* Metabolic M 500^°^ 100% Class III: Mutants sensitive to killing YMl013c-a Unknown S 230^°^ 60% SHE4 Cell S 600^°^ 30% TRP2 Metabolic HS 250^°^ 60% ^†-^HM -- hypermutable, M -- more mutable than wild-type, HS -- hypersensitive, S -- more sensitive than wild-type, see first paragraph od the Results Section for the explanation. ^\#^-- 25 μg/ml ^1^-- \'DNA\' -- includes genes involved in the control of DNA precursor metabolism, purine salvage, DNA repair. ^2^-- \'Cell\' -- includes genes involved in cytoskeleton organization, cell walls and organelles. ^3^-- \'Metabolic\' -- includes genes involved in general metabolic pathways. \* -- These mutants were also sensitive to the mutagenic or toxic effect of AHA (see Fig. 3). ^°^-- small difference from the wild-type strain was reproducible and statistically significant by Wilcoxon-Mann-Whitney test. ::: Eleven mutants fall into Group II. Mutants of this class were sensitive to the mutagenic effect of HAP, but their growth was not severely inhibited on HAP-containing medium. As a result, there is a smaller difference, in comparison to mutants of the group I, or no difference in the number of Can^r^mutants induced by 10 and 100 μg/ml of HAP in spot-tests (Fig. [3C](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}, column 4). These strains produced some HAP-induced Can^r^mutants at 10 μg/ml of HAP, whereas the parental strain was non-mutable at this HAP concentration. We also did not detect any substantial drops of viability after growth in liquid media containing 25 μg/ml of HAP (Table [1](#T1){ref-type="table"}, column 6). Group II was not homogeneous in respect to HAP mutability and sensitivity. The six most sensitive mutants in the group are vip1, vid27, ade1, ade5,7, ade6, ade8, ipk1and rim101(see Table [1](#T1){ref-type="table"}). These mutants showed a decrease in the number of Can^r^mutants in a qualitative test when concentration of HAP increased ten-fold (up to 100 μg/ml), which is an indication of some cell death at very high doses of HAP (Fig. [3C](#F3){ref-type="fig"}). Mutants yjl055w, ygr035c, and ade3were more resistant to HAP, since the number of Can^r^colonies was similar at 10 and 100 μg/ml of HAP. As could be seen from the results of quantitative assay, mutant classification is quite conditional and there is substantial variation in responses between mutants of the Group II, but all of them were more mutable that the wild-type strain. Finally, Group III includes mutants she4, trp2, and yml013c-a;which were sensitive to HAP-induced killing, but not to HAP-induced mutagenesis (Fig. [3D](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}, column 6). In quantitative tests the yml013c-aand trp2mutants were even less sensitive to the mutagenic action of HAP than the wild-type strain (Table [1](#T1){ref-type="table"}). The she2mutant was slightly more mutable that the wild-type strain only in quantitative test. The existence of such a type of mutants suggests that the toxic effect of HAP in yeast may be not only due to the induction of lethal mutations, but also due to inhibition of certain metabolic pathways. In the present study, we have also characterized three AHA-sensitive mutants, aah1, ade12, and yjl055w(Fig. [3E](#F3){ref-type="fig"} and [3F](#F3){ref-type="fig"}). Remarkably, all of these mutants were also sensitive to HAP (Fig. [3C](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}). Two of those mutants, aah1and yjl055w, were AHA-hypermutable; whereas ade12strain did not mutate in the presence of AHA, but was sensitive to AHA-induced killing. In a quantitative test, 50 μg/ml of AHA did not inhibit survival of the wild-type strain, aah1and yjl055wstrains; whereas survival of the ade12mutant was reduced to 50%. In the aah1and yjl055wstrains, 50 μg/ml of AHA induced canavanine-resistant mutants with the frequencies 120 × 10^-7^and 270 × 10^-7^, respectively, that was 4--9-folds higher than observed in the wild-type and ade12strains (30 × 10^-7^, in both strains). Discussion ========== In this study, we elaborated the technology for testing base analog-induced mutability and killing of thousands of yeast strains in microtiter format (see Fig. [2](#F2){ref-type="fig"}). We found that the method is quite sensitive and reliable. Next, we screened the library of haploid yeast strains carrying deletions in all nonessential ORFs for the sensitivity to mutagenic base analogs HAP and AHA. We have found 16 novel HAP-sensitive mutants that fall into several groups, based on the sensitivity profiles (Fig. [3](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}). One group comprises the mutants that are HAP-hypermutable and grow poorly in the presence of HAP. Another class comprises strains with elevated HAP-mutability that grow normally on medium containing HAP. Finally, the third group includes the mutants sensitive to HAP-induced killing, but not to HAP-induced mutagenesis. We have also isolated three AHA-sensitive mutants. All of them were HAP sensitive as well. We summarized the properties of the genes found in our screening in [Additional file 1](#S1){ref-type="supplementary-material"}. One interesting result from our study is that none of the genes involved in the control of base analogs sensitivity except two genes, YML013c-a and ade12, were found in screenings for genes controlling sensitivity to the other types of mutagens, MMS, UV, and ionizing radiation \[[@B23]-[@B27]\] or for elevated spontaneous mutagenesis \[[@B28]\]. It was previously reported that deletion of the YML013c-aand ADE12open reading frame specifically enhanced sensitivity to killing (as in case of HAP) induced by γ-radiation and bleomycin, but did not affect sensitivity to MMS, UV, and hydroxyurea \[[@B23],[@B30]\]. It is known that there are overlaps of the sets of genes detected in genome screenings for the MMS-, UV- or ionizing radiation-sensitive strains \[[@B29]\]. Usually, those spectra include genes controlling DNA replication, recombination, and DNA repair. In our study, we did not find any of those genes. This is consistent with our earlier observation that the mutagenic action of HAP in S. cerevisiaedoes not depend on nucleotide excision repair, mutagenic repair, and mismatch repair \[[@B16]\]. A system to repair non-canonical purines and, probably HAP and AHA, in DNA has been recently characterized in E. coli\[[@B9],[@B10]\]. It is possible that yeast S. cerevisiaelacks such a repair system. Based on our data, we propose that, in yeast, the major base analogs protective mechanism is a control of the quality of DNA precursor pool that prevents incorporation of base analogs into DNA. This mechanism may work at several levels: transport of analogs into cells, detoxification of analogs by metabolic enzymes, maintenance of nucleotide pools, and fidelity control of DNA replication (Fig. [4](#F4){ref-type="fig"}). HAP and AHA are likely transported into the yeast cell by the same permeases, which are involved in transport of natural purines. One candidate is purine-cytosine permease, Fcy2p, a major purine (adenine, guanine, and hypoxanthine) and cytosine transporter in yeast \[[@B31]\]. According to our unpublished data, fcy2mutants are resistant to HAP. Thus, the active transport of HAP is the first critical step in the HAP mutagenic pathway. The next step is a conversion of the base analog to the corresponding ribonucleoside monophosphate by enzymes of the purine salvage pathway. Previously we observed that the inactivation of the APT1gene, encoding adenine phosphoribosyl transferase, led to a severe decrease of the mutagenic effect of HAP \[[@B16]\], suggesting that this enzyme plays a key role in the biosynthesis of HAP-riboside-5\'-monophosphate (HAPMP). HAPMP then may be converted to the corresponding nucleoside triphosphate, which could be ambiguously incorporated into DNA by DNA polymerases and provoke replication errors in the subsequent replication cycles \[[@B13]\]. The mechanism preventing HAP- and AHA-induced toxicity most likely involves the conversion of base analogs to non-mutagenic metabolites by purine salvage enzymes. The main argument for this hypothesis is that HAP could be utilized by yeast cells as a sole purine source \[[@B13],[@B18],[@B32]\]. The first enzyme in this HAP and AHA detoxification pathway is adenine aminohydrolase, encoded by the AAH1gene. Aah1p from several microorganisms has been biochemically characterized (see \[[@B33]\]). Is has a broad substrate specificity and is capable of converting both adenine and its six-substituted analogs into hypoxanthine in vitro. According to our data, yeast Aah1p may convert HAP to hypoxanthine \[[@B12]\] and AHA to guanine in vivo, since the inactivation of the AAH1leads to a defect in this conversion that is readily detected by UV spectroscopy of yeast medium (unpublished observation). In the aah1mutant, the base analog intracellular concentration increases. We propose that this causes the elevation of base analog-induced mutagenesis (Fig. [3B](#F3){ref-type="fig"} and [3E](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}). Interestingly, in our screening we did not detect the amd1mutant, deficient in AMP aminohydrolase, that catalyze deamination of AMP to IMP \[[@B34]\]. We disrupted this gene by the kanMXcassette in several yeast strains and did not see any effect on HAP-induced mutagenesis anywhere. Thus, deamination of HAP at the mononucleotide level does not play an important role in HAP detoxification. This can be due to several reasons: the inability of the Amd1p to hydrolyse HAPMP, the minor role of the AMD1gene in yeast, or the short life-time of the HAPMP in yeast cells. We have found that inactivation of adenilosuccinate synthase (ASS or Ade12p) encoded by the ADE12gene strongly enhanced HAP-induced mutagenesis and AHA-induced killing (see Fig. [3B](#F3){ref-type="fig"} and [3F](#F3){ref-type="fig"}, and Table [1](#T1){ref-type="table"}). The primary function of this enzyme is the conversion of IMP to SAMP in the pathway of AMP biosynthesis de novo(Fig. [4](#F4){ref-type="fig"}). We propose that the reason for this hypersensitivity is the dysregulation of purine biosynthesis, as follows. First, the blocking of this step of purine biosynthesis causes accumulation of IMP in the cell. The excess of IMP is probably toxic for the cell, since ade12mutants have a slow-growth phenotype that can be rescued by blocks of the earlier steps of the de novopurine biosynthesis pathway (Dr. A. M. Zekhnov {St-Petersburg State University}, personal communication). Accumulated IMP can be phosphorylated to ITP by nucleotide phosporylases. Thus, we propose that, in the ade12mutant, an excess of ITP may saturate Ham1p triphosphatase, an essential enzyme for the destruction of HAPTP and dHAPTP, which leads to increased HAP-sensitivity (see Fig. [3](#F3){ref-type="fig"}). The data obtained recently in bacteria are consistent with this hypothesis. It was shown that, in vitro, E. coliHam1p homologues protein, RdgB, is a triphosphatase that acts to hydrolize non-canonical DNA precursors, dIPT and dXTP. The Ham1p protein was shown to possess a similar activity on dITP, dXTP, and dHAPTP substrates (Kozmin and Pavlov, unpublished; Burgis and Cunnigham, personal communication; and see refs. \[[@B19]-[@B21]\]). In E. coli, the rdgBmutation is synthetically lethal with recA. As proposed, absence of RdgB leads to a dramatic increase of hypoxanthine and xanthine in DNA. Accordingly, base excision repair of such modified bases occurring in opposite strands may generate double strand breaks that require the RecA function to be repaired. However, over-expression of adenilosuccinate synthase PurA (homolog of yeast Ade12p) suppresses this lethality \[[@B9]\]. This suggests a critical role of ASS in the regulation of intracellular concentration of genotoxic hypoxanthine-containing nucleotides. We found that certain mutations affecting IMP biosynthesis de novoenhance HAP-induced mutagenesis (Table [1](#T1){ref-type="table"}). Seven of sixteen newly identified genes controlling HAP and AHA sensitivity are involved in AMP biosynthesis (see [Additional file 1](#S1){ref-type="supplementary-material"} and Fig. [3](#F3){ref-type="fig"}). It is known that the regulation of the AMP biosynthesis pathway by adenine is mediated by SAICAR, one of the precursors in adenine biosynthesis de novo\[[@B35]\]. The accumulation of certain purine biosynthesis by-products may play a role in the regulation of the nucleotide pool. A defect in endogenous purine biosynthesis probably alters nucleotide pools to favour dHAPTP mis-incorporation into DNA or leads to HAP toxicity. In respect to this hypothesis, it is important that there is a difference in the level of HAP-induced mutagenesis among the strains carrying mutations in the genes of AMP biosynthesis. The less sensitive mutant is ade3(Fig. [3](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}). The ADE3does not directly control any steps of purine biosynthesis. It encodes the C1-tetrahydrofolate synthase that provides C1-tetrahydrofolate, an indispensable precursor for AMP, histidine, thymidylate, and methionine biosynthesis (see Fig. [4](#F4){ref-type="fig"}). In this respect it is of interest that ade4, ade16and ade17mutants, also defective in IMP biosynthesis, were not found in our screen and were not sensitive to HAP or AHA when constructed de novoand tested directly (data not shown). The ADE4stands apart because PRPP, a substrate of the product of the ADE4gene, serves as a precursor for additional biosynthetic pathways. This prevents by-product accumulation and might be one of the explanations of lack of HAP sensitivity of the ade4mutants. The single ade16and ade17mutants also do not lead to byproduct accumulation because ADE16and ADE17are isozymes and the inactivation of one gene does not block the pathway. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### *Purine salvage and purine biosynthesis de novoin yeast Saccharomyces cerevisiae**. Intermediates of the purine biosynthesis de novoare designated in blue. The salvage pathway is presented in black. Genes, whose deletions lead to HAP and/or AHA sensitivity are highlighted by red boxes. The proposed conversions of HAP and AHA are represented in brackets below the adenine and guanine metabolites, respectively. Dashed arrows represent hypothetical pathways that were not demonstrated experimentally for a given substrate. Abbreviations:Purine biosynthesis de novo: PRPP -- 5-phospho-ribosyl-1α-pyrophosphate, PRA -- 5-phospho-β-D-ribosylamine, GAR -- 5-phosphoribosylglycinamide, FGAR -- 5\'-phosphoribosyl-N-formyl glycinamide, FGAM -- 5\'-phosphoribosyl-N-formylglycinamidine, AIR -- 5\'-phosphoribosyl-5-aminoimidazole, CAIR -- 5\'-phosphoribosyl-5-aminoimidazole-4-carboxylate, SAICAR -- 5\'-phosphoribosyl-4-(N-succinocarboxamide)-5-aminoimidazole, AICAR -- 5\'-phosphoribosyl-4-carboxamide-5-aminoimidazole, FAICAR -- 5\'-phosphoribosyl-4-carboxamide-5-formamidoimidazole, SAMP -- adenylosuccinate, 5,10-methylene-THF -- 5,10-methylenetetrahydrofolate, 10-formyl-THF -- 10-formyltetrahydrofolate. Salvage: Ade -- adenine, AdeRib -- adenosine, Hyp -- hypoxanthine, Gua -- guanine, Xan -- xanthine. ::: ![](1471-2156-6-31-4) ::: We have found that 6 genes detected in our screening, VIP1, VID27, IPK1, YGR035c, YML013c-aand SHE4, are putatively involved in cell organization or genetically interact with cell-cycle control genes (see [Additional file 1](#S1){ref-type="supplementary-material"}). This observation provides new perspectives on the mechanisms of base analogs-induced mutagenesis. It is possible that there is a specific structural route, mediated by cell cycle and cytoskeleton components, initiated by penetration of the analog inside the cell to its final target. We have also identified several hypothetical genes critical for the HAP and/or AHA resistance. This may be an initial clue to their functional significance. Finally, we would like to mention that out of the 18 genes we found to be involved in HAP and AHA sensitivity control, 11 (60%) have orthologs in all groups of organisms, including mammals. Therefore, the results have relevance to higher eukaryotes and humans as well (see [Additional file 1](#S1){ref-type="supplementary-material"}). Conclusion ========== We identified novel mutants sensitive to mutagenic and toxic effects of purine base analogs. AHA sensitivity was not previously described for three of the mutants identified in this study. The results reveal a complex control of base analogue mutagenesis by genes encoding the components of metabolic pathways and cytoskeleton. Methods ======= Yeast strains and media ----------------------- We have used a set of 4,823 S. cerevisiaemutants carrying deletions of all non-essential ORFs created in the haploid strain BY4742 (MATα his3Δleu2Δlys2Δura3Δ). The information about the deletion strains set is available from the Yeast Deletion Project site: <ftp://ftp.resgen.com/pub/deletions/mat_alpha_041902.txt>. Deletion strains were constructed by replacement of the ORF\'s with the kanMX4*cassette, which confers resistance to geneticin \[[@B36]\]. The standard yeast complete media (YPD) and minimal synthetic media (SD) were used \[[@B37]\]. Deletion strains were cultivated on YPD medium, supplemented with 200 μg/ml of geneticin. Sensitivity to HAP and AHA was examined on YPD media containing analogs in concentrations 100 μg/ml, 10 μg/ml and 1 μg/ml. SD medium containing 40 μg/ml of L-canavanine was used for the selection of Can^r^mutants. Base analogs sensitivity tests ------------------------------ HAP was purchased from MP Biomedicals (Irvine, California, USA). AHA was custom- synthesized by Dr. I. Kuchuk at the Department of Chemistry of Indiana University (Bloomington, Indiana, USA) by the method of Janion \[[@B38]\]. Both chemicals were dissolved in DMSO (Fisher, USA) with mild heating. A single colony of each tested strain was inoculated into a well of 96-well microtitre plate containing 200 μl of liquid YPD medium (see Fig. [2](#F2){ref-type="fig"}). Microtiter plates were incubated for 2 days at 30°C with agitation, to reach a stationary phase (approximately 10^8^cells/ml). For mutagenesis assay, cells were then plated by a multiprong replicator device (approximately 5 μl of cell suspension per prong) to the YPD plates containing various concentrations of HAP or AHA, as shown on Fig. [2](#F2){ref-type="fig"}. After one day of incubation, the plates were replica-plated on SD minimal-medium plates containing L-canavanine. The plates were incubated 5 days and inspected for induction of Can^r^mutants. For the survival test, cell cultures from the microplates were diluted in water in series of 96-well microplates, using a multichanel pipette (see Fig. [2](#F2){ref-type="fig"}). Diluted cells suspensions were plated to YPD plates containing HAP or AHA by a multiprong replicator device. After 2--3 days of incubation, the number of colonies was recorded. Strains that produced smaller colonies or a smaller number of colonies on the HAP-containing medium, relatively to wild-type strain, were classified as sensitive to killing. Quantitative assay of the base analog-induced mutagenesis --------------------------------------------------------- For each strain and each concentration of base analogs to be tested, we prepared six independent cultures by inoculating a single colony into 1 ml of liquid YPD medium with or without mutagens. After two days incubation in the roller drum, the mutant frequencies were determined by plating of appropriately diluted cell suspensions on minimal-medium SD plates supplemented with L-canavanine (to determine the number of canavanine-resistant cells per culture), and on YPD plates (to obtain the total number of cells per culture). Then the frequency of mutants was calculated as described \[[@B14]\]. Each experiment was repeated at least three times. We have used several doses of HAP and found that in this type of test the most reproducible results are obtained at dose 25 μg/ml. The statistical significance of differences between variants was estimated by Wilcoxon-Mann-Whitney nonparametric criterion. Authors\' contributions ======================= Youri Pavlov and Elena Stepchenkova designed this study. Elena Stepchenkova performed the experimental work and wrote the initial draft of the manuscript. Vladimir Alenin, Youri Pavlov, Stanislav Kozmin, and Elena Stepchenkova analyzed the data set and wrote the final version of the paper. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Annotation of the genes whose deletion results in HAP and AHA sensitivity ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We are grateful to Polina Shcherbakova for comments during this study and for critical reading of the manuscript. This work was supported in part, by NE DHHS LB506 grant \#9934 for YIP, and by the CRDF grant provided by The Russian Ministry of Education \# ST-012-0.	PubMed Central	2024-06-05T03:55:58.375806	2005-6-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173102/", "journal": "BMC Genet. 2005 Jun 2; 6:31", "authors": [ { "first": "Elena I", "last": "Stepchenkova" }, { "first": "Stanislav G", "last": "Kozmin" }, { "first": "Vladimir V", "last": "Alenin" }, { "first": "Youri I", "last": "Pavlov" } ] }
PMC1173103	Background ========== The antigenic regions of protein recognized by the binding sites of immunoglobulin molecules are called B-cell epitopes \[[@B1]\]. These epitopes can be classified into two categories; i) conformational/discontinuous epitope, where residues are distantly separated in the sequence and brought into physical proximity by protein folding and, ii) linear/continuous epitope, comprised of a single continuous stretch of amino acids within a protein sequence that can react with anti-protein antibodies \[[@B2],[@B3]\]. Most of the B-cell epitopes were thought to be discontinuous. However, in late 1980s it was shown that this conformational restriction is not a necessary condition for the production of protein-reactive anti-peptide antibodies \[[@B4]\]. The designing of the conformational epitopes is difficult and so experimental B-cell epitopes largely include linear epitopes. These linear epitopes can be exploited in the development of synthetic vaccines and disease diagnosis. A number of vaccines based on B-cell epitopes are currently under clinical phase trials against viruses \[[@B5]\], bacteria \[[@B6]\] and cancer \[[@B7]\]. These epitopes are also important for allergy research and in determining cross-reactivity of IgE-type epitopes of allergens \[[@B8]\]. A large number of B-cell epitopes have been reported in the literature in last two decades. There is a need to collect and compile these epitopes to evaluate the performance of existing B-cell epitope prediction methods and to further develop better methods \[[@B9],[@B10]\]. We have observed that the performance of the existing physico-chemical scales is not very high \[[@B11]\]. Recently, Blythe and Flower examined 484 amino acid propensity scales in predicting of B-cell epitopes and found that even the best set of scales and parameters performed only marginally better than random \[[@B12]\]. The evaluation of the existing scales indicates that there is a need to develop better methods by using artificial intelligence techniques. The collection of B-cell epitopes should help in deriving new scales for accurate in silico prediction of linear epitopes. Also it will help the immunologist to understand the complex nature of immunogenic peptides and for the development of vaccines. There are many databases available on T-cell epitopes \[[@B13]-[@B15]\]. In contrast, there are limited number of databases on B-cell epitope for example JenPep \[[@B16],[@B17]\] and HIVDB \[[@B18]\]. Recently, JenPep has been superseded by AntiJen 2.0, which has included peptides bound to MHC ligand, TCR-MHC Complexes, T cell epitope, TAP, B cell epitope molecules and immunological protein-protein interactions. AntiJen also contains peptide library, copy numbers and diffusion coefficient data. Though AntiJen provides information about different types of peptides (\>24000 entries) from a single source but it provides limited information about B-cell epitopes and tools to analyze and retrieve the data. Recently, we have created a comprehensive database, Bcipep, of B-cell epitopes. Latest version of Bcipep contains 3031 entries, where each entry provides detailed description of a B-cell epitope. Currently, we have covered only the continuous B-cell epitopes. The aim of this database is to assist the scientific community working in the areas of synthetic peptide vaccines (based on B-cell epitopes) and allergy research. The database will complement the existing databases such as AntiJen \[[@B16],[@B17]\]. Availability ------------ The database is available at <http://www.imtech.res.in/raghava/bcipep>, <http://bioinformatics.uams.edu/mirror/bcipep/> (Mirror Site) and <http://srs.ebi.ac.uk/srs6bin/cgi-bin/wgetz?-page+LibInfo+-id+1X2XW1JU5_L+-lib+BCIPEP> (SRS version). Database construction ===================== The PostgreSQL relational database management system (RDBMS) has been used for storing, retrieval and managing the data. The scripts, which provide interface between user and database, were written in PERL, CGIPerl and Pgperl. B-cell epitopes were collected from the literature (PubMed, <http://www.ncbi.nlm.nih.gov/pubmed/>; ScienceDirect, <http://www.sciencedirect.com/>). A large number of HIV epitopes were extracted from a book \[[@B18]\]. Database description ==================== The aim of Bcipep database is to provide; i) comprehensive information about B-cell epitopes, ii) tools for extraction and analysis of this information and, iii) hyperlinks to related databases. The overall architecture of Bcipep database is shown in Figure [1](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A schematic representation of Bcipep database. ::: ![](1471-2164-6-79-1) ::: Database information ==================== General ------- This database provides compehensive information about the linear B-cell epitopes which includes; i) amino acid sequence of epitope, ii) source protein from which epitopes were obtained, iii) experimental methods used in accessing the immunogenic potential of epitopes, iv) pathogen group (e.g., bacteria, virus, fungi, protozoa) of source protein and, v) miscellaneous information in the \'Comment\'field. Immunogenicity and model organisms ---------------------------------- The immunogenicity of a peptide is a semi-quantitative measure of its immunogenic potency. In Bcipep it is divided into three categories; i) immunodominant, if it increases 2--3 folds anti-peptide antibodies in comparison to reference or control (carrier protein, eg., BSA or KLH), ii) immunogenic, if it enhances anti-peptide antibodies by one-fold in comparison to reference, and iii) null-immunonogenic, where no difference was observed when compared to reference. This information is very important for developing B-cell epitope prediction method. The database also provides information about \'Model organism\'used for immunization. Antibodies and neutralization ----------------------------- The database provides full information about monoclonal or polyclonal antibodies produced against an epitope. The information includes, isotypes of immunoglobin and name/number of monoclonal antibodies. The database also contains information about neutralization potential of anti-peptide antibody, which is crucial for considering a peptide for synthetic vaccine design. Links to databases ------------------ The Bcipep provides hyperlinks to various sequence databases in order to provide detailed information about peptides in database. The \'database reference\' field consists name/code of protein available in SWISS-PROT \[[@B19]\]. The \'Antigenic structure\' field consists of PDB codes \[[@B20]\] of protein structures having matching peptides. These PDB codes are linked to OCA browser <http://pdb.tau.ac.il/> in order to provide detailed structure information of these proteins. It also provides structure information about 242 antibodies, where each antibody is hyperlinked to PDB database through the OCA browser. The \'Publication reference\' field provides full information about related publications with link to PUBMED \[[@B21]\]. Bcipep is also linked to MHCBN database \[[@B15]\] in order to identify the peptides that are B-cell as well as T-cell epitopes. Web tools ========= Bcipep has following major web-based tools for retrieval and analysis of information in Bcipep database (Figure [1](#F1){ref-type="fig"}). These web tools have been designed to facilitate the user in retrieving information from database. Keyword search -------------- This option allows users to perform search on all fields of the database (\'Peptide Sequence\', \'Source Protein\', \'Publication Reference\', \'Database Reference\'). One can restrict the keyword search on any specific field. It also allows users to select the fields to be displayed. An example of keyword search is shown in Figure [2a](#F2){ref-type="fig"}, where key word \'P26694\' is searched in any filed of database. The output/result of this keyword search is shown in Figure [2b](#F2){ref-type="fig"}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The typical display of Bcipep database for keyword search; a) input page of keyword search; and b) output of keyword search. ::: ![](1471-2164-6-79-2) ::: Peptide search -------------- The database provides option to search a peptide in Bcipep. The tool will display full information about the peptides included in Bcipep. The server also permits users to search their query sequence in any pathogen group. Search can be restricted on the basis of immunogencity that is immunodominant, immunogenic or null-immunogenic. An example of input and output of peptide search is shown in Figures [3a](#F3){ref-type="fig"} and [3b](#F3){ref-type="fig"} respectively. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### This example illustrate peptide search on Bcipep; a) Peptide search page; and b) result of peptide search. ::: ![](1471-2164-6-79-3) ::: Mapping of T-cell epitopes -------------------------- This server allows searching of peptide in Bcipep against MHCBN database. The MHCBN database provides information about components of cell-mediated immunity like MHC binders/non-binders, T-cell epitopes and TAP binders \[[@B14]\]. The peptides related to cell-mediated immunity can be mapped on resultant B-cell epitopes obtained from keyword/peptide search by clicking on \'Peptide Sequence\'field (Figures [2b](#F2){ref-type="fig"} and [3b](#F3){ref-type="fig"}). Thus, the server is useful in identifying the potential B-cell epitopes having T-cell epitopes (or MHC binders). The example of mapping of MHCBN peptides on B-cell epitope (by clicking on peptides sequence field of Figure [3b](#F3){ref-type="fig"}) is shown in Figure [4a](#F4){ref-type="fig"}. The full information of each map peptide can be obtained by clicking on the mapped sequence. One such example is shown in Figure [4a](#F4){ref-type="fig"} and [4b](#F4){ref-type="fig"}. The mapping allows user to detect the regions in B-cell epitope having promiscuous MHC binders (peptides that can bind to large number of MHC alleles) or T-cell epitopes. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Mapping of peptide in MHCBN database; a) mapping of MHCBN peptides on B-cell epitope, and b) full information about a MHCBN peptide. ::: ![](1471-2164-6-79-4) ::: Peptide mapping --------------- The peptides of Bcipep can be mapped on query sequence using this option. The full information about mapped peptide can be obtained by clicking on it. The tool will assist the researchers in gaining knowledge about the known immunogenic or non-immunogenic regions in target protein of interest. The example input and output of peptide mapping is shown in Figures [5a](#F5){ref-type="fig"} and [5b](#F5){ref-type="fig"} respectively. The users can specify the pathogen group and/or immunogenicity level of peptides to be mapped on query sequence. As shown in Figure [5b](#F5){ref-type="fig"}, the graphical mapping of peptides/epitopes on query allows one to easily detect the regions that bind to large number of B-cell peptides. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Mapping of B-cell epitopes on antigen sequence, a) submission page of B-cell epitope mapping, and b) mapping results ::: ![](1471-2164-6-79-5) ::: BLAST search ------------ This tool allows users to search their query protein against antigenic proteins maintained at Bcipep. The sequence of 1070 antigenic proteins has been obtained from SWISS-PROT. The similarity search is performed using the GWBLAST server <http://www.imtech.res.in/raghava/gwblast/>. The GWBLAST also allows users to analyze the BLAST output like multiple alignments, phylogenetic analysis. Online data submission ====================== The database has the facility to submit data to Bcipep on-line via Internet. Users can submit information about their experimentally determined B-cell epitopes. We hope that immunologists will submit their information on B-cell epitope in Bcipep, similar to the sequence data at GenBank and SWISS-PROT. The Bcipep team will add more data from literature to maintain B-cell data up-to-date and cross check the data submitted by users to maintain quality of the data. Potential utility and limitations ================================= One of the major challenges in the field of subunit vaccine design is to identify the antigenic regions (B and T cell epitopes) that can generate antigen specific memory cells. Thus, the identification of regions/stretches on an antigen from the data pool of known epitopes is an important step in vaccine design. The Bcipep database would be very useful as it consists of comprehensive information about experimentally verified linear B-cell epitopes and tools for mapping these epitopes on an antigen sequence. In case query antigen contains known epitopes, this database might aid in the wet experimentation and lower the cost by reducing the overlapping repeats. This strategy is frequently used for the screening of transgenic proteins by searching linear IgE-binding epitopes \[[@B22]\]. Unlike discontinuous epitopes, the linear epitopes are easy to design, as they do not require tertiary structure information. Bcipep also provides information on neutralizing B-cell epitopes where an antibody generated against a B-cell epitope neutralizes the parent antigen. The current version of Bcipep provides neutralizing information on about 1309 such B-cell epitopes. This information is very important for selecting functional B-cell epitopes. This database also provides a link with MHCBN to search for overlapping regions of MHC binders and T-cell epitopes in the B-cell epitopes. Thus, the user can identify both antigenic regions that can activate B-cell and T-cell, which can lead to the development of better vaccine. The epitopes in Bcipep can be used to derive the rules for predicting B-cell epitopes. The aim of designing synthetic linear peptides as epitope-vaccine is to induce neutralizing antibodies against the pathogen \[[@B23]\]. There are many reports that the linear B-cell epitopes were characterized as neutralizing antibodies as in Clostridium botulinum neurotoxin type A (Btx A)\[[@B24]\]. In Bcipep, there are 748 neutralizing anti-peptide antibodies entries. However, in some cases these linear epitope(s) fail to produce neutralizing antibodies and do not give protective immunity. For instance, it has been shown in the past that the antibodies against the synthetic peptides and short recombinant proteins of approximately 100 amino acids of hepatitis E virus (HEV) do not neutralize, suggesting that the HEV neutralization epitope(s) is conformation dependent \[[@B25]\]. The elicitation of a bactericidal and protective immune response to Borrelia burgdoferidecorin binding protein requires a properly folded conformation for the production of functional antibodies \[[@B26]\]. Recently, Corcoran et al, 2004, observed that B-cell memory is established and maintained against conformational epitopes of Parvovirus VP2 and against linear epitopes of VP1 but not against linear epitope VP2 \[[@B27]\]. Thus, it is not necessary that the linear B-cell epitope will always give rise to memory cells. One should also check the neutralizing information of B-epitopes, as only 748 B-cell epitopes out of 1309 in Bcipep were able to neutralize the parent protein. For an effective use of Bcipep, it is important to understand the limitation of linear B-cell epitopes and data in Bcipep. The few limitations of current version of Bcipep are; i) it does not cover discontinuous epitopes, ii) it has limited number of unique peptides (1590) in 3031 entries and, iii) it contains peptides having only natural amino acids. One should be careful in using linear B-cell epitopes in developing epitope based subunit vaccine. The organism used for immunization (information included in the database) should also be taken into consideration, since immune response is T-helper cell (MHC-II-peptide complex) dependant and B-cell epitope alone may not generate protective antibodies \[[@B28]\]. In some cases, the nature of the adjuvant used and the route of immunization (information not included in the database) might also play important roles in the induction of protective anti-peptide antibody response against the pathogen \[[@B29]-[@B31]\]. Authors\' contributions ======================= SS collected and compiled the data as well as developed the web server. MB helped in designing website and stored data in PostgreSQL. GPSR conceived the idea and supervised the work. Acknowledgements ================ Authors are thankful to Dr Grish C. Varshney for critically reading the manuscript. We are also thankful to Council of Scientific and Industrial Research (CSIR) and Department of Biotechnology (DBT), Govt. of India, for financial assistance.	PubMed Central	2024-06-05T03:55:58.378766	2005-5-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173103/", "journal": "BMC Genomics. 2005 May 29; 6:79", "authors": [ { "first": "Sudipto", "last": "Saha" }, { "first": "Manoj", "last": "Bhasin" }, { "first": "Gajendra PS", "last": "Raghava" } ] }
PMC1173104	Background ========== Dinoflagellates play critical roles in marine ecosystems as primary producers and grazers of other bacterial and eukaryotic plankton \[[@B1]\]. Approximately one-half of the ca. 4,000 species of dinoflagellates contain plastids, although many are mixotrophic \[[@B2]\]. Many taxa produce potent toxins and form harmful algal blooms, or \"red tides\", resulting from populations of more than 20 million cells per liter of seawater. The toxins cause a variety of poisonings that affect humans and marine wildlife \[[@B1]\] and have a significant impact on coastal ecosystems throughout the world \[[@B3]\]. Yet, other dinoflagellates, like Symbodinium, are central contributors to the health of reef ecosystems as the symbionts of corals \[[@B4]\]. Loss of the dinoflagellate symbiont results in coral bleaching. In addition to their ecological role, dinoflagellates display some fascinating and unique aspects of cell biology. One intriguing character is nuclear biology. The nucleus of dinoflagellates is unlike that of any other eukaryote because the chromosomes are condensed throughout the cell cycle except during DNA replication \[[@B5]\]. The morphologically similar chromosomes are attached to the nuclear envelope and can number in the hundreds \[[@B6]\]. Dinoflagellates also lack nucleosomes \[[@B7]\], instead the nuclear DNA is associated with basic proteins that are moderately similar to bacterial histone-like proteins (HLPs \[[@B8],[@B9]\]). Dinoflagellates were thought to lack histones \[[@B10]\], but in a recent gene expression study, a putative histone H3 was annotated in Pyrocystis lunula, although the sequence was not analyzed further \[[@B11]\]. The general lack of nucleosomes raises many questions about transcription and gene regulation in these organisms. Dinoflagellate nuclei also contain vast amounts of DNA compared to other eukaryotes. Estimates range from 3 -- 250 pg·cell^-1^, or approximately 3,000 -- 215,000 megabases (MB) \[[@B12]\]. In comparison, human nuclei contain 3.2 pg·cell^-1^(3,180 MB). The dinoflagellate nucleus contains such a high concentration of DNA that it exists in a liquid crystal state, which is responsible for the unique morphology \[[@B13],[@B14]\]. The DNA to basic protein ratio of dinoflagellate chromosomes has been estimated to be 10:1, which is dramatically higher than the 1:1 ratio observed in most eukaryotes. This indicates that very little basic protein is associated with dinoflagellate chromosomes and that the crystal structure is the primary cause of the unusual morphology. Dinoflagellates are also the only eukaryotes to contain hydroxymethyluracil, a deaminated nucleotide that can be produced by oxidative damage of DNA, which replaces 12 -- 70% of the thymidine \[[@B15]\]. The role of polyploidy or potentially, genome amplification within particular life history stages remains to be clarified for dinoflagellates. It is highly unlikely, however, given their relatively simple morphology that the immense DNA content is explained solely by gene content. The most widespread plastid in dinoflagellates contains the unique photopigment peridinin. The \"peridinin plastid\" is remarkably different from this organelle in other eukaryotes because it lacks a typical genome. Plastids normally contain a circular genome of about 150 kb that encodes 100 -- 200 genes that are necessary for plastid function. In peridinin-containing dinoflagellates, the plastid genome has been broken into minicircles that encode a single, or a few genes per circle. However, only 16 genes have been identified thus far on minicircles \[[@B16],[@B17]\]. Recent studies show that most of the plastid genes have been transferred to the nucleus \[[@B18],[@B19]\] with 15 of these genes found exclusively on the plastid genome in all other photosynthetic eukaryotes \[[@B18]\]. The peridinin dinoflagellates encode therefore the smallest number of plastid genes of any photosynthetic eukaryote, making them a model for understanding organellar gene transfer. Nuclear-encoded plastid proteins are targeted to the plastid using a tripartite N-terminal targeting signal \[[@B20]\]. As in Euglena, nuclear-encoded plastid proteins are co-translationally inserted into the endoplasmic reticulum and embedded in this membrane using a stop-transfer sequence in the N-terminus. Through algal endosymbioses, the dinoflagellates have been able to acquire four other types of plastids from distantly related evolutionary lineages including the haptophytes, cryptophytes, diatoms, and prasinophytes \[[@B1],[@B21]\]. This aspect of their evolutionary history highlights the unmatched ability of dinoflagellates to capture and retain foreign plastids. Alexandrium tamarenseis one of the best-studied dinoflagellates. This species forms toxic blooms and causes paralytic shellfish poisoning through saxitoxin production. It has a peridinin-containing plastid and in North America, A. tamarenseblooms from Alaska to Southern California in the Pacific and along the Canadian and New England coasts in the Atlantic. There has been a recent increase in blooms of A. tamarenseand other Alexandriumspecies in other parts of the world making this genus of high importance to the world\'s fisheries. We undertook a gene discovery project with this organism using expressed sequence tag (EST) data to investigate dinoflagellate evolution and to create a genomic resource for scientists working on different aspects of A. tamarenseand dinoflagellate biology. The EST method was the most reasonable approach in this case because haploid A. tamarensecells contain approximately 143 chromosomes and have a genome size of 200 pg/cell (ca. 200,000 Mb \[Erdner and Anderson unpublished data\]). Our EST results comprise the first extensive high-throughput, genome-wide data set for a dinoflagellate. Results and discussion ====================== Clustering and sequence analyses -------------------------------- The collection of 11,171 ESTs comprised of single-pass 3\'-reads (483 from the start library and 10,688 from the normalized library) from A. tamarensewas assembled into 6,723 clusters. The normalized library showed a high degree of complexity, with a novelty rate of 60.18% and about 52% of the sequenced clones contained inserts that were longer than the single sequence read (ca. 750 bp). Clustering of the total EST set showed that most of the reads were singletons (4,618 sequences) and the largest cluster was comprised of 46 ESTs that are closely related to HLPs (Table [1](#T1){ref-type="table"}). Other highly represented transcripts were those encoding luciferin-binding protein (a protein involved in the regulation of bioluminescence) and several photosynthetic proteins (e.g., Rubisco, ATP synthase C chain, light harvesting proteins). Several large clusters were transcripts that lacked a similarity (e-value \< e^-5^) to known proteins. One of these ESTs has an open reading frame that encodes a protein with a potential plastid-targeting signal (Figure [1A](#F1){ref-type="fig"}). Interestingly, database searches against NCBI\'s nr and dbEST returned hits only to other dinoflagellate ESTs. Another of the largest clusters only had hits to ESTs from other dinoflagellates (Figure [1B](#F1){ref-type="fig"}). These two proteins are therefore candidates for dinoflagellate-specific proteins. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Cluster size and frequency of the A. tamarenseESTs. ::: Cluster Size Frequency Cluster Size Frequency Best BLAST hit(s) -------------- ----------- -------------- ----------- ---------------------------------------------------------------- 1 4618 14 7 2 1249 15 1 unknown 3 427 16 1 HSP90 4 176 17 4 peridinin-chl a protein, Cytochrome C6, EF1-alpha, unknown 5 81 18 1 ATP synthase C chain 6 44 19 2 Form II Rubisco, unknown putative dino. specific protein 7 32 21 1 fucoxanthin chlorophyll a/c binding protein like 8 15 22 1 Unknown putative plastid protein 9 21 23 1 Unknown 10 13 24 3 peridinin-chlorophyll a protein, ATP synthase C chain, unknown 11 10 29 1 luciferin-binding protein 12 7 46 1 histone-like protein/basic nuclear protein 13 6 ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Putative dinoflagellate-specific proteins.Amino acid sequence alignments of putative dinoflagellate-specific proteins. A) putative plastid protein that was highly represented in the A. tamarensecDNA library (cluster size = 22). A. tamarensesequences 1, 2, and 3 correspond to clones GC1-aba-e-13, GC1-abh-e14, and GC1-abd-o-22, respectively, and are aligned with highly similar ESTs from the dinoflagellates L. polyedrum(CD809498) and P. lunula(BU582532). The boxed region indicated a possible plastid targeting sequence. B) Putative dinoflagellate specific protein with significant blast hits only to other dinoflagellate ESTs. The Alexandriumsequence corresponds to clone UI-D-GC1-abh-f-23-0-UI. ::: ![](1471-2164-6-80-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### *GO category assignment of A. tamarenseESTs.Classification of 1,203 A. tamarenseESTs into the GO categories. ::: ![](1471-2164-6-80-2) ::: Each cluster was searched against the SwissProt protein database using blastx. A total of 515 hits with an e-value less than 1e-20 were identified that terminated within 10 amino acids of the end of the SwissProt entry. From these hits, we estimated that the 3\'-UTRs ranged in length from 25 -- 620 nt with a mean length of 155 nt. This is shorter than the average length observed for fungi (\~200 nt) and metazoans (300--600 nt) \[[@B22]\]. However, this analysis is likely to be an underestimate of the average 3\'-UTR length because only ESTs that were sequenced into the coding region were included in the analysis. The 3\'-UTRs of A. tamarensecDNAs are also interesting because of their apparent lack of a polyA signal. Both simple n-mer searches (e.g. hexamer, pentamer) and the Gibb\'s sampler were used to assay the canonical region from -11 to -30 preceding the polyadenylation site in search of a polyadenylation signal. We were unable to find a single or a related set of hexamers or pentamers that are enriched in the 3\'-UTRs (data not shown). Clearly, polyadenylation of transcripts occurs in A. tamarense, however, the mechanism by which this process takes place apparently does not involve a typical polyA signal. These ESTs were also analyzed for GC-content and codon usage. Coding region GC-content was 60.8%, whereas GC-content in the 3\'-UTR was slightly less at 57.6%. The GC-content is reflected in the codon usage (Table [2](#T2){ref-type="table"}), whereby 3^rd^positions are strongly biased towards Gs or Cs. The stop codon TGA is also significantly favoured over TAG and TAA (frequencies of 411, 71, and 25 occurrences, respectively). The accession numbers of SwissProt hits with an e-value of 9e^-10^and below (1,292 sequences) were submitted to the ProToGo server for GO category assignment \[[@B23]\]. A total of 1,203 of the SwissProt accession numbers could be assigned to GO categories. The results are summarized in Figure [2](#F2){ref-type="fig"}. The functional distribution of the A. tamarenseESTs that could be placed among GO categories is typical of other eukaryotes. However, the overall small number (i.e., 20%) of significant hits to GenBank is surprising, suggesting that many A. tamarenseproteins may be either highly diverged and/or encode novel dinoflagellate-specific functions (e.g., Figure [1](#F1){ref-type="fig"}), or the sequence does not extend into the coding region of the transcript. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Codon Usage in the A. tamarenseESTs. ::: ------- ------ -------- ------- ------ ------- -------- ------ ------- -------- ------ -------- TTT F 703 23.1% TCT S 482 10.0% TAT Y 372 18.8% TGT C 251 15.6% TTC F 2335 76.9% TCC S 1348 27.9% TAC Y 1612 81.2% TGC C 1356 84.4% TTA L 61 0.9% TCA S 413 8.6% TAA \ 29 5.6% TGA \* 411 79.8% TTG L 1118 15.7% TCG S 926 19.2% TAG \* 75 14.6% TGG W 1051 100.0% CTT L 902 12.7% CCT P 751 17.6% CAT H 464 25.7% CGT R 475 9.6% CTC L 2296 32.3% CCC P 1382 32.4% CAC H 1340 74.1% CGC R 1779 35.8% CTA L 139 2.0% CCA P 829 19.4% CAA Q 433 14.6% CGA R 426 8.6% CTG L 2596 36.5% CCG P 1307 30.6% CAG Q 2535 85.4% CGG R 1128 22.7% ATT I 715 19.1% ACT T 542 13.1% AAT N 508 21.0% AGT S 344 7.1% ATC I 2770 74.1% ACC T 1442 34.9% AAC N 1915 79.0% AGC S 1310 27.2% ATA I 253 6.8% ACA T 638 15.4% AAA K 415 8.5% AGA R 253 5.1% ATG M 2096 100.0% ACG T 1510 36.5% AAG K 4485 91.5% AGG R 910 18.3% GTT V 686 11.2% GCT A 1195 15.2% GAT D 1117 24.9% GGT G 943 13.8% GTC V 2214 37.8% GCC A 2899 36.8% GAC D 3371 75.1% GGC G 3957 58.1% GTA V 268 4.6% GCA A 1559 19.8% GAA E 750 13.8% GGA G 767 11.3% GTG V 2694 46.0% GCG A 2218 28.2% GAG E 4682 86.2% GGG G 1142 16.8% ------- ------ -------- ------- ------ ------- -------- ------ ------- -------- ------ -------- Analysis is of 515 proteins (81,893 codons). Third position nucleotide usage was T = 12.8%, A = 9.3%, C = 40.7%, G = 37.2%. The asterisk (\) indicates a stop codon. ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Analyses of A. tamarensehistone H2A.X.A) Alignment of A. tamarenseH2A.X with eukaryotic homologs. The alignment is shaded according to the level of conservation. The symbols above the alignment indicate the location of functional residues (T = trypsin cleavage site, \^ = arginines that contact the DNA helix, \ = H2A-H2B interaction sites, U = ubiquitination site). The annotation below the alignment indicates conserved structural features including the α-helices, loops, and the SQ(E/D)Φgotif. B) A ML tree of H2A and H2A.X. The numbers above and below the branches are the results of ML and NJ bootstrap analyses, respectively. The thick branches indicate \> 0.95 posterior probability from Bayesian inference. Only bootstrap values ≥ 50% are shown. Branch lengths are proportional to the number of substitutions per site (see scale bar). ::: ![](1471-2164-6-80-3) ::: Dinoflagellate gene content and gene families --------------------------------------------- Of species with sequenced genomes, the apicomplexan Plasmodium falciparumis the most closely related organism to A. tamarense. Both of these species are members of the alveolate lineage with dinoflagellates and apicomplexans forming a monophyletic clade that is sister to the ciliates (e.g., \[[@B23]\]). Sequence comparisons using BLAST revealed that 609 of the 6723 A. tamarenseESTs had a significant hit (e-value less than 1e^-10^) to P. falciparumproteins. The top 20 most significant hits are shown in Table [3](#T3){ref-type="table"}. The most highly conserved proteins between these organisms include many \"housekeeping\" proteins such as α-tubulin and heat shock protein 70. Despite their close evolutionary relationship, there are however likely to be substantial differences between A. tamarenseand P. falciparumwith respect to gene content. Due to the apicomplexan intracellular lifestyle, P. falciparumhas lost most of the genes related to plastid function as well as other metabolic genes. Many of these same proteins appear in the list of the top BLAST hits against the nr database of GenBank (Table [4](#T4){ref-type="table"}). There were 1,349 hits to the nr database that were better than 1e^-10^. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Top 20 A. tamarenseEST blast hits against the genome of the apicomplexan P. falciparum. ::: **A. tamarenseEST E-Value GI Number Protein Description** ------------------------ ------------- --------------- ----------------------------------------------------------- UI-D-GC1-aao-m-13-0-UI 1.00E-112 23613558 α-tubulin UI-D-GC1-aav-f-09-0-UI 6.00E-86 23508137 flavoprotein subunit of succinate dehydrogenase UI-D-GC1-aad-d-15-0-UI 9.00E-86 23509363 serine/threonine protein phosphatase UI-D-GC0-aae-b-08-0-UI 2.00E-85 23509135 actin UI-D-GC1-aaz-h-12-0-UI 3.00E-85 23507885 26S proteasome regulatory subunit 4 UI-D-GC1-abe-o-23-0-UI 8.00E-85 23510155 bifunctional dihydrofolate reductase-thymidylate synthase UI-D-GC1-abh-e-16-0-UI 1.00E-84 23612827 hsp70 UI-D-GC0-aae-p-02-0-UI 2.00E-84 23613232 adenosylhomocysteinase UI-D-GC1-aay-i-10-0-UI 3.00E-82 16804988 helicase UI-D-GC1-aau-b-16-0-UI 1.00E-80 23509325 eukaryotic translation initiation factor 2 gamma subunit UI-D-GC0-aae-h-03-0-UI 8.00E-78 23509820 glyceraldehyde-3-phosphate dehydrogenase UI-D-GC1-aao-o-20-0-UI 4.00E-77 23508006 ADP ribosylation factor 1 UI-D-GC0-aae-f-01-0-UI 1.00E-76 23509545 calmodulin UI-D-GC1-abb-n-18-0-UI 2.00E-76 23510206 eukaryotic initiation factor UI-D-GC1-abf-g-07-0-UI 4.00E-75 23612467 HSP86 UI-D-GC1-abd-m-07-0-UI 2.00E-74 23612587 40S ribosomal protein S5 UI-D-GC0-aae-b-08-0-UI 3.00E-74 23509345 actin II UI-D-GC1-aab-m-24-0-UI 4.00E-74 23509670 ribosomal protein S2 UI-D-GC1-aar-f-11-0-UI 3.00E-72 23509852 protein serine/threonine phosphatase UI-D-GC1-aao-b-16-0-UI 1.00E-69 23509877 RNA helicase 1 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Top 20 hits of the A. tamarenseESTs to the GenBank nr database. ::: **A. tamarenseEST E-Value GI Number Protein Description Organism** ------------------------ ------------- --------------- ------------------------------------------------------------- -------------------------- UI-D-GC1-abg-i-22-0-UI 1.00E-110 845405 ribulose 1,5-bisphosphate carboxylase Gonyaulax polyedra UI-D-GC1-aao-m-13-0-UI 1.00E-109 135433 alpha tubulin Oxytricha granulifera UI-D-GC1-abh-e-16-0-UI 2.00E-98 20143982 hsp70 Crypthecodinium cohnii UI-D-GC1-abe-o-23-0-UI 1.00E-96 1169423 bifunctional dihydrofolate reductase-thymidylate synthase Arabidopsis thaliana UI-D-GC0-aae-p-02-0-UI 1.00E-91 4416330 S-adenosyl-homocysteine hydrolase like protein Alexandrium fundyense UI-D-GC0-aae-h-11-0-UI 2.00E-91 21913167 oxygen evolving enhancer 1 precursor Heterocapsa triquetra UI-D-GC1-abh-d-23-0-UI 4.00E-91 32307578 glutamate 1-semialdehyde 2,1-aminomutase Bigelowiella natans UI-D-GC1-abe-e-15-0-UI 1.00E-89 27450753 proliferating cell nuclear antigen Pyrocystis lunula UI-D-GC1-abb-n-18-0-UI 3.00E-88 28277876 Similar to DEAD box polypeptide 48 Danio rerio UI-D-GC1-aav-f-09-0-UI 3.00E-87 15240075 succinate dehydrogenase flavoprotein subunit, mitochondrial Arabidopsis thaliana UI-D-GC1-abc-o-16-0-UI 8.00E-85 13560096 ALA dehydratase Gonyaulax polyedra UI-D-GC1-aao-o-20-0-UI 1.00E-83 7025460 ADP ribosylation factor 1 Toxoplasma gondii UI-D-GC0-aae-b-23-0-UI 5.00E-83 1076185 luciferin-binding protein Gonyaulax polyedra UI-D-GC1-aay-i-10-0-UI 9.00E-83 18416493 DEAD/DEAH box helicase, putative Arabidopsis thaliana UI-D-GC1-aau-b-16-0-UI 1.00E-82 4503507 eukaryotic translation initiation factor 2, subunit 3 gamma Homo sapiens UI-D-GC1-aad-d-15-0-UI 5.00E-81 1346753 Serine/threonine protein phosphatase PP1 isozyme 2 Acetabularia cliftonii UI-D-GC1-aaz-h-12-0-UI 1.00E-77 23507885 26S proteasome regulatory subunit 4, putative Plasmodium falciparum UI-D-GC1-abc-m-19-0-UI 1.00E-77 32307576 geranyl-geranyl reductase Bigelowiella natans UI-D-GC1-abj-e-13-0-UI 2.00E-76 4033509 Calmodulin Tetrahymena pyriformis UI-D-GC1-abd-m-07-0-UI 3.00E-75 6831665 40S ribosomal protein S5 Cicer arietinum ::: As previously mentioned, our bioinformatic analyses identified 6,723 clusters of unique genes. However, this is likely to be a conservative estimate of the number of unique transcripts that were sequenced. A combination of short 3\'-UTRs and highly conserved coding regions caused many related transcripts to be assembled together, even though their 3\'-UTRs contained sequence differences. For example, two large clusters comprise ESTs that correspond to the plastid atpH gene that encodes the ATP synthase C chain. This gene is normally plastid encoded in other photosynthetic eukaryotes. These two clusters form closely related, but clearly distinct sets of transcripts. An additional atpH-encoding transcript was identified by a single EST. Together, the three clusters contain 43 ESTs, 16 of which are unique. The N-terminal extensions, which encode the tripartite plastid-targeting signals, share an average 74.3% nucleotide and 68.6% amino acid identity, respectively. Similar to many other species, the dinoflagellate transit peptides appear to be under selection to maintain hydrophobicity rather than a conserved amino acid sequence. This may explain why the nucleotide conservation is greater than that of the encoded amino acids. Five hydrophobic amino acids (phenylalanine, leucine, isoleucine, methionine, and valine) are, for example, encoded by codons with a T in the second position. This combined with the high GC-content at third positions results in higher conservation at second and third positions than at first positions. In addition, the high proportion of alanine (28.6%), leucine (10.2%), and valine (11.8%) rather than phenylalanine (2.4%), isoleucine (3.6%), methionine (4.3%, excluding starting methionine), and tyrosine (0.3%) in the N-terminal extensions may reflect the underlying GC-richness, because alanine, leucine, and valine are encoded by GC-rich codons. It is unclear if these amino acids are evolutionarily selected for specifically, or if they are selected for the combination of their hydrophobic character and the GC-content of their codons. In contrast, the conserved core of the protein shared an average 88.4% nucleotide and 98% amino acid identity, respectively, which corresponds to the more typical pattern of third position variation resulting from selection. The 3\'-UTRs of the atpH genes show substantial variation and were difficult to align. There are several groups of more closely related 3\'-UTRs that may be the result of recently duplicated genes. In all, there are five alignable groups of UTRs (and one singleton) that may have originated from more closely related genes. Histone and histone-like proteins in dinoflagellates ---------------------------------------------------- A significant finding of this study is the identification of two rare (2/11,171) ESTs that encode a partial histone H2A.X. The longest cDNA isolated from the library using PCR was predicted to encoded a protein of 169 amino acids that shares high sequence identity to eukaryotic histone H2A.X (Figure [3A](#F3){ref-type="fig"}). This clone putatively lacked only the start codon at the N-terminus. The divergent N-terminus of A. tamarenseH2A.X is somewhat longer than in other homologs but the remainder of the sequence is conserved (in particular the α-helices of the histone fold). Several functional residues from the known crystal structure are also present in A. tamarenseH2A.X including the lysine at the trypsin cleavage site, the arginines in the loops that interact with the DNA α-helix, and the lysine ubiquitination site \[[@B24]\]. The sites of interaction with histone H2B are also present. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Analysis of dinoflagellate HLPs.A) HLPs from dinoflagellates (red taxa names) and bacteria (blue) and HU proteins from bacteria (black). B) The predicted secondary structure of HLPs from A. tamarenseand B. pertussisaligned with the known secondary structure of E. coliHU. Curled lines indicate α-helices and jagged lines indicate β-strands. The arrow indicates the position of a conserved lysine. The asterisk indicates the proline that intercalates into the DNA in HU proteins. C) An ML tree of HU and HLP proteins from bacteria and eukaryotes. The numbers above and below the branches result from ML and NJ bootstrap analyses, respectively. The thick branches indicate \> 0.95 posterior probability from Bayesian inference. Only bootstrap values ≥ 50% are shown. Branch lengths are proportional to the number of substitutions per site (see scale bar). ::: ![](1471-2164-6-80-4) ::: H2A.X proteins are closely related to the canonical H2A except for the C-terminus which contains the distinctive SQ(E/D)Φ motif (where Φ is a hydrophobic residue). H2A.X plays an important role in the recognition and repair of double-strand DNA breaks by non-homologous end-joining. At the site of double-strand breaks, the serine of the SQ(E/D)Φ motif is rapidly phosphorylated \[[@B25]\]. The phosphorylation signal spreads a large distance down the chromosome around the breaks, signalling the recruitment of the DNA repair proteins Rad50, Rad51, and BRCA1 \[[@B26],[@B27]\]. We also identified histone H2A and H2A.X from the haptophyte Emilania huxleyithrough high-throughput EST sequencing of this alga (J. D. H. and D. B. unpublished data). Phylogenetic analysis places A. tamarenseH2A.X in its predicted position (with moderate bootstrap support) as sister to the E. huxleyihomolog within a group of chromalveolates that includes haptophytes, stramenopiles, and apicomplexans (Figure [3B](#F3){ref-type="fig"}). H2A.X from A. tamarense, E. huxleyi, and Toxoplasma gondiido not, however, form a monophyletic group suggesting multiple origins within chromalveolates. This is not surprising because H2A.X appears to have arisen independently many times during eukaryotic evolution \[[@B28],[@B29]\]. We tested the strength of these results using the Approximately Unbiased (AU-) statistical test. A 16-taxon ML backbone tree was generated without A. tamarenseH2A.X and then we made a set of 17-taxon trees by placing this sequence on every possible branch (29 in total). This analysis provides good support for the position shown in Figure [3B](#F3){ref-type="fig"} (P = 0.827), however, many alternative positions were included in the 5% confidence set of trees (i.e., as sister to Thalassiosira pseudonana, Phaeodactylum tricornutum, Homo sapiens, or Drosophila melanogaster, and at the base of or sister to either of the land plants). The lack of robust phylogenetic signal for the divergence point of A. tamarenseH2A.X likely reflects the short length and high conservation of these histones. Dinoflagellate chromosomes do not contain nucleosomes, instead the DNA is associated with HLPs \[[@B10],[@B30],[@B31]\]. The similarity between dinoflagellate HLPs and bacterial HU and HLPs has only recently been noted and these proteins have not yet been subjected to phylogenetic analysis with a broad taxon sampling \[[@B32]\]. In our A. tamarenseEST data, HLPs were the most highly represented transcripts (45/11,171 ESTs) and encoded 5 closely related proteins. Alignment of the HLPs from A. tamarenseand other dinoflagellates with HLPs and HU proteins from bacteria and eukaryotes showed moderate sequence similarity (a representative alignment is shown in Figure [4A](#F4){ref-type="fig"}). This alignment was constructed using information from secondary structure predictions (discussed below). One group of proteins (referred to here as bacterial HLPs) is more closely related to dinoflagellate HLPs and includes Bph2 from Bordetella pertussis. Bph2 has a role in virulence gene expression and shares limited (likely convergent) sequence similarity with histone H1 \[[@B33]\]. The dinoflagellate and bacterial HLPs also contain an N-terminal extension in comparison to HU proteins. This extension is rich in alanine, lysine, and proline, which is reminiscent of the C-terminus of histone H1. The dinoflagellate HLP N-termini are however, also enriched in methionines. Compared to the bacterial HLPs, this N-terminal region is generally shorter in the dinoflagellates, although there is variability among species in both groups (Figure [4A](#F4){ref-type="fig"}). In contrast to the primary sequence, secondary structure predictions for these three classes of proteins are remarkably similar. The crystal structure of E. coliHU has been determined (PBD ID: 1MUL) and the known secondary structure was compared to the predicted secondary structures of B. pertussisBph2 and an A. tamarenseHLP (Figure [4B](#F4){ref-type="fig"}). Both types of HLPs are predicted to have two α-helices that are identical in size and spacing to the N-terminal helices in E. coliHU, followed by two β-strands that are similar in size and position. We conclude from this analysis that dinoflagellate HLPs show structural similarity to HU proteins from bacteria, however, it is unclear if these proteins are functional homologs. It is also apparent that dinoflagellate HLPs are distantly related to bacterial HU proteins. The dinoflagellates have one putatively homologous functional residue corresponding to Lys3 (arrow in Figure [4A](#F4){ref-type="fig"}) of HU proteins, which interacts with the DNA and is involved in wrapping the DNA around the protein \[[@B34]\]. A proline residue (asterisk in Figure [4A](#F4){ref-type="fig"}), which intercalates into the DNA during HU binding, appears to be conserved among HU proteins and bacterial HLPs, but is not present in the dinoflagellate HLPs \[[@B35]\]. However, there are several prolines conserved among dinoflagellates in the C-terminal end of the protein. The C-terminal arms of HU are critical for the interactions that bend the DNA. Given the low level of sequence similarity and the absence of a homologous proline in this region, it is unclear if the dinoflagellate HLPs are able to interact with DNA in the same manner as HU proteins. In our phylogenetic analyses, the proteobacterial HLPs form a well-supported monophyletic group with the dinoflagellates (Figure [4C](#F4){ref-type="fig"}) suggesting an origin of the dinoflagellate gene through lateral transfer (followed by several rounds of gene duplication). It is also noteworthy that dinoflagellates are the only eukaryotes to possess a proteobacterial form II rubisco \[[@B36]\]. The position of the dinoflagellate HLPs is distinct from that of other eukaryotic HU proteins. These latter proteins group with the canonical HU proteins from bacteria and have likely originated through intracellular transfer from the mitochondrial or plastid endosymbiont. Statistical support for the monophyly of the dinoflagellate and proteobacterial HLPs was tested using the AU-test. In these analyses (details not shown), a sister group relationship between the HLPs was the most highly favored topology (P = 0.659) and all other positions for the dinoflagellates (except branching inside the bacterial HLP clade) had significantly lower probabilities (P \< 0.05). Dinoflagellates no longer use the nucleosome as the major DNA packaging protein complex. Chromosomal DNA strands in these taxa are smooth, in contrast to the \"beads on a string\" conformation in other eukaryotes \[[@B12]\]. The chromosome structure is also unique in that they are uniform in size and morphology, remain condensed throughout the cell cycle, and are birefringent, indicating a liquid crystal state \[[@B5],[@B14],[@B37]\]. Transcription is thought to take place in DNA loops that protrude from the condensed chromosome \[[@B38]\]. It appears that dinoflagellates have acquired DNA binding proteins from a proteobacterium possibly to facilitate the compaction of their immense genomes. HU and related proteins from bacteria induce sharp bends in DNA strands and some models suggest that HLPs are responsible for creating DNA bends at the periphery of the chromosomes \[[@B39],[@B40]\]. Immunolocalization shows dinoflagellate HLP to be associated with the periphery of chromosomes \[[@B41]\]. However, the HLP concentration is very low relative to DNA content. Dinoflagellate chromosomes have a 1:10 protein:DNA ratio (in contrast to the 1:1 ratio in other eukaryotes). The HLP concentration may therefore be too low to function in DNA compaction, rather they may act as transcriptional regulators \[[@B41],[@B42]\]. In summary, our discovery of H2A.X in A. tamarenseshows that, whereas dinoflagellates appear to no longer use nucleosomes for DNA packaging, at least one histone has been retained and is weakly expressed. Interestingly, in a recent paper, histone H3 appears in a table of redox-regulated genes in the dinoflagellate Pyrocystis lunula\[[@B11]\]. Until now, only these two histones have been identified in dinoflagellates and it is unclear if all dinoflagellates possess either of these two genes, or others that have not yet been found. If other histones are present (which is likely), they may however also be expressed at a low level (as is the case for A. tamarenseH2A.X). This would render difficult their identification using the EST-based approach unless comprehensive sequencing of normalized and subtracted cDNA libraries is used. In metazoans, replication-dependant canonical histone (H2A, H2B, H3 and H4) mRNAs are not polyadenylated, raising the possibility that they have been excluded from this poly-A primed cDNA library \[[@B43]\]. However, these histone mRNAs are polyadenylated in plants, apicomplexans, and ciliates, suggesting that if they are present, they may be in dinoflagellates as well \[[@B44]-[@B46]\]. Given the important role that H2A.X plays in DNA repair, we speculate that this gene may have been maintained specifically to perform this function. Consistent with this idea, the core region of A. tamarenseH2A.X is highly conserved, indicating that it may still be able to interact with DNA in a manner similar to H2A in other species. Conclusion ========== This collection of ESTs is the most extensive genomic resource for a toxic dinoflagellate species to date and provides a useful glimpse into its nuclear genome. These data will be instrumental to future research to understand the unique and complex cell biology of these organisms and for understanding the method of toxin production in these species. We have likely not yet exhausted the gene discovery potential using the EST approach (i.e., note the high discovery rate of our normalized library). In the future, we will use serial subtraction of cDNA libraries to improve/maintain the novelty rate of our cDNA library and create cDNA libraries from A. tamarenseunder various growth conditions and life history stages to get generate a more complete catalog of the gene content of this important organism. Methods ======= Library construction -------------------- Total RNA from a culture of the toxic dinoflagellate Alexandrium tamarense(CCMP 1598) was extracted using Trizol (GibcoBRL) and mRNA purified using the Oligotex mRNA Midi Kit (Qiagen). This culture strain was produced by isolating a single cyst, a diploid resting stage that produces haploid vegetative cells by meiosis. However, it is unknown if a single or multiple vegetative cells were isolated after antibiotic treatment to make the culture axenic. If a single vegetative cell was isolated, the culture would be clonal. The culture was grown at 20°C on a 13:11 hour light:dark cycle (80 μEinsteins of light) in L1 media. Start and normalized directionally cloned (3\' NotI-5\'EcoR1) cDNA libraries were constructed as previously described \[[@B47]\]. ESTs were sequenced from the 3\' end to maximize clustering accuracy using the 3\' untranslated region (UTR). All ESTs were processed as previously described \[[@B48]\]. Identification of a total of a non-redundant \"unigene\" set of 6,723 unique clusters from 11,171 sequences was accomplished using using UIcluster v3.0.5 \[[@B49]\]. Phylogenetic analyses --------------------- Data was gathered from GenBank (including the recently released Karenia brevisEST data, Frances Van Dolah, unpublished data) using blast searches. Maximum likelihood (ML) analyses were done with PHYLIP using the JTT model of protein evolution with gamma corrected rates (JTT + Γ) with 5 random additions \[[@B50]\]. ML bootstrap analyses (100 replications) were done as described with either 5 (histone H2A) or 1 (HLPs) rounds of random taxon addition. Bayesian analyses were done using MrBayes V3.0b4 \[[@B51]\]. Four chains (1 cold, 3 heated) were run for 1 million generations, sampled every 1000 generations, using the JTT + Γ model. The first 500 trees were discarded as burn-in. Neighbor joining (NJ) bootstrap (500 replicates) analyses were done with PHYLIP using the JTT + Γ model. Minimum evolution (ME) analyses done with PHYLIP using the JTT + Γ model with global rearrangements and 10 rounds of random taxon addition (1 round was used in the bootstrap analysis). The Approximately Unbiased test was done using CONSEL \[[@B52]\]. ML trees without the groups of interest were generated as described above. A pool of trees was then generated by adding the group of interest (A. tamarenseH2A.X or dinoflagellate HLPs) to every possible branch in the ML tree. For the HLP analyses, a reduced taxon set was used that included Bordetella, Ralstonia, Xylella, Pasteurella, Nostoc, Synechocystis, Agrobacterium, Rikettsia, Escherichia, Guillardia, Cyanidioschyzon, Sorghum, Toxoplasma, Xenopus, and Homo. A. tamarense1 and C. cohniiHCC2 were added as a monophyletic group to every branch in this reduced ML tree. Secondary structure prediction was done using Jpred \[53, 54\]. The consensus secondary structures were used in the comparison to the know structure of E. coli HU (PDB ID: 1MUL). Authors\' contributions ======================= JDH constructed the cDNA libraries and did the sequence and phylogenetic analyses, the Blast and GO analyses on the EST dataset, the histone and HLP analyses, and drafted the manuscript. TES did many of the other global sequence analyses of the EST dataset. HSY contributed intellectually to the manuscript. Library construction and high-throughput EST sequencing was done in the laboratory of MBS and was supervised by MFB. The bioinformatics group led by TLC did the EST sequence processing and clustering. DB conceived of and supervised this study and contributed to the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ JDH was supported by an Institutional NRSA (T 32 GM98629) from the National Institutes of Health. This work was supported by grants from the National Science Foundation awarded to DB (DEB 01-07754, MCB 02-36631). TES was partially supported by a Career Development Award from Research to Prevent Blindness.	PubMed Central	2024-06-05T03:55:58.380180	2005-5-29	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173104/", "journal": "BMC Genomics. 2005 May 29; 6:80", "authors": [ { "first": "Jeremiah D", "last": "Hackett" }, { "first": "Todd E", "last": "Scheetz" }, { "first": "Hwan Su", "last": "Yoon" }, { "first": "Marcelo B", "last": "Soares" }, { "first": "Maria F", "last": "Bonaldo" }, { "first": "Thomas L", "last": "Casavant" }, { "first": "Debashish", "last": "Bhattacharya" } ] }
PMC1173105	Background ========== Measurement of foot pressure distribution (FPD) is clinically useful because it can identify anatomical foot deformities \[[@B1]\], guide the diagnosis and treatment of gait disorders and falls, as well lead to strategies for preventing pressure ulcers in diabetes. Age-related anatomical and physiological changes in foot bone and ligament structure affect FPD during gait \[[@B1]\]. Gait analysis of healthy elderly people has revealed decreased stride length, reduced step force and increased variability in gait parameters. These findings indicated that unsteadiness during walking is increased in the community-dwelling elderly people, posing a risk for falls \[[@B2]\]. Age was independently associated with lower pressure under the heel, midfoot, and hallux in the multivariate analysis \[[@B3]\]. Foot pressure studies during walking have focused on specific pathology and deformity \[[@B4]-[@B6]\] specific anatomical areas \[[@B7]\], exercise \[[@B8]\] and younger subjects \[[@B9]-[@B11]\]. Knowledge of the plantar FPD map during normal walking in healthy elderly people is lacking. It is not known if distribution of plantar pressure, force, and load across several anatomical regions of the foot during walking is different between young and old. To determine the effects of normal aging on FPD, we evaluated the anatomical distribution of plantar pressure, force, and relative loads in healthy young and old subjects during walking. Methods ======= Subjects -------- The study was performed in the Syncope and Falls in the Elderly (SAFE) Laboratory at the Beth Israel Deaconess Medical Center, and all subjects signed informed consent approved by the Institutional Review Board. Subject characteristics are shown in the Table [1](#T1){ref-type="table"}: 9 young subjects (5 men and 4 women; mean age 30 ± 5.2 years, mean ± SD, range 23--39 years), and 6 elderly subjects (4 men and 2 women, mean age 68.7 ± 4.8, range 63--74 years). All subjects were screened with a detailed medical history, physical activity questionnaire, electrocardiogram, and were not treated for any systemic disease. Inclusion criteria were: age range 20--40 or 60--80 years, ability to walk a flight of stairs, and ability to walk for 12 minutes. All subjects were normotensive by medical history, confirmed by measurements of the sitting and standing heart rate and blood pressure. Height and body mass were measured and used in FPD analysis. Shoe size and foot imprints were taken to determine the insole size. Exclusion criteria were: neurological or musculoskeletal abnormalities affecting gait, peripheral neuropathies, diabetes, past hip/leg/foot trauma or surgery, dementia, epilepsy, alcohol, smoking and other drug abuse, current pregnancy, history or physical evidence for coronary artery disease, abnormal electrocardiogram (ST-T wave changes, old myocardial infarction, arrhythmias, bundle branch block), history of more than one fall or syncope in the past year, orthostatic hypotension, systemic disease requiring continuous medical treatment for \>1 month, malignant neoplasms, hepatic, renal, heart disease or failure, hypertension, and body mass index (BMI) \>35. The arch indexes of the subjects were not measured. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics ::: Demographics Young Old P --------------------------------- ------------- ------------- ---------- Age (years) 30 ± 5.2 68.7 ± 4.4 \<0.0001 Men/Women 4/5 2/4 Mass (kg) 72.3 ± 7.5 70.3 ± 13.6 NS Height (cm) 175 ± 9.5 169.9 ± 7.8 NS BMI 23.2 ± 4.6 24.2 ± 3.1 NS NWS (m/s) 1.25 ± 0.25 1.2 ± 0.2 NS Stride (s) 1.1 ± 0.2 1.04 ± 0.1 NS Step (s) 0.6 ± 0.1 0.6 ± 0.1 NS Stance (s) 0.6 ± 0.1 0.6 ± 0.1 NS Swing (s) 0.5 ± 0.1 0.4 ± 0.04 NS Rate Perceived Exertion (RPE) 2.5 ± 1.4 4.2 ± 1.2 0.004 Δ RPE Beginning-End 0.22 ± 0.7 0.1 ± 0.4 NS Physical Activity Questionnaire 4.8 ± 1.1 4.9 ± 1.1 NS Mean ± SD NWS -- normal walking speed BMI -- body mass index ::: Experimental protocol and data acquisition ------------------------------------------ Foot pressure distribution was measured during walking on the treadmill at individual normal walking speed. A treadmill was used to ensure a consistent speed, despite of its artificial milieu, as plantar pressure and force vary at different gait speeds \[[@B6]\]. Self-paced normal walking speed (NWS) was determined with hallway walking for 12 minutes. During walking, the subject was followed by the investigator for safety and timing purposes. Total distance (m) was divided by the walking time to determine individual NWS. Gait characteristics at NWS (step, stride, stance and percentage of the initial and terminal double stance) were not different between the groups (Table [1](#T1){ref-type="table"}). The treadmill walk started at 0.8 mph for all subjects and increased 0.2 mph every 30 seconds until NWS was achieved. All subjects walked at their NWS for 6 minutes. FPD analysis was done on 100 steps (50 left and 50 right) selected from steady data segments after 2 minutes of sustained treadmill walking. At the end of walking, the treadmill decelerated over 1-minute and stopped. Subjects rated perceived exertion using 10-point Borg Rating Perceived Exertion scale \[[@B12]\] before the treadmill walk started, at the end of the speed increase period, and at the end of normal walking. Subjects sat on a chair while heart rate and blood pressure were monitored for six additional minutes during the post-walk period. Foot pressure distribution was measured using the shoe insoles with 99 capacitive sensors, connected to a small portable data acquisition device that sampled pressure for each sensor at 50 Hz (Pedar Mobile, Novel Electronics Inc., GmbH Munich, Germany). The insoles were calibrated regularly using a \'Trublu\' calibration device (Novel GmbH, Munich, Germany). Two insole sizes (size WW = European shoe size 40/41, size XW = European 41/42) were used to account for differences in foot sizes. To control for differences in personal footwear, all subjects were provided with a standard, thin pair of slippers. Data analysis ------------- All data were visually inspected prior to analysis to assure high quality of data acquisition. Figure [1A](#F1){ref-type="fig"} shows distribution of maximum pressure for one step for all sensors. Time-series pressure measurements for all sensors were grouped into nine anatomical masks \[[@B5],[@B13],[@B14]\] (Figure [1B](#F1){ref-type="fig"}). These masks corresponded to the following anatomical areas: medial calcaneus, lateral calcaneus, medial arch, lateral arch, first metatarsal, metatarsals two and three, metatarsals four and five, hallux, and toes. The following 5 variables were calculated for the each mask: maximum pressure, maximum force, mean pressure, mean force, and relative load. All variables were calculated for each step and then averaged over the 50 steps for each foot. Maximum pressure was defined as the greatest pressure any single sensor in each mask measured in a single step, and these values were averaged separately for each mask over 50 steps. Mean pressure was defined as the average of all activated sensors in a mask for a single step. To calculate maximum and mean forces, the pressure time-series data were converted to force by multiplying each pressure value with the cross-sectional area of the corresponding sensor. All sensors in a defined mask were added together for each time frame to give the summed time-series for force, which was the total force for each mask. The maximum force was defined as the greatest force exerted for each mask in a single step. The mean force was defined as the average force exerted in each mask for a single step. Body weight was significantly different between men and women (p \< 0.0001). All variables were normalized by body weight (BW) and the area of each mask, to account for these factors. Relative load was defined as the ratio of the total force in a specific mask to the total force of all masks combined, expressed as a percentage \[[@B5]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Foot pressure distribution. A. Maximum pressure distribution on all sensors during stance for one subject. B. The nine anatomical masks superimposed on the insole (MC = medial calcaneus, LC = lateral calcaneus, MA = medial arch, LA = lateral arch, MT1 = first metatarse, 3 = second and third metatarse, 4 = fourth and fifth metatarse, H = hallux, and T = toes). ::: ![](1471-2318-5-8-1) ::: Statistical analysis -------------------- The maximum and mean pressure and force were compared between the groups for all masks. In addition, we compared the mean and maximum pressure and force in the medial masks (medial calcaneus, medial arch, first metatarsal, and hallux) to the lateral masks (lateral calcaneus, lateral arch, second and third metatarsal, and toes) and between the groups. We also compared the anterior masks (hallux, toes, first metatarsal, second and third metatarsal, and fourth and fifth metatarsals) to the posterior masks (medial arch, lateral arch, medial calcaneus, and lateral calcaneus) and the anterior masks (hallux and first metatarsal) similarly. Within group comparisons between masks were done using nonparametric ANOVA (Friedman test). Comparisons between groups were done using Wilcoxon nonparametric test using JMP 5.0.1 software (SAS Institute 2003). Results ======= Foot pressure distribution was highly significantly different between masks for the young and old groups for all variables (maximum and mean pressures p \< 0.00001, normalized maximum and mean pressures \<0.00001 and maximum and mean force p \< 0.00001 and normalized maximum and mean force p \< 0.00001). Differences in the foot pressure distribution between the young and old groups for the maximum and mean pressures were confined to the calcaneus region and to the medial masks of the foot. Figure [2](#F2){ref-type="fig"} shows differences in maximum pressure distribution (normalized for body weight (BW)) for all 9 anatomical regions. Elderly subjects had lower normalized maximum pressure in the medial (3.2 ± 0.5 vs. 4.6 ± 1.1 %BW, p = 0.001) and lateral (2.8 ± 0.6 vs. 3.4 ± 0.8 %BW, p = 0.036) calcaneal masks, for all medial masks combined (2.6 ± 0.2 vs. 3.3 ± 0.2 %BW/cm^2^, p = 0.019) and marginally reduced in the hallux mask (2.5 ± 1.2 vs. 3.7 ± 1.6%BW/cm^2^, p = 0.07). In the medial calcaneus region, the elderly group also had lower maximum pressure (22.2 ± 6.3 vs. 32.9 ± 11.8 N/cm^2^, p = 0.01), as well as lower mean (6.2 ± 1.8 vs. 8.9 ± 3.1 N/cm^2^, p = 0.01) and normalized mean pressure (0.9 ± 0.2 vs. 1.2 ± 0.2 %BW/cm^2^, p = 0.0006) (Figure [3A](#F3){ref-type="fig"}). The elderly subjects had also lower maximum force (240.9 ± 77.9 vs. 328.4 ± 138.7 N, p = 0.049) and normalized maximum force (34.3 ± 6.2 vs. 45.6 ± 8.5 %BW, p = 0.001) and the mean force (126.3 ± 34.9 vs. 178.8 ± 71.0 N, p = 0.02) and the normalized mean forces (18.3 ± 3.5 vs. 24.6 ± 4.2 %BW/ cm^2^, p = 0.0006) (Figure [3B](#F3){ref-type="fig"}). In the medial region masks, the elderly group had reduced maximum pressure (17.7 ± 1.6 vs. 23.2 ± 1.4 N/cm^2^, p = 0.02), reduced normalized maximum pressure (2.6 ± 0.2 vs.3.3 ± 2.0 %BW, p = 0.019); and reduced normalized mean pressure (0.5 ± 0.1 vs. 0.7 ± 0.1 % BW, p = 0.037) and borderline mean pressure (3.5 ± 0.5 vs. 4.8 ± 0.4 N/cm^2^, p = 0.06). In all anterior region masks, the elderly group exerted lower mean pressure (4.1 ± 2.3 vs. 4.8 ± 2.2 N/cm^2^, p = 0.046), borderline reduced mean force (57.3 ± 36.0 vs. 67.2 ± 33.6 N, p = 0.052) and borderline reduced normalized mean force (8.2 ± 4.5 vs. 9.8 ± 5.0 %BW, p = 0.055). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Pressure distribution by anatomical region. Normalized maximum pressure distribution for the young (white bar) and elderly (black bar) group for each anatomical region (medial calcalneus mask p = 0.0001, lateral calcaneus mask p = 0.03). ::: ![](1471-2318-5-8-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Medial calcaneus mask. A. The average maximum and mean pressures, and normalized mean and maximum pressures for medial calcaneus mask for young and old subjects (\\p = 0.01, \\\* normalized maximum pressure p = 0.001, \\\* normalized mean pressure p = 0.0006). B. The average mean and maximum forces and normalized mean and maximum forces for medial calcaneus for the young and old groups (mean ± SD, maximum force \* p = 0.05 \\\* normalized maximumn force p = 0.001, mean force \* p = 0.02, \\\* normalized mean force p = 0.0006). ::: ![](1471-2318-5-8-3) ::: In the anterior medial region (hallux and first metatarsal), the elderly group displayed reduced mean pressure (3.6 ± 2.2 vs. 4.7 ± 2.1 N/cm^2^, p = 0.03), reduced normalized mean pressure (0.5 ± 0.03 vs. 0.7 ± 0.3 % BW, p = 0.027). The relative load in the medial masks (11.0 ± 8.8 vs.12.0 ± 7.8%) and the lateral masks (11.2 ± 6.8 vs.10.4 ± 6.2%) were not different. The relative load over the medial and lateral arches was not different between groups. Elderly subjects lower medial pressure values compared to young subjects, indicates that older people had tendency for greater weight bearing on the lateral mask relative to young subjects. The arch height, has not been measured, however, the contact mid-foot area was not different between the groups. Rate perceived exertion was higher in the older subjects at the end of treadmill speed increment (beginning of normal walk) (p \< 0.01), as well as at the end of normal treadmill walking (p \< 0.004). However, the difference in perceived exertion between the beginning and end of normal walk was not significant between groups. There was no significant difference between groups in the physical activity questionnaire on the scale 0--10 (Table [1](#T1){ref-type="table"}). Discussion ========== Our study has shown that elderly people exert less pressure and force under the medial masks of the foot (medial calcaneus, hallux, anterior and posterior medial masks) during heel touch and toe-off phase. This implicates that elderly subjects preferentially bear weight on the lateral foot during normal walking. Lateralization of foot pressure suggested that medial weight bearing from heel-strike to toe-off is limited in older people compared to younger subjects. Well-distributed weight bearing and foot pressure compensate for the forces and heavy loads imposed on the foot during normal walking. Treadmill walking is different than normal walking due to an inability to change speed voluntarily and reduced stride variability. Although an artificial pace and walking environment are imposed by use of a treadmill, it was a tool used to maintain experimental control. Because foot pressure distribution is affected by walking speed and stride variability, \[[@B6]\] it was deemed necessary to control the speed using treadmill walking. In the posterior masks, the older subjects exerted lower maximum pressure and force on the calcaneus region when normalized for body weight, indicating that, along with the results above, maximum pressure at heel strike is also lower in old subjects than in young subjects. These findings may indicate that forces needed to stabilize the ankle during heel touch phase are reduced in older people. In the anterior masks, the elderly subjects also exerted lower normalized mean pressure and lower normalized mean force compared to young subjects. These findings, supported by results in the hallux mask, support the notion that old subjects have lesser ability to push-off in anticipation of the swing phase. Walking may present a challenge to elderly people, and several age-related gait changes have been identified \[[@B2],[@B3]\]. Morag et al. 1997\[[@B3]\] found that age correlated with heel pad stiffness, but to a lesser degree with walking speed, soft tissue characteristics, and height of the medial longitudinal arch. Our study did not confirm an age-related arch-flattening phenomenon to the extent of altered FPD, as forces in the medial arch area were not different between groups. Normal walking speed, stride intervals, timing within the gait cycle and the relative load in the arch area were similar between young and old subjects. These findings rebuke the notion that age-related decline in pressure is due to flattening of the longitudinal arch or that the stride length would be the primary factor underlying the reduced pressure and forces at heel strike. Anatomical foot structure, including soft tissue thickness and arch height, account for 35% of plantar pressure differences during gait \[[@B6]\]. Pressure values under the heel and midfoot are predominantly affected by weight bearing at the heel strike and midstance, whereas pressures in the anterior regions are determined to a greater extent by flexibility, muscle strength, and muscle recruitment \[[@B5]\]. Therefore, age-related soft tissue and bony structure degradation may reduce the capability of the plantar foot to deflect load \[[@B15]\]. FPD pattern in older people was similar to the pattern of experimentally reduced plantar sensation by cooling, emphasizing that decline in proprioception with aging may contribute to these results \[[@B11]\]. Older runners exhibited significantly more knee flexion at heel strike, but the range of motion and peak maximal vertical forces were reduced. The ground impact force and the initial rate of loading at heel strike were greater, indicating loss of shock absorbing capacity in older people \[[@B8]\]. Limitations ----------- The number of subjects in our study was relatively small, but was comparable to other studies of gait and foot pressure \[[@B5],[@B9]\]. However, our study used the large number of steps during steady walking compared to previous studies. Moreover, selection of steps from the middle portion of the walk minimized inconsistencies that may accompany gait initiation and termination \[[@B5]\]. The foot shape (high vs. flat arch) has not been assessed. However, the midfoot contact area and the relative load in the medial and lateral arch masks were not different between groups. Conclusion ========== Healthy aging affects the dynamics of foot pressure distribution during normal walking. The forces and pressures under the medial foot masks were reduced in elderly people, resulting in lower propulsion during the step from the heel-touch to the toe-off phases. Clinically, lateralized foot pressure and lessened propulsion may affect walking ability in elderly people, posing difficulties in balance, forward thrust, and terrain adaptation. List of abbreviations ===================== BMI = body mass index BW = body weight FPD = foot pressure distribution NWS = normal walking speed Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MH -- the recipient of American federation on Aging Scholarship, designed and performed data analysis and wrote the first version of the manuscript. MV -- participated in the experiments, software development and analysis. JL -- participated in the experiments and analysis. KH -- participated in the statistical analysis. LL -- contributed to data interpretation and manuscript preparation. VN -- designed the study, conducted the experiments, participated in the analysis, data interpretation, and manuscript preparation. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2318/5/8/prepub> Acknowledgements ================ This study was sponsored by The Older American Pepper Center Grant 2P60 AG08812-11; Research Resource Cores A and B and the General Clinical Research Center grant MO1-RR01302. The authors wish to thank Shiva Gautam PhD for consultation on statistical methods.	PubMed Central	2024-06-05T03:55:58.384462	2005-5-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173105/", "journal": "BMC Geriatr. 2005 May 19; 5:8", "authors": [ { "first": "Mary Josephine", "last": "Hessert" }, { "first": "Mitul", "last": "Vyas" }, { "first": "Jason", "last": "Leach" }, { "first": "Kun", "last": "Hu" }, { "first": "Lewis A", "last": "Lipsitz" }, { "first": "Vera", "last": "Novak" } ] }
PMC1173106	Background ========== Patient choice is a key part of current U.K. National Health Service (NHS) Strategy \[[@B1]\]. At a time when the population has access to 24 hour banking and to supermarkets which are open 24 hours a day, specialist consultations in U.K. hospitals for non-emergency conditions remain largely confined to the hours of 9-5 pm. In this regard it would appear that the NHS has failed to keep pace with a society which generally demands fast, convenient and 24/7 services. The Department of Health\'s 10 year plan has echoed these sentiments and specifically states that the NHS has been too slow to change its ways of working to meet modern expectations in this consumer age. While it recognises that the NHS is under funded, money alone can only be a starting point. Cost-effectiveness, it states, is the key to success and this can only be achieved if available resources are used to achieve real benefits for patients. Clearly, the redundancy of most outpatient clinics after 5 pm and during the weekend is not an efficient use of resources, nor is it a modern way of working. Medical practice nowadays strives to be patient-centred \[[@B2]\]. There is good evidence that involving patients in decision-making and listening to the patient voice improves health outcomes and patient satisfaction. Hence, when we undertook an extensive literature review, we were surprised to find no published work concerning medical patients\' views on out of hours clinics in secondary care. Given the current government proposals we felt it necessary to explore patients\' views about out of hours clinics amongst patients attending medical outpatients\' clinics in a central London teaching hospital. After embarking on our study of medical outpatients we subsequently found two other reports on patients\' opinions on out of hours ophthalmology clinics. The first is a study of patients with ophthalmological conditions attending a hospital in Leeds and Bristol which looked at whether patients found their current appointments and surgery times convenient or inconvenient \[[@B3]\]. They found that 89% of their patients were apparently happy with their current appointment times but when asked about alternatives would find Saturday mornings convenient. The second study was a critique of the first and reported the responses of 100 consecutive patients attending Saturday morning ophthalmology clinics at an outer London hospital \[[@B4]\]. They were asked whether they preferred a Saturday morning clinic, a regular 9-5 appointment, or whether they didn\'t mind either. They found that 50% had no preference, 41% preferred a Saturday morning and only 9% preferred a 9-5 pm weekday appointment. Both these studies have their limitations in that the first essentially asked about convenience not preference and the second may be affected by bias in that verbal responses alone were reported of a select group of patients i.e. only those already attending a Saturday morning clinic. These studies are based in ophthalmology clinics and may not be representative of all patients attending hospital outpatient clinics. Our study is based on respiratory and cardiology outpatient clinics. These two disease areas account for a major part of the burden of medical illness in the UK with 8 million people having lung disease alone. Methods ======= Centre of data collection: Charing Cross Hospital, London Clinics: Respiratory, Monday morning, Wednesday morning and Thursday afternoons; Cardiology, Thursday morning Duration of data collection: 6 months Patient Profile: Outpatients routed through the central booking office (excluding rapid or open access respiratory and cardiology clinics). 300 sequential attendees at Cardiorespiratory clinics were asked to complete a questionnaire \[see [additional file 1](#S1){ref-type="supplementary-material"}\] to determine their opinions regarding access to out of hours (defined as after 5.00 pm) and weekend outpatient clinics. They were then asked to give up to five reasons as to why they might be interested in such clinics. The patients included a mixture of patients attending for the first time and those attending for follow-up appointments. The reception staff were specifically asked to give the questionnaire to all patients who had not previously received them. The items selected for inclusion in the questionnaire had been raised repetitively in another study (not yet published) looking at reasons why patients did not attend clinics (forgetfulness, long waits, seeing a different doctor each time, nowhere to park, time off work, no one available for childcare). Data was analysed using the statistical package SPSS (version 12.0), however as this was an observational study power calculations were not calculated and data was only subjected to descriptive analysis Results ======= 264 completed questionnaires were obtained (response rate 88%) from 124 males and 122 females with a mean age (SD) of 58 years (16.94). 144 of the patients were attending with a respiratory condition (amongst which were 52 with asthma, 18 with sleep related breathing disorders, 13 with unexplained cough) 65 with a cardiac condition (25 with ischaemic heart disease, 12 with arrhythmias, 9 with hypertension), and the remainder did not state their condition or had an ill-defined medical problem. 165 patients (62.5%) wanted either an out of hours clinic or a clinic on a Saturday or Sunday. Of the reasons given, 81 patients (49.1%) said it was the greater flexibility that such clinics offered with 64 (38.8%) specifically stating that it was because they worked during the day. Forty one (24.8%) responded that it was easier for them to get to the hospital in the evening because someone could drive them there. The same number responded that it was easier to get someone to accompany them in the evening, and 22 (13.3%) stated that the reason was because they had family commitments during the day. 18 patients listed other reasons (see Table [1](#T1){ref-type="table"}) which included important practical issues such that parking restrictions in the environs of the hospital would not apply in the evening and at weekends. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### \'Other\' Reasons volunteered by patients for wanting out of hours clinics ::: Reason Number of patients -------------------------------------------------------------------------------------- -------------------- I am a shiftworker or self-employed so out of hours clinics would be more convenient 4 Less traffic and less stressful to get to the hospital 3 Parking restrictions would not apply 3 I would not have to finance child care at these times 3 Waiting times would probably be reduced 2 ::: When the responses were sub-analysed according to whether the respondent was over or under the age of 65 years, 152 patients were seen to be under the age of 65. Of those 152 patients, 110 (72.4%) expressed a preference for out of hours clinics, whereas amongst the 100 patients aged over 65, 49 (49%) expressed such a wish. Not surprisingly amongst those aged under 65, the commonest reason stated for why they were interested in an out of hours clinic was that they worked during the daytime, and this response was given by 58 out of the 110 of the patients (53%) The responses also varied according to the medical condition of the patient. Of the 144 patients attending with a respiratory condition, 97 (67.4%) wanted out of hours clinics, compared to 36 out of the 65 (55.4%) patients attending with a cardiac condition. However, the mean age of respiratory patients was 55 years compared to 65 years in the cardiac group. Of those patients who wanted out of hours clinics, weekend clinics appeared to be consistently more popular than evening clinics and this trend was upheld even when the responses were analysed according to age, gender and medical condition. Discussion ========== The majority of medical outpatients we questioned would like more flexible timing of clinics. The reasons expressed to justify such a preference appear to be entirely reasonable, such as work commitments during the day. An extension of hours for routine consultations may therefore enhance care for those who are in employment. It also represents a logical fuller use of outpatient consultation facilities. However, out of hours clinics involve more than a doctor-patient consultation. Additional reception and nursing staff are likely to be needed and if the consultation is to be truly effective for the patient they would need to be able to have simple investigations like blood tests, ECG and imaging tests at the same time. Further economic evaluation of the cost of making such services available would be necessary before the widespread introduction of out of hours clinics. However, not all outpatients require investigations and it may not be difficult to construct a list of conditions where patient attending for follow-up can be offered out of hours services without other departments needing to be available. The patients surveyed in this study were patients with medical conditions attending an inner city London teaching hospital. In a patient centred health service similar studies should probably be done amongst those with other conditions living in other areas. The response rate for our study was high and reasons for non response are not known but could include the patient having gone straight into see the doctor and having no waiting period in which to complete the form. In a multicultural society some may have had difficulty with a questionnaire in the English language and up to 15% of patients attending some UK out patient clinics may be functionally illiterate \[[@B5]\]. The response rate is unlikely to have altered the validity of the conclusions which are similar to one study in an ophthalmology clinic \[[@B4]\] but dissimilar to another \[[@B3]\]. The differences almost certainly reflect the question which is asked (stating actual preferences versus stating satisfaction with what you have) and how it is asked and by whom. Conclusion ========== We feel that this is an interesting and important area of research with the results suggesting that the hours of opening of traditional medical outpatients clinics should be extended. Each hospital and each speciality should probably survey their own clientele, but to enable comparisons to be made some form of standardised wording should be used and the questionnaire itself should be included in any subsequent report. Competing interests =================== MRP, CF and NJR do not have any significant competing interests (financial or non-financial). Authors\' contributions ======================= MRP developed the study design and protocol. NJR and CF were responsible for implementing the study and analysing the results. All contributed to drafting the report and approved final manuscript. [Additional file 1](#S1){ref-type="supplementary-material"}: Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/47/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A copy of the questionnaire used for this study. ::: ::: {.caption} ###### Click here for file :::	PubMed Central	2024-06-05T03:55:58.386451	2005-6-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173106/", "journal": "BMC Health Serv Res. 2005 Jun 24; 5:47", "authors": [ { "first": "Claire L", "last": "Feeney" }, { "first": "Nicola J", "last": "Roberts" }, { "first": "Martyn R", "last": "Partridge" } ] }
PMC1173107	Background ========== The cancellous bone microarchitecture corresponds to the spatial organization and morphology of the trabecular network. Sugita et al. \[[@B1]\] observed that the mechanical behavior of the cancellous bone varied following the testing direction; these variations were interpreted as an anisotropic feature of the bone stress. Trabecular bone anisotropy corresponds to the preferential orientation(s) of trabeculae. Anisotropy is constituted under the influence of prefererential oriented strength applied to bone \[[@B2]\] and permits to establish resistance to these strengths in a given preferential direction. Sugita et al. \[[@B1]\] concluded that anisotropy of the cancellous bone should be considered to predict the fracture risk. The determination of collagen and crystal orientation in connective tissues at molecular scale has been studied, for instance by diffraction \[[@B3]-[@B6]\]. Amorphous materials such as reinforced concrete have been characterized by x-rays and the orientation of fibers investigated by the Fourier transform method \[[@B7]\]. Defossez et al. assessed several methods on images of the femoral neck region: co-occurrence and run length matrix, spatial-frequency and fractal techniques to determine the trabecular direction \[[@B8]\]. These methods exhibited a high degree of accuracy, suggesting a possibility of trabecular orientation characterization in clinical practice \[[@B8]\]. Different methods are available to characterize the structural anisotropy on bone radiographs. In 1970, Singh et al. \[[@B9]\] developed a semi-quantitative index applied to femoral neck radiographs. This index is based on the existence of several arches of trabeculae in the femoral neck. Some arches preferentially disappear with age and osteoporosis, and the count of these arch systems can be used to determine the Singh index. Aggarwal et al. \[[@B10]\] developed a similar index for the calcaneus. Whitehouse et al. \[[@B11]\] introduced the Mean Intercept Length (MIL) defined as the average length between two bone/marrow interfaces. The MIL polar diagram generates an ellipse for 2D slice images \[[@B11]\] or projectional images such as radiographs \[[@B12]\], and an ellipsoid for 3D images \[[@B13]\]. A MIL tensor can be calculated \[[@B13]\] by fitting the MIL values and can be used to determine the Degree of Anisotropy (DA) \[[@B14]\]. The MIL method requires a previous binarization of the gray-level images, then it is a suitable method for the calculation of DA from 3D images obtained by Quantitative Computed Tomography (QCT) \[[@B15],[@B16]\] or Magnetic Resonance Imaging \[[@B17]\]. Therefore, the binarization on 2D radiographs remains a critical step due to partial volume effect. The MIL method was used by Luo et al. \[[@B18]\] to compare the anisotropy of the 3D structure and of the 2D projected image of one in vitrosample and 14 simulation models derived from this sample. The authors reported a very good correlation (r = 0.99) between anisotropy values assessed on the 3D trabecular structure and the 2D projection images. Luo et al. \[[@B18]\] insisted on the interest of the estimation of architectural fabric from plain radiographs. Few methods evaluating the trabecular structure have been developed on radiographs. Caldwell et al. \[[@B19]\] developed a method characterizing trabecular bone structure on vertebral digitized radiographs. An evaluation of the orientation named the fraction orientation edges is obtained from the histogram of the magnitude of the edge gradient versus the direction of the edges. Geraets et al. \[[@B20]\] developed an index called Line Fraction Deviation (LFD) derived from the tresholded radiographs and based on the fraction of bone pixels. A diagram plotting the standard deviation of this fraction in all directions could reflect the orientation of the trabecular bone. This method has been found more sensitive than MIL method on bone radiographs to evaluate anisotropy \[[@B12]\]. Jiang et al. \[[@B21]\] developed a method for a non invasive evaluation of bone mechanical properties: textural features, the global Minkowski dimension and trabecular orientation were determined using Minkowski dimension analysis. Jiang et al. demonstrated the contribution of normalized Bone Mineral Density (BMD), structural features and patient age to bone mechanical properties. In 1993, Oxnard reported slight visible microarchitectural changes from the Fourier transform on bone radiographic images but no parameter calculation was performed \[[@B22]\]. The same group worked on vertebral bodies T1 to L5 in seven male columns and applied the FFT to the radiographic images of these vertebrae. They characterized the orientation and size of the architectural elements of vertebral cancellous bone, and underline the potential of this technique to define the effects of ageing \[[@B23]\]. Wigderowitz et al. considered the properties of Fast Fourier Transform (FFT) to evaluate trabecular bone structure: the directional pattern of frequencies with high-magnitude can identify the orientation of trabeculae. He defined three indices including spectral trabecular index, longitudinal and transversal trabecular indices \[[@B24]\]. These indices approximate the evaluation of the trabecular structure in terms of trabeculae spacing and orientation. Wigderowitz et al. concluded that this quantification detects structural changes occurring with age and may be useful in osteoporosis studies \[[@B24]\]. Caligiuri et al. \[[@B25]\] also used the FFT to perform a structural analysis of the spine trabecular bone. The hypothesis was that the magnitude and the first moment of the power spectrum could correspond to the coarseness or the fineness of the texture pattern. The two texture measurements were compared to BMD. Caligiuri et al. concluded that this texture analysis might lead to a better prediction of the osteoporotic fracture risk \[[@B25]\]. Recently, Gregory et al. have proposed a FFT-based analysis of trabecular bone structure on hip radiographs from patients with and without hip fracture. They worked on FFT profiles parallel and perpendicular to the preferred orientation. Principal conponents analysis was used to generate scores from the profiles and was able to discriminate fracture and control groups better than fractal dimension \[[@B26]\]. Our research group has previously experienced fractal analysis on bone radiographs \[[@B27]-[@B29]\]. The measured parameter (Hurst coefficient) offers a global evaluation of the image irregularity, of its complexity. It is considered as an indicator of roughness. The Maximum Likelihood Estimator is applied first on one direction, then the process is repeated in 36 directions by 10 degrees steps. The directional results can be expressed on a polar diagram that can be itself be fitted to an ellipse; the shape of this polar diagram can be characterized in order to determine the textural anisotropy \[[@B30],[@B31]\]. When comparing radius and calcaneus in the same subjects, Lespessailles et al. have shown significant differences \[[@B31]\] but the interpretation is equivocal because this technique characterizes the variations of grey level roughness following directions, and not directly the trabecular network orientation. The osteoporotic fracture risk prediction is important especially in postmenopausal women. BMD is currently measured in clinical practice and textural parameters can be assessed on trabecular bone radiographs \[[@B19]-[@B22],[@B24],[@B25],[@B31]\], but the anisotropy was not clearly quantified in these texture analyses. To address this question, we have worked on a quantitative evaluation of anisotropy on bone radiographs. The development of a structural parameter such as DA from radiographs is very promising; this parameter could be complementary of the BMD and of fractal texture parameters in the explanation of bone strength. We describe here a new quantitative method based on the FFT to obtain anisotropy indices on bone radiographs of the calcaneus and a validation on synthetic images. The intra and inter-observer reproducibility and a pilot clinical study comparing osteoporotic fracture cases to control cases are presented. Methods ======= Population of the pilot study ----------------------------- Postmenopausal women were recruited from a cross-sectional unicenter case control study. The protocol screened 400 women: 349 were enrolled. The inclusion and exclusion criteria have been previously described in details \[[@B28]\]. For this preliminary study, a sample of 22 osteoporotic fracture cases was randomly selected, and was secondarily age-matched with 44 control cases (for each age, we randomly selected two control cases with an equal age ± 5 years). The distribution of fracture cases was 1 hip fracture, 5 wrist fractures, 11 vertebral fractures and 5 metatarsus fractures, all these fractures were considered as low energy fractures. The ages of the control and fracture cases ranged from 45 to 87 years. The mean age was 69.2 ± 11.9 years for fracture cases and the mean age was 69.0 ± 11.7 years for control cases. The BMD was measured by dual energy x-ray absorptiometry (Hologic^®^4500 device) at lumbar spine and femoral neck. The mean lumbar spine BMD was respectively 0.814 ± 0.09 g.cm^-2^and 0.895 ± 0.16 g.cm^-2^(p \< 0.05) for fracture cases and control cases. The mean femoral neck BMD was respectively 0.650 ± 0.09 g.cm^-2^and 0.686 ± 0.12 g.cm^-2^(ns) for the fracture cases and control cases. Study design ------------ ### Image realization This study was performed on trabecular bone radiographic images of the calcaneus. The calcaneus radiographic images were performed following a standardized procedure reported elsewhere \[[@B27]\]. Briefly, the X-ray tube voltage (48 kV), the exposure conditions (18 mAs) and the focal-calcaneus distance (1 meter) were fixed for all patients. Kodak single emulsion Min RG films were used and all developed by the same film processor at fixed developer and fixer temperatures. In order to obtain digitized images of the calcaneus trabecular bone, the radiographic films were digitized with an AGFA Duoscan scanner (AGFA GEVAERT N.V., Morstel, Belgium) with 256 gray-levels. Calcaneus is known to be a heterogeneous site of trabecular bone \[[@B32]\]. For this reason we have selected a large region of interest (ROI) of 2.7 × 2.7 cm^2^(256 × 256 pixels with a pixel size of 105 μm). The ROI of 256 × 256 pixels (Figure [1](#F1){ref-type="fig"}) was defined from anatomic marks \[[@B27]\]. The basis of the ROI was positioned on a line linking the plantar aponeurosis insertion to the superior end of the Achilles tendon insertion, the ROI comprising both compressive and tensile trabecular network \[[@B10]\]. ### Noise filtering The low frequency noise of an image corresponds to the gray-value variations over large distances, due to the radiological artifacts and to the fat tissues projections on the radiograph. In order to remove the low frequency noise and to take into account only trabecular components of the image, we used a convolution filter previously described by Geraets \[[@B33]\]. A kernel box was used; for each pixel, the average gray-value of the box was allocated to the box middle pixel. The new image obtained is the low-frequency image. The window must be small enough to extract the low frequency noise and large enough to prevent the trabecular pattern leaking into the low frequency region. For our images, the optimal size of the box was 25 × 25 pixels. The filtered image was obtained by subtracting the low-frequency image from the original images, this filter also removed cross of high intensity centered on the central pixel due to the finite bondaries of the ROI. ### Fourier transform The Fourier transform represents a signal in spatial frequency space. An image can be considered as a repartition of bright intensities in a (xOy) plane and can be expressed as a two-dimensional function f(x, y). The Fourier transform is expressed by a function F(μ,υ) with the two variables μ and υ corresponding to spatial frequencies in (μ Oυ) plane. The 2D Fourier transform spectrum of an image is expressed by the following formula: ![](1471-2342-5-4-i1.gif) where μ, υ and x, yare the respective variables of the frequency and the spatial domains and N the size of the image. Any periodic structure in the original spatial-domain image is represented by peaks in the frequency-domain image at a distance corresponding to the period and a direction at right angle of the original orientation. If a mild degree of disorientation is introduced in an oriented periodic structure, the frequencies of the Fast Fourier transform (FFT) spectrum are spread over an angle corresponding to the deviation of the structure in the original image. By analogy, we hypothesized that the periodic structure is represented by trabeculae projection, and the degree of disorientation by anisotropy. The magnitude of the transform corresponds to: ![](1471-2342-5-4-i2.gif) where Re= real part of the FFT and Im= the imaginary part of FFT. The FFT was calculated on the gray level filtered images of the trabecular bone using the Visilog 5.1 software (Noesis). Then the magnitudes of the frequency images were divided to the total magnitude of the transform to normalize the contrast in the images according to the following formula: ![](1471-2342-5-4-i3.gif) ### Measurements on FFT spectrum Trabecular bone was assimilated to an oriented structure with two main directions, the ROI including longitudinal and transversal trabeculae (Figure [1](#F1){ref-type="fig"}). All the FFT spectra of the calcaneus radiographs involved horizontal and vertical conponents (Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}). The horizontal components were indicative of the longitudinal trabeculae fabric and the vertical components were indicative of the transverse trabeculae fabric. We have to notice that the so-called horizontal and vertebral trabeculae were so described on the ROI, which was slightly rotated by comparison to the original image. The limit between high energy area representing trabeculae and low energy area corresponding to the noise on the FFT spectrum were visually determined by the operator. We measured the spreading angles of the longitudinal and the transversal trabeculae named respectively the Dispersion Longitudinal Index (DLI) and the Dispersion Transverse Index (DTI). In order to obtain an average value, the DTI and DLI parameters were measured in the two symmetrical parts of the FFT spectrum conponents (Figures [2b](#F2){ref-type="fig"} and [3b](#F3){ref-type="fig"}). An index relative to the trabecular fabric or degree of anisotropy was derived from the measured parameters DLI and DTI. The Degree of Anisotropy (DA) was defined as: ![](1471-2342-5-4-i4.gif) For a perfectly isotropic structure, the FFT spectrum has a disc shape and DA is equal to unity. More isotropic trabecular bone structures will have DA values closer to unity. Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"} illustrate two examples of digitized radiographic images and FFT spectra of trabecular bone at calcaneus. Figure [2](#F2){ref-type="fig"} corresponds to a control case and Figure [3](#F3){ref-type="fig"} to a vertebral fracture case. ### Synthetic images To test the reliability of DA assessment, we tested this method on projected volume with known disorientations. We synthesized 6 series of 8 images composed of beam like structures more or less aligned following two directions in order to obtain structures close to trabecular bone of calcaneus. For the less anisotropic structure, horizontal and vertical disorientations varied from 0 to 80° (Figure [4a](#F4){ref-type="fig"}), and for the most anisotropic structure, the disorientation of the beam like structures varied from 0 to 17° (Figure [4c](#F4){ref-type="fig"}). These synthetic images comprised a wide variety of structures, from anisotropic to almost isotropic, a variety much wider than expected in the clinical X-rays. The projections of these volumes were analyzed with the same software as bone radiographs. Data analysis ------------- ### Intra and inter-observer reproducibility of the measurements The two measured parameters DLI and DTI and the derived parameter DA were determined to calculate the intra and inter-observer reproducibility. To determine the intra-observer reproducibility, a single observer performed two sets of measurements on the FFT spectra of 20 subjects with a one day interval between each set. To determine the inter-observer reproducibility, two sets of measurements were performed on the FFT spectra of 20 subjects by two observers. The observers were blinded for each sets of measurements. The intra-observer and inter-observer reproducibilities were calculated for nsubjects with the root mean square RMS average according to the following formula \[[@B34]\]: ![](1471-2342-5-4-i5.gif) where SD~j~is the standard deviation for the subject jand ![](1471-2342-5-4-i6.gif) is the average of the measurements for the subject j. ### Clinical evaluation Results in these two groups were compared using Student\'s t-test for comparisons of the means after checking Gaussian distribution. The area under ROC curves was calculated for BMD measurements and DA. Results ======= The results of synthetic images are represented on Table [1](#T1){ref-type="table"}. The DA varied from 1.3 to.3.6, respectively corresponding from more anisotropic (Figures [4a](#F4){ref-type="fig"} and [4b](#F4){ref-type="fig"}) to almost isotropic structures (Figures [4c](#F4){ref-type="fig"} and [4d](#F4){ref-type="fig"}). Table [2](#T2){ref-type="table"} shows the intra-observer and inter-observer RMSCVs. The RMSCVs of DA were respectively 1.5% and 3.1 % Table [3](#T3){ref-type="table"} presents the results of the fabric indices obtained on the 22 osteoporotic fracture cases and 44 control cases. Results of DLI and DTI were statistically significant lower in osteoporotic cases (p \< 0.01, p \< 0.05, respectively) leading to a significantly higher DA. In control cases DA was closer to 1 due to a large spreading of frequencies in the FFT spectra while DA was higher in osteoporotic cases in relation to a narrower frequency spreading. Comparing the DA values from Table [3](#T3){ref-type="table"} to Table [1](#T1){ref-type="table"}, it appears that the controls (DA = 1.7) were in the 0--55° category and the fractures cases (DA = 1.9) were in the 0--45° category. The difference of the mean of DA between cases and controls was 8.7 % and close to the figured variation ![](1471-2342-5-4-i7.gif) corresponding to the 95% confidence interval of the measurements \[[@B35]\]. Differences in DA determined by spectral analysis (p \< 0.01) and lumbar spine BMD (p \< 0.05) were significant (Figure [5](#F5){ref-type="fig"}). Whereas the difference between fracture cases and controls for femoral neck BMD was not significant. From the ROC analysis, the Area Under Curve (AUC) were respectively 0.65, 0.62, 0.64, 0.77 for lumbar spine BMD, femoral neck BMD, total femoral BMD and DA. There is a trend to higher AUC for DA comparatively to BMD measurements, but according to the small size of the population the statistical significance was not reached. Discussion ========== We have developed a new index of trabecular bone anisotropy on radiographs based on a spectral analysis of the gray-level images. In this pilot study, this new index applied to radiographs of the calcaneus has shown a reasonable reproducibility for a technique at its first step of development. As it was demonstrated with synthetic images, the DA of an almost isotropic structure is close to one and the DA of an anisotropic structure is greater than one. We hypothesized that the trabeculae did not randomly disappear, the effect might be a reduction of frequencies on the FFT spectra in preferential directions leading to an increase of anisotropy. The results of this pilot study involving 22 osteoporotic fracture cases and 44 controls suggest that the DA parameter may be potentially useful to distinguish fracture cases from control cases. The main interest of this technique is that it can be applied directly to gray level radiographic images without previous binarization. The radiographic technique provides a projection image. However a good 2D-3D correlation has been described concerning microarchitecture \[[@B29]\] and anisotropy \[[@B18]\] from gray level analysis. The significantly higher DLI and DTI in control cases suggest that there are more various orientations of the elemental structures around the two main orientations (longitudinal and transversal) in control cases than in osteoporosis cases with vertebral fractures. The larger range of orientations in controls corresponds to a lesser anisotropic structure than in osteoporosis. DTI was the best discriminant parameter between fracture patients and control cases but also the less reproducible. There are only few transversal trabeculae comparatively to longitudinal ones and they disappear first with osteoporosis due to a less contribution in bone strength. The difference between fracture cases and controls was close to errors attributed to the intra or inter-observer reproducibility. The intra-observer reproducibility is close to the 1 to 2% of reproducibility found in usual bone densitometry method as dual x-ray absorptiometry \[[@B37]\]. The poor inter-observer reproducibility is due to the difficulty to identify the limit between high energy area of the spectrum (corresponding to trabeculae) and low energy area (corresponding to noise) in relation to the few numbers of trabeculae projections. Since the population size is small, further studies are required with larger groups. We could expect more stringent results with a more homogenous population of fractured patients with vertebral fracture, for instance. The higher anisotropy found in osteoporosis cases is in accordance with the findings of Newitt et al. \[[@B38]\] who report that the increase of bone resorption in osteoporosis leads to a loss of thinner trabeculae first, resulting in an increase of anisotropy. The study of Newitt et al. was performed with the MIL method on 3D magnetic resonance images at the radius. The MIL method is widely used to determine the trabecular 3D structure anisotropy but some authors \[[@B39],[@B40]\] discuss its reliability since it reflects the boundary orientation rather than the real anisotropy of the structural elements. This concept of transversal and longitudinal systems of trabeculae must be cautiously interpreted in our study. Indeed the ROI on the calcaneus radiographic images is tilted around 45 degrees (Figure [1](#F1){ref-type="fig"}). The trabeculae were named longitudinal and transversal in reference to the radiographic image (Figure [1](#F1){ref-type="fig"}) and not to the ROI orientation (Figures [2a](#F2){ref-type="fig"} and [3a](#F3){ref-type="fig"}). Longitudinal trabeculae correspond to the compressive trabecular network extended from the subtalar joint and the transverse to the tensile trabecular network sweeping backwards and upwards the great tuberosity \[[@B41]\]. The anisotropy of trabecular bone is different according to the skeletal sites: in a study comparing the properties of calcaneus, distal femur, proximal femur and vertebrae on human specimens, Majumdar et al. \[[@B17]\] found the highest anisotropy of trabecular bone at the calcaneus followed by distal femur and proximal femur, vertebrae constituting the least anisotropic site. Our results corroborate the studies of Geraets et al. \[[@B33]\] on hip and radius, Wigderowitz et al. \[[@B24]\] on wrist radiographs, Mosekilde et al. \[[@B42]\] on vertebrae and Ciarelli et al. \[[@B43]\] on femoral head samples. The Line Fraction Deviation was developed by Geraets on binarized radiographic images, and leads to the determination of the preferential orientations. There was no calculation of an anisotropy index but it was possible to quantify the anisotropy from the deviation of polar diagram from a circle. The lower values of the Line Fraction Deviation index found in osteoporotic subjects were consistent with the early loss of the secondary compressive trabecular group of the hip \[[@B44]\]. Moreover, Geraets et al. showed in a study of the distal radius that the Line Fraction Deviation values decreased along the transversal direction with age whereas the orientation along the axial direction remains stable during the entire life \[[@B33]\]. Wigderowitz et al. \[[@B24]\] found using the spectral analysis at the distal radius that the transversal trabeculae are preferentially absorbed, thinned and spared with age. Mosekilde et al. in vertebrae \[[@B42]\] showed that compressive strength was greater in the vertical direction than in the transversal direction. This anisotropy in vertebral strength increased with age and indicated according to the authors that the transverse trabeculae were selectively removed. Buck et al. have found that the oblique conponents declined in the cranio-caudal direction particularly for age superior to 60 years leading to variation in anisotropy \[[@B23]\]. Ciarelli et al. \[[@B43]\] also found that fracture cases have proportionally more trabeculae aligned along the primary load axis (and thus proportionally fewer transverse trabeculae) than control cases; the authors hypothesized that the loss of transverse trabeculae leads to the difference in anisotropy between groups. Zhao et al. characterized iliac trabecular bone by micro-QCT and showed that trabeculae thinning led to a more isotropic structure in the first postmenopausal years whereas the structure became more anisotropic in the later years \[[@B45]\]. They hypothesized that, in the later years, the remaining trabeculae would be more widely separated, less connected and some more thickened leading to an increase of anisotropy. It has been well established that the calcaneus structure is heterogeneous \[[@B32],[@B46]\]. Lin et al. \[[@B32]\] have studied the calcaneus microarchitecture on MRI images. They analyzed 20 to 25 ROIs 1 × 1 cm^2^in each calcaneus. They found a spatial heterogeneity in the posterior region of 40 %. In our study the ROI was 2.7 × 2.7 cm^2^and represented a much larger area; it was accurately defined by anatomic marks, this point avoiding large variation in fractal analysis \[[@B27]\]. Furthermore our ROI contained both transversal and longitudinal trabeculae, if transversal trabeculae disappeared, it should be possible to calculate DA and a high value will be obtained. As the DLI and DTI measurements were not yet automated, the parameters were measured in the two symmetrical parts of the FFT spectrum in order to obtain an average value. In a near future the automation of our method could be performed, allowing applications to large sets of images. In spite of this lack of automatization the reproducibility of the DA parameter was acceptable and allowed for the accurate characterization of osteoporosis changes in small series. At this first step of development, this clinical study must be considered as a preliminary study evaluating the potential of this anisotropy evaluation. Conclusion ========== This study has shown that the DA can be determined on plain radiographs using spectral analysis. The reproducibility of the DA values may be improved by automating the method. The distinction between fracture cases and control cases is very promising, but further studies are necessary to know if the DA evaluation could improve the osteoporotic fracture risk determination when combined with BMD and other textural parameters such as fractal analysis. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= BBI participated in the design of the study, carried out the measurements and the manuscript preparation, G L carried out measurements of reproducibility and synthetic images, CC performed statistical analysis and participated to the manuscript preparation, R H supervised the application of the Fourier Transform and CLB carried out the design of the study and supervised the manuscript preparation. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2342/5/4/prepub> Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Radiographic image of the calcaneus showing the localization of the Region of Interest (ROI). Inside the ROI, longitudinal trabeculae correspond to the compressive trabecular network extended from the subtalar joint and the transverse to the tensile trabecular network sweeping backwards and upwards the great tuberosity. Arrows show anatomical landmarks corresponding to the bottom of the ROI. ::: ![](1471-2342-5-4-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Control case a: ROI of the digitized radiographicimage of calcaneus trabecular bone. b: FFT spectrum A: corresponds to longitudinal trabeculae B: corresponds to transverse trabeculae DLI: Dispersion Longitudinal Index, spreading angle of the longitudinal trabeculae DTI: Dispersion Transverse Index, spreading angle of the transverse trabeculae ::: ![](1471-2342-5-4-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Vertebral fracture case aRegion of Interest (ROI) of the digitized radiographic image of calcaneus trabecular bone. bFFT spectrum DLI : Dispersion Longitudinal Index, spreading angle of the longitudinal trabeculae DTI : Dispersion Transverse Index, spreading angle of the transverse trabeculae ::: ![](1471-2342-5-4-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Synthetic images a: Synthetic image of astructure with orientations between 0 and 80° in 3 dimensional space b: Fast Fourier Transform (FFT) spectrum of the image 4a c: Synthetic image of a structure with orientations between 0 and 17° in 3 dimensional space d: FFT spectrum of the image 4c ::: ![](1471-2342-5-4-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Box plots of lumbar spine BMD and DA. A T-test was used to compare controls and fractured patients. ::: ![](1471-2342-5-4-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### DA assessment on synthetic images composed of cylinders more or less aligned following two directions leading to isotropic and anisotropic structures. ::: Synthetic images ------------------ -------- -------- -------- -------- -------- -------- Angles 0--80° 0--69° 0--52° 0--40° 0--28° 0--17° DA 1.3 1.3 1.8 1.9 2.5 3.8 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Root Mean Square Standard Deviation Coefficient Variation RMSCV(%) for measured indices DLI and DTI and the degree of anisotropy. ::: Intra-observer reproducibility Inter-observer reproducibility ---------------------- -------------------------------- -------------------------------- Measured parameter DLI 2.4 3.4 DTI 2.4 4.1 Calculated parameter DA 1.5 3.1 DLI: Dispersion Longitudinal Index DTI: Dispersion Transversal Index DA: Degree of Anisotropy ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Fabric indices (mean ± SD) from the fabric parameters in osteoporotic fracture cases and control cases. ::: Osteoporotic fractures Controls Statistical significance p ----- ------------------------ ------------- ---------------------------- DLI 55.6 ± 6.2 59.1 ± 6.6 0.04 DTI 41.3 ± 5.9 46.8 ± 6.7 0.002 DA 1.87 ± 0.15 1.72 ± 0.18 0.001 :::	PubMed Central	2024-06-05T03:55:58.387634	2005-5-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173107/", "journal": "BMC Med Imaging. 2005 May 31; 5:4", "authors": [ { "first": "Barbara", "last": "Brunet-Imbault" }, { "first": "Gerald", "last": "Lemineur" }, { "first": "Christine", "last": "Chappard" }, { "first": "Rachid", "last": "Harba" }, { "first": "Claude-Laurent", "last": "Benhamou" } ] }
PMC1173108	Background ========== The detection of an unusual, highly conserved sequence motif \"His-phi-His-phi-His-phi-phi\" (phi representing hydrophobic amino acids) in a variety of organisms of all kingdoms led to the definition of a new family of proteins named HIT ([hi]{.underline}stidine triad) \[[@B1]\]. This family has three main branches: the fragile histidine triad (FHit)-related proteins found in animals and fungi, which act as di-adenosine polyphosphate hydrolases and function as tumor suppressors in humans and mice \[[@B2]\] (although the tumor suppressing function is not dependent on ApppA hydrolysis \[[@B3]\]), the GalT (galactose 1 phosphate uridylytransferase) homologues, which have a modified \"His-phi-His-phi-Gln\" motif, which transfers nucleoside monophosphates to phosphorylated secondary substrates rather than hydrolyzing them \[[@B4]\], and the [Hi]{.underline}stidine-[t]{.underline}riad [n]{.underline}ucleotide binding (HinT) homologues, which in eukaryotes are intracellular receptors and hydrolases of purine mononucleotides \[[@B5]\]. Although HinT homologues are found in all kingdoms and this family is the most ancient and widespread branch of the HIT proteins, the cellular function, the substrates and the interaction partners of HinT proteins are largely unknown. In prokaryotes, knowledge of HinT proteins is generally restricted to sequence analyses. In the cell wall-less prokaryote Mycoplasma hominis, the cytoplasmic HinT protein interacts with a surface localized membrane complex by binding to the P80 moiety. Interestingly, the genes encoding P80 and P60, the domains of the membrane complex, form an operon with the HinT gene \[[@B6]\]. The identification of a homologous hitlocus in M. pulmonisand access to several sequenced prokaryotic genomes enabled us, in this study, to hypothetically identify interaction partners and thus propose a functional role for HinT in bacteria with small genomes. Results ======= To find out more about the function of prokaryotic HinT proteins we first analyzed the hitloci of bacteria with a restricted genome, with the view that they might represent a model for organisms possessing the minimal genetic make-up essential for life as a free-living organism. As polycistronically organized genes often encode proteins that are functionally related (e.g. in a protein-complex formation or as part of a common pathway) we ran a search of known genome sequences for genes closely neighbouring or overlapping the hitgene. Species from the Mollicutesand the Chlamydiaceaefulfill these requirements and were thus analyzed. Mollicutesare phenotypically distinguished from other bacteria by their minute size and total lack of a cell wall. They have evolved as a branch of gram-positive bacteria by a process of reductive evolution. The significant genome \"condensation\" (for example the genome of M. genitaliumis only 580 kbp) was made possible by adopting a parasitic behavior. The primary habitats of human and animal mycoplasmas are the mucous surfaces which they colonize during infection \[[@B7]\]. While the intracellular localization of Mollicutesin insect tissues is well established, cell entry of human or animal mycoplasmas seems to be rare and possibly mediated by a site-directed, receptor-mediated event found in chlamydia \[[@B8]\]. However, Chlamydiaceaehave a number of features in common with mycoplasmas: their small genomes predict a limited metabolic capability, their major target tissue are mucous membranes and they also cause ocular and sexually transmitted diseases. In contrast, the chlamydia are obligate-intracellular bacteria that undergo a unique developmental cycle in which is an alteration in size between the small metabolically inactive infectious elementar body and the relatively large, metabolically active reticular body which is adapted for intracellular growth \[[@B9]\]. Bearing in mind, that in M. hoministhe HinT-interacting protein is a secreted antigen and thus predicted to play a role in the pathogenicity of mycoplasmas we expand this analysis to include a family of obligate-intracellular organisms known to secrete antigens in the inclusion body in the infected cell. Organization of the hit-locus in Mollicutes ----------------------------------------------- Hitloci with a genomic organization comparable to that of M. hominis\[[@B6]\], with the hitAB genes encoding two membrane proteins and hitL encoding the cytoplasmic HinT, were detected in nearly half of the mollicute genomes analyzed (Figure [1A](#F1){ref-type="fig"}). The highest similarity was found with the hitlocus of M. pulmonis. The MYPU\_0080 encoded protein had 44.3 % identity to the M. hominisHinT protein, and the predicted MYPU\_0060 and MYPU\_0070 proteins were 23.8 % and 26.6 % identical to the membrane proteins P80 and P60 of M. hominis, which were encoded by hitAB. MYPU\_0060 had structural features similar to those of P80, with an amino-terminal signal sequence with a predicted signal peptidase I (SPase I) cleavage site and a predominantly alpha-helical structure \[[@B10]\]. MYPU\_0070 began with an amino-terminal signal sequence of transmembrane helix (from AA 5 to AA 20) and a signal peptidase II recognition site with a lipoprotein attachment site at position 27. Thus, MYPU\_0070 of M. pulmonisappears to encode a P60 homologue, a cysteine-anchored lipoprotein. The order of genes within the hitloci of M. mycoidessubsp. mycoidesSC (MSC), Mesoplasma florum(MF) and M. mobile(MMOB) was similar with two genes predicted to encode membrane proteins and a downstream hitL gene. The similarities between the predicted sequences of MSC\_0500, MF\_235 and MMOB\_910, and P80 of M. hominiswere quite low and the proteins would be significantly larger than P80. While MMOB\_910 contained a signal peptidase I recognition site, like P80, MSC\_0500 and MF\_235 were pro-lipoproteins with an amino-terminal SPase II recognition sequence. Only MSC\_500 was predicted to have a predominantly alpha-helical structure, MF\_235 being predicted to have a secondary structure of alternate alpha helical and beta sheet regions and MMOB\_910 to consist mainly of beta sheets (data not shown). The proteins encoded by the gene next to hitL had little similarity with P60 of M. hominiswhen the whole sequence was compared. However, when the P60 protein region from AA 164 to AA 177 (CS1; ELQKMLLAKLYLLK) was used, identities of 43 % (MMOB\_900) to 57% (MYPU\_0070) were detected and scrutiny of the sequence from AA 226 to AA 234 (CS2; LYLMKYLVE) revealed 60 % (MYPU\_0070) to 70 % identity (MMOB\_900). The corresponding proteins of Mesoplasma florumand M. mycoidessubsp. mycoidesSC did not contain these conserved sequences. The different HinT homologues had the highest identity with the HinT protein of M. hominis, ranging from 44.3 % (MF\_233) to 55.7 % (MMOB\_890). Hitloci with a different genomic organization were identified in the mollicutes Ureaplasma parvum,M. pneumoniaeand M. genitalium(Figure [1B](#F1){ref-type="fig"}). While UU271 of U. parvum, the gene immediately upstream of hitL encoded a putative P60 homologue possessing the P60 consensus sequences CS1 (35.7 % identity) and CS2 (30% identity), it differed from all other P60 homologues in possessing a SPase I recognition sequence and two other transmembrane helices (AA 355 to AA 375 and AA 407 to AA 427). The UU273 gene, immediately downstream of hitL, encoded a protein with two transmembrane spanning helices, but with no significant similarity to other known proteins. The deduced protein sequence of UU270 did not have any sequence similarity with P80. With six transmembrane segments, it appeared more likely to form a pore than to be surface exposed. The organization of the hitlocus of U. parvumappeared intermediate between the hitloci described above and those of M. pneumoniaeand M. genitalium. The sequence identity between the ureaplasma HinT and the HinT proteins of M. pneumoniae(42.7 %) and M. genitalium(47.4 %) was higher than that between HinT of U. parvumand M. hominis(39.2 %), and the genes immediately upstream of hitL encoded proteins with seven (MPN\_274) or five (MG\_133) transmembrane domains (Figure [1B](#F1){ref-type="fig"}). MPN\_274 is predicted to encode a permease of an ABC transporter \[[@B11]\] suggesting a comparable function in M. genitaliumand probably also in U. parvum. Thus, these hitloci of Mollicutesanalyzed appear to have a hitL gene flanked by genes that are predicted to encode membrane-anchored proteins. As in M. gallisepticum, the hitL gene is not flanked by other genes on the same strand, also exceptions of a polycistronic organization of the hitL gene seem to exist within the Mollicutes. Organization of the hitloci in Chlamydiaceae ------------------------------------------------ In the obligately intra cellular Chlamydiaceaethe order of genes within the hitloci and the function of the encoded proteins appeared to be highly conserved, but distinctly different from that of the Mollicutes(Figure [1C](#F1){ref-type="fig"}). In all chlamydial species analyzed, the gene upstream of hitL encoded a putative cytoplasmic protein with the signature sequence of a metal-dependent protein hydrolase and a large number of metal binding residues (IPR003226/UPF0160), probably indicative of a phosphoesterase function, and an [o]{.underline}ligonucleotide/oligosaccharide-[b]{.underline}inding OB fold (IPR008994). The deduced protein sequence of the gene located downstream of hitL contained a 37--47 AA long tandemly repeated ARM repeat fold (IPR008938), which forms a right-handed superhelix and has been implicated in the mediation of protein-protein interactions \[[@B12]\]. Cp267 contained an RGD motif, which plays a role in cell adhesion \[[@B13]\]. However, the presence of RGD in a sequence alone is not sufficient to suggest a biological function for this motif \[[@B14]\]. The presence of two transmembrane helices (AA 314 to AA 334 and AA 428 to AA 448) suggested that it is more likely that Cp267 may interact with the bacterial cell membrane. In all other chlamydial species analyzed, the Cp267 homologues did not contain domains suggestive of membrane attachment. Are the hitloci genes co-expressed? ------------------------------------- In bacteria, overlapping genes on the same coding strand may indicate the polycistronic organization of these genes, and co-expressed proteins are often related in function. As the hitlocus of M. hominisis an operon containing three genes \[[@B6]\] the next step was to analyze whether the polycistronic organization of hitgenes was conserved within the Mollicutesand whether this also occurred in Chlamydiaceae. Initially, we chose M. pulmonis, the species with the greatest similarity to M. hominis, for reverse transcription (RT) PCR analysis. To identify the likely boundaries of an mRNA containing MYPU\_0060, MYPU\_0070 and MYPU\_0080, two different primers upstream of the MYPU\_0060/hitA gene and three primers downstream of the hitL gene were used (Fig. [2A](#F2){ref-type="fig"}). Amplification occurred with primers binding just upstream of the predicted promoter regions 1 or 2 and just downstream of the TAA stop codon of the hitL gene (Fig. [2B](#F2){ref-type="fig"}, lanes A and B) indicating a polycistronic mRNA. Amplification was not seen with a primer hybridizing to the sequence downstream of a predicted hairpin loop structure after the TAA stop codon of hitL (Figure [2B](#F2){ref-type="fig"}, lanes C1 and C2) suggesting that the mRNA terminates at the predicted hairpin structure. Next, we examined the genes flanking hitL in U. parvum. Amplicons which spanned the intergenic regions were obtained with primers hybridizing to the center of UU270 and to the 3\'-end of UU272 (Fig. [3C](#F3){ref-type="fig"}), and with primers hybridizing to the 5\'-end of UU271 and to the 3\'-end of UU273 (Fig. [3D](#F3){ref-type="fig"}). No amplification occurred with a primer hybridizing to the 5\'-end of UU270 and the 3\'-end of hitL (Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). These data suggest that HinT is expressed with the flanking genes UU271 and UU273. These findings are in accordance with the predicted termination of transcription by a hairpin loop with a stem energy of -9.1 kcal mole^-1^located 33 nt downstream of the UU273 gene \[[@B15]\]. Of the organisms analyzed so far U. parvumwas the most demanding organism of the Mollicutesin terms of DNA-free, full-length total RNA preparation. Thus the detection of a common RNA from the center of UU270 up to UU273 (Fig. [3D](#F3){ref-type="fig"}) may be due to an unstable RNA or indicatory for UU270 not taking part in the operon structure. To characterize the organization of hitloci genes within the Chlamydiaceaewe analyzed the locus in Chlamydophila pneumoniae. The genes Cp266 (encoding HinT), Cp267 and Cp268, which is predicted to encode a solute symporter family protein \[[@B16]\], were predicted to comprise an operon. No amplification occurred when a primer pair that hybridized upstream of Cp265 and downstream of Cp267 was used (data not shown). Amplification was detected using primers binding within each gene (Figure [4A,B,C](#F4){ref-type="fig"}) and with primers binding within Cp265 and Cp266 (Fig. [4D](#F4){ref-type="fig"}) and within Cp266 and Cp267 (Fig. [4E](#F4){ref-type="fig"}). The predicted co-expression of Cp267 and Cp268 downstream of which several putative transcription terminators were detected, was not supported by RT-PCR analysis (Fig. [4F](#F4){ref-type="fig"}), with amplification being obtained when using genomic DNA as a template but not with cDNA. Interactions between the proteins encoded by the hitlocus of C. pneumoniae ------------------------------------------------------------------------------ As the stability of an RNA is increased in regions with pronounced secondary structures, such as transcriptional terminators, and is often decreased in regions lacking such protective structures \[[@B17]\], partial degradation of the chlamydial mRNA may have occurred. The results from RT-PCR analysis of the mRNA of CP265, Cp266 and Cp267 suggest an operon producing a transient mRNA thus suggesting a physical interaction between the products of these genes. Therefore we analyzed the interactions between Cp265, Cp266 and Cp267 using the yeast two-hybrid system \[[@B18]\]. The system is based on the modular organization of the transcription factor GAL4, which has a DNA-binding domain (DB) and an activation domain (AD). When GAL4 binds (via DB) to its cognate binding site, the activation domain AD is brought close to the promoter and activates transcription of the reporter genes lacZand HIS3. In the yeast two-hybrid system these two domains, which alone are not able to activate transcription, are separately cloned into pGADT7 (AD) and pGBKT7 (BD) and expressed in fusion with the proteins under investigation. Growth of a histidine dependent yeast strain on histidine deficient media will only occur after transformation with both plasmids if the AD and DB domains are apposed by an interaction between the fusion proteins, which will result in transcription of the reporter genes. The plasmids pGADT7-T and pGBKT7-53, which encode the SV40 large T-antigen and the murine p53 protein, respectively served as a positive control (Fig. [5](#F5){ref-type="fig"} K+; Fig. [6A](#F6){ref-type="fig"}). HinT is known to dimerize \[[@B19],[@B20]\]. We introduced the coding region of M. hominisHinT into both pGADT7 and pGBKT7 and transformed the histidine dependent yeast strain AH109 with them. Quantification of transcriptional activation of the reporter gene was achieved using a liquid β-galactosidase assay measuring the hydrolysis of 4-methylumbelliferyl-β, D-galactopyranoside (MUG), yielding the fluorescent molecule 4-methylumbelliferone (4 MU). As expected, dimerization of the fused HinT peptides led to apposition of the AD and DB domains generating a functional transcription factor. The transcription of the reporter genes resulted in growth of the yeast on histidine-deficient agar plates (Fig. [5](#F5){ref-type="fig"}, top row, left plate) and production of β-galactosidase (Fig. [6C](#F6){ref-type="fig"}). Next we analyzed whether the proteins encoded by the hitlocus of C. pneumoniaeinteracted. The coding regions of Cp265, Cp266 and Cp267 were introduced into pGADT7 and pGBKT7 and expressed in yeast in all possible combinations. As shown in Figure [5](#F5){ref-type="fig"}, large colonies, comparable in size to those of the positive control, were produced by yeast that expressed Cp265 fused to both the binding and the activation domain. This suggested a strong interaction between the molecules of the Cp265 protein. All other combinations of plasmids encoding Cp265, Cp266 and Cp267 only led to the growth of small colonies, a finding that was difficult to interpret. To differentiate between strong and weak interactions, the β-galactosidase assay was used. As shown in Fig. [6](#F6){ref-type="fig"}, the measurement of β-galactoside activity indicated an interaction between Cp267 and Cp265, Cp266 and Cp267 itself, independent of the fusion partner (the AD or DB domains of the GAL4 protein). Surprisingly, dimerization of Cp266 (HinT) was not detected, although Cp265 interacted strongly with itself. Cp267 did not activate transcription by itself (Fig. [6A](#F6){ref-type="fig"}. 267/BD and 267/AD) nor did it interact with the unrelated protein SV40 large T antigen (T/267) or murine tumor suppressor p53 (267/p53) to activate transcription, as in all these cases the β-galactosidase activity was at background levels (Fig. [6A](#F6){ref-type="fig"}). Immune co-precipitation of Cp265, Cp266 and Cp267 ------------------------------------------------- To confirm the findings derived from the two-hybrid analyses we performed protein-binding assays in vitro. Protein C-tagged Cp265, Cp266 and Cp267 were expressed in E. coliand purified by affinity chromatography (Fig. [7A](#F7){ref-type="fig"}). Each protein was then incubated with equal amounts of \[^35^S\]-labeled Cp265, Cp266 or Cp267 in vitroand precipitated with anti-Protein C antibodies. As shown in Fig. [7B](#F7){ref-type="fig"}, the supernatants of the different samples contained comparable amounts of the respective isotope labeled proteins. As expected the \[^35^S\] labeled Cp proteins alone were neither precipitated by unrelated membrane proteins, such as Protein C-tagged OppA of M. hominisnor by the anti-Protein C-sepharose itself (Fig. [7C](#F7){ref-type="fig"}). In contrast, an interaction of Cp266 with Cp267 was suggested by co-precipitation of significant amounts of Cp266 \[35S\] with Cp267^C^, and Cp267 \[35S\] with Cp266^C^(Fig. [7C](#F7){ref-type="fig"}). The other findings from the two-hybrid analyses were also confirmed as both Cp265 and Cp267 were found to dimerize and there was no interaction between Cp265 and Cp266. Interestingly dimerization of Cp266 (HinT) which was not detected by the two-hybrid system, was demonstrated in this co-precipitation assay. Thus, we can conclude that the chlamydial proteins Cp265, Cp266 (HinT) and Cp267 have a tendency to dimerize and Cp267 interacts with both Cp265 and Cp266. Discussion ========== A remarkable feature of HinT proteins is their ubiquity. They are found in all kingdoms ranging from Mycoplasma genitalium, the smallest prokaryote, to one of the most complex eukaryotes, the human \[[@B1],[@B19]\]. However the function of this homo-dimer in the cytoplasm of these different organisms has only been closely examined in eukaryotes, where it has been characterized as an intracellular receptor for purine ribonucleotides \[[@B4],[@B5]\], especially for hydrolases of 5\'-monophosphoramide substrates such as AMP-lysine \[[@B21]\], with the histidine triad motif forming the α-phosphate binding site \[[@B20]\]. HinT has been shown to associate with the protein microphthalmia, an important transcription factor that controls growth and function in mast cells and melanocytes \[[@B22]\], to interact with the human cyclin-dependent kinase 7 (Cdk7), a subunit of the RNA polymerase II C-terminal domain kinase Cdk7/Kin28 \[[@B23]\], and to act as a positive regulator of the yeast Cdk7 homologue Kin28 \[[@B21]\]. That it may not be the key regulator of Cdk7 activity was recently suggested by analysis of HinT^-/-^knock-out mice \[[@B24]\]. This led to the hypothesis that eukaryotic HinT homologues are involved in the regulation of transcriptional processes by hydrolyzing related adenylyl-modified proteins \[[@B25]\]. The observation that HinT homologues are ubiquitous and that they are even encoded by the smallest bacterial genomes, suggests that HinT was present at the cellular root of the tree of life and that its preservation is advantageous for the survival of cells \[[@B20]\]. However the function of bacterial HinT homologues seems to be quite different from those of eukaryotes. In M. hominisHinT had been demonstrated to interact both physically and genetically with a surface-localized membrane complex by binding to the P80 domain \[[@B6]\]. The data presented here suggest that the co-expression of HinT with membrane proteins is prevalent within the Mollicutes. Comparable organization of the hit-loci was found in Mycoplasma hominis, M. pulmonis, M. mycoidessubsp. mycoidesSC, M. mobileand Mesoplasma florum. Analysis of further genomes will most likely lengthen this list. A genomic DNA fragment of M. bovisencoding the amino-terminal end of a P80 homologue has been sequenced and this homologue is predicted to possess an SPase I cleavage site. In M. hyorhinis, a nearly complete hitlocus, encoding P80, P60 and HinT homologues has been found adjacent to a genomic region encoding the high affinity transport system (personal communication of Michael Calcutt, Columbia, MO, USA) \[[@B26]\]. The organization of the hitABL genes in an operon, as recently demonstrated for M. hominis\[[@B6]\] and shown here for M. pulmonis, suggests polycistronic expression of the hitloci genes. However, it remains to be elucidated whether the encoded proteins have comparable functions to those already shown in M. hominis. P80 of M. hominiswas recently shown to reside in the membrane as a precursor protein and to be secreted into the extra cellular milieu as a 10 kDa smaller antigen. Processing of P80 was suggested to be initiated by SPase I cleavage \[[@B27]\]. While the P80 homologues in M. mobileand M. bovisalso contained signal sequences for SPase I cleavage, those in M. mycoidessubsp. mycoidesSC and Mesoplasma florumwere putative pro-lipoproteins. Secretion of lipoproteins has been observed in M. hominis\[[@B6]\] and other bacteria \[[@B28]\], indicating that similar function may be still possible. M. pneumoniaeand M. genitaliumeach had only one gene adjacent to hitL encoding a pore-forming protein with homologies to ABC permeases, suggesting they are part of a distinct phylogenetic branch. Interestingly, in M. penetransthe gene upstream of hitL also encodes a protein of an ABC transporter, an ATPase \[[@B29]\]. The hitlocus of U. parvum, which contained a P60 homologue and a gene encoding a permease, may have functions that are a hybrid of those of M. hominisand the M. pneumoniaegroups. RT-PCR analyses infer the presence of a mRNA encoding HinT and UU271, a P60-homologue, and a membrane protein of unknown function. Thus, the interaction of HinT with membrane proteins seems to be a common phenomenon in Mollicutes. A quite different situation was detected in the Chlamydiaceae. Although we were not able to establish definitely the presence of a polycistronic mRNA derived from the three genes comprising hitlocus, the overlap of the hitL gene with the upstream gene, as well as the conserved order of the three genes within the hitloci of Chlamydiaceae, suggested a relationship between the gene products. Analysis of Cp265, Cp266 (HinT) and Cp267 in the yeast two-hybrid system and immune co-precipitation assays confirmed our hypothesis that these proteins interacted and demonstrated that they form homo-dimers. Cp267, a protein with large areas of ARM repeats, which are known to mediate protein-protein interactions \[[@B30]\], was shown to interact with HinT, a protein known in eukaryotes to influence transcriptional activation, as well as with Cp265, a putative metal-dependent hydrolase with a binding fold for short, single stranded nucleic acids. These findings suggest that in Chlamydiaceae, the function of HinT may be more closely related to intracellular processes than to interactions with the extracellular milieu, as suggested by the findings on Mollicutes. Conclusion ========== The data presented here demonstrate that HinT proteins of the Chlamydiaceaeassociate with probable cytosolic proteins likely to function in the regulation of cellular processes, such as nucleotide metabolism. This function would be similar to that of homologues in eukaryotes, where HinT has been shown to influence transcription. The finding that HinT proteins of the Mollicutesinteract both physically and genetically with membrane proteins may reflect a phylogenic shift towards a different function in this group, or may indicate an additional function of bacterial HinT proteins. Our future work will focus on decoding the intra- and extra-cellular processes that bacterial HinT proteins are involved in. Methods ======= DNA manipulations ----------------- All routine DNA manipulation techniques, including plasmid preparation, restriction endonuclease analysis, ligation and transformation of E. coliwere performed as described by Sambrook et al. \[[@B31]\] or according to the manufacturers\' recommendations (Qiagen, Hilden, Germany, and Roche Applied Science, Mannheim, Germany). Bacterial strains and plasmids ------------------------------ The plasmids pXB (Roche Applied Science, Mannheim, Germany), pGADT7 and pGBKT7 (BD Bioscience Clontech, Palo Alto, USA) were used as expression vectors for the heterologous expression of Cp265, Cp266 and Cp267. The pXB plasmids were propagated in Escherichia coliDH5α F\'(Life Technologies), pGADT7 and pGBKT7 in yeast AH109 cells (BD Biosciences Clontech, Palo Alto, CA USA). Sequence analysis ----------------- Analysis of the DNA and protein sequences and the design of oligonucleotides were facilitated by the Lasergene software (DNA Star Inc., Madison Wisc.). The genomes listed in table [1](#T1){ref-type="table"} were analyzed using the web site of NCBI <http://www.ncbi.nlm.nih.gov/genomes/MICROBES/Complete.html>. Computer-based prediction of transcription termination was performed using the GeneBee service (online <http://www.genebee.msu.su/services/rna2_reduced.html>). Distant relation-ships between HinT protein sequences of the analyzed Mollicuteswere determined by using the BLAST method through Molligen 1.4 \[[@B32]\]. Conserved domains of proteins were predicted using the Conserved Domain Database (CDD) or the InterPro service (<http://www.ncbi.nlm.nih.gov/Structure/cdd/cdd.shtml>, <http://www.ebi.ac.uk/interpro/>). RT-PCR ------ RNA was prepared from exponential phase cultures of M. pulmonisand U. parvumas described by Chomczynski and Sacchi \[[@B33]\]. Chlamydophila pneumoniaerespiratory isolate GiD was used for all experiments \[[@B34]\]. Chlamydial RNA was prepared from HEp-2 cells 2 days post-infection as described by Mölleken et al. (Mölleken et al., manuscript in preparartion). Before use as a template for RT-PCR, contaminating traces of DNA were digested with DNase I as described previously \[[@B35]\]. Overlapping regions of the Ureaplasmaand Chlamydophilagenes were amplified using the Expand Template PCR system (Roche Applied Science, Mannheim, Germany) by standard PCR conditions (initial cycle of 5 min at 95°C; 35 cycles of 1 min at 95°C, 1 min at 50°C (U. parvum) or 58°C (C. pneumoniae), 2.5 min at 68°C). The hitoperon of M. pulmoniswas amplified using the Expand Long Template PCR system by standard PCR conditions (initial cycle of 5 min at 95°C; 10 cycles of 1 min at 95°C, 1 min at 44°C, 4 min at 68°C; addition of 0.5 U polymerase; then 20 cycles of 1 min at 95°C, 30 sec at 44°C, 4 min (+ 20 sec per cycle) at 68 °C). Southern blot analysis was performed as described previously \[[@B6]\]. Construction of plasmids ------------------------ The regions encoding Cp265, Cp266 and Cp267 in C. pneumoniaeGiD \[[@B34]\] were amplified by PCR with the primers Cp1 to Cp6 (Table [2](#T2){ref-type="table"}). The amplicons were cut at the primer-introduced restriction endonuclease cleavage sites for BamHI and SacI, and ligated in-frame into the expression vector pGADT7. The EcoRI/SacI inserts in the pGADT7 constructs were then ligated in-frame into EcoRI/PstI-digested pGBKT7. For heterologous expression of the protein-C tagged proteins, CP265^C^, Cp266^C^and Cp267^C^, in E. coli, the Cp inserts of the respective pGADT7 plasmids were digested with SacI, the ends bluntened by incubation with 3 units S1 nuclease (Amersham Bioscience, Freiburg, Germany) for 30 min at 37°C and subsequently cut with EcoRI. The blunt-end/EcoRI digested Cp265, Cp266 and Cp267 fragments were cloned in-frame into the blunt-end/EcoRI digested plasmid pXB and propagated in E. coliDh5α F\'. All plasmid constructs were confirmed by DNA sequencing (ABI 373 A machine) \[[@B36]\]. Expression and purification of Cp265^C^, Cp266^C^and Cp267^C^ ------------------------------------------------------------- One liter of LB broth (Gibco BRL, Life Technologies Inc., Gaithersburg, USA) containing ampicillin (100 μg/ml) was inoculated with the respective E. coliDH5α F\' clone and incubated for 16 h at 37°C with vigorous shaking. The cells were harvested by centrifugation (15,000 × g, 20 min, 4°C) and frozen at -70°C. After thawing on ice, the cells were resuspended in 40 ml buffer A (120 mM NaCl, 1 mM CaCl~2~, 20 mM Tris-HCl pH 7.5) and disrupted by three repeated freeze-thaw cycles in liquid nitrogen, followed by three bursts of sonication on ice (5 min bursts at 95 W with a 1 min cooling period between each burst). Insoluble material was sedimented (15,000 × g, 20 min, 4°C) and the supernatant was transferred to an anti-Protein C affinity matrix (Roche Applied Science, Mannheim, Germany). Bound proteins were eluted from the anti-Protein C affinity matrix with 2 mM EDTA. SDS-PAGE and immunostaining of proteins --------------------------------------- Proteins were separated on 12 or 15 % polyacrylamide gels \[[@B37]\], transferred to nitrocellulose (Schleicher and Schüll, Dassel, Germany) using a semi-dry blotting apparatus (Phase, Mölln, Germany) \[[@B38]\], and immunostained using anti-Protein C peroxidase (Roche Applied Science, Mannheim, Germany), anti HA-antibody (pGADT7) or anti c-myc antibody (pGBKT7) (BD Bioscience Clontech, Palo Alto, USA). Immune co-precipitation assay ----------------------------- The \[^35^S\]-labeled Cp265, Cp266 and Cp267 proteins were generated from pGADT7 and pGBKT7 fusion vectors by in vitrotranslation using a T7 coupled transcription/translation system in the presence of \[^35^S\]-methionine (Promega, Madison, USA). Two micrograms of purified Protein C-tagged protein were incubated with 10 μl of the \[^35^S\]-labeled protein in buffer B (20 mM Tris/HCl (pH 7.5), 75 mM KCl, 0.1 mM EDTA, 2.5 mM MgCl~2~, 0.05 (v/v) % Igepal, 2 mM DTT, 10 (v/v) % glycerol and 2 mM CaCl~2~) for 1 h \[[@B39]\]. Thereafter, a 5 μl aliquot of a slurry of the anti-Protein C matrix, equilibrated in buffer B containing 5 μg BSA, was added to each sample and the mixture further incubated for 1 h. The Protein C-matrix was washed eight times in buffer B, resuspended in 20 μl of SDS-PAGE sample buffer and heated at 95°C for 5 min. The proteins were then analyzed by 15 % SDS-PAGE and autoradiography using a phosphorimager (Fujifilm FLA 3000). Yeast two-hybrid assays ----------------------- The MATCHMAKER Gal4 two-hybrid system (BD Bioscience Clontech, Palo Alto, USA) was used for interaction assays. AH109 cells were co-transformed with the Cp265, Cp266 or Cp267 expressing pGADT-T7 and pGBKT7 plasmids, grown in small scale cultures and plated onto histidine deficient agar plates, followed by a β-galactosidase colony lift filter assay. All procedures were carried out as outlined in the BD Bioscience Clontech protocol. The β-galactosidase assay was performed after the inoculation of 7.5 ml YPD Medium with 0.5 ml overnight culture in selective medium and incubation for five hours at 30°C until the culture reached an OD~600~of 0.4 to 0.6. The cells were harvested by centrifugation at 14,000 × g for 1 min and re-suspended in 300 μl Buffer Z (Na~2~HPO~4~+ 7 × H~2~O, NaH~2~PO~4~+ 4 × H~2~O, KCl, MgSO~4~× 7 H~2~O, pH7.0). Lysis of the cells was achieved with four cycles of freezing in liquid nitrogen and thawing at 37°C. Soluble and insoluble components were separated by centrifugation at 14000 × g for 2 min. The β-galactosidase activity was measured using a FluorAce™ β-galactosidase Reporter Assay kit (BioRad, Hercules, USA). Authors\' contributions ======================= MH carried out the immunoassays and Yeast TwoHybrid assays. BH carried out the molecular genetic studies. BH and JH participated in the design of the study and drafted the manuscript. All authors have read and approved of the final manuscript. Acknowledgements ================ This work was supported by the Deutsche Forschungsgemeinschaft (DFG-HE 2028/3-2). We thank Michael Calcutt (Columbia, MO, USA) for information about the hitloci in M. bovisand M. hyorhinis, Marzena Wyschkon for excellent technical assistance and Colin MacKenzie for critically reading the manuscript. Katja Mölleken is thanked for preparation of C. pneumoniaeRNA and genomic DNA. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *Schematic of the hitloci in Mollicutesand Chlamydiaceae.A schematic representation of genes within the hitloci of the following species: A) M. pulmonis, M. mycoidessubsp. mycoidesSC, Mesoplasma florumand M. mobilecontaining genes homologous to hitABL of M. hominis; B) U. parvum, M. pneumoniaeand M. genitalium, each containing a gene upstream of hitL predicted to encode an integral pore-forming protein; and C) C. pneumoniae, C. trachomatis, C. muridarumand C. caviaecontaining hitL flanking genes predicted to encode, upstream, a protein with the signature sequence of a metal-dependent hydrolase (light blue region), with an OB nucleic acid binding fold (dark blue region) and, downstream, a protein with ARM repeats (red regions), which are known to mediate protein-protein interactions. Triangles indicate signal peptidase recognition sites of SPase I and SPase II. Transmembrane regions are depicted as striped regions. The position of the RGD tri-peptide is marked by a dotted region. ::: ![](1471-2180-5-27-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### RT-PCR analysis of M. pulmonis.A. The regions amplified by RT-PCR are shown below the genomic region encompassing MYPU\_0060, MYPU\_0070 and MYPU\_0080. The primers used and the lengths of the amplicons are indicated. B. Amplicons were separated on a 0.8 % agarose gel and subjected to Southern blot analysis, here shown for the MYPU\_0070 specific probe, using digoxigenin (DIG)-labeled probes hybridizing to each of the three genes, detected using chemiluminescence. Bands of lower length as expected are degradation products. M, DIG-labeled DNA molecular weight marker VII (Roche Biochemicals). ::: ![](1471-2180-5-27-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### RT-PCR analysis of U. parvum.A. The positions of the different amplicons are shown below the schematic of the hitlocus genes of U. parvum. The primers used (Table 2) and the lengths of the amplicons are indicated. B. The PCR products (A -- D) for genomic DNA (g), cDNA (c) and RNA (r) were separated on a 0.6 % agarose gel and stained with ethidium bromide. Southern blot analysis was performed with digoxigenin (DIG)-labeled probes hybridizing to one of the four genes, (here shown for UU272 probing), and detected using chemiluminescence. M, Gene Ruler 1 kb DNA ladder (Fermentas). ::: ![](1471-2180-5-27-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### RT-PCR analysis of C. pneumoniae.**A. Below the schematic of the hitlocus of C. pneumoniaethe positions of the PCR amplicons (A -- F) are shown. Genomic DNA (g), cDNA (c) and RNA without RT reaction (r) are used as templates. B. The PCR products were separated on a 0.8 % agarose gel and subjected to Southern blot analysis with DIG-labeled probes hybridizing to Cp265 (A, D), Cp266 / hitL (B, E) Cp267 (C, F) with visualization using chemiluminescence. In E, the signals of lower length as 1.4 kb may be due to primer dimerization. M, DIG-labeled DNA molecular weight marker VII (Roche Biochemicals) ::: ![](1471-2180-5-27-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### The yeast-two hybrid assay.AH109 were co-transformed with the pGADT7 (AD) and pGBKT7 (BD) constructs containing hitL of M. hominis, Cp265, Cp266 or Cp267 from C. pneumoniae, pGADT7-T, expressing the SV40 large T-antigen and pGBKT7-53, expressing the murine p53 protein (K+), or pGADT7 and pGBKT7 without fusion partners (K-). Yeast cultures were incubated at 30°C on histidine-deficient agar plates for 4 days. ::: ![](1471-2180-5-27-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Quantitation of β-galactosidase activity.In yeast cells, co-transformed with pGADT7 (AD) and pGBKT7 (BD) constructs as indicated, β-galactosidase activity was quantified using 4-methylumbelliferyl-galactopyranoside (MUG) hydrolysis to the fluorescent molecule, 4-methylumbelliferone (4 MU). The fluorescence of 4 MU was excited at 360 nm and emission measured at 460 nm. ::: ![](1471-2180-5-27-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Immune co-precipitation.A. Affinity purified proteins, Cp265 (265^C^), Cp266 (266^C^), Cp267 (267^C^) and OppA^C^, were analyzed on 15 % SDS-polyacrylamide gels by Coomassie staining (C) or on Western blots by immunostaining with an anti-Protein C antibody (W). \[^35^S\]-labeled Cp265, Cp266 and Cp267 (Captives) were incubated at room temperature for 1 h with Cp265^C^, Cp266^C^, Cp267^C^and OppA^C^(Catcher). Immune complexes were precipitated with anti-Protein C-matrix, and the supernatants (B.) and precipitates (C.) analyzed on 15 % SDS-polyacrylamide gels by autoradiography using a phosphorimager. As a negative control, the \[^35^S\]-labeled proteins were subjected to the Protein C-matrix without a catcher (Sepha.^αC^). M, prestained low-molecular weight marker (Biorad) ::: ![](1471-2180-5-27-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Organisms used for genome sequence analyses ::: ----------------------------------------------- ------------ Chlamydophila caviaeGPIC NC\_003361 Chlamydia muridarum* NC\_002620 Chlamydophila pneumoniaeAR39 NC\_002179 Chlamydia trachomatisD/UW-3/CX NC\_000117 Mesoplasma florumL1 NC\_006055 Mycoplasma gallisepticumR NC\_004829 Mycoplasma genitaliumG-37 NC\_000908 Mycoplasma mobile163 K NC\_006908 Mycoplasma mycoides spp.mycoides SCstr. PG1 NC\_005364 Mycoplasma pneumoniaeM129 NC\_000912 Mycoplasma penetransHF2 NC\_004432 Mycoplasma pulmonisUAB CTIP NC\_002771 Parachlamydia sp.UWE25 NC\_005861 Ureaplasma parvumserovar 3 str. ATCC 700970 NC\_002162 ----------------------------------------------- ------------ ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Primers used in PCR ::: Primer Length Sequence (5\'-3\') ------------- ------------ ---------------------------------------------- MP1 27 ATTTATTTTTGATTACTATTTTTGAAC MP2 23 AATTTGTTTTATTATTTTGTTTA MP8 20 TTAATGTTCATGTCCTTCTT MP9 21 AATCCAGAAAGTTTTTTTATT MP12 24 ATTTGTTATTGCTAAAAGATTGTA UU1 22 ATGATCAATAAAAACAAAAAAT UU2 19 GCGCCGTTAGTGATGTTAG UU6 32 TGAGCTCTTCCATATTTTTTGATTATACCATA UU7 23 AAAAACGAATTGCCTCTATGTAT Cp1 21 CATGATGCGCGTGGAGTAGGT Cp2 25 TCCAAGAAGCATTGGTACTCACGAT Cp3 22 CGGTTTCCATGGCTTCTCTGAC Cp4 29 TTATCGATGGATTGATAGATTGTGAAAAG Cp5 22 GGACGCCCACCTAAAAGATGAA Cp6 21 TTTCGCTTGGTCCCTCTCCTG Cp7 25 GCTCTTTCAATATCTTCACGGCTCA Cp8 28 TACGGTGCTGAGCTTAAAGATAGAAATG Cp\_265 up 27 [ATGGATCCTA]{.underline}ATGGAGGATTGGCTAAG Cp\_265 low 29 [TTGAGCTC]{.underline}TCATATGATCCCTCTATCTTG Cp\_266 up 27 [TTAGGATCCTC]{.underline}ATGCCTACGTGCTTTG Cp\_266 low 29 [TTTGAGCTC]{.underline}ATCAGGCTATAGCACCTAAA Cp\_267 up 23 [ATTGGATCC]{.underline}CCATGTTCGGCTCG Cp\_267 low 30 [TTAGAGCTC]{.underline}ACTTGAAAATAGAGAAAAGAG Cp1 BamHI 27 [ATGGATCC]{.underline}ATGGAGGATTGGCTAAG Cp2 BamHI 27 [TTAGGATCC]{.underline}ATGCCTACGTGCTTTG Cp3 BamHI 23 [ATTGGATCC]{.underline}ATGTTCGGCTCG :::	PubMed Central	2024-06-05T03:55:58.391092	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173108/", "journal": "BMC Microbiol. 2005 May 18; 5:27", "authors": [ { "first": "Miriam", "last": "Hopfe" }, { "first": "Johannes H", "last": "Hegemann" }, { "first": "Birgit", "last": "Henrich" } ] }
PMC1173109	Background ========== The elucidation of the P. aeruginosastrain PAO1 genome sequence in 2000 \[[@B1]\] opened the doors for genome- and proteome-scale research with this important opportunistic pathogen. These include the availability of commercial (Affymetrix, Santa Clara, CA) P. aeruginosaGeneChips^®^for transcriptome analyses \[[@B2]\], as well as the P. aeruginosaPAO1 gene collection for high-throughput expression and purification of recombinant proteins \[[@B3]\]. A comprehensive transposon mutant library was generated containing over 30,000 sequence-defined mutants, corresponding to an average of five insertions per gene \[[@B4]\]. However, despite the development of many genetic tools for P. aeruginosaover the past decade \[[@B5],[@B6]\], isolation of defined deletion mutants is still a relatively tedious process, which relies on construction of deletion alleles, most often tagged with an antibiotic resistance gene, on a suicide plasmid, followed by recombination of the plasmid-borne deletions into the chromosome, usually after conjugal transfer of the suicide plasmid (reviewed in \[[@B6]\]). Merodiploids arising from chromosomal integration of the suicide plasmid are resolved by utilization of a counterselectable marker, most often Bacillus subtilis sacB(sucrose counterselection) \[[@B7]\] or, less frequently, rpsL(streptomycin counterselection)\[[@B8]\]. In cases where the antibiotic selection markers are flanked by site specific recombination sites, e.g., the Flp recombinase target (FRT) \[[@B9]\] or Cre recombinase (loxP) \[[@B10]\] sites, they can subsequently be deleted from the chromosome, resulting in unmarked deletion mutants. Unfortunately, the previously described method for one-step inactivation of chromosomal genes in Escherichia coliusing PCR products \[[@B11]\] does not work in P. aeruginosa, although we have used a modification \[[@B12]\] of this approach where the target gene is first cloned into a plasmid, followed by λ RED-mediated recombination of a PCR-generated mutated copy of the gene \[[@B13]\]. Here, we describe a more streamlined method for generation of chromosomal deletion mutants. In this method, a \"mutant\" fragment containing the 5\' and 3\' flanking regions of the target gene plus the antibiotic resistance gene is assembled by a modified \[[@B14]\] PCR overlap extension protocol \[[@B15]\] and then cloned into a suicide vector using the Gateway recombinational cloning system, similarly to a previously described method \[[@B16]\]. The plasmid-borne deletion allele is then transferred to P. aeruginosautilizing a rapid transformation procedure and recombined into the chromosome. The transformants are often merodiploids which are then resolved by sucrose counterselection against the undesired wild-type sequences. The antibiotic resistance marker is subsequently deleted from the chromosome by Flp recombinase. Results ======= Construction of a Gateway-compatible suicide vector --------------------------------------------------- We previously described the pEX family of relatively small, fully sequenced suicide plasmids which are now widely used for gene replacement experiments in P. aeruginosaand related bacteria \[[@B9]\]. Wolfgang et al. \[[@B16]\] recently described a Gateway-compatible pEX18Gm derivative. Here, we modified pEX18Ap for use as a Gateway destination vector for cloning of mutations marked with a gentamycin resistance (Gm^r^) encoding gene. We prefer using the Gm^r^marker because it is more easily transferable between P. aeruginosachromosomes by transformation with chromosomal DNA fragments when compared to other selection markers \[[@B17]\]. The resulting vector pEX18ApGW (Fig. [1](#F1){ref-type="fig"}) contains the functional sequences for recombination-based cloning (attR1and attR2) and the ccdBcounterselection marker, while maintaining the sacBcounterselection marker for downstream resolution of merodiploids. Generation of a mutant fragment by PCR overlap extension -------------------------------------------------------- The amplification of each mutant fragment was achieved in a two step PCR that required 8 PCR primers, 4 of which were unique for each fragment to be mutated and 4 of which were common primers (Table [1](#T1){ref-type="table"}). The gene-specific portions of the unique primers were designed to yield T~m~\'s of 55--57°C and PCR fragments ranging from \~200 to 600 bp. These parameters enabled utilization of the same PCR conditions for different target genes, while at the same time provided for amplification of sufficient chromosomal DNA to allow for efficient homologous recombination of plasmid-borne sequence with the chromosome. The basic strategy for generation of the mutant fragment by PCR overlap extension is illustrated in Fig. [2](#F2){ref-type="fig"}. In PCR1, the four gene-specific primers are used to amplify the 5\' and 3\' regions of the target gene. Simultaneously, two of the common primers, Gm-F and Gm-R, are used to amplify the Gm^r^cassette flanked by FRTsites from a pPS856 template. Since the Gm^r^fragment is used for all constructs, it can be prepared in larger amounts ahead of time and stored until needed. Fig. [3A](#F3){ref-type="fig"} illustrates typical results of PCR1 utilizing the example of P. aeruginosagene PA1520, encoding a putative transcriptional regulator of the GntR family. When we adhered to the prescribed PCR conditions, the amplified fragments were very clean. It was especially prudent to follow the conditions for PCR amplification of the Gm^r^FRTfragment since the FRTsites possess significant secondary structures. The PCR fragments were gel-purified and equal amounts (50 ng) of each were combined for PCR2. PCR2 was allowed to proceed for 3 cycles in the absence of exogenous primers, which for PA1520yields the intermittent 1,709 bp \'attB1-PA1520\'-FRT-Gm-FRT-\'PA1520-attB2\' fragment illustrated in Fig [3A](#F3){ref-type="fig"}. The common Gateway primers GW-attB1and GW-attB2were then added and the PCR continued for another 25 cycles to obtain the final, recombination-proficient DNA fragment, which constituted the predominant PCR product at this stage (Fig. [3B](#F3){ref-type="fig"}). For PA1520, the resulting 1,735 bp attB1-PA1520\'-FRT-Gm-FRT-\'PA1520-attB2fragment was gel-purified and used for the BP clonase reaction for construction of an entry clone. BP and LR clonase reactions --------------------------- The BP and LR clonase reactions (Fig. [4](#F4){ref-type="fig"}) were performed according to Invitrogen\'s (Carlsbad, CA) Gateway cloning manual. For the BP clonase reaction pDONR221 was used, but only half of the recommended BP clonase enzyme mix. Gm^r^and kanamycin resistant (Km^r^) DH5α or HPS1 transformants were selected and the presence of the correct plasmids was confirmed by XbaI digestion (the FRTsites flanking the Gm^r^gene each contain a single XbaI site). The insert of a correct pDONR-Gene::Gm entry clone was then recombined into the destination vector pEX18ApGW using the LR clonase protocol, again using only half of the recommended amount of LR clonase mix. Recombinants were transformed into DH5α and Ap^r^colonies were selected. We noticed that, at least under the conditions we used, many transformants exhibited a Ap^r^Km^r^phenotype, indicating that they either contained both pDONR-Gene::Gm and pEX18ApGW-Gene::Gm or, more likely, a cointegrate of both plasmids. The presence of the correct pEX18ApGW-Gene::Gm in transformants that were only Ap^r^was confirmed by XbaI digestion. Gene replacement ---------------- To expedite gene replacement, an improved electroporation method \[[@B17]\] was used for transfer of the suicide plasmid pEX18ApGW-Gene::Gm into P. aeruginosainstead of the traditionally employed conjugation method. Although after transfer into P. aeruginosadouble cross-over events can occur frequently (\>50%), they can also be rare (\<1--5%). In the latter instance, merodiploids formed via integration of the suicide plasmid by a single cross-over event (Fig. [4](#F4){ref-type="fig"}). The merodiploid state was then resolved via sucrose selection in the presence of gentamycin, resulting in deletion of the wild type gene. For generation of unmarked deletion mutants, the Gm^r^marker was subsequently removed by deletion of a 967 bp fragment using Flp recombinase. The presence of the desired deletion in either the marked or unmarked PA1520mutant was verified by colony PCR utilizing the gene-specific up and down primers (Fig. [5](#F5){ref-type="fig"}). Deletion of 25 genes encoding transcriptional regulators of the GntR family --------------------------------------------------------------------------- We are interested in characterizing the activator involved in regulation of expression of the fabABoperon, which encodes two essential enzymes of de novofatty acid biosynthesis \[[@B18]\]. Expression of the fabABoperon is repressed by exogenously added fatty acids, most effectively by oleic acid, and preliminary results from our laboratory indicate that this may be due to relief of activation by a transcriptional regulator which binds to a 30 nucleotide site located in the fabABpromoter region (Choi and Schweizer, unpublished results). This 30 bp region is only found and conserved in Pseudomonasspp. Its deletion dramatically reduces transcription of the fabABoperon and the residual transcriptional activity is no longer fatty acid responsive (Fig. [6](#F6){ref-type="fig"}). In E. coli, transcriptional activation of the unlinked fabAand fabBgenes is achieved by FadR, a regulator belonging to a group of transcriptional regulators with the GntR family signature \[[@B19],[@B20]\]. The PAO1 chromosome encodes more than 30 regulatory proteins with this signature. Of these, 27 have not been assigned function. Using the methodology described above for one of these genes, PA1520, we deleted 25 of the 27 genes (we could not delete the presumably essential PA1285and PA2299) in less than 4 weeks in strain PAO1 with a chromosomally integrated fabA\'-lacZtranscriptional fusion and analyzed β-galactosidase expression in the resulting mutant strains (Fig. [6](#F6){ref-type="fig"}). In exponential phase cells, the overall pattern of fabAexpression was indistinguishable from the parental strain in all 25 mutants, indicating that none of these genes encodes the fabABactivator. We would have expected an expression pattern similar to that observed in the PAO1::fabAΔ30\'-lacZcontrol strain which contains a deletion of the 30 nucleotide putative activator-binding site. Discussion ========== In our experience, the method described here for generation of unmarked deletion mutants is the fastest available to date. From start to finish, an individual unmarked P. aeruginosamutant can be generated and characterized in \~14 days, which is up to one week faster than using our previously described method \[[@B9]\]. However, the main advantage of the method does not lie in individual mutant generation but in the speed by which collective deletion mutant pools can be generated. For example, we isolated unmarked deletions in 25 of the 27 genes encoding transcriptional regulators of the GntR family in less than 4 weeks. Two genes could not be deleted, presumably because they encode essential proteins. Besides speed, what are some of the other unique features of this method? First, the deletion of gene sequences is not restricted by the natural availability or synthetic generation of restriction sites that are subsequently used to delete intervening sequences. With proper primer location, the PCR-based method allows clean deletion of more-or-less the entire coding sequence without fear of generating truncated genes which potentially produce products with undesired side effects. Second, Flp excision of the antibiotic resistance gene cassette leaves behind a short 85 bp FRTscar which does not exhibit any known polar effects on downstream genes. The method is therefore uniquely suited to study genes organized in polycistronic units. Flp excision is a very straightforward procedure with efficiencies approaching 100% \[[@B9]\]. Third, the method is economical, especially in a more high-throughput format. PCR overlap extension requires only four gene specific primers of reasonable length which keeps oligonucleotide costs to a minimum. In comparison to traditional methods, it also requires fewer hands-on steps and gene-specific reagents (e.g., restriction enzymes). Conclusion ========== Although not yet as fast as the PCR fragment-directed gene replacement method described for E. coli, the PCR- and recombinational cloning based method described opens the door for high-throughput generation of P. aeruginosadeletion mutants on a genome-wide scale. With appropriate modifications, i.e., choice of appropriate selection markers and Gateway-compatible gene replacement vectors, the method should be applicable to the many other bacteria in which the pEX vector system can be applied \[[@B6]\]. Methods ======= Media ----- Escherichia coliand P. aeruginosastrains were maintained on Luria-Bertani medium (LB; 10 g per liter tryptone, 5 g per liter yeast extract, 10 g per liter NaCl; Becton, Dickinson & Co., Sparks, MD). For plasmid maintenance in E. coli, the medium was supplemented with 100 μg per ml ampicillin, 25 μg per ml chloramphenicol, 35 μg per ml kanamycin or 15 μg per ml gentamycin. For marker selection in P. aeruginosa, 200 μg per ml carbenicillin and 30 μg per ml of gentamycin was used, as appropriate. β-galactosidase activity assays ------------------------------- Cells were grown at 37°C with shaking to exponential phase (optical density at 540 nm \~0.4--0.8) in LB medium with 0.05% Brij 58 +/- 0.05% oleic acid. β-galactosidase activity was measured in chloroform/sodium dodecyl sulfate-permeabilized cells and its activity calculated as previously described \[[@B21]\]. Standard DNA procedures ----------------------- Routine procedures were employed for manipulation of DNA \[[@B22]\]. Plasmid DNAs were isolated using the QIAprep Mini-spin kit (Qiagen, Valencia, CA) and P. aeruginosachromosomal DNA was obtained using the QIAamp DNA Mini Kit. DNA fragments were purified from agarose gels utilizing the QIAquick gel extraction kit. Construction of the Gateway-compatible gene replacement vector pEX18ApGW ------------------------------------------------------------------------ The previously described gene replacement vector pEX18Ap \[[@B9]\] was modified by cloning of a 1,755 bp HindIII-KpnI fragment from pUC18-mini-Tn7T-Gm-GW (GenBank accession number AY737004), followed by transformation into the gyrAstrain DB3.1 (Invitrogen). 1^st^round PCRs --------------- ### PCR-amplification of the gentamycin resistance gene cassette A 50 μl PCR reaction contained 5 ng pPS856 \[[@B9]\] template DNA, 1x HiFi Platinum Taqbuffer, 2 mM MgSO~4~, 200 μM dNTPs, 0.2 μM of Gm-F and Gm-R (Table [1](#T1){ref-type="table"}) and 5 units of HiFi Platinum Taqpolymerase (Invitrogen). Cycle conditions were 95°C for 2 min, followed by 30 cycles of 94°C for 30 s, 50°C for 30 s, and 68°C for 1 min 30 s, and a final extension at 68°C for 7 min. The resulting 1,053 bp PCR product was purified by agarose gel electrophoresis and its concentration determined spectrophotometrically (A~260\ nm~) in an Eppendorf Biophotometer (Hamburg, Germany) and using the 50--2000 μl Eppendorf UVettes ### PCR-amplification of 5\' and 3\' gene fragments Two 50 μl PCR reactions were prepared. The first reaction contained 20 ng chromosomal template DNA, 1x HiFi Platinum Taqbuffer, 2 mM MgSO~4~, 5% DMSO, 200 μM dNTPs, 0.8 μM of PA1520-UpF-GWL and PA1520-UpR-Gm (Table [1](#T1){ref-type="table"}) and 5 units of Platinum Taqpolymerase. The second reaction contained the same components as the first, but 0.8 μM of PA1520-DnF-Gm and PA1520-DnR-GWR (Table [1](#T1){ref-type="table"}). Cycle conditions were 94°C for 5 min, followed by 30 cycles of 94°C for 30 s, 56°C for 30 s, and 68°C for 30 s, and a final extension at 68°C for 10 min. The resulting PCR products were purified by agarose gel electrophoresis and their concentrations determined spectrophotometrically (A~260\ nm~). 2^nd^round PCR -------------- A 50 μl PCR reaction contained 50 ng each of the PA15205\' and 3\' purified template DNAs, and 50 ng of FRT-Gm-FRTtemplate DNA prepared during 1^st^round PCR. The reaction mix also contained 1x HiFi Platinum Taqbuffer, 2 mM MgSO~4~, 5% DMSO and 200 μM dNTPs, and 5 units of HiFi Platinum Taqpolymerase. After an initial denaturation at 94°C for 2 min, 3 cycles of 94°C for 30 s, 55°C for 30 s, and 68°C for 1 min were run without added primers. The third cycle was paused at 30 s of the 68°C extension, primers GW-attB1and GW-attB2were added to 0.2 μM each, and the cycle was then finished by another 30 s extension at 68°C. The PCR was completed by 25 cycles of 94°C for 30 s, 56°C for 30 s, and 68°C for 5 min, and a final extension at 68°C for 10 min. The resulting major PCR product was purified by agarose gel electrophoresis and its concentrations determined spectrophotometrically (A~260\ nm~). The identity of the PCR fragment was confirmed by XbaI digestion (each FRTsite of the FRT-Gm-FRTfragment contains a XbaI site). ### BP and LR clonase reactions The BP and LR clonase reactions for recombinational transfer of the PCR product into pDONR221 and pEX18ApGW, respectively, were performed as described in Invitrogen\'s Gateway cloning manual, but using only half of the recommended amounts of BP and LR clonase mixes and either E. coliDH5α \[[@B23]\] or HPS1 \[[@B24]\] as host strains. The presence of the correct fragments in transformants obtained with DNA from either clonase reaction was verified by digestion with XbaI because each FRTsite flanking the Gm^r^gene contains a XbaI site. However, before plasmid isolation from transformants obtained with DNA from the LR clonase reaction, 25--50 transformants were i) patched on LB+Km and LB+Ap plates, and ii) simultaneously purified for single colonies on LB+Ap plates. This was necessary to distinguish between those colonies containing only the desired pEX18ApGW-Gene::Gm from those containing this plasmid and the frequently contaminating pDONR-Gene::Gm (pEX18Ap-derived plasmids confer Ap^r^and pDONR plasmids confer Km^r^). Plasmids were then isolated from Ap^r^Km^s^transformants and analyzed by XbaI digestion. In those extremely rare instances where all patched transformants contained a mixed plasmid population (\<1% of to date performed LR clonase reactions), retransformation with Ap^r^selection was necessary to obtain colonies containing only pEX18ApGW-Gene::Gm. ### Transfer of plasmid-borne deletions to the P. aeruginosachromosome A rapid electroporation method described elsewhere \[[@B17]\] was used to transfer the pEX18ApGW-borne deletion mutations to P. aeruginosa. Briefly, 6 ml of an overnight culture grown in LB medium was harvested in 4 microcentrifuge tubes by centrifugation (1--2 min, 16,000 × g) at room temperature. Each cell pellet was washed twice with 1 ml of room temperature 300 mM sucrose and they were then combined in a total of 100 μl 300 mM sucrose. For electroporation, 300-500 ng of plasmid DNA was mixed with 100 μl of electrocompetent cells and transferred to a 2 mm gap width electroporation cuvette. After applying a pulse (settings: 25 μF; 200 Ohm; 2.5 kV on a Bio-Rad GenePulserXcell™), 1 ml of LB medium was added at once, and the cells were transferred to a 17 × 100 mm glass or polystyrene tube and shaken for 1 h at 37°C. The cells were then harvested in a microcentrifuge tube. 800 μl of the supernatant was discarded and the cell pellet resuspended in the residual medium. The entire mixture was then plated on two LB plates containing 30 μg per ml Gm (LB+Gm30). The plates were incubated at 37°C until colonies appeared (usually within 24 h). Under these conditions, the transformation efficiencies were generally 30--100 transformants per μg of DNA. A few colonies were patched on LB+Gm30 plates and LB+Cb200 plates to differentiate single- from double cross-over events. To ascertain resolution of merodiploids, Gm^r^colonies were struck for single colonies on LB+Gm30 plates containing 5% sucrose. Gm^r^colonies from the LB-Gm-sucrose plates were patched onto LB+Gm30+5% sucrose, as well as LB plates with 200 μg per ml carbenicillin (LB+Cb200). Colonies growing on the LB-Gm-sucrose, but not on the LB-carbenicillin plates were considered putative deletion mutants. The presence of the correct mutations was verified by colony PCR. To do this, a single, large colony (or the equivalent from a cell patch) was picked from a LB-Gm-sucrose plate, transferred to 30 μl H~2~O in a microcentrifuge tube and boiled for 5 min. Cell debris was removed by centrifugation in a microcentrifuge fuge (2 min; 16,000 × g), and the supernatant was transferred to a fresh tube which was placed on ice. 5 μl of the supernatant was used as source of template DNA in a 50 μl PCR reaction containing Taqbuffer, 1.5 mM MgSO~4~, 5% DMSO, 0.6 μM each of PA1520-UpF-GWL and PA1520-DnR-GWR, 200 μM dNTPs and 5 units Taqpolymerase (Fermentas, Hanover, MD). Cycle conditions were 95°C for 5 min, followed by 30 cycles of 95°C for 45 s, 55°C for 30 s, and 72°C for 2 min, and a final extension at 72°C for 10 min. PCR products were analyzed by agarose gel electrophoresis. ### Flp-mediated marker excision Electrocompetent cells of the newly constructed mutant strain were prepared as described in the preceding paragraph and transformed with 20 ng of pFLP2 \[[@B9]\] DNA as described above. After phenotypic expression at 37°C for 1 h, the cell suspension was diluted 1:1,000 and 1:10,000 with either LB or 0.9% NaCl, and 50 μl aliquots were plated on LB+Cb200 plates and incubated at 37°C until colonies appeared. Transformants were purified for single colonies on LB+Cb200 plates. Ten single colonies were tested for antibiotic-susceptibility on LB ± Gm30 plates and on a LB+Cb200 plate. Two Gm^s^Cb^r^isolates were struck for single colonies onto a LB+5% sucrose plate and incubated at 37°C until sucrose-resistant colonies appeared. Ten sucrose-resistant colonies were retested on a LB+5% sucrose (master) plate and a LB+Cb200 plate. Finally, two sucrose-resistant and Cb^s^colonies were struck on LB plates without antibiotics, and their Cb^s^and Gm^s^phenotypes confirmed by patching on LB ± Cb200 and LB ± Gm30 plates. Deletion of the Gm^r^marker was assessed by colony PCR utilizing the conditions and primers described in the preceding paragraph Authors\' contributions ======================= KHC, performance of experiments, writing of manuscript. HPS, design and coordination of the study, data evaluation, direct supervision of experimental work, writing of manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table 2 -- Sequences of PCR primers used to amplify genes encoding transcriptional regulators of the GntR family ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was supported by NIH grant AI058141 to HPS. The authors wish to thank Dr. Donald Moir (Microbiotix, Inc.) for helpful advice regarding use of the PCR overlap extension technology with GC-rich DNA. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Map of the Gateway-compatible pEX18ApGW. This vector was derived by cloning a Gateway conversion fragment (grey) into the multiple cloning site of pEX18Ap (black). Only selected restriction enzyme cleavage sites are shown. Abbreviations: attR1and attR2, bacteriophage λ recombination sites; bla, β-lactamase-encoding gene; cat, chloramphenicol acetyl transferase-encoding gene; ccdB, gene encoding gyrase-modifying enzyme (CcdB poisons host DNA gyrase by forming a covalent complex with the DNA gyrase A subunit and thus serves as a counter-selectable marker in gyrA^+^cloning hosts \[25\]); ori, ColE1-derived replication origin; oriT, origin of conjugal transfer; sacB, Bacillus subtilislevansucrase-encoding gene. The sequence of this plasmid was deposited in GenBank and assigned accession number AY928469. ::: ![](1471-2180-5-30-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Schematic illustration of mutant fragment generation by overlap extension PCR. During first round PCR (PCR1), the 5\' and 3\' ends of the target genes, as well as the gentamycin (Gm) resistance cassette are amplified using four gene-specific primers (G-UpF-GWL, G-UpR-Gm, G-DnF-Gm and G-DnR-GWR) and the common Gm-specific primers (Gm-F and Gm-R). This generates three fragments with partial overlaps either to each other (indicated by the blue boxes signifying Gm overlap) or the attB1and attB2λ recombination sites (indicated by the green and pink boxes). These purified fragments are then assembled in vitroby overlap extension during second round PCR (PCR2) using common primers GW-attB1 and GW-attB2, resulting in a recombination-proficient mutant PCR fragment. ::: ![](1471-2180-5-30-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### *PCR amplification of the Gm FRTcassette, PA1520gene fragments and the overlap extension product. A.First round PCR fragments. The left panel illustrates amplification of the 1,053-bp Gm^r^fragment from pPS856 which contains 24 bp (right) and 25 bp (left) overlaps with the PA1520\' and \'PA1520fragments (blue boxes). The right panel illustrates amplification of the PA1520\' and \'PA1520fragments. The 5\' fragment contains 388 bp of chromosomal DNA, 25 bp overlapping the left side of the Gm^r^fragment and 16 bp overlapping the GW-attB1primer. Similarly, the 3\' fragment contains 236 bp of chromosomal DNA, 24 bp overlapping the right side of the Gm^r^fragment and 16 bp overlapping the GW-attB2primer. The sequences of the gene-specific and common primers are listed in Table 1. B.Second round PCR. The purified fragments shown in panel Awere used in the second round PCR illustrated in Fig. 2 to derive the indicated attB1-PA1520\'-FRT-Gm-FRT-\'PA1520-attB2fragment. The entire 50 μl second round PCR reaction was subjected to agarose gel electrophoresis. The desired DNA fragment constituting the major product marked by the arrow was excised from the gel, purified and then used for the BP clonase reaction. Lanes labeled M in both panels contained Hi-Lo molecular size markers from Minnesota Molecular (Minneapolis, MN). ::: ![](1471-2180-5-30-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Gateway-recombinational cloning and return of the plasmid-borne deletion allele to the P. aeruginosachromosome. The mutant DNA fragment generated by overlap extension PCR is first cloned into pDONR221 via the BP clonase reaction to create the entry clone pDONR221-Gene::Gm, which then serves as the substrate for LR clonase-mediated recombination into the destination vector pEX18ApGW. The resulting suicide vector pEX18ApGW-Gene::Gm is then transferred to P. aeruginosaand the plasmid-borne deletion mutation is exchanged with the chromosome to generate the desired deletion mutant. Please note that, as discussed in the text, gene replacement by double-crossover can occur quite frequently, but it can also be a rare event in which case allele exchange happens in two steps involving homologous recombination. First, the suicide plasmid is integrated via a single-crossover event resulting in generation of a merodiploid containing the wild-type and mutant allele. Second, the merodiploid state is resolved by sacB-mediated sucrose counterselection in the presence of gentamycin, resulting in generation of the illustrated chromosomal deletion mutant. An unmarked mutant is then obtained after Flp recombinase-mediated excision of the Gm marker. ::: ![](1471-2180-5-30-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### PCR analysis of marked and unmarked P. aeruginosaPA1520deletion strains. Colony PCR was performed on either wild-type PAO1 and its PA1520mutant derivatives, either containing a marked (PA1520::FRT-Gm-FRT) or unmarked (PA1520::FRT) PA1520deletion. The sizes of the expected PCR fragments are indicated. Note that the short deletion removes 41 bp of the PA1520coding sequence, corresponding to codons 131 to 145, but replaces these sequences with a 85 bp FRTscar. Lane M contained Hi-Lo molecular size markers of the indicated sizes from Minnesota Molecular. ::: ![](1471-2180-5-30-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Effect of deletions of P. aeruginosaGntR homologs on fabAexpression. The indicated genes were deleted in strain PAO1 containing a chromosomally integrated fabA\'-lacZfusion (labelled PAO1). The control was PAO1 with a fabAΔ30\'-lacZfusion; this strain harbours the same fabA\'-lacZfusion but contains a deletion of the putative 30 nucleotide activator-binding site in the fabApromoter region. Strains were grown to mid-log phase in LB medium containing 0.05% Brij 58 +/- 0.05% oleic acid and β-galactosidase expression was monitored in triplicate samples. The dotted line marks expression levels observed in the putative activator binding-site mutant and similar levels would be expected in an activator mutant. ::: ![](1471-2180-5-30-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### PCR oligonucleotides ::: Name* Sequence (5\'→3\')^1^ ----------------------- ----------------------------------------------- Gene-specific primers G-UpF-GWL^2^ TACAAAAAAGCAGGCTccgatcagttgcaacacatc G-UpR-Gm^2^ TCAGAGCGCTTTTGAAGCTAATTCGttgttcctgcaccaccatct G-DnF-Gm^2^ AGGAACTTCAAGATCCCCAATTCGccggcgagttccacctgaa G-DnR-GWR^2^ TACAAGAAAGCTGGGTggttgagcttgctgtcgata Common primers Gm-F CGAATTAGCTTCAAAAGCGCTCTGA Gm-R CGAATTGGGGATCTTGAAGTTCCT GW-attB1 GGGGACAAGTTTGTACAAAAAAGCAGGCT GW-attB2 GGGGACCACTTTGTACAAGAAAGCTGGGT ^1^Sequences in capital letters are common for all genes amplified and overlap with the Gm or attBprimer sequences. Lower-case letters indicate gene-specific sequences, here PA1520. PCR oligonucleotide sequences for all 27 genes amplified in this study are given in Table 2 (see [Additional file 1](#S1){ref-type="supplementary-material"}). ^2^G, gene specific sequence, here PA1520. :::	PubMed Central	2024-06-05T03:55:58.394580	2005-5-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173109/", "journal": "BMC Microbiol. 2005 May 23; 5:30", "authors": [ { "first": "Kyoung-Hee", "last": "Choi" }, { "first": "Herbert P", "last": "Schweizer" } ] }
PMC1173110	Background ========== The evolutionary pressures that have driven Hepatitis B virus (HBV) variation remain incompletely understood. Using whole HBV genotype sequencing, this variability can usefully be classified into at least eight families (genotypes) with a characteristic geographic distribution (reviewed in \[[@B1]\]). Alternatively, HBV strains can be classified serologically on the basis of antibody to surface antigen (subtypes). These two classifications broadly correlate, although some subtypes appear in more than one genotype. The extent of genetic diversity reflects the evolutionary history of the virus and the rate of genomic mutation, as well as gene specific selection forces. Several models of HBV evolution have been proposed (reviewed in \[[@B2],[@B3]\]) but fundamental parameters, such as the rate of interspecies transmission or the rate of nucleotide mutation (the molecular clock) remain unresolved \[[@B3]\]. Nevertheless, it is generally assumed that the emergence of HBV families may reflect adaptation to the genotype of the prevalent human host population \[[@B4]\]. The clinical course of HBV infection is very variable. Acute infections in adults are usually effectively controlled, but occasionally lead to fulminant hepatitis and death. In a proportion of individuals however, infection leads to chronic viral replication, which can lead to severe liver damage or hepatocellular carcinoma. Host factors including immune status clearly play a major role in determining clinical outcome. For example, perinatal transmission leads to up to 90% chronic carriership, while the figure is less than 10% for adults. However, pathogenicity has also been linked to virus genotype and several different mechanisms have been proposed for this observation \[[@B5],[@B6]\]. Sequence changes occurring during the course of infection (longitudinal diversity) have also been extensively documented. One common example is the introduction of a stop codon in the precore region which results in downregulation of secretion of a soluble form of HBV core protein (HBVe) whose function remains unclear \[[@B7],[@B8]\]. Interestingly, the downregulation of HBVe secretion is often associated with the appearance of anti-HBVe antibodies in serum, suggesting the protein itself may induce some form of immunological tolerance \[[@B7],[@B9]\]. The role of adaptive immunity both in determining the course of HBV infection and in driving HBV evolution is of special interest. Although pre-existing antibody to HBV surface protein (HBVs) (for example in vaccinated individuals) clearly provides strong protection, antibody to this antigen in natural infection is a late event, usually subsequent to effective control for viremia. In contrast, antibody to HBV core (HBVc), although this protein is internal to the virion, occurs early in infection in almost all infected individuals, irrespective of their ability to control viral replication \[[@B10]\]. T helper and cytotoxic responses to several proteins of HBV have also been detected, and the presence of a higher frequency of HBV specific CTL in liver is associated with the ability to control viremia \[[@B11]\]. As might be predicted CTL responses are not limited to structural proteins, but recognise several non-structural viral proteins. Virus specific T cell immune responses are most readily detected in individuals who effectively control viral replication. In chronically infected individuals, these responses are often much more difficult to detect, suggesting that the chronic state is associated with the establishment of some form of immunological tolerance \[[@B12]\]. HBVc antigen is a small protein, whose three-dimensional structure has been determined by X-ray crystallography \[[@B13]\], and whose immunogenicity in terms of both antibody and T cell responses has been studied rather extensively in both mouse and man. It thus represents an ideal starting point for studies aimed at relating antigen structure to immune response. Indeed longitudinal studies on small groups of HBV infected individuals have suggested that variation is more common in B cell and T helper cell epitopes, suggesting a possible immune driven escape mechanism \[[@B14]\]. As a basis for further functional exploration we have documented HBVc variation in detail. In this study we have collected several hundred protein sequences of HBVc from public databases, and have re-examined variation in relation to structure, immunogenicity and genotype. Results and discussion ====================== 742 protein HBVc sequences were retrieved from the NCBI protein database after some manual curation of very short sequence fragments. The mature HBVc protein sequences 1--185 (not including the precore region) were aligned and the number of polymorphisms found at each position (defined as amino acids occurring in more than 1% of the sequences at each site) was determined (see [Additional file 1](#S1){ref-type="supplementary-material"}). A plot of the number of variants at each position is shown in fig [1a](#F1){ref-type="fig"} and the major structural features, including the four alpha helices, the proline rich loop and the C-terminal arm \[[@B13]\] of the protein are shown in fig [1b](#F1){ref-type="fig"}. Over three quarters (141/185) of positions are completely invariant. Variation clustered within certain regions of the protein, with 18 of the 45 variable amino acids found within positions 59-100. Sequence variation in core protein may reflect the overall genotypic variation among HBV strains driven by drift or other unknown factors (\"hitchhiking\"). Alternatively, specific selection pressures may operate on HBVc driving diversity independently. In order to approach this question, the relationship between HBVc variation and HBV genotype was explored. Using text querying of the data base, a subset of 402 sequences were selected which had been assigned genotype A-D (these were the most frequent assigned genotypes within this set) by analysis of whole genome sequence or sequences outside the core region. These 402 core protein amino acid sequences were aligned and compared to each other and to the overall consensus sequence using Protdist, a Phylip program using the Jones-Taylor-Thornton model \[[@B15]\]. The distribution of distances between each sequence and the consensus was then plotted for each viral genotype as given in the database record (fig [2](#F2){ref-type="fig"}). The distance distribution profiles of the different genotypes were largely overlapping, suggesting that there was no correlation between distance from consensus and genotype. In an alternative approach, the relationship between all the different core sequences was determined using Fitch-Margoliash least squares analysis, and plotted as an unrooted tree (fig [3](#F3){ref-type="fig"}). The different genotypes of each virus are colour coded. Although some broad clustering of genotypes was evident, a significant amount of \"mixing\" can be observed. Similar qualitative results were obtained when phylogeny was determined by Neighbour Joining analysis (using Phylip program Neighbor) or by parsimony (using Phylip program ProtPars). This phenomenon was analysed in more detail in a subset of 40 full length core protein sequences selected from various sections of the tree illustrated in fig [3](#F3){ref-type="fig"} (see Table [1](#T1){ref-type="table"}). The phylogenetic relationship between these forty sequences was analysed again, using Neighbour-Joining (a less computation intense method) incorporating a 1000 fold bootstrap replicate in order to validate the tree topology. The results of this analysis are shown in fig [4a](#F4){ref-type="fig"}. In this smaller subset, genotypes A and D are reasonably well resolved, but genotypes B and C are extensively jumbled. One instance of this is illustrated by two identical sequences (1 and 40 in fig [4](#F4){ref-type="fig"}) derived from genotypes B and C respectively. The DNA sequences corresponding to each of these forty protein sequences were then obtained from GenBank, and analysed using the same bootstrapped Neighbour-Joining procedure. The tree obtained is shown in fig [4b](#F4){ref-type="fig"}. As expected, the bootstrap values are in general higher for the DNA tree (since three times as much sequence information is being analysed). The DNA sequences are somewhat more efficient at classifying B and C as separate grouping. In particular the two identical protein sequences (1 and 40 in figure) are well separated and correctly classified by the DNA phylogeny. Overall, however, significant misclassification remains, reflecting either incorrect genotyping, recombination between viral strains \[[@B16],[@B17]\], or simply insufficient discrimination based on these relatively short viral sequences. Taken together, this data suggested the forces driving the evolution of core were partially independent of the evolutionary forces driving diversification of overall genotype. The evolutionary pressures on sequence diversity were analysed in more detail by analysing synonymous (S) /nonsynonymous (N) nucleotide variation within the 40 viral sequences shown in Table [1](#T1){ref-type="table"}. The overall distribution pattern of synonymous /nonsynonymous mutation rates is shown in fig [5a](#F5){ref-type="fig"}. The protein shows evidence of strong purifying selection throughout the sequence, with dN/dS ratios (using a sliding window of 36 base pairs) mostly below 0.1, The apparent dS rate was not homogenous along the length of the gene. The sharp decrease in substitution rate from around position 110 most likely reflects the start of the overlapping open reading frame of the polymerase, and this area was therefore excluded from further analysis. The identification of site-specific positive or neutral selection operating within an area of overall purifying selection has received considerable attention \[[@B18]-[@B21]\], and a number of methodologies have been adopted. We applied maximum-likelyhood analysis as implemented using Bayes Empirical Bayes method in the software package PAML \[[@B22]\] and compared a number of models of site-specific selection distribution \[[@B18]\]. The best likelihood value was obtained using model 8 (see Methods) which includes a positive selection subset. The posterior probabilities for the positive (ω = 1.38) classes are plotted in fig [5b](#F5){ref-type="fig"}. Interestingly, only one position showed a probability of \>90 of being positively selected, with two more approaching 50% probability. In contrast 133 out of 140 codons showed a probability of \>90% of negative purifying selection. Thus there appears to be only very limited amount of positive selection within the HBV core sequence. The sequence information in [Additional file 1](#S1){ref-type="supplementary-material"}/fig [1](#F1){ref-type="fig"} was rationalised on the three dimensional structure of the capsid obtained from the Brookhaven Data Base. As evident in fig [6](#F6){ref-type="fig"} the variation in sequence is concentrated on the spikes at the outer surface of the capsids. All the amino acids with three or more variants are found on the outer surface of the capsid (fig [6b](#F6){ref-type="fig"} upper panel) while the internal surface facing the capsid lumen are relatively conserved (fig [6b](#F6){ref-type="fig"}, lower panel). A detailed alanine mutagenesis study has been carried out mapping those amino acids critical to proper capsid formation, and/or required for envelopment and virion formation \[[@B23]\]. Mutation of 24 amino acids was found to block capsid formation, virion formation or both. The position of these mutations on the 3D structure is shown in fig [7](#F7){ref-type="fig"}. The position of this set of mutations is quite widely distributed over the structure, suggesting multiple essential interactions are absolutely required either for proper capsid assembly, or for envelope and virion assembly. However, interestingly, all but one of these 24 amino acids were found to be invariant in the data set analysed in this study. The only exception observed was at position 129, where changing proline to alanine was found to block both capsid and virion formation. Both glutamine and threonine are found in a proportion of virus sequences at this position, and further mutagenesis will be required to clarify the constraints imposed by the requirements of virion formation on the sequence at this particular position. Thus the structural requirements of virion assembly seem to impose a significant restraint on HBVc diversity. However, several positions which were found to tolerate alanine mutagenesis in terms of capsid/virion assembly were nevertheless invariant in the set of sequences examined here. The overall high degree of conservation in HBVc therefore probably reflects the multiple functions required from this protein, including control of intracellular targeting \[[@B5]\], pregenome/DNA polymerase packaging, capsid disassembly and viral maturation. Pressure by the host immune system is one obvious candidate driving variation in the protein sequence of HBVc. CD8 cytotoxic cells are believed to play a key role in controlling virus replication during HBV infection \[[@B24],[@B25]\]. Several CD8 T cell epitopes have been characterised in detail (e.g. \[[@B26]-[@B28]\]) by the use of T cell clones and lines or by elution from HLA \[[@B29]\]. Some of the sequences are shown in fig [7](#F7){ref-type="fig"}, together with the known variant sequences in the dataset analysed in this study. Although several of the epitopes lie within the more conserved inner region of the capsid, two of the epitopes show substantial polymorphism (fig [8](#F8){ref-type="fig"}). Interestingly, a previous paper did not find any evidence for emergence of mutations within the major CD8 epitopes during chronic HBV infection \[[@B30]\]. The evidence implicating CD4 T cells in HBV control remains much less clear. Furthermore, although regions containing CD4 epitopes have been described, many of the epitopes are not very well characterised. One putative \"immunodominant\" CD4 T cell epitope (amino acids 50-69) does contain a number of variable amino acids (fig [9a](#F9){ref-type="fig"}), and indeed changes in its sequence have been related directly to changes in T cell response in vivo \[[@B31]\]. A second epitope (core region 147-156) identified as a major target in HLADR13 individuals has also been examined in some detail \[[@B32]\]. Interestingly this epitope also shows considerable variation (fig [9a](#F9){ref-type="fig"}) including a mutation at position 151 shown to be essential for T cell recognition. The epitope also spans the region of a two amino acid insertion which is found exclusively in viruses of genotype A. Further detailed mapping of CD4 T cell recognition sites, in relation to natural sequence variation would be seem to be an area of great interest. Finally, considerable information has been accumulated on the interaction of antibodies to HBV capsids. As shown in fig [9b](#F9){ref-type="fig"}\[[@B10]\] the major defined antibody specificities lie on the outside of the capsid structure, particularly at the tip of the spikes and at the junctions between adjacent spikes. These regions do indeed contain the majority of the HBVc sequence variation, although the most variable amino acids themselves have not been identified as known antibody contacts \[[@B33]\]. However, the contribution of anti-HBVc antibody to protection remains unclear, particularly since the capsid in intact virions is presumably largely shielded from antibody by the HBV envelope. Conclusion ========== This study makes use of the large number of HBVc sequences now available in public databases to characterise sequence variation in HBVc. One limitation of such an approach is that detailed clinical information associated with infection is not available, and in particular, it is not possible to examine variability in the context of the longitudinal course of an HBV infection. This is likely to be an important factor since mutations are often found to arise late in infection, associated with a variety of clinical outcomes (e.g. \[[@B34],[@B35]\]). In addition few of the available sequences have been checked for their ability to make competent infectious virus, and some sequences may therefore represent non-functional proteins. However, despite these limitations, the data available does allow some general conclusions. Overall, HBVc contains a large proportion of invariant amino acids, and a strong over representation of synonymous versus non-synonymous mutations at almost every codon. Both features suggest the presence of strong constraining forces on sequence diversity. Virion assembly is likely to provide one major constraining force \[[@B23]\]. As reported previously (e.g. \[[@B14]\]) sequence diversity appears to be clustered, and mapped predominantly to the spike and external surface of the capsid. These positions may allow greater flexibility in terms of virion assembly. One significant consequence of the strong purifying selection is that protein sequence is a poor predictor of genotype for this gene. DNA sequence which reflects predominantly synonymous mutations, is a better discriminator, particularly in resolving genotypes B and C. Longer sequence analysis, however, is clearly necessary to obtain reliable genotyping information. The putative role of immune selection in driving HBV core diversity is much more unclear. Direct evidence for positive selection, at least using the analysis presented here, identifies only a single amino acid (position 74 at the tip of the viral spike) as showing evidence for positive selection of diversity. Nevertheless, it is clear that several polymorphic positions lie within T or B cell epitopes. Hence, while the overall effect of immune selection on HBV sequence diversity may be small, sequence diversity may have a significant effect on the immune response at an individual level. The data analysis given here will help inform further analysis of the HBV-specific immune response. The combination of T cell and antibody recognition studies with directed mutagenesis of HBVc should determine more precisely the relationship between structure, immunity and pathogenicity. Methods ======= Human HBVc sequences were retrieved from the NCBI protein sequence database \[[@B36]\], limiting the search to organism = Hepatitis B virus and searching for core protein. Additional classification into genotypes A-D was done using the text search tool to look for \"genotype X\". A proportion of hits were verified by manual inspection. The initial 780 hits were aligned using the EMMA program on EMBOSS (a version of Clustal)(see\[[@B37]\] using the BLOSUM 62 similarity matrix. The alignment was further refined by manual inspection using the sequence editor Bioedit, and very short or badly aligned sequences removed. The frequency of amino acids at each position was determined using the EMBOSS program PROPHECY. The matrix observed was converted into polymorphism frequency by setting a cut-off of 1% frequency at each position. A phylogenetic tree of the HBVc sequences was created using the Phylip program Kitsch \[[@B38]\]), which uses a Fitch-Margoliash criterion based on a distance matrix obtained using the Phylip program Protdist. The tree was displayed using Treeview \[[@B39]\], copied in Adobe Illustrator and color coded according to genotype. Because this method is extremely processor intensive (the best tree is analysed at each iteration) it was not possible to bootstrap. Analyses of the same data were also done using nearest neighbor analysis (using the Phylip program Neighbor) and parsimony using the EMBOSS (loc cit) program EPROTPARS. Although the fine details of the trees varied between methods the overall qualitative conclusion were the same. In order to further validate the conclusions of the phylogeny, 40 sequences (shown in Table [1](#T1){ref-type="table"}) chosen manually to cover the major branches of the tree shown in fig [3](#F3){ref-type="fig"} were reanalysed using the Nearest Neighbour with bootstrap option (1000 bootstraps) of ClustalW \[[@B40]\]. The consensus tree set was plotted in Treeview and coloured in Illustrator. Similar analysis was carried out on an alignment of the DNA sequences corresponding to each protein sequence. The analysis of synonymous/nonsynonymous mutations rates was carried out initially using the program DNASP3.0 \[[@B41]\] (using the Nei/ Gojobori algorithm) with a 36 base pair sliding window shifted by 9 nucleotides. A more detailed analysis at individual codons was carried out using the program PAML version 3.14 (using maximum likelihood Bayes Empirical Bayes inference, as described in \[[@B20]\]). Analysis was carried out using a variety of selection models, but the output represented in fig [5](#F5){ref-type="fig"} used model 8 \[[@B22]\]. The models assume that the selection pressure (measured as ω) operating at each codon can fall within a range of different classes. Model 8 assumes a distribution of negative selection values (for all of which ω \<1), or a positive selection class with ω = 1.38. This model was found to give the best likelihood estimate. The program then calculates the posterior probability (p value) that each codon within a sequence falls within a particular class. The crystal structure of HBVc was retrieved as a pdb file from the Brookhaven database, and displayed and coloured using RasMol software (version 2.7.1.1 for Windows). Figures [6](#F6){ref-type="fig"} and [7](#F7){ref-type="fig"} show the structure of four identical monomers, to illustrate the spikes and their interaction, while figs [8](#F8){ref-type="fig"} and [9](#F9){ref-type="fig"} show a single monomer for clarity. Abbreviations ============= Hepatitis B virus HBV; Hepatitis B core protein HBVc; Hepatitis B surface protein HBVs Authors\' contributions ======================= RM provided bioinformatics support and expertise, and carried out the phylogenetic analysis shown in fig 4. BMC initiated the project and was responsible for the text, and the data and analysis shown in the other figures. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Amino acid polymorphisms along the HBVc sequence. Table containing all the polymorphic amino acid residues found at each position along the 742 HBV sequences analysed. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ I am grateful for much useful discussion with many colleagues at UCL, in particular to Paul Kellam for help with the bioinformatics, to Ziheng Yang for initial help setting up PAML and to Richard Tedder, Antonio Bertoletti, Mala Maini and Nikolai Naoumov for advice on HBV. I am grateful to Dr. Volker Bruss (University of Goettingen) for his help and advice. I am also very grateful to the staff at the UK HGMP Resource Centre for a lot of help and patience with running the various Bioinformatic programs. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Sequence variation along the HBVc sequence.a)The number of polymorphisms (variants occurring in more than 1 % sequences examined) found at each position of the HBVc sequence. See [additional file 1](#S1){ref-type="supplementary-material"} for full data. b)Major features of secondary structure along HBVc \[13\]. Blue : α-helix Green : Proline rich loop; Red : C-terminal arm. ::: ![](1471-2180-5-33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Relationship between HBVc sequence and genotype.The 403 sequences from viruses of known genotype were aligned and compared to the consensus. The distance of each sequence from the consensus was calculated using maximum likely-hood model (Jones-Taylor-Thornton). The frequency distribution of distance from consensus was then plotted for each genotype A-D. ::: ![](1471-2180-5-33-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Analysis of the relationships between HBVc sequences.All the sequences were classified and plotted as an unrooted tree using the Fitsch Margoliash criterion (Kitsch, see Materials and Methods). Where known the genotype is indicated by colour coding (Yellow = A, Green = B, Blue = C, Red = D). Sequences of unknown genotype are in black. ::: ![](1471-2180-5-33-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Analysis of 40 protein and DNA HBVc sequences.The 40 sequences detailed in Table 1 were analysed using Neighbor Joining analysis of both protein (panel a) and DNA (panel b) and plotted using TreeView. The bootstrap values (as percentage) are shown in smaller font. The accession number corresponding to each numbered sequence is shown in Table 1. The genotype is indicated by colour coding (Yellow = A, Green = B, Blue = C, Red = D). ::: ![](1471-2180-5-33-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Synonymous/nonsynonymous mutation rates and selection in HBVc.Panel a shows the distribution of synonymous (dS) and non-synonymous (dN) substitutions (the number of such substations per site), and the ratio dN/dS along the sequence of HBVc. Gaps indicate dS = 0. Panel b shows the posterior probability p^+^that dN/dS\>1 for each position of HBVc, as calculated by PAML. The probability that dN/dS\<1 (purifying selection) is not plotted for clarity but is given by (1- p^+^). ::: ![](1471-2180-5-33-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Sequence variation in the context of capsid structure.The crystal structure formed by four HBVc subunits was displayed and coloured using RasMol software. Positions containing 2 polymorphisms are coloured blue, 3 polymorphisms are coloured orange, and 4 or above are coloured red. Grey amino acids are invariant. a) The structure is displayed in ribbon form, showing a vertical section though the capsid, with two spikes projecting upwards, and the internal face of the capsid shown at the bottom of the picture. b) The structure is displayed in space fill. Upper Panel The structure is displayed so that the spikes and outward surface of the capsid are shown towards the viewer, and only the outer surface of the capsid is visible. Lower Panel The structure is rotated by 180 degrees so only the lower (inner) face of the capsid is visible. ::: ![](1471-2180-5-33-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Structural constraints and HBV diversity.The crystal structure formed by four HBVc subunits was displayed and coloured using RasMol software. The position of the amino acid identified as essential for virion formation (violet) or for both capsid and virion assembly (green) \[23\] are shown. The rest of the colour coding is as shown in fig 6. ::: ![](1471-2180-5-33-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Variation within CD8 T cell epitopes of HBVc.Five of the best defined class I T cell epitopes from HBVc (for references see text) are shown together with alternative amino acids found at each position. The position of the epitopes (green) within the HBVc three dimensional structure is shown in the right hand panel. The rest of the colour coding is as shown in fig 6. ::: ![](1471-2180-5-33-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Variation with CD4 and antibody epitopes of HBVc.a) Two of the best defined CD4 epitopes within HBVc (taken from \[31,32\]) are shown together with alternative amino acids found at each position. The position of the epitope 50\#150;69 is shown in the context of the HBVc three dimensional structure above(in green). The rest of the colour coding is as shown in fig 6. Epitope 147\#150;156 is not shown since this region of HBVc was excluded in the crystallographic study. b) The best defined antibody epitopes (taken from \[33\]) are shown (in green) in the context of the HBVc three dimensional structure above. The rest of the colour coding is as shown in fig 6. ::: ![](1471-2180-5-33-9) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Accession numbers and genotype for HBV sequences analysed in fig 4 ::: Number Accession number Genotype -------- ------------------ ---------- --- Protein DNA 1. AAK81691 AY040627 C 2. AAO41301 AY167091 C 3. AAP06648 AY206392 C 4. AAL31832 AF324119 C 5. AAP06544 AY206375 B 6. AAP06549 AY206376 C 7. AAL31776 AF323466 B 8. AAL31802 AF324088 B 9. AAP06642 AY206391 B 10. AAP79869 AY233296 D 11. AAP31176 AY269059 D 12. AAR19326 AY230115 D 13. BAC57440 AB090269 D 14. AAL31837 AF324124 A 15. AAF24670 AF121240 D 16. BAD02318 AB126581 D 17. AAL31826 AF324112 D 18. AAP31178 AY269061 D 19. BAA85376 AB033559 D 20. AAP31169 AY269052 D 21. BAC92696 AB106885 B 22. BAC92693 AB106884 B 23. AAR19322 AY230111 A 24. AAL31839 AF324126 A 25. AAP79778 AY233283 A 26. AAL31785 AF324071 A 27. AAP79724 AY233275 A 28. AAP79715 AY233274 A 29. AAP79729 AY233276 A 30. AAP79736 AY233277 A 31. AAL31856 AF324143 A 32. AAG49707 AF223961 C 33. AAP06636 AY206390 B 34. AAF24738 AF121250 B 35. BAA32843 AB014363 C 36. AAP31199 AY269084 C 37. AAL31815 AF324101 C 38. AAL31819 AF324105 C 39. AAL31781 AF324066 B 40. AAF24705 AF121245 B :::	PubMed Central	2024-06-05T03:55:58.396937	2005-5-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173110/", "journal": "BMC Microbiol. 2005 May 27; 5:33", "authors": [ { "first": "Benjamin M", "last": "Chain" }, { "first": "Richard", "last": "Myers" } ] }
PMC1173111	Background ========== The vast majority of fast synaptic transmission in the mammalian central nervous system is glutamatergic. Proper expression, localization, and regulation of glutamate receptors are critical for brain development and plasticity. Not surprisingly, the molecular mechanisms controlling glutamate receptor expression, localization, and stability are of great interest. The most common recent approach to understanding these mechanisms has been biochemical: Proteins are identified based on a biochemical interaction with a glutamate receptor subunit. This approach has identified a number of important candidates, some of which have been subsequently shown to play important roles in glutamate receptor trafficking and stability at the PSD (for review see \[[@B1],[@B2]\]). An alternative approach is forward genetics: Mutant animals are screened for alterations in glutamate receptor localization. Perhaps the most well-known glutamate receptor trafficking protein identified via genetics is stargazin, which is disrupted in stargazer mutant mice \[[@B3]\]. The identification of stargazin was serendipitous; mammalian genetic screens are unfortunately still relatively time-consuming and expensive (though this may change as RNAi techniques advance). Most forward genetic screens glutamate receptor mutants have used C. elegans. Studies using C. eleganshave highlighted the importance of ubiquitination, CamKII, and PDZ proteins in controlling glutamate receptor number, and identified a novel protein required for nematode glutamate receptor function \[[@B4]-[@B7]\]. Unfortunately, glutamatergic synapses in C. elegansare accessible for electrophysiology only with great difficulty. The subunit composition of C. elegansreceptors in vivo has also not yet been determined, which hinders the study of subunit-specific trafficking mechanisms. Drosophilaare not only amenable to powerful genetics, but also contain glutamatergic neuromuscular junctions (NMJs) that are individually identifiable and accessible throughout development to high-resolution electrophysiological and microscopic techniques. These techniques have revealed a significant amount of information concerning this popular model synapse, including the exact subunit composition of glutamate receptors in vivo. DrosophilaNMJs contain two subtypes of postsynaptic glutamate receptor, which are molecularly, pharmacologically, and spatially distinct \[[@B8]-[@B11]\]. The \'A\' receptor subtype contains the subunit GluRIIA, in combination with subunits GluRIIC, GluRIID, and GluRIIE. The \'B\' receptor subtype contains the subunit GluRIIB, in combination with the subunits GluRIIC, GluRIID, and GluRIIE. Both receptor subtypes are most similar in sequence to mammalian kainate receptors. The Drosophilagenome also encodes receptor subunits with high similarity to mammalian AMPA, delta, and NMDA receptor subunits, but these proteins are not found postsynaptically in the NMJ. Virtually nothing is known about the molecular mechanisms that control Drosophilaglutamate receptor expression, localization, and stability. To determine the molecular mechanisms controlling Drosophilaglutamate receptor expression, localization, and stability, we are screening transposon insertion mutants generated by the Berkeley DrosophilaGene Disruption Project \[[@B12]\]. This mutant collection contains insertions in over 40% of the entire genome, and the insertion site for almost every mutant has already been identified by inverse PCR. With the addition of transposon insertions from other collections, it is likely that Drosophilists will have access to insertion mutants of almost every gene in the fly genome within a few years. Theoretically, one could quickly and efficiently identify the complete list of genes required for any particular process simply by examining a non-redundant set of mutants for a phenotype of interest. We are testing this idea and using this approach to define the broad categories of genes required for glutamate receptor cluster formation. We identified 57 lethal mutants with qualitatively abnormal glutamate receptor clusters. Here, we present this list and discuss the types of genes that are, and are not, represented. Because of the high prevalence of background mutations even in transposon mutant collections, the role of individual genes must be extensively confirmed before a role for any particular protein is assumed. However, this problem may be circumvented, and perhaps more insight gained, by focusing on the types of genes identified rather than the identity of individual genes. We do so here. Results ======= We sought to identify Drosophilamutants with abnormal glutamate receptor cluster development. Elimination of DrosophilaNMJ glutamate receptors results in paralysis and embryonic lethality \[[@B9]\]. Mutations that reduce (but do not eliminate) NMJ glutamate receptors allow hatching, but typically cause larval or pupal lethality \[[@B8]-[@B10]\]. Any effective screen for glutamate receptor cluster formation mutants must therefore include examination of homozygous lethal mutants. However, proper examination of glutamate receptor clusters in embryos and larvae requires technically challenging dissection techniques and time-consuming confocal microscopy. Therefore, we sought to minimize examination of mutants that do not have glutamate receptor cluster defects. To do this, we made the assumption, based on the studies cited above, that mutants with severe defects in glutamate receptor cluster formation are more likely to be homozygous lethal. Thus, the first step in our screen for mutants with reduced or eliminated glutamate receptors was identification of recessive lethal transposon mutants (e.g. insertions lines with no homozygous viable adults). We concentrated on the collection of GT1and SuPor-Pinsertion mutants, since these transposons were engineered for maximal gene disruption \[[@B13],[@B14]\]. Of 2185 mutants (representing insertions in approximately 16% of the entire genome), 220 insertion lines contained lethal mutations. Because a prerequisite for NMJ glutamate cluster formation is development of neuromusculature, we examined dechorionated embryos from each of the 220 stocks to ensure that homozygous mutants developed into \'morphologically mature\' late stage embryos typical of wildtype animals 16--17 hr after egg laying, AEL, which is when NMJs begin to form \[[@B15]-[@B18]\]. Morphological maturity was based on the presence of characteristics typical of late stage 17: clear segmentation, mouthhooks, condensed CNS, malphigian tubules, and visible trachea. 205 of 220 (93%) of the homozygous lethal mutants formed morphologically mature embryos. 202 of these mutants were rebalanced using a GFP-tagged balancer chromosome for unambiguous identification of homozygous mutant animals. We were unable to rebalance 3 mutant lines using chromosome-appropriate GFP-tagged balancers, possibly because the insertion mutant chromosome and balancer chromosome both contained lethal mutations that fail to complement. We then verified that all 202 GFP-balanced stocks did indeed carry recessive lethal mutations and determined the latest stage to which homozygous (non GFP) embryos and larvae lived. For each of the 202 balanced lines, several (typically 5--10) homozygous mutant animals, along with control heterozygous siblings, were manually dissected at the latest viable stage to expose NMJs on ventral longitudinal muscles, then fixed and stained using anti-HRP antibodies that recognize all neuronal membrane (including peripheral axons and NMJs) and anti-GluRIIA antibodies. After dissection, fixation, and staining, each of the 202 rebalanced mutant lines was examined for changes in NMJ morphology and glutamate receptor expression using laser-scanning confocal fluorescent microscopy. 57 of the 202 rebalanced lethal mutants (28%) displayed consistent (most NMJs in several animals) defects in glutamate receptor expression without severe presynaptic morphological abnormalities (e.g. 57 mutants were identified in which NMJs formed, but glutamate receptor clusters were altered). Examples of some of these phenotypes, from both embryonic and larval NMJs, are shown in Fig. [1](#F1){ref-type="fig"}. All phenotypes could be classified into one of two broad categories: 1) loss of glutamate receptors (fewer glutamate receptor clusters or smaller individual clusters), or 2) gain of glutamate receptors (more clusters or larger clusters). We relied on cluster size and number because cluster size and number measurements (compared to fluorescence intensity measurements) avoid a requirement for fluorescence intensity calibration or problems associated with potential differences in background immunofluorescence between genotypes. Assuming constant receptor density, cluster size should be directly proportional to the number of clustered receptors. In support of this assumption, immunoreactive cluster sizes correlate well with high-resolution patch-clamp electrophysiological measurements between genotypes \[[@B11],[@B19],[@B20]\] and throughout embryonic/larval NMJ development (Featherstone, unpublished observations). Most (84%) of the mutants with disruptions in glutamate receptor clusters showed a qualitative loss in receptors. Severe loss of GluRIIA was always associated with embryonic/L1 lethality. One example is P{SUPor-P}KG00333, which shows a complete loss of A-type receptors despite the presence of morphologically normal presynaptic terminals (Fig. [1A](#F1){ref-type="fig"}, left). Some mutants with loss of receptors survived until pupation. An example of a third instar viable mutant, P{SUPor-P} Chro^KG03258^, with reduced GluRIIA is shown in Fig. [1B](#F1){ref-type="fig"} (middle column). A minority (16%) of third instar viable mutants showed a qualitative increase in GluRIIA immunoreactivity. An example of one of these mutants, P{SUPor-P} vri^KG01220^, is shown in Fig. [1B](#F1){ref-type="fig"} (right column). Except for P{SUPor-P}KG00212, all of the mutants that displayed an increase in GluRIIA immunoreactivity were viable as third instar larvae and also displayed an increase in the number of presynaptic boutons. This presynaptic phenotype is consistent with previous Drosophilastudies showing that overexpression of postsynaptic GluRIIA causes presynaptic overgrowth \[[@B21]\]. BG and KG transposons are designed to generate loss of function mutations. Therefore, the isolation of mutants with increased receptor cluster size suggests that receptor insertion and clustering per se are not rate-limiting. There are molecular mechanisms, revealed by our screen, that actively restrain the number of synaptic glutamate receptors. For all of the mutants identified in our screen, the Gene Disruption Project has precisely determined the genomic insertion site of the transposon using inverse PCR. In the vast majority of cases, inverse PCR results were consistent with a unique insertion, and in most cases the flanking genomic sequence revealed that the insertion was in an annotated gene. For each of the mutants identified in our screen, we used the BDGDP P-screen database <http://flypush.imgen.bcm.tmc.edu/pscreen/> and/or BLAST searches with flanking genomic sequence from iPCR results to identify which gene was mutated by the inserted P-element. Putative functions were assigned to each of these genes based on previous publications, FlyBase annotations, and/or Genbank BLAST searches. We binned each of the insertion site genes into one of the following functional categories: 1) extracellular matrix proteins (mucins), 2) cell adhesion proteins 3) cytoskeleton proteins, cytoskeletal regulators, and adaptor proteins, 4) kinases and phosphatases, 5) ubiqitination proteins (ubiquitin ligases), 6) transporters/pumps, 7) proteins involved in gene expression and protein translation, 8) enzymes, and 9) proteins of unknown/novel function. These categories, and the relative number of proteins in each category, are illustrated by Figure [2](#F2){ref-type="fig"}. All categories shown in Figure [2](#F2){ref-type="fig"} were represented by at least two mutant genes. Table [1](#T1){ref-type="table"} contains the complete list of all mutants identified in our screen, listed by functional category. For each mutant, Table [1](#T1){ref-type="table"} indicates whether the phenotype was loss (down arrow) or increase (up arrow) in postsynaptic glutamate receptors. Table [1](#T1){ref-type="table"} also lists the Drosophilagene in which each transposon is inserted, and the mouse homolog of each of those genes (identified by BLASTP against the mouse genome refseq protein database; <http://www.ncbi.nlm.nih.gov/genome/seq/MmBlast.html>). A definitive test of whether any particular gene is required for glutamate receptor cluster development requires precise excision of each P-element insertion, or transgenic rescue, followed by re-examination of the NMJ at the same developmental stage -- a task which is not practical on a genomic scale. However, we performed two broad types of checks to identify potential caveats in our results: expression analysis and complementation tests for lethality. Drosophilaglutamate receptors are expressed in neurons and muscles. Thus, any gene required for expression and/or localization of glutamate receptors should be expressed in neurons and/or muscles. The Drosophilagene expression database (accessed at: <http://www.fruitfly.org/cgi-bin/ex/insitu.pl>) describes expression patterns of approximately 3300 different Drosophilagenes (\~24% of the genome), as determined by embryonic in situ cDNA hybridization \[[@B22]\]. Of these 3300 genes, which presumably represent a random \'sampling\' of the genome, 2350 (71%) are expressed in the nervous system and/or body wall muscles. The gene expression database contains data for 28 (49%) of the genes identified in our study. 24 (86%) of these genes identified by our screen are expressed in the nervous or muscular systems. 4 (14%) of the genes identified in our screen are annotated as not detectably expressed in the nervous or muscular systems. Thus, most (86%) of the genes identified in our screen have expression patterns consistent with the conclusion that these genes are important for glutamate receptor cluster development, but we did not identify neuronal and muscle genes significantly above the level expected by a random sampling of the genome -- possibly because the fraction of genes expressed in neurons and muscles (71%) is quite high to begin with. There may also be cell non autonomous roles for some genes with regard to receptor expression and/or localization. Although all of the mutants identified in our screens displayed a receptor phenotype, and all of the mutants contain a unique P-element insertion, an important genetic caveat is that the receptor phenotype may not actually be caused by the transposon insertion. For example, the P-element mobilizations that generated each of the P-element alleles for the gene disruption project may have also created second-site mutations in an unknown locus. Spontaneous mutations can also be easily stabilized and propagated in a stock carrying a balanced lethal mutant chromosome such as those screened here. To estimate the frequency of background mutations in this transposon mutant collection, we performed complementation tests to determine whether the P-element insertions failed to complement lethal alleles of the same insertion locus. Of the 57 mutants identified in our screen, 20 were tested for complementation to deficiencies that remove the insertion site. Ten (50%) of the complementation tests failed, suggesting that lethality in half of the mutants may be caused by a mutation other than the P-element insertion. These results are consistent with those from a related screen for presynaptic morphology mutants (Liebl FLW, Werner KM, McCabe BD, Featherstone DE: A Genome-Wide P-element screen for Drosophila synaptogenesis mutants. Manuscript in preparation), which included complementation tests to over 80 genes. Complementation did not appear to be biased for any particular type of gene (in this or the related Liebl et al study). We did not systematically test for complementation of the receptor cluster phenotypes. Discussion ========== In an effort to identify new genes required for glutamate receptor cluster development, we screened lethal transposon insertion mutants for alterations in postsynaptic glutamate receptor clusters. 202 lethal insertion lines, representing insertions in \~1.4% of the genome, were manually dissected and glutamate receptor clusters were examined using immunocytochemistry and confocal microscopy. This screen identified 57 mutations in 56 different loci. Transposon mutageneses are becoming increasingly popular, because a transposon insertion simultaneously mutates and \'tags\' a gene. An assumption of all transposon mutagenesis (and subsequent screens) is that the insertion locus is responsible for any observed phenotype. Expression analysis showed that the expression pattern of insertion loci identified in our screen is broadly consistent with that expected from genes involved in glutamate receptor cluster development, but also not compellingly different (86% vs. 71%) from that expected from a random sampling of the genome. Furthermore, complementation tests suggest that approximately one-half of the lethality observed in our screen was not due to disruption of the insertion locus. Thus, it is impossible to definitively say which particular genes are important without extensive validation of each candidate. And even with extensive validation sufficient to confirm that a particular locus is involved, it is difficult to know whether a gene\'s role in receptor cluster formation is indirect with regard to glutamate receptor cluster formation (i.e. whether a gene product regulates another protein which in turn clusters receptors). Nevertheless, one cannot discount our results without also casting aside the fact that decades of forward genetic screens have successfully identified many genes critical for many different processes \[[@B23]\]. But how does one decide which screen \'hits\' to trust, and is there any way to usefully interpret results of forward genetic screens? Should we exclude from consideration all mutants in which lethality was complemented (and conversely accept all mutants in which lethality failed to be complemented)? This depends on the degree of coupling between two different phenotypes: lethality and abnormal glutamate receptor cluster development. To enrich for mutants with receptor cluster defects, we examined only homozygous lethal mutants, on the assumption that GluRIIA cluster defects and lethality are strongly linked. Indeed, all of the mutants showing severe loss of GluRIIA were embryonic/first instar lethal. Nevertheless, this assumption probably does not hold for all mutants. At the time this screen was initiated, it was assumed that all DrosophilaNMJ glutamate receptors contained the subunit GluRIIA, and that the viability of GluRIIA null mutants was due to substitution by the alternate subunit GluRIIB \[[@B24]\]. Thus, the presence and localization of GluRIIA was considered an accurate marker for all NMJ receptors. Subsequently, we and others have determined that the DrosophilaNMJ contains two independently assembled and localized subtypes of postsynaptic glutamate receptors: A-type receptors, that contain the subunit GluRIIA (plus GluRIIC, GluRIID, and GluRIIE, but not GluRIIB), and B-type receptors, which contain the subunit GluRIIB (plus GluRIIC, GluRIID, and GluRIIE, but not GluRIIA) \[[@B8]-[@B10]\]. GluRIIA therefore serves as a tag for only one-half to two-thirds (depending on developmental stage; Featherstone, unpublished) of fly NMJ glutamate receptors. Complete elimination of A-type receptors does not result in lethality \[[@B24],[@B25]\]. Conversely, lethality can obviously be caused by many defects other than loss of NMJ glutamate receptors. Since viability and glutamate receptor cluster development may be only loosely coupled, complementation tests for lethality indicate relatively little about the reliability with which our screen identified genes required for proper glutamate receptor cluster formation (although these complementation results do give important insights into the properties of this mutant collection, and transposon mutageneses in general). Similarly, the large fraction of genes expressed in the neuromusculature makes it difficult to judge, based on expression, whether any particular gene has simply been randomly selected. Thus, we are left with uncertainty regarding any particular gene until that gene\'s role in synaptogenesis can be extensively validated -- a prospect that is not practical on a genomic scale. However, the sequencing of several metazoan genomes, along with the realization that the function of homologous proteins tends to be conserved across genomes, opens the possibility of another approach, as demonstrated in our study: \'functional category analysis.\' Functional category analysis involves screening a non-redundant collection of mutants for the phenotype of interest, assigning putative functions bioinformatically, and categorizing the \'hits\' by function. The goal of this approach is not to generate a definitive list of individual proteins involved in a process, but to gain insight into the types and relative numbers of proteins required. All of the categories shown in Figure [2](#F2){ref-type="fig"}, for example, are represented by several mutants. Therefore, even with a 50% accuracy rate (which would be a worst case estimate; the real accuracy is probably substantially higher -- see below), the gene function categoriesare likely to be correctly identified. Functional category analysis recognizes that not all hits will be valid, that not all genes will play roles specific to a particular process, and that many proteins are only indirectly required. Functional category analysis reveals unexplored areas of relevant biology and provides a broad roadmap for further study. For example: instead of studying the individual genes identified in Table [1](#T1){ref-type="table"} (which is arguably a hit-and-miss endeavor unlikely to shed much global insight into the process), we plan to focus research toward understanding how RNA regulation and regulated translation are involved in glutamate receptor cluster formation. Similarly, we are directing efforts toward understanding the interactions of receptors with the actin and microtubule-based cytoskeletons. Preliminary results suggest that this approach is insightful and effective \[[@B20]\]; Liebl, Karr, & Featherstone, unpublished observations). What percentage of the genome is involved in glutamate receptor cluster development? Our results allow this question to be addressed (at least with regard to the DrosophilaNMJ). If glutamate receptor cluster development and lethality are completely uncoupled, then our results imply that approximately 28% (57/202) of the entire Drosophilagenome is required for glutamate receptor cluster development. Depending on the percentage of insertions that were too hypomorphic to show qualitatively detectable phenotypes via immunocytochemistry, that percentage could be higher. Glutamate receptor cluster formation and lethality, however, are not unrelated phenotypes. Therefore, 28% is likely to be a gross overestimate. Complete loss of DrosophilaNMJ glutamate receptors results in paralysis and embryonic lethality \[[@B9]\], and mutations that reduce (but do not eliminate) NMJ glutamate receptors allow hatching, but typically cause larval or pupal lethality \[[@B8]-[@B10]\]. In the present study, all severe loss of function phenotypes were associated with embryonic/early larval lethality. If all GluRIIA cluster mutants are lethal, then mutants with normal glutamate receptor clusters would have been eliminated in the selection for lethality early in the screen (indeed, this was the rationale for this step). In this case, our results suggest that approximately 2.6% (57/2185) of the genome is required for glutamate receptor cluster formation. This estimate, however, is also likely to be flawed; although severe disruptions in glutamate receptor cluster formation cause lethality, not all lethality is likely due to disruptions in glutamate receptor cluster formation. Mutations that trigger loss of GluRIIA are not necessarily lethal \[[@B24],[@B25]\]. Lethality and GluRIIA cluster phenotypes are not absolutely coupled, and therefore 2.6% is likely too low of an estimate. Given the uncertainties above, we can say only that somewhere between 2.6 and 28% of the fly genome (360--3900 genes) is required for NMJ glutamate receptor cluster formation. This is a wide range, which in any case represents a surprisingly large number of genes. Is this reasonable? Do all of these genes represent specific machinery for glutamate receptor cluster expression and clustering? To answer these questions, it is helpful to consider the types of genes identified in the screen. As noted earlier, the role of each individual gene needs to be validated before placing too much emphasis on any particular protein\'s role. But a general discussion of the genes implicated is helpful for evaluation of the general results. In support of the idea that our screen correctly identified genes required for glutamate receptor cluster formation, some of the types of genes identified in our screen have been previously identified as important for postsynaptic development. Polo, for example, was identified in our screen, and mammalian polo-like kinases are receiving increasing attention as important players in synapse development \[[@B26]\]. We also identified several other kinases and phosphatases. Activation of the mitogen activated protein kinase (MAPK) pathway facilitates AMPAR surface expression \[[@B27],[@B28]\], and interaction of AMPAR GluR2 subunits with GRIP and PICK1 is dependent upon the phosphorylation of the GluR2 subunit \[[@B29],[@B30]\]. Our screen also isolated mutants in two different ubiquitin ligases. Consistent with this, ubiquitination is known to regulate glutamate receptor number and synapse development \[[@B6],[@B31]\]. Our screen identified a fly neuroligin family member, and mammalian neuroligins were recently implicated in postsynaptic development \[[@B32]\]. One of the largest groups of genes identified in our screen encodes proteins that comprise or regulate the actin and/or microtubule cytoskeletons. Glutamate receptors, like many other proteins, are thought to be transported along microtubules, and anchored to the synaptic actin cytoskeleton \[[@B33]-[@B35]\]. In support of this, we\'ve recently found that GluRIIA-containing receptors are specifically linked to the actin cytoskeleton via the 4.1 homolog coracle, which interacts directly with GluRIIA \[[@B20]\]. We\'ve also found that regulation of synaptic microtubules affects fly NMJ glutamate receptor cluster development (see below). The cytoskeleton might not only be important for receptor protein localization; trafficking and localization of synapse-specific mRNAs (see below) probably also relies on the cytoskeleton. Consistent with this, untranslated regions of GluRIIA appear to be required for proper synaptic receptor localization in the fly NMJ (Karr & Featherstone, unpublished). Other types of genes identified by the screen are consistent with what one might expect. For example, many of the genes identified in our screen are involved in gene expression or protein translation. Mutation of a transcription factor or component of the translation machinery would be expected to disrupt many downstream things, including production of the cellular machinery required for postsynaptic development. Since our screen specifically excluded mutants which did not develop to the later stages of embryogenesis, and which did not form neuromuscular junctions, our screen may have highlighted components of pathways specific for a subset of cell differentiation steps that includes glutamate receptor cluster formation. Some of this machinery is possibly localized to the synapse. DrosophilaGluRIIA mRNA is localized to the NMJ and locally translated \[[@B36]\]. Mammalian glutamate receptors may also rely on local translation and editing for surface expression \[[@B37]-[@B39]\]. Preliminary results also suggest that DrosophilaNMJ glutamate receptor cluster formation depends on a burst of receptor subunit transcription that follows contact between pre and postsynaptice cells (Karr & Featherstone, unpublished). Our screen also revealed some important surprises. For example, we identified insertions in two different putative mucin-encoding genes. Mammalian mucins are secreted glycoproteins that are widely implicated in tumor cell adhesion but have no previously identified role in the nervous system \[[@B40]\]. Are these false positives or important insights into postsynaptic development? Interestingly, the \'mucin\' genes identified in our screen are also similar to the C. eleganspolycystin gene lov-1, which is localized to the ciliated sensory endings of dendrites required for male mating behavior, where it may be critical for regulating localization of other transmembrane proteins \[[@B41],[@B42]\]. Our screen also identified the exocyst protein sec-8. Sec-8 has recently been implicated in NMDA receptor trafficking but has not been shown to regulate non-NMDA receptor localization, and the mechanism by which it (and other sec proteins) functions remains unclear \[[@B43],[@B44]\]. We are particularly interested in validating and studying the most surprising candidates from out screen, and have done so for sec 8. This work (Liebl FLW, Chen K, Karr J, Sheng Q, Featherstone DE: *Altered Synapse Development in DrosophilaSec 8 Mutants. Manuscript in preparation, and Liebl & Featherstone, unpublished) reveals that Drosophilasec 8 regulates the synaptic microtubule cytoskeleton to facilitate transmembrane protein localization (Thus the inclusion of sec 8 as a \'cytoskeletal regulator\' in Table [1](#T1){ref-type="table"}). Thus, even the surprises identified by our screen are, so far, apparently reasonable. But note that even validated candidates, such as sec 8, may not work directly or specifically on glutamate receptors. Approximately 12% of the genes identified in our screen encode proteins with unknown or novel function. A better understanding of these genes is important for understanding both those protein families and synaptogenesis -- assuming these proteins are really regulators of receptor cluster formation. Even with a 50% success rate in identifying genes involved in receptor cluster formation -- and the arguments and data above suggest the success rate was much higher -- it is clear that novel genes encode a large fraction of proteins required for glutamate receptor cluster development in the fly NMJ. Therefore, major unexplored areas apparently may exist with regard to postsynaptic development. Interestingly, our screen did not isolate any PDZ-domain or MAGUK proteins, which are widely regarded as essential trafficking and scaffold proteins at mammalian glutamatergic synapses \[[@B45]\]. However, this may not reflect a complete lack of importance for these proteins in fly NMJ glutamate receptor localization. Drosophiladiscs-large (DLG) is the sole fly representative of the mammalian DLG/SAP 97/SAP102/PSD-95 protein family; DLG is important for formation of fly NMJ glutamate receptor clusters that contain the subunit GluRIIB, but not those containing GluRIIA \[[@B11]\]. Because we screened only for alterations in GluRIIA immunoreactivity, we would not have isolated DLG mutants. However, our results do support the idea that PDZ proteins are not predominant components of the glutamate receptor localization/stabilization machinery in DrosophilaNMJs. Many of the genes implicated by our screen probably do not work directly or specifically. Kinases only have to regulate something that in turn regulates receptor cluster formation. Transcription and translation factors probably control expression of multiple proteins required for PSD formation. Given the fact that many proteins definitively required for glutamate receptor cluster formation may not work directly, it is reasonable that a large percentage of the genome appears to be required. In other words, it is not difficult to identify proteins required for receptor cluster development, and we feel that the demonstration of such by itself does not give real insight into receptor cluster formation. An alternative approach, which we call \'functional category analysis\' and introduce here, is to focus on the typesof genes identified rather than the identities of individual genes, in an effort to gain larger insights into the entire process. Subsequent work can then be directed at investigating the processes represented by these gene types, rather than validation of individual genes that may or may not work directly or be applicable to other synapses and organisms. For example, many microtubule and actin regulators were identified in our screen. The target for these proteins, obviously, is likely to be cytoskeletal proteins instead of receptors. But their identification tells which types of cytoskeleton might be important for glutamate receptor cluster formation. Using these clues, we subsequently determined that microtubules are important for fly glutamate receptor trafficking (Liebl FLW, Chen K, Karr J, Sheng Q, Featherstone DE: Altered Synapse Development in DrosophilaSec 8 Mutants. Manuscript in preparation, and Liebl & Featherstone, unpublished), and that A-type receptors are anchored via the 4.1 protein coracle to postsynaptic actin \[[@B20]\]. Conclusion ========== We identified 57 transposon mutants with qualitative differences in glutamate receptor expression and localization. Mutant gene identities need to be validated despite the fact that mutant genes are tagged. Focus on the types of genes identified (\'functional category analysis\') may provide more useful insight into the process of glutamate receptor cluster formation, compared to focus on individual genes. Our results suggest that glutamate receptor cluster formation involves cell adhesion and signaling, extensive and relatively specific regulation of gene expression and RNA regulation, the actin and microtubule cytoskeletons, and many novel/unexplored processes such as those involving mucin/polycystin-like proteins and proteins of unknown function. Methods ======= NMJ staining and microscopy was performed as previously described \[[@B9],[@B19]\]. Briefly, animals were dissected and fixed for 30--60 min in Bouin\'s fixative. Late stage embryos were dechorionated in bleach and then manually devitellinated and dissected. Mouse monoclonal anti-GluRIIA (\'8B4D2\' Iowa Developmental Studies Hybridoma Bank, Iowa City, IA) was used at 1:100. Fluorescently conjugated anti-HRP (Jackson Immunoresearch Labs, West Grove, PA) was used at 1:100. Goat anti-rabbit or goat anti-mouse fluorescent (FITC or TRITC) secondary antibodies (Jackson Immunoresearch Labs, West Grove, PA) were used at 1:400. Confocal images were obtained using an Olympus FV500 laser-scanning confocal microscope. Image analysis and quantification was performed using ImageJ software. Complementation analysis was performed by crossing GFP-balanced P-element mutant stocks to a balanced stock containing a deficiency that removes the insertion site. The F1 generation of each cross was examined for the presence or absence of adult flies carrying neither balancer chromosome. Thus, P-element insertion chromosomes were tested for their capacity to complement the viability of the lethal mutations in the gene carrying the P-element insertion. All P-element stocks were obtained from Bloomington Stock Center <flystocks.bio.indiana.edu>. \'Control\' genotypes in all experiments are w^1118^. Authors\' contributions ======================= FLWL performed most of the immunocytochemistry and microscopy. Analysis was jointly performed by FLWL and DEF. Acknowledgements ================ Thanks to Qi Sheng and Pei-San Ng for technical assistance. This work was supported by NIH/NINDS R01 NS045628 to D.F Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Example phenotypes identified in the screen.A: NMJs on embryonic ventral longitudinal muscles 7, 6, and 13, visualized using anti-HRP antibodies (which stain all neuronal membrane) and anti-GluRIIA antibodies (which stain postsynaptic glutamate receptors). HRP immunoreactivity is red; GluRIIA immunoreactivity is green. In wildtype NMJs (A, left column), GluRIIA immunoreactivity appears as puncta, representing clusters of postsynaptic glutamate receptors. In homozygous P{SUPor-P}KG00333mutants (A, right column), GluRIIA immunoreactivity is eliminated. B: NMJs on third instar larval longitudinal muscles 7 and 6, visualized using anti-HRP and anti-GluRIIA antibodies. In wildtype NMJs (B. left column), GluRIIA immunoreactivity appears as large blobs, representing developmentally merged puncta, at sites of innervation. In homozygous P{SUPor-P}Chro \[KG03258\]mutants (B, middle column), synaptic GluRIIA immunoreactivity is dramatically decreased, although extrasynaptic receptors remain prominent. In contrast, synaptic GluRIIA immunoreactivity is dramatically increased in homozygous P{SUPor-P}vri \[KG01220\]mutants (B, right column). Scale bars: 10 micrometers. ::: ![](1471-2202-6-44-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Pie chart showing the different types of proteins encoded by the insertion loci identified in the screen.The size of the pie wedge represents the proportion of genes of each type identified. Functional categories were based on previously published studies of the Drosophilagene and/or sequence similarity to functionally annotated mouse genes. Table 1 contains a complete list of the genes represented in this chart. ::: ![](1471-2202-6-44-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Genes affected in glutamate receptor expression/localization mutants. Complete list of the mutants identified in the screen. Table 1 lists all of the mutants identified in the screen, organized by putative function of the protein encoded by the gene in which the transposon is inserted. The qualitative glutamate receptor phenotype in the Drosophilamutants is indicated by an arrow pointing upwards (for an increase in immunocytochemically detectable GluRIIA) or downwards (for a decrease in immunocytochemically detectable GluRIIA). The third column lists the Drosophilagene that is mutated by the transposon insertion; superscript letters represent complementation test results: \'f\' = transposon insertion failed to complement a deficiency covering the region, and \'c\' = insertion complemented a deficiency. Note that the confirmed presence of a background lethal mutation (denoted by \'c\') does not indicate whether the NMJ phenotype is also complemented. The fourth column lists the mouse homolog, as determined by best BLAST match to an annotated gene in the mus musculusrefseq database. ::: Allele* Phenotype Mutant gene Mouse protein ----------------------------------------------------------------- --------------- ----------------- ------------------------------- Extracellular matrix P{SUPor-P}CG18819 \[KG01657\] ↓ CG32814 mucin P{SUPor-P}CG14713 \[KG05924\] ↓ CG14713 mucin Cell adhesion P{SUPor-P}mys \[KG02930\] ↓ mys beta integrin P{SUPor-P}KG00212 ↑ Nrt neuroligin Cytoskeleton, cytoskeletal regulators, and adaptor proteins P{SUPor-P}CG10540 \[KG02261\] ↓ CG10540^c^ actin capping protein P{SUPor-P}KG01009 ↓ Tkr actinfilin P{SUPor-P}pnut \[KG00478\] ↑ pnut^f^ septin P{SUPor-P}didum \[KG04384\] ↓ didum^c^ myosin V P{SUPor-P}neb \[KG05913\] ↓ neb^f^ kinesin P{SUPor-P}KG05547 ↓ Cdic dynein intermediate chain P{SUPor-P}CG2095 \[KG02723\] ↓ CG2095^f^ sec 8 P{SUPor-P}lap \[KG06751\] ↓ lap synaptosomal protein 91 P{SUPor-P}noodle \[KG03815\] ↓ CG3210 dynamin Kinases & Phosphatases P{SUPor-P}KG06341 ↓ KP78a^f^ MAP/MT affin.-reg. kinase P{SUPor-P}SNF4Aγ \[KG00325\] ↓ SNF4γ AMP-activated kinase P{SUPor-P}trbl \[KG02308\] ↓ trbl AMP-activated kinase P{SUPor-P}KG04591 ↑ CG15072^c^ SNF-like kinase P{SUPor-P}KG02006 ↓ Rheb RAS-homolog P{SUPor-P}polo\[16-1\] ↓ polo polo-like kinase P{SUPor-P}KG00853 ↓ CG10082^c^ inositol hexaphosphate kin. P{SUPor-P}CG32666 \[KG03058\] ↓ CG32666 serine-threonine kinase P{SUPor-P}KG00564 ↓ ia2 rec. tyrosine phosphatase P{SUPor-P}l(1)G0003 \[KG02485\] ↓ l(1)G003^f^ Rab coupling protein Ubiquitination P{SUPor-P}Cbl \[KG03080\] ↓ Cbl E3 ubiquitin ligase P{GT1}sina \[BG02648\] ↓ sina sina (ubiquitin ligase) Transporters P{SUPor-P}CG5802 \[KG01634\] ↑ CG5802^f^ UDP-galactose translocator P{SUPor-P}KG02272 ↓ CG8029^f^ lysosomal H+ ATPase P{SUPor-P}Vha44 \[KG00915\] ↓ Vha44 vacuolar H+ ATPase Unknown P{SUPor-P}oaf \[KG03408\] ↑ oaf^c^ D130038B21 RIKEN cDNA P{SUPor-P}KG03591 ↓ CG32345 hyp. Protein XP\_355833 P{SUPor-P}cmp44E \[KG03925\] ↓ cmp44E^c^ \'similar to KIAA0953\' P{SUPor-P}CG2185 \[KG02712\] ↑ CG2185 calcium binding P22 P{SUPor-P}KG00333 ↓ CG15358 cd209e antigen P{SUPor-P}CG31694 \[KG04350\] ↑ CG31694^f^ interferon development reg. P{GT1}l(2)35Di \[BG02008\] ↓ l(2)35Di^c^ \[none\] P{SUPor-P}KG06339 ↓ \[none\] Enzymes P{SUPor-P}CG4825 \[KG06018\] ↓ CG4825 phosphatidylserine synth. P{SUPor-P}TppII \[KG03294\] ↓ TppII^c^ tripeptidyl peptidase II Gene expression & Translation P{SUPor-P}KG02920 ↓ hsrω^f^ ladybird homeobox-1-like P{SUPor-P}KG02514 ↓ hbn aristaless-rel. homeobox P{SUPor-P}KG03852 ↓ Ntf-2 nuclear transport factor-2 P{SUPor-P}Trn-SR \[KG04870\] ↓ CG2848 transportin 3 P{GT1}CG10689 \[BG01776\] ↓ CG10689 DEAH box polypeptide P{SUPor-P}CG11107 \[KG02727\] ↓ CG11107 DEAH box polypeptide P{SUPor-P}pnt \[KG04968\] ↓ pnt avian leukemia oncogene P{SUPor-P}KG00625 ↓ CG18591 snrpE P{SUPor-P}RpL3 \[KG05440\] ↓ RpL3^f^ ribosomal protein L3 P{SUPor-P}KG03101 ↓ Ef2b translation elong. factor 2 P{SUPor-P}crp \[KG00953\] ↓ crp transcription factor AP-4 P{SUPor-P}E2f \[KG03332\] ↑ E2f E2f transcription factor P{SUPor-P}armi \[KG04664\] ↓ armi^c^ Moloney leukemia virus 10 P{SUPor-P}vri \[KG01220\] ↑ vri interleukin3 reg. nuc. factor P{SUPor-P}Kr-h1 \[KG00354\] ↓ Kr-h zinc finger protein P{SUPor-P}Bgb \[KG03779\] ↓ Bgb core binding factor beta P{SUPor-P}KG06256 ↓ Chro^c^ Domino (histone deactyl.) P{SUPor-P}Chro \[KG03258\] ↓ Chro Domino (histone deactyl.) P{SUPor-P}EcR \[KG04522\] ↓ EcR nuclear receptor 1, group H :::	PubMed Central	2024-06-05T03:55:58.400207	2005-6-28	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173111/", "journal": "BMC Neurosci. 2005 Jun 28; 6:44", "authors": [ { "first": "Faith LW", "last": "Liebl" }, { "first": "David E", "last": "Featherstone" } ] }
PMC1173112	Background ========== There has been an apparent increase in the incidence of coeliac disease (coeliac disease) over the past 30 years \[[@B1]\]. For instance, the number of reported cases in the Netherlands increased from 0.18 per 1000 live births from 1975--1990, to 0.54 per 1000 live births 1993--1994 \[[@B1]\]. There is also increasing recognition that symptomatic coeliac disease may be the tip of the iceberg, with many more asymptomatic cases in the community \[[@B2]-[@B4]\]. It is unclear whether the changing prevalence of reported cases of coeliac disease is due to a real increase in the number of cases, enhanced awareness of disease or more reliable serological testing. Dietary changes, such as earlier introduction of gluten, have been suggested as contributing to the changing presentation of coeliac disease in some populations \[[@B5],[@B6]\]. Breast feeding, in particular breast feeding after the introduction of gluten, is protective against the development of coeliac disease \[[@B7]\]. There are few data on the clinical features, incidence and prevalence of coeliac disease in Australian populations. In a population based screening study from a rural community of Western Australia, 10/3001 were positive for anti-endomysial antibodies. This community was of predominantly Anglo-Celtic origin \[[@B4]\]. Australians come from diverse ethnic backgrounds, thus the prevalence reported in this community cannot be generalised to the Australian population. The primary aim of this study was to describe the clinical features of children diagnosed with coeliac disease at one tertiary paediatric centre in Australia. Our secondary aim was to determine the presentation patterns of coeliac disease in the local community at different ages. Methods ======= A retrospective cross sectional study was conducted at Sydney Children\'s Hospital, one of two tertiary referral paediatric hospitals in Sydney, Australia. The study population consisted of infants and children who had been referred to the Gastroenterology Department at Sydney Children\'s Hospital for investigation of possible coeliac disease between 1997 and 2002. Subjects were identified through records kept by the Departments of Nutrition and Dietetics, Gastroenterology, and Anatomical Pathology. Ethics approval was granted from the South East Sydney Area Health Service Research Ethics Committee-Eastern Section. Cases and clinical features --------------------------- Cases were defined as those with histological confirmation of coeliac disease \[[@B8]\]. A single pathologist reviewed all slides. Where the histological diagnosis of coeliac disease was uncertain, clinical and laboratory features, including response to gluten free diet, were included. The clinical details of each patient were extracted from the medical records retrospectively. Features of interest included presenting symptoms as well as the symptoms and signs elicited during consultation with a paediatric gastroenterologist. Weight and height measurements at diagnosis were documented. SD (standard deviation) scores were calculated using reference data \[[@B9]\]. The ethnic mix for the reference population of children admitted to Sydney Children\'s Hospital was described using child\'s country of birth. Laboratory investigations ------------------------- The results of haemoglobin, iron studies, serum IgA, anti-gliadin (AGA) and anti-endomysial antibodies (EMA) were obtained from each patient\'s medical records. AGA and EMA tests had been performed at SEALS (South Eastern Area Laboratory Service, Kogarah, NSW). The SEALS laboratories measure EMA using indirect immunoflurescence, anti-human FITC conjugate IgA against monkey distal oesophagus. The slides are prepared by MeDICa (Encinitas, CA 92024). A value of 10 is used for as positive cut off. AGA were measured by commercial immunoassay kits (Analytica Ltd, Castle Hill NSW Australia). The cut off values used for children were IgA\>25 and IgG \>46. Low haemoglobin values for age were defined as those less than the reference range for age used by the hospital laboratory (SEALS). The low end of the normal range for children 6 months to 2 years was 104 g/L; 2--4 years 107 g/L; 4--8 years 110 g/L; 8--12 y 113 g/L; \>12 y 130 g/L. Patients were considered iron deficient on the basis of a microcytic blood film, low serum iron, low serum ferritin, and elevated iron binding capacity. Statistical analysis -------------------- Data were analysed by simple descriptive analysis performed manually and using Excel (Microsoft Office 97). Fisher\'s Exact tests were used to compare independent data using STATA \[[@B10]\]. Where data was not available, summary statistics were performed using only the data available. Results ======= Subjects -------- A total of 119 patients were identified over the six year period. Forty-three patients were excluded as there was another medical reason to explain their histological features or reason for dietetic referral. The medical records were unable to be located for two patients, leaving 74 subjects with coeliac disease available for analysis. The male: female ratio of patients was 1:2. The age at diagnosis ranged from 11 months to 14 years with median of 5.5 years. Seven patients (9% of total group) were less than 2 years of age (Figure [1](#F1){ref-type="fig"}). The ethnic mix (based on country of birth) of all admissions to SCH consists of approximately 94% Australian, 2% Asian, 1% UK and Europe, 0.05% Africa, and 0.02% Pacific Islands (data compiled by MS using admission data for Sydney Children\'s Hospital for 2002). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Age at Presentation of Coeliac disease in the cohort of 74 subjects. This figure illustrates the age distribution of the 74 subjects with coeliac disease in this study. Infants less than 2 years of age represented only 12% of the population. 27% of subjects were aged between 4 and 5 years, 36% were aged between 5 and 10 years, 26% were more than 10 years. ::: ![](1471-2431-5-11-1) ::: Clinical features ----------------- Sixty nine of the seventy four children had presented with symptoms (Table [1](#T1){ref-type="table"}). Full details of the presenting symptoms were available in 65 of the patients. Twenty nine of these children were aged less than five (preschool) and forty were aged five years or greater. The most common presenting symptoms in younger children (those \<5 years) were diarrhoea (59%), irritability (34%) and weight loss (38%). In older children (≥ 5 years), the most common presenting feature was abdominal pain (55%), followed by diarrhoea (26%). Twenty-three percent of the symptomatic children had \> 3 presenting symptoms. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Clinical features of 69 symptomatic children with Coeliac disease ::: Age \< 5 years n = 29 (% of group) Age ≥ 5 years n = 40 (% of group) --------------------------------------------- ---------------------------------------- --------------------------------------- Abdominal pain 6 (20) 25 (62.5)\* Diarrhoea 17 (59) 12 (30)\* Tiredness/lassitude 9 (31) 9 (22.5) Abdominal distension 16 (55) 9 (22.5)\* Constipation 4 (14) 5 (12.5) Irritability 10 (34) 6 (15)\* Mouth Ulcers 2 (7) 6 (15) Anaemia 5 (17) 10 (25) Weight loss or lack of expected weight gain 11 (35) 6 (15)\* Anorexia and vomiting 7 (24) 0\* (\*p \< 0.05) This table displays the number of subjects with each symptom. The numbers in brackets describe the percentage of the age group with that symptom. ::: Five patients had no gastrointestinal symptoms at diagnosis. Four of these patients had positive coeliac serology during routine complication screening for type 1 diabetes (T1DM). The fifth patient had coeliac serology performed as part of investigation into recurrent infections. Overall, ten children were noted to have other medical problems potentially related to the development of coeliac disease. Five patients had T1DM, two patients had Trisomy 21, two patients had juvenile rheumatoid arthritis, and one patient had Raynaud\'s phenomenon. Fourteen of the 79 (18%) patients had a family history of coeliac disease in a first degree relative. The mean weights and heights of the subjects with coeliac disease were not significantly different to the reference sex and age matched population (mean weight SD -0.45 ± 0.14; mean height SD -0.4 ± 0.17). However, 5% of the children with coeliac disease had a weight SD score of more than 2 SD below the mean and 12% of children had a height SD score of more than 2 SD below the mean. Laboratory data --------------- Seventeen of 64 (26%) patients with haemoglobin levels available had haemoglobin levels below the reference range for age (mean haemoglobin 118 g/L ± 0.2). Thirty two of 37 (86%) patients who had iron studies available were deemed iron deficient (low serum iron and ferritin). Of those who were biochemically iron deficient, 8 (25%) had normal haemoglobin values for their age. IgG AGA was elevated in 97% of cases. IgA AGA was elevated in 94% of cases. There were no patients with IgA deficiency. EMA were elevated in 97% of this group of children. There were two patients with positive EMA antibodies who had negative AGA, on the other hand there was one patient (aged 9 years) with negative EMA who had positive AGA. Discussion ========== This study describes the clinical features at presentation in a group of Australian children diagnosed with coeliac disease during a time of increasing awareness of this disease and widespread availability of reliable laboratory screening tests. The children included in this retrospective review presented with a wide variety of clinical features. The most common clinical features of coeliac disease in younger children were the so-called classical symptoms of diarrhoea, irritability and weight loss. However, older children more commonly presented with abdominal pain. There were few cases without gastrointestinal symptoms in our cohort. The clinical features of the children in this study were similar to the clinical features described from populations in New Zealand \[[@B11]\], South Yorkshire \[[@B12]\] and Sweden \[[@B13]\]. Although abdominal pain is a common symptom of older children with coeliac disease, there is no association between classical recurrent abdominal pain and coeliac disease \[[@B14]\]. Anaemia was identified in approximately one quarter of the cases. Furthermore, many children had laboratory evidence of iron deficiency but were not anaemic at the time of diagnosis. Unfortunately only 86% of subjects had a full blood count taken, and 50% had blood taken for iron studies. Thus the true prevalence of anaemia and iron deficiency is uncertain. Cross sectional studies have identified a high prevalence of coeliac disease in patients with iron or folate deficiency \[[@B15],[@B16]\]. The incidence of coeliac disease in patients in the community presenting only with iron or folate deficiency has been reported at 4.7% \[[@B15]\]. Anaemia was reported in only 5% of children with coeliac disease from the Netherlands \[[@B1]\]. In the United Kingdom, iron deficiency anaemia was reported in approximately 20% of adults with coeliac disease. Although the current data may have underestimated the true incidence of anaemia and iron deficiency, it is likely higher than that reported in these previous cohorts. Growth failure in height and weight was not as common in this cohort as was described in coeliac children from the Netherlands, where 49.7% of children had poor growth \[[@B1]\]. In our study, 5% of children had a weight more than 2 SD below the mean, and 12% of children had a height more than 2 SD below the mean. Growth failure is more commonly seen in children diagnosed at a younger age than in our study \[[@B17]\]. The discrepancy in reported growth failure may also be due to different definitions of growth failure. In addition, because this was a retrospective study, it was not possible to obtain adequate information about growth velocity or of the children\'s genetic growth potential. The children in our study were older at the time of diagnosis than previous surveys of children with coeliac disease. In a cohort of patients diagnosed between 1950 and 1969, 70 of 91 were less than 2 years at the time of diagnosis and only 3 were more than 6 years. In a smaller cohort of children diagnosed between 1972 and 1975, 32 of 42 were less than 2 years at the time of diagnosis and 6 were greater than 6 years \[[@B18]\]. In the Netherlands, the proportion of children with coeliac disease aged \< 2 yrs has remained at about 60% from 1975 to 1994 \[[@B1]\]. An epidemic in coeliac disease in Swedish Children in the 1980\'s paralleled the change in feeding practices at the time, in particular increased exposure to wheat, rye and barley and decreased duration of breast feeding \[[@B5]\]. A prospective study of coeliac disease in infants showed that a longer duration of breast feeding, smaller amounts of gluten in the diet in infants less than 12 months \[[@B17]\] and breast feeding after gluten is introduced to the diet reduces the risk of coeliac disease in infants less than 2 years \[[@B7]\]. Thus, the later age of acquiring coeliac disease in the Australian population may be due to the delayed introduction of solids in the diet of Australian children. The current recommendations in Australia are to introduce solids at 6 months (NHMRC Dietary Guidelines for Children and Adolescents in Australia 1993). In this study, there were relatively few cases of coeliac disease detected by the screening of at risk groups. This contrasts with the worldwide trend of an increased prevalence of coeliac disease detected by screening \[[@B2],[@B19]-[@B22]\]. A large multi-centered study of children and adults in the United States of America showed that the prevalence of celiac disease was 1:22 in first degree relatives, 1:56 in symptomatic individuals or a disorder known to be associated with celiac disease (type 1 diabetes, down syndrome, anemia, arthritis, osteoporosis, infertility and short stature), and 1:133 in the not-at-risk population \[[@B23]\]. In our study, the best screening test in children aged over 2 years with symptoms of, or who are at risk of, coeliac disease was IgA EMA. Other studies have shown that this test has a sensitivity of 95--98%, specificity of 94--95%, positive predictive value of 91--95% and negative predictive value of 96--98% \[[@B24]\]. For children less than 2 years of age, IgA AGA may be more reliable \[[@B24]\]. Increasingly over the last few years, reports suggest that tissue-transglutaminase (tTG) antibody may be even more reliable than EMA as a screening test \[[@B25],[@B26]\], and in addition is more reproducible and reliable. This test was not available during the period of this review. Although the utility of these newer tests is improved compared to AGA antibodies, they are not yet accepted as diagnostic tests. Any child with symptoms suspicious of coeliac disease, even if negative serology, warrants further review by a paediatric gastroenterologist, as a small bowel biopsy remains the gold standard test for diagnosis of coeliac disease. Conclusion ========== Coeliac disease in children does not always present with classical clinical features, especially in older children. The possibility of coeliac disease should be considered in any child presenting with diarrhoea, irritability and weight loss, and particularly in older children with abdominal pain. With the widespread availability of minimally invasive screening tests for coeliac disease, identifying cases in at risk groups by screening should also be considered. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MS conceived the study, performed the data collection and statistical analysis and drafted the manuscript; KW for assistance in collecting the data and reviewing the manuscript; VT performed the histological examination of the biopsy specimens; AD and TB were involved in the study design, writing and reviewing of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2431/5/11/prepub> Acknowledgements ================ Dr Phillip Emder for his support and review of the manuscript.	PubMed Central	2024-06-05T03:55:58.403675	2005-5-21	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173112/", "journal": "BMC Pediatr. 2005 May 21; 5:11", "authors": [ { "first": "Monique L", "last": "Stone" }, { "first": "Timothy D", "last": "Bohane" }, { "first": "Kylie E", "last": "Whitten" }, { "first": "Vivienne H", "last": "Tobias" }, { "first": "Andrew S", "last": "Day" } ] }
PMC1173113	Background ========== The concern for health disparities extends to residents of rural areas, who are recognized to face distance barriers when seeking to access health services. Concerns about disparities apply to mental as well as physical health. Studies addressing rural residence as a risk factor for poor mental health are important to verify the existence of a disparity and to clarify its nature. Singh and Siahpush studied rural residence as a risk factor for suicide. They reported that suicide risk increases for males with increasing levels of rurality and that the rural-urban differentials are increasing over time. Rural suicide rates also were higher for women but the rural-urban differential was found to be decreasing over time \[[@B1]\]. Recently, national experts have emphasized the recognition of mental health as a public health problem \[[@B2]-[@B4]\]. Suicide, though important, may not be the most important focus for a public health approach to mental health. Community health programs typically are targeted at moderately healthy populations, not the seriously ill, because the goal is prevention so as to avoid the need for treatment. Less serious mental health concerns affect a large fraction of the population. The more common mental health problems often are preventable and may be amenable to health promotion programs involving self-help strategies. Population surveys such as the one reported here naturally are more relevant to the less severe, more common mental health issues. Rural health advocates frequently make the case that rural residents have worse health than residents of urban areas, implying that residing in a rural area is a risk factor for poor health \[[@B5]\]. However, it is also possible that the causes of poor health are the same regardless of where one lives, and thus programs should be targeted not at rural locations but at populations that are afflicted with the genuine risk factors (e.g., poverty, low educational attainment, and age). The purpose of the study reported here was to investigate the relationship between rural residence and frequent mental distress, which can be described as poor self-rated mental health. This study differs from some earlier studies in that it includes a variable representing rural or urban residence in an analysis of data drawn from the Behavioral Risk Factor Surveillance System (BRFSS). The BRFSS does not ordinarily contain a variable representing rural or urban residence of the respondent. The Texas Department of State Health Services recently created such a variable, permitting comparisons that previously were not possible. Methods ======= Data were obtained from the 2003 Texas Behavioral Risk Factor Surveillance (BRFSS) survey. The Texas Behavioral Risk Factor Surveillance System (BRFSS), sponsored by the Texas Department of State Health Services in partnership with the Centers for Disease Control and Prevention (CDC), is an ongoing random-digit dialed telephone survey that collects information from the non-institutionalized, civilian adult population. BRFSS surveys adhere to the highest scientific standards for telephone researchand pose questions relating to health status, personal health habits, and use of preventive health services. The 2003 survey data were collected over a 12-month period frominterviews with 6,035 Texas residents. The overall cooperation rate, or proportion of individuals actually contacted on the phone who completed the survey, was 70%. The more conservative response rate favored by the Council of American Survey Research Organizations (CASRO), which adjusts for the estimated number of eligible respondents in the samplewho could not be contacted due to technical or other barriers, was 41%. Missing values were excluded from the analyses if they constituted less than 5% of cases for an independent variable. Missing dummy variables were created for independent variables for which 5% or more of the cases had missing values (household income and body mass index). The data set contained 5,757 observations. Measures -------- The dependent variable was frequent mental distress, which was assessed with the following question: \"Now, thinking about your mental health, which includes stress, depression and problems with emotions, for how many days during the past 30 days was your mental health not good?\" Frequent mental distress was defined as having 14 or more days of \"not good\" mental health. The primary independent variable of interest was whether the individual lived in a metropolitan, suburban, or non-metropolitan/rural area. Many prior studies have categorized the area of residence as metropolitan vs. non-metropolitan. However, a dichotomous classification may mask important differences between residents of highly urban and suburban areas. Thus, we chose a more detailed classification system of metropolitan-central city, metropolitan-suburban, and non-metropolitan areas based on respondents\' reports of their county of residence. Metropolitan-central city was defined as an urban core county containing 50,000 or more persons. Metropolitan-suburban was defined as a metropolitan county adjacent to a core metropolitan-central city area. All other counties were considered non-metropolitan or rural. Additional independent factors included demographics and body mass index. Demographic variables were age category (18--24, 25--34, 35--44, 45--54, 55--64, and 65+), race/ethnicity (non-Hispanic white, Hispanic, black, and other), gender, marital status (single/never married/divorced/widowed vs married/member of a couple), educational status (less than a high school degree, a high school degree, some college, and college graduate), and household income (\<\$25,000, \$25,000 to \$75,000, ≥\$75,000, and income missing). Body mass index (BMI) categories were defined as obese (BMI ≥ 30), overweight (BMI 25--29.99), normal weight (BMI \<25 and ≥ 18.5), underweight (BMI \< 18.5), and a BMI missing category. Analyses -------- The percentage of respondents with frequent mental distress (FMD) was calculated by urban/rural residence and all other independent variables. Multivariate logistic regression analyses were conducted to estimate adjusted odds ratios (AORs) for selected risk factors. Data were weighted to reflect the demographic distributions of the sample. Because we were interested in testing whether the predictors of FMD differ between residents of metropolitan-central city, metropolitan-suburban, and non-metropolitan areas, we then stratified the analyses by metropolitan status. STATA 8.0 was used to conduct descriptive and multivariate analyses to account for the complex sampling scheme and population weights \[[@B6]\]. Results ======= Frequent mental distress was reported by 9.79% of the respondents. The first column of Table [1](#T1){ref-type="table"} describes respondents\' demographic, social, economic, and body mass characteristics. Approximately 62% and 21% of respondents lived in metropolitan-central city and metropolitan-suburban areas, respectively, whereas about 17% resided in non-metropolitan counties. The second column describes the percentage of respondents with frequent mental distress by each selected risk or independent variable. Residents of metropolitan-suburban areas had the highest rate of FMD (11.92%). Those residing in metropolitan-central city and non-metropolitan areas had similar rates of FMD (9.34 and 8.77%). After adjustment for other variables, persons in metropolitan-suburban areas had a significantly higher odds of FMD compared to metropolitan-central city residents (AOR = 1.37). Non-metropolitan residents did not have a significantly different adjusted odds of FMD as compared to individuals residing in metropolitan-central city areas. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Prevalence of frequent mental distress (FMD) and adjusted odds ratios (AOR) for selected risk factors, State of Texas, 2003, (N=5,757) ::: Risk variable Overall % \% with FMD (95% CI) AOR\* (95% CI) N= 5,757 Pvalue ---------------------------- ----------- ---------------------- --------------------------- ---------- Residence Metropolitan-central city 61.50 9.34 (8.24--10.44) Reference Metropolitan-suburb 21.23 11.92 (9.77--14.07) 1.37 (1.06--1.77) 0.016 Non-metropolitan 17.27 8.77 (6.59--10.96) 0.86 (0.63--1.18) 0.350 Age group 18--24 14.66 11.92 (8.80--15.05) Reference 25--34 20.23 6.79 (5.04--8.54) 0.92 (0.62--1.34) 0.635 35--44 21.09 9.26 (6.91--11.60) 1.17 (0.78--1.69) 0.488 45--54 18.08 11.40 (9.32--13.49) 1.26 (0.86--1.85) 0.227 55--64 11.88 10.39 (8.29--12.50) 0.91 (0.59--1.41) 0.675 65+ 14.07 8.57 (6.84--10.32) 0.51 (0.33--0.78) 0.002 Race/ethnicity non-Hispanic white 58.94 9.54 (8.61--11.68) Reference Hispanic 28.91 9.22 (7.51--10.95) 0.56 (0.41--0.75) \<0.001 Non-Hispanic Black 8.20 13.58 (9.47--17.68) 0.90 (0.62--1.31) 0.587 Other 3.95 9.72 (5.20--14.24) 1.15 (0.67--0.98) 0.607 Sex Female 50.91 12.06 (10.80--13.31) Reference Male 49.09 7.44 (6.16--8.73) 0.60 (0.48--0.76) \<0.001 Marital status Single/never married 38.65 13.28 (11.58--14.98) Reference Married 61.35 7.61 (6.61--8.61) 0.66 (0.54--0.82) \<0.001 Educational status \< high school degree 17.52 13.61 (10.96--16.26) Reference high school degree 27.44 12.84 (10.77--14.91) 0.80 (0.59--1.10) 0.172 some college 26.51 8.72 (7.20--10.24) 0.53 (0.37--0.74) \<0.001 College graduate 28.53 5.50 (4.41--6.60) 0.38 (0.26--0.55) \<0.001 Household income \<\$25,000 30.01 15.79 (13.68--17.90) Reference \$25,000 to \$75,000, 39.27 7.50 (6.28--8.72) 0.52 (0.40--0.67) \<0.001 ≥\$75,000 18.85 6.40 (4.75--8.06) 0.49 (0.33--0.71) \<0.001 Missing 11.87 7.58 (5.41--9.75) 0.50 (0.35--0.71) \<0.001 Body mass index Obese 23.02 12.66 (10.58--14.75) 1.56 (1.19--2.05) 0.001 Overweight 34.61 8.76 (7.31--10.21) 1.22 (0.93--1.60) 0.159 Normal 34.07 8.18 (6.75--9.61) Reference Underweight 2.08 24.38 (15.00--33.76) 3.15 (1.82--5.45) \<0.001 BMI missing 6.22 8.85 (5.42--12.27) 0.88 (0.53--1.44) 0.602 CI = confidence interval. \Adjusted for all other variables in Table 1. N= 5,757 prior to weighting. ::: Age differences in FMD were revealed as those 65 years and older had a significantly lower adjusted odds of FMD than individuals between 18 and 24 years of age (AOR = 0.51). In regard to race/ethnicity, a higher percentage of Blacks/African Americans (non-Hispanic) reported FMD than non-Hispanic whites, Hispanics, or other non-Hispanic racial groups. However, when adjusting for other factors, only Hispanics\' odds of FMD differed from non-Hispanic whites\' (AOR = 0.56). Males exhibited a significantly lower adjusted odds of FMD (AOR = 0.60) as compared to females. Married or living-together persons had a significantly lower adjusted odds of FMD than single persons (AOR = 0.66). Educational attainment and household income, two indicators of socioeconomic status, were also associated with FMD. Compared with those with less than a high school degree, individuals with some college education as well as those who were college graduates had lower adjusted odds of FMD (AOR = 0.53 and 0.38, respectively). Similarly, individuals who had higher household incomes experienced a lower adjusted odds of FMD. Body mass index categories of obesity and underweight were associated with substantially higher odds (AORs = 1.56 and 3.15, respectively) of FMD as compared to normal weight persons. We also performed analyses of the predictors of FMD by metropolitan-central city, metropolitan-suburban, and non-metropolitan status (Table [2](#T2){ref-type="table"}). The findings for each model were similar with a few notable exceptions. Among those residing in metropolitan-central city areas, Hispanics had a significantly lower odds (AOR = 0.53, 95% CI = 0.37--0.77) of FMD than non-Hispanic whites. However, among residents of metropolitan-suburban and non-metropolitan areas, the odds of FMD did not differ between Hispanics and non-Hispanic whites. Similarly, male gender was significant in the model for metropolitan-central city residents (AOR = 0.50, 95% CI = 0.37--0.68) but insignificant in the remaining two models. Married or living-together persons were less likely to have FMD among metro-central city respondents, but marital status was not significant elsewhere in the state. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Adjusted odds ratios (AOR) of frequent mental distress (FMD), stratified by metropolitan status ::: Risk variable Metropolitan-Central City N= 3,505 Pvalue Metropolitan-Suburban N= 1,251 Pvalue Non-Metropolitan N= 1,001 Pvalue ------------------------ -------------------------------------- ---------- ---------------------------------- ---------- ----------------------------- ---------- AOR \ (95% CI) AOR \* (95% CI) AOR \* (95% CI) Age group 18--24 Reference Reference Reference 25--34 1.25 (0.77--2.03) 0.366 0.40 (0.19--0.85) 0.016 0.94 (0.32--2.72) 0.904 35--44 1.40 (0.85--2.30) 0.186 0.63 (0.28--1.40) 0.256 1.73 (0.64--4.67) 0.283 45--54 1.42 (0.87--2.30) 0.157 0.96 (0.46--1.98) 0.904 1.49 (0.50--4.46) 0.476 55--64 1.16 (0.65--2.09) 0.614 0.59 (0.25--1.37) 0.218 0.93 (0.31--2.77) 0.892 65+ 0.60 (0.33--1.08) 0.091 0.26 (0.11--0.59) 0.001 0.77 (0.29--2.02) 0.586 Race/ethnicity Non-Hispanic white Reference Reference Reference Hispanic 0.53 (0.37--0.77) 0.001 0.49 (0.23--1.03) 0.058 0.88 (0.41--1.88) 0.738 Non-Hispanic Black 0.90 (0.59--1.39) 0.648 0.88 (0.31--2.46) 0.804 0.90 (0.33--2.50) 0.845 Other 1.23 (0.62--2.43) 0.547 0.81 (0.25--2.63) 0.725 1.57 (0.37--6.65) 0.540 Sex Female Reference Reference Reference Male 0.50 (0.37--0.68) \<0.001 0.73 (0.47--1.14) 0.171 0.77 (0.44--1.36) 0.368 Marital status Single/never married Reference Reference Reference Married 0.56 (0.43--0.74) \<0.001 0.81 (0.50--1.30) 0.384 0.98 (0.56--1.70) 0.931 Educational status \< high school degree Reference Reference Reference high school degree 0.75 (0.51--1.12) 0.165 0.85 (0.44--1.61) 0.623 0.90 (0.42--1.96) 0.800 some college 0.59 (0.38--0.92) 0.018 0.36 (0.17--0.77) 0.008 0.71 (0.31--1.62) 0.420 College graduate 0.42 (0.26--0.67) \<0.001 0.41 (0.20--0.83) 0.013 0.15 (0.05--0.50) 0.002 Household income \<\$25,000 Reference Reference Reference \$25,000--\$75,000 0.53 (0.38--0.74) \<0.001 0.56 (0.32--0.97) 0.040 0.34 (0.17--0.71) 0.004 ≥\$75,000 0.51 (0.33--0.80) 0.003 0.36 (0.16--0.82) 0.016 0.94 (0.38--2.42) 0.905 Income missing 0.45 (0.28--0.71) 0.001 0.62 (0.27--1.39) 0.245 0.41 (0.18--0.96) 0.040 Body mass index Obese 1.96 (1.37--2.80) \<0.001 1.04 (0.61--1.79) 0.881 1.28 (0.58--2.87) 0.542 Overweight 1.19 (0.82--1.72) 0.357 1.16 (0.70--1.92) 0.562 1.69 (0.78--3.63) 0.181 Normal Reference Reference Reference Underweight 2.98 (1.54--5.78) 0.001 1.26 (0.37--4.29) 0.710 11.94 (3.66--38.97) \<0.001 BMI missing 0.22 (0.67--2.23) 0.513 0.35 (0.11--1.16) 0.087 0.74 (0.19--2.83) 0.656 CI = confidence interval. \* Adjusted for all other variables in Table 1. ::: Discussion ========== The theory that residing in a rural location rather than an urban area is a risk factor for frequent mental distress is not supported by our data. Instead, persons living in suburbia were found to be at risk for FMD after adjusting for other risk factors. We suspect that some stressors may be at work in the suburbs that are not found in rural areas or central cities. This warrants further investigation. In our data, lifestyle variables (obesity and underweight), Hispanic ethnicity, female gender, being unmarried, and having less education are more important risk factors for FMD than rural residence. Because obesity and underweight are related to frequent mental distress, public health programs designed to achieve normal body weight might improve mental health \[[@B7]\]. The potential for population-level health promotion strategies to reduce the prevalence of mental health problems is important, since the medical care system has been chastised for its insensitivity to underlying mental health problems among patients who present with physical symptoms \[[@B8]-[@B11]\]. Efforts directed at improving the quality of mental health services delivered in primary care settings have had mixed results \[[@B9]\]. Clearly, this is another case where population-based preventive strategies by public health agencies might be more effective than individualized medical treatment of mental health problems, at least for less severe or incipient cases \[[@B7],[@B11]\]. The widespread and increasing \[[@B12]-[@B14]\] prevalence of frequent mental distress in the population (8.6 percent of adults nationally a decade ago and ten percent in 2001 \[[@B13]\]) necessitates a public health response. In addition, the psychological consequences of any future terrorist attacks or natural disasters will require population-based responses that can reach large numbers of people in a short time \[[@B15]\]. Conclusion ========== This study examined how rural residence was related to frequent mental distress in the general population in a single state. Our findings are a direct result of the modeling strategy employed. Accordingly, replication of these results by other investigators is important. Limitations of the study include a modest participation rate, the cross-sectional design, and the loss of some cases due to missing data. The cross-sectional nature of the study makes causal conclusions impossible. Furthermore, the use of a self-reported single item to measure frequent mental distress does not equate to providing diagnostic information. Another important limitation is that most seriously mentally ill persons are likely to be missed by a survey such as this because they at increased risk of being homeless or otherwise unavailable for telephone interviews. Nevertheless, the single item has been accepted by public health researchers for use in population surveys \[[@B11]-[@B14],[@B16]-[@B18]\]. Health status in rural populations is believed to be lower than in urban areas, partly due to higher rates of poverty \[[@B5]\]. This assertion specifically relates to rates of chronic disease, infant mortality, injury rates, trauma mortality, and overall mortality. However, most government sponsored surveys do not include a variable that reflects urban or rural residence. Our study is the first to use a Behavioral Risk Factor Surveillance System (BRFSS) survey with a variable reflecting rural residence. Our results do not support the concern about rural residence being a risk factor for poor mental health. The risk factors for poor mental health are found in urban as well as rural areas, although the delivery of services may be more difficult in rural areas. When Singh and Siahpush reported that suicide rates were higher in rural areas, they attributed this finding partly to social isolation \[[@B1]\]. If this theory was correct, then we would expect frequent mental distress also to be more prevalent among rural respondents in our data. However, we did not find rural respondents to be at greater risk of frequent mental distress than urban respondents. We question the generalization that rural people experience greater levels of social isolation, since rural communities sometimes may be more integrated and supportive than urban neighborhoods. Community health programs that are supported by strong evidence will, for the most part, focus on three strategies: policy changes, mass communication, and targeted health education \[[@B19]\]. Applied to risk factors for frequent mental distress, examples of policy changes might be zoning practices that promote physical activity, labeling of unhealthy menu items in restaurants, and increases in liquor and cigarette taxes. Media campaigns could be employed to raise awareness about anxiety, depression, stress, and healthy coping strategies. The third approach to community mental health is to target health education programs at persons who are at high risk. In Texas, this would be the elderly, the poor, and persons who are undereducated. Since none of these population-based strategies involve personal health services or require substantial investments in physical plant, they could each be carried out without regard to whether the targeted populations reside in rural or urban areas. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= TFB carried out the statistical analysis, wrote the methods section, and wrote the results section. JB wrote the section on sampling and the BRFSS in Texas. JER conceived of the study, planned the design, and wrote the background, discussion and conclusions. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/46/prepub>	PubMed Central	2024-06-05T03:55:58.405448	2005-5-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173113/", "journal": "BMC Public Health. 2005 May 16; 5:46", "authors": [ { "first": "James E", "last": "Rohrer" }, { "first": "Tyrone F", "last": "Borders" }, { "first": "Jimmy", "last": "Blanton" } ] }
PMC1173114	Background ========== Dietary supplements are defined in the United States Dietary Supplement Health and Education Act of 1994 (United States DSHEA) as any product (other than tobacco) intended to supplement the diet that bears or contains one or more of the following ingredients: a vitamin, a mineral, an herb or other botanical, an amino acid, a supplement used by man to supplement the diet by increasing the total dietary intake, or a concentrate, metabolite, constituent, extract, or combination of any ingredient described above \[[@B1]\]. Hankin \[[@B2]\] and Radimer et al.\[[@B3]\] used the term \'nonvitamin, nonmineral (NVNM) supplements\' to differentiate a new class of dietary supplements. NVNM supplements include products such as chrondroitin, sulfate, kava kava, ginseng, echinacea, ginkgo biloba, garlic herbals, botanicals, protein and amino acids, as well as Brewer\'s yeast and shark cartilage \[[@B3]-[@B5]\]. Although use of NVNM supplements is increasing in popularity, patterns of use for these supplements are not well known, and what is available is not always consistent, clear, or easily accessible \[[@B3]\]. People have reported a variety of reasons for taking dietary supplements, including decreasing their susceptibility to health problems such as stress, colds, heart attacks, osteoporosis, neural tube defects, dental caries and cancer, as well as to increase energy \[[@B6],[@B7]\]. The supplements typically can be considered as falling into three distinct categories: supplements that are believed to add nutrients to a system that owing to inadequate dietary practices might otherwise be lacking \[[@B8]\]; supplements that allege rapid weight loss and maintenance of the loss \[[@B9]\], or conversely, supplements that are believed to result in weight gain or muscle development \[[@B10]\]. Furthermore, the rise in popularity of freely available substances purported to enhance performance, such as androsterone (a steroid precursor), and creatine (an amino acid derivative), has also been documented \[[@B11],[@B12]\]. To date, however, the taking of nutritional preparations has not consistently indicated a competitive advantage other than a possible placebo effect, or one resulting from the treatment of concurrent nutritional deficiencies \[[@B13]\]. Consequently, according to the United States Commission on Dietary Supplement Labels, it is important for health and nutrition professionals to become more knowledgeable about all types of dietary supplements in order to help consumers make appropriate choices \[[@B14]\]. Variables that have been found to have a relationship with supplement use are race or ethnicity, age, education, income and lifestyle variables such as drinking, smoking, and exercising \[[@B15],[@B16]\]. There is documented evidence that the use of NVNM supplements in western society is high, and its use is increasing worldwide \[[@B17]-[@B20]\]. In some countries, such as Germany, many botanical products are classified as drugs, and thus are subject to quality, safety, and efficacy regulations \[[@B21]\]. In the United States, although it is acknowledged that some dietary supplements, particularly botanicals, are used to prevent and treat diseases, DSHEA distinguishes these products from drugs and food additives and thus they are not subject to the same regulations \[[@B1],[@B4],[@B5]\]. Use of NVNM supplements increased substantially with the passage of the DSHEA \[[@B1]\], which gave manufacturers greater freedom to market more products as dietary supplements and to provide information about their purported benefits in package labeling and advertising. Although the United States Food and Drug Administration (FDA) regulates additives and drugs, premarket review of dietary supplements is minimal \[[@B22]\]. The use of NVNM supplements has increased recently in many countries, a fact also reflected in our country. However, while the worldwide rate of consumption of these supplements is near to 50% \[[@B17]-[@B19]\], it is still rather low in developing countries. This may be put down to indications that those with higher education and income levels seem to be comparatively more likely to use supplements \[[@B15]\]. In Turkey, there is no regulation of NVNM supplements in that they are widely available over the counter from such places as pharmacies and markets without a prescription. There is no regulation or health policy on the use of NVNM supplements, and furthermore, data on the prevalence and use of NVNM supplements are limited or non-existent \[[@B23]\]. Finally, there is a need to understand the frequency with which people use supplements and what variables are associated with particular frequencies of use. In general, although we consider Turkish NVNM supplement consumption to fit into the prevalence patterns for developing countries, to our knowledge, no data is available either concerning use in Turkish students or the population as a whole. For this reason, the objective of this survey was to quantify the prevalence of NVNMS usage among university students; to identify supplements consumed and rationale for usage; to identify sources of supplement information; and to relate usage to selected demographic characteristics. Methods ======= Sampling -------- The survey was conducted between September 2004 and December 2004 at Osmangazi University, a public university located in the west of Turkey. All students from the schools of medicine, engineering and architecture, science and literature, economics, education, and the college of health services were invited to participate. Participation was voluntary and anonymous, and the Director of the Institution approved the survey. Subjects -------- Subjects were randomly selected with a stratified sampling method. Osmangazi University is an urban, mid-sized university with six schools, all of which were included in the survey. At least one class from the first, second, third and fourth years in each department were randomly selected to participate in the survey. The sample size for the survey was determined by multiplying at least 10% by the number of the students: the total numbers of the students in each department and in our survey, respectively, were 1097 and 203 for the school of medicine; 4212 and 561 for the school of engineering and architecture; 3556 and 403 for the school of science and literature; 1394 and 178 for the school of economics; 858 and 177 for the school of education; and 302 and 212 for the college of health services. Students were excluded from the survey due to: unwillingness to participate (n = 186), handing in incomplete questionnaires (n = 87), and non-attendance of class (n = 109). Thus, the response rate for sample respondents was 83% (1871/2253) of 2253 subjects, with ages ranging from 17 to 28, leaving a total of 1871 students (962 women, 909 men). Questionnaire and interview schedules ------------------------------------- Completion of the questionnaire was self-reporting on the part of the students. The dates on which the study would be conducted were determined in cooperation with class teachers in the schools concerned. The students completed questionnaires in the presence of a member of the research team. ### The first section of questionnaire This section identified consumption of NVNM supplements during the survey or the past year, previous use of dietary supplements, and the probability of students considering the use of NVNM supplements in the near future. If participants reported having used a NVNM supplements during the survey period or the year prior to the study, they were further questioned on a number of topics related to consumption: the number used; the name of the product(s), 73 of which were written on the questionnaire; frequency of use of each product; total product consumption; length of product use; the reasons for using NVNM supplements, revised from previous studies \[[@B24],[@B25]\]; and general demographic information such as age, sex, cigarette smoking, body height and weight, affiliation to which school of the university, class, marital status, parents\' education level, and parents\' total income level. ### The second section of questionnaire This section included the items related to nutritional beliefs about NVNM supplements. Subjects\' views about NVNM supplement use and the health benefits associated with it were used to assess the motivational factors related to supplement use in this population. Items were based on those of previous studies \[[@B24]-[@B28]\], and also contained the most frequently consumed food items in the Turkish population. Twenty two opinions were expressed, consisting of such items as \'vitamins and minerals can provide energy\', \'maintain health\', increase longevity\', \'reduce stress\' \'prevent colds\', \'improve mental function\', \'reduce the risk of chronic diseases\', or \'aid in recovery from fatigue\', all of which were assessed to be commonly-held assumptions connected with use of the supplements. For this section, a five-point Likert scale was used, and responses were scored from one (disagree very strongly) to five (agree very strongly). Thus, the higher score showed a stronger belief on the students\' part that NVNM supplements could provide specific health benefits. Applying it to a sample of 59 randomly selected students from same schools assessed the reproductivity of the questionnaire. The students who participated in the test/retest procedure were not included in the final analyses. Statistical analysis -------------------- Frequency, mean, and standard deviation (SD) were calculated, and the ttest compared scores on nutritional beliefs between users and non-users, or women and men. The chi-square test (x^2^) examined the relation between general characteristics and NVNM supplements use. The percentages of the reasons cited for using NVNM supplements arranged according to sex, were calculated by first extracting out of the total figure male and female students\' numbers for NVNM supplements usage, and the significance between men and women was evaluated using cross tables. The reliability of the internal consistency of data concerning nutritional beliefs collected from the pretest was calculated by Cronbach\'s alpha and was found to be 0.96, indicating a high internal consistency across nutritional-belief statements. Consequently, belief statement use was used for the final survey. Statistical analyses were performed using SPSS for Windows (Version 10.0, SPSS Inc., Chicago, Illinois). The standard used for statistical significance was p ≤ 0.05. The data were analyzed using descriptive statistics including means, standard deviations, and frequency distributions. Results ======= Of the participants, 909 (48.6%) were men and 962 (51.4%) women. The average age of the participants was 19.9 years (range = 17 to 28 years). Nearly half of the students (49.9%) were between the ages of 19--20. Most (74.8%) were of normal weight. More than 50% of the students (53.9%) were from the schools of engineering and architecture, and science and literature. Most students (72.4%) were in their freshman or sophomore years. Almost 100% (97.9%) were single or divorced. The proportion of students whose mothers had an education level of secondary school and lower was 61.0%, with the figure of 35.0% reported for students\' fathers. Most students\' level of family income (47.0%) was low. The majority of the students reported living in villages or cities (85.1%), and 26.0% were current smokers. The percentage of those participating in sport was 15.4%, and 18.0% reported exercising more than 3 times per month. Most (74.3%) rated their health as \'good/excellent\'. There were significant differences between men and women according to all the descriptive information. The more detailed general characteristics of users and non-users of NVNM supplements are shown in Table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### General characteristics of users and non-users of nonvitamin/nonmineral supplements ::: --------------------------------------------------------------------------------------------------- Non-users\ Users\ Total\ n(%)\ n(%)\ n(%)\ 1563(83.5) 308(16.5) 1871(100.0) --------------------------------------- ------------------------- --------------- ----------------- Sex x^2^= 5.594; p = 0.018 Men 758(83.4) 151(16.6) 909(48.6) Women 805(83.7) 157(16.3) 962(51.4) Ages x^2^= 35.732; p = 0.000 ≤ 18 358(86.7) 55(13.3) 413(22.1) 19--20 789(84.6) 144(15.4) 933(49.9) 21--22 292(85.1) 51(14.9) 343(18.3) ≥ 23 124(68.1) 58(31.9) 182(9.7) BMI values x^2^= 73.392; p = 0.000 Underweight (\<18.5) 199(67.7) 95(32.3) 294(15.7) Normal weight (≥ 18.5 -- ≤ 24.99) 1212(86.6) 187(13.4) 1399(74.8) Overweight (≥ 25 -- ≤ 29.9) 121(90.3) 13(9.7) 134(7.2) Obese (≥ 30 -- ≤ 39.9) 31(70.5) 13(29.5) 44(2.4) Type of school x^2^= 23.573; p = 0.000 Medicine 164(79.2) 43(20.8) 207(11.1) Engineering and architecture 536(88.4) 70(11.6) 606(32.4) Science and literature 332(82.4) 71(17.6) 403(21.5) Economics 204(79.7) 52(16.9) 256(13.7) Education 144(77.0) 43(23.0) 187(10.0) College of health services. 183(86.3) 29(13.7) 212(11.3) Year in school ns Freshman 570(85.6) 96(14.4) 666(35.6) Sophomore 556(80.8) 132(19.2) 688(36.8) Junior 168(86.2) 27(13.8) 195(10.4) Senior 269(83.5) 53(16.5) 322(17.2) Marital status ns Single 1520(83.7) 296(16.3) 1816(97.1) Married 33(82.5) 7(17.5) 40(2.1) Divorced/widow(er)/separated 10(66.7) 5(33.3) 15(0.8) Mother\'s education level x^2^= 11.940; p = 0.008 Primary school and lower 848(84.8) 152(15.2) 1000(53.4) Secondary school 106(74.1) 37(25.9) 143(7.6) High school 395(84.9) 70(15.1) 465(24.9) College or university 214(81.4) 49(18.6) 263(14.1) Father\'s education level x^2^= 7.807; p = 0.050 Primary school and lower 416(87.6) 59(12.4) 475(25.4) Secondary school 149(83.2) 30(16.8) 179(9.6) High school 537(81.7) 120(18.3) 657(35.1) College or university 461(82.3) 99(17.7) 560(29.9) Family income x^2^= 22.971; p = 0.000 Low 763(86.7) 117(13.3) 880(47.0) Moderate 566(83.4) 113(16.6) 679(36.3) High 234(75.0) 78(25.3) 312(16.7) Most permanent place of residence x^2^= 13.642; p = 0.001 Village/town 141(79.7) 36(20.3) 177(9.5) City 1207(85.3) 208(14.7) 1415(75.6) Metropolis 215(77.1) 64(22.9) 279(14.9) Smoking status x^2^= 15.435; p = 0.000 Nonsmoker 1093(85.7) 183(14.3) 1276(68.2) Current smoker 390(80.1) 97(19.9) 487(26.0) Former smoker 80(74.1) 26(25.9) 108(5.8) Participation in sport x^2^= 9.193; p = 0.010 No 724(82.5) 154(17.5) 878(46.9) Former 580(82.4) 124(17.6) 704(37.6) Yes 259(89.6) 30(10.4) 289(15.4) Exercise x^2^= 12.205; p = 0.002 Never 440(83.5) 87(16.5) 527(28.2) Fewer than 3 times per month 862(85.6) 145(14.4) 1007(53.8) More than 3 times per month 261(77.4) 76(22.6) 337(18.0) Self-reported health ns Fair/poor 400(83.2) 81(16.8) 481(25.7) Good-excellent 1163(83.7) 227(16.3) 1390(74.3) --------------------------------------------------------------------------------------------------- ::: The duration of supplement use was cited as \'between 1 and 6 months\' by 65.3%, followed by \'between 6 and 12 months\' and \'more than one year\' (23.1% and 11.7%, respectively). More than 50% of supplement users (53.6%) were taking just one supplement at the time of this survey, followed by those using 2 supplements (38.6%) and 3 or more (7.8%). The average number of supplement taken was 1.6 (standard deviation = 0.6). Most students (68.2%) rated the frequency of supplement use as \'once/twice a day\' (37.0%) or \'once every other day\' (31.2%), followed by \'once/twice a week\' (20.5%) and \'once/twice a month\' (11.4%). No difference was revealed between men and women by supplement use (Unshown data). Table [2](#T2){ref-type="table"} presents reasons cited for using NVNM supplements. The four most commonly given reasons for use were \'improvement of energy and vitality\' (78.6%), \'promotion of weight loss\' (71.1%), followed by \'enhancement of athletic performance\' (64.3%) and \'retardation the onset of aging\' (63.3%). Women were more likely to be using supplements to promote weight loss, burn-up fat, prevent colds, improve memory, and relieve stress; whereas men were more likely to use supplements to enhance athletic performance, retard the onset of aging, build muscle, and improve sexual function. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Reasons cited for using nonvitamin/nonmineral supplements by sex ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------- Men\ Women\ Total\ *x^2^; Pvalues* n(%)\ n(%)\ n(%)\ 151(49.0) 157(51.0) 308(100.0) ---------------------------------------------------------------------------------------- --------------- --------------- ---------------- --------------------- Improve energy and vitality 128(49.6) 114(50.4) 242(78.6) ns Promote weight loss 92(42.1) 127(57.9) 219(71.1) 5.59; 0.018 Enhance athletic performance 117(59.1) 81(40.9) 198(64.3) 6.54; 0.011 Retard aging 124(63.6) 71(46.4) 195(63.3) 14.41; 0.000 Burn-up fat 55(35.1) 102(64.9) 157(50.9) 14.07; 0.000 Prevent(threat) colds(sore throat) 41(35.4) 75(64.6) 116(37.6) 9.97; 0.002 Promote skin(hair) health 41(41.8) 57(58.2) 98(31.8) ns Build muscle 71(74.3) 24(25.7) 95(30.8) 23.25; 0.000 Supplement inadequate diet due to nutrition deficiency 48(53.9) 41(46.1) 89(28.9) ns Improve memory 33(39.3) 51(60.7) 84(27.3) 3.86; 0.05 Relieve stress (improve mood) 29(37.7) 48(62.3) 77(25.0) 4.67; 0.03 Improve sexual function 55(79.7) 14(20.3) 69(22.4) 24.36; 0.000 Enhance sleep 22(52.4) 20(47.6) 42(13.6) ns Reduce dangers of cigarette smoking 18(52.9) 16(47.1) 34(11.1) ns Improve circulation 18(58.1) 13(41.9) 31(10.1) ns Gain weight 18(62.1) 11(37.9) 29 (9.4) ns Prevent illnesses such as cancer, osteoporosis, menopause, hypertension, kidney stones 9(42.9) 12(57.1) 21 (6.8) ns --------------------------------------------------------------------------------------------------------------------------------------------------------------- Note: Subjects could list more than one reason for nonvitamin, nonmineral supplement usage ::: In general, when the scores obtained from all the nutritional beliefs between men and women were taken into consideration, women agreed on the effects obtained from taking NVNM supplements more strongly than men in the categories: \'important in maintaining health\', \'stress reduction\', and \'enhancement sleep\', whereas men agreed more strongly on \'improvement sexual function\' than women (Unshown data). Twenty-six of the 308 NVNM users (26/308, 8.4%) reported having had an adverse reaction to an NVNM supplement. Of 26, the two most adverse effects were nausea and vomiting (61.5%, 30.7%, respectively), followed by gastrointestinal disturbances (15.4%), flashing (11.5%) and liver insufficiency (7.7%), Subjects could list more than one complication cited for NVNM usage (Unshown data). The most frequent used sources for obtaining information about supplements were television (76.3%), magazines/newspapers (41.5%) and internet websites (37.3%); followed by family (20.1%), friends (19.8%), school and teachers (16.5%), books (13.3), radio (5.8%), health food stores (0.5%) and agricultural engineers (0.3%). The rates of those citing dietetic professionals, and doctor and nurse as their sources were only 4.2% and 6.8%, respectively (Unshown data). Of 1871, 308 students (16.5%) reported using NVNM supplements during the past year, 23.0% students indicated that they were considering NVNM use in the near future. About 10% (9.7%) reported past usage. The data revealed no difference between men and women according to supplement usage. There were also no differences between men and women in the categories \'use of NVNM during the past year\', \'considering NVNM supplement use in the future\', and \'past usage of NVNM supplements\' (p \> 0.05 for each one) (Unshown data). Of the 1690 students reporting no use of NVNM supplements in the past, 44.3% indicated that they had not considered using supplements, 34.7% reported that they had only thought about using supplements, and 11.6% reported that they had been unaware of the presence of supplements. There was a 9.4% non-response rate (Unshown data). Table [3](#T3){ref-type="table"} summarizes information regarding the prevalence of NVNM supplement use. The three most commonly used supplements were reported as echinacea, ginseng, and gingko biloba (38.6%, 36.4%, and 32.8%, respectively). Since all of the 73 most commonly purchased supplements in Turkey and the world were listed on the questionnaire, the students reported using no supplements other than those listed. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Prevalence of use of nonvitamin, nonmineral supplements ::: ---------------------------------------------------- Total\ n(%)\ 308(100.0) ----------------------------------- ---------------- Echinacea 119(38.6) Ginseng 112(36.4) Gingko biloba 101(32.8) Protein powder and or Amino acids 89(28.9) Fish oil 84(27.3) Bee pollen 78(25.3) Garlic 64(20.8) Green Tea 61(19.8) Bee pollen 57(18.5) St. John\'s wort 51(16.5) Ginger 47(15.2) Creatine 45(14.6) Aloe 27 (8.7) Lecithin 25 (8.1) Flax 21 (6.8) Chestnut seed 19 (6.2) Valerian 16 (5.2) Burdock root 15 (4.9) Coenzyme Q 15 (4.9) Goldenseal 12 (3.9) Cranberry 11 (3.6) Guarana 9 (2.9) Chamomile tea 8 (2.6) Sam-E 7 (2.3) Kava kava 7 (2.3) Melatonin 6 (1.9) Dong Quai 4 (1.3) Angelica 3 (0.9) Cayenne 2 (0.6) Borage 2 (0.6) Astragalus 1 (0.3) Evening primrose 1 (0.3) ---------------------------------------------------- The percentages total to more than 100% because some respondents reported taking more than one NVNM supplement ::: Table [4](#T4){ref-type="table"} shows that NVNM supplement use correlated to nutritional beliefs. Of all the beliefs, the links between \'supplement use and recovery from fatigue\' and \'supplement use and prevention/threat colds/sore throat\' were the ones with which users and non-users most agreed. The lower degree of perception of any \'relationship between supplement use and improved mental health\' was similar between users and non-users. In general, when all the nutritional beliefs between users and non-users were taken into consideration, users agreed on the effects obtained from taking NVNM supplements more strongly than non-users except for the categories of \'prevention/threat of colds/sore throat\', \'prevention of mental illnesses\', \'burning-up fat accumulating in the body\', \'promotion of more weight loss\', and \'reduction of the harmful effects of cigarette smoking\'. Of the 22 beliefs, the score from only one item, \'burning-up fat accumulating in the body\' was found to be lower in users than in the non-users. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Nutritional beliefs about nonvitamin/nonmineral supplements of supplement users and non-users\* ::: ----------------------------------------------------------------------------------------------------------- Belief statement Non-users†\ Users†\ **Pvalues‡ n(%)\ n(%)\ 1563(83.5) 308(16.5)** -------------------------------------------------------- ----------------- --------------- ---------------- Improve energy and vitality 2.8 ± 0.8 3.2 ± 0.8 0.000 Important in maintaining health 2.9 ± 0.8 3.3 ± 0.8 0.000 Provide living longer 2.7 ± 0.8 2.8 ± 0.7 0.000 Help recovery from fatigue 3.0 ± 0.8 3.3 ± 0.7 0.000 Reduce stress 2.8 ± 0.7 3.0 ± 0.8 0.012 Prevent/threat colds/sore throat 3.1 ± 0.8 3.2 ± 0.7 ns Prevent mental illnesses 2.7 ± 0.8 2.8 ± 0.8 ns Improve mental health 2.6 ± 0.8 2.7 ± 0.8 0.016 Reduces risk of chronic illnesses 2.8 ± 0.7 3.0 ± 0.8 0.000 Retard aging 2.7 ± 0.9 2.9 ± 0.8 0.000 Burn fat accumulating in the body 2.9 ± 0.8 2.8 ± 0.8 ns Promote weight loss more 2.8 ± 0.7 2.9 ± 0.8 ns Build muscle more 2.9 ± 0.8 3.1 ± 0.7 0.000 Enhance athletic performance 2.9 ± 0.8 3.1 ± 0.8 0.001 Enhance sleep 2.8 ± 0.8 3.0 ± 0.8 0.000 Improve sexual function 2.8 ± 0.8 3.0 ± 0.8 0.000 Improve circulation 2.9 ± 0.8 3.2 ± 0.7 0.000 Supplement inadequate diet due to nutrition deficiency 2.9 ± 0.8 3.2 ± 0.8 0.000 Improve memory 2.8 ± 0.8 3.0 ± 0.8 0.001 Promote skin/hair health 2.9 ± 0.8 3.1 ± 0.8 0.000 Gain weight 3.0 ± 0.8 3.1 ± 0.8 0.009 Reduce dangers of cigarette smoking 2.7 ± 0.9 2.8 ± 0.9 ns ----------------------------------------------------------------------------------------------------------- \Values are shown as mean ± standard deviation † A five-point Likert scale was used; disagree very strongly (one point) to agree very strongly (five points) ‡ ttest ::: Discussion ========== To our knowledge, this is the first survey on NVNM supplements use conducted in Turkey. This survey allowed the assessment of the prevalence of NVNM supplement use in a sample of university students from different departments. Our results indicate that about one in six students (16.5%) consumed one or more NVNM supplement during the past year. While our finding is compatible with some other study results, it is not with others: Radimer et al. \[[@B3]\] found that between 3.3 and 12% of respondents to their survey reported using NVNM supplements in 1994 and 1995, respectively. In a survey on a 1000 non-patient student population by Perkin et al. \[[@B25]\], a NVNM supplement usage rate of 26.3% was observed. Eisenberg et al. \[[@B29]\] reported this proportion as 12.1%. Furthermore, in a survey relevant to NVNM supplement use over a 12-month period by adult members of a large health maintenance organization, an estimated 32.7% of adults used at least one NVNM supplement \[[@B30]\]. Newberry et al.\[[@B31]\] reported that out of 272 college students canvassed, 48.5% had taken an NVNM supplement during the previous year and in a survey in the United States, it was found that 14.5% of the adults reported having used a NVNM supplement during the previous year \[[@B32]\]. The aforementioned studies show that the rate of NVNM supplement usage ranges from 3.3 to 48.5%. One explanation for these differences in reported usage rates could be the variation in how questions were asked regarding time frame and the types of supplements used. A further possibility could be relevant to individuals\' sociodemographic characteristics, such as race or ethnicity, age, education, income and lifestyle variables such as drinking, smoking, and exercising \[[@B15],[@B16]\]. As in other populations, echinacea, ginseng, gingko biloba, and protein powder and or amino acids were reported by this population to be the most popular supplements \[[@B3],[@B33]-[@B35]\]. Wingate \[[@B33]\] found that in 1997 the three most reportedly used supplements were garlic, ginseng, and ginkgo biloba. Similarly, the three most commonly consumed NVNM supplements reported in the Slone Survey \[[@B36]\], conducted from 1998 to 1999, were ginseng, ginkgo biloba extract, and allium sativum. This high use of supplements in Turkey and the rest of the world may be due to the fact that those products are prevalent in the market. In this survey, most students reported that they were healthy. This was reflected in the rationale offered by the students for supplement usage, such as increasing energy and enhancing athletic performance. In this survey, a positive relationship was seen to exist between exercising and NVNM supplement usage. Both behaviors may be perceived as congruent with disease prevention. NHANES III \[[@B3]\] also found a positive relationship between exercise and NVNM usage. The current survey found that former smokers used NVNM supplements most, which is in line with the NHANES III survey \[[@B3]\]. One explanation for this is that usage could be viewed as offsetting a negative behavior \[[@B25]\]. In this survey, 9.7% of the students reported past usage, 16.5% currently usage and 23.0% potential for use in the future. That these findings show increasing trends in the use of supplements with a corresponding increase in time may be due to media reports or advertisements in local newspapers about provision of energy and vitality, an increase in athletic performance, the retarding of aging, and the reduction of the risk of chronic diseases such as hypertension, colon cancer, breast cancer, kidney stones, and osteoporosis \[[@B37]-[@B39]\]. A further explanation may be that the use of NVNM supplements increased substantially with the passage of the United States Dietary Supplement Health and Education Act of 1994 (DSHEA), which gave manufacturers greater freedom to market more products through purported benefits reported in package labeling and adverting \[[@B1]\]. Twenty-six of the 308 NVNM users (26/308, 8.4%) reported having experienced an adverse reaction to an NVNM supplement. This rather high rate is in line with results published by Newberry et al.\[[@B31]\], which reported a reaction rate of 6.9%. This may be due to the fact that the majority of the students did not seek nutrition information from reliable sources such as doctors, nurses or dietetic professionals. In this survey, only 11.0% of supplement users cited doctors, nurses, and dietetic professionals as sources of information. However, it is important that students get nutrition information from reliable sources and receive proper nutrition education. To put this another way, that 76% of the respondents obtained information about supplementation from television may indicate that television has enormous implications for educational and public policy initiatives, and that television should be used to get more health information about supplementation. The nutritional beliefs that subjects questioned had concerning supplements were an important factor for their use of supplements. Supplement users agreed more strongly than non-users on the health benefits of supplement use for the majority of items. Others reported similar findings \[[@B9],[@B26]\]. In this survey, NVNM supplement use was higher in men than in women, in line with Radimer et al.*\' survey \[[@B3]\]. However, these findings are not concordant with data reported from Malaysia (40), where the prevalence of supplement use was higher in women (14.9% vs. 9.7%, respectively). No apparent reasons were disclosed for such differences; it is possible that NVNM supplement use might be related to a desire to increase athletic performance in men \[[@B41]\], but further studies are needed to better assess this result. In this survey, nutritional knowledge average scores were low in both sexes (63.3 ± 12.4 for a maximum score of 110), probably because of the absence of nutritional education in primary, secondary, high schools or university in our country. NVNM supplement use by the students was different according to sex, age, Body Mass Index (BMI) values, school type, mother and fathers\' education levels, family income, most permanent place of residence, smoking status, participating in sport, and exercise. This finding shows that the rationale for supplements use changes according to subjects\' general characteristics. Of further interest is the trend for more female students in the current sample to indicate use of significantly more supplements than males. Parallel to our survey, this finding highlights the fact that female students are more likely to be dieting to lose weight, whereas male students are more likely to be controlling their diet to gain weight and build muscle, which is compatible with some other studies \[[@B42],[@B43]\]. We are well aware of the limitations of the present survey. Firstly, it was performed in a single institution, and therefore the sample may not be representative of all Turkish students. Another limitation on the data was the use of the self-reporting system employed in this survey. This may have given rise to bias. Conclusion ========== This descriptive analysis helps to provide a clearer picture of NVNM supplement use. Not only does it indicate that students are indeed using or considering using supplements as a viable means to achieve their short-term fitness and or appearance goals, but also which products are most likely to be used, and indeed which individuals are most in need of education and information concerning the use and abuse of certain dietary aids. Further research, therefore, needs to focus on how NVNM nutritional supplements use information can be applied to specific user groups. Monitoring the use of NVNM supplements within the community or university students should be of interest because these substances may affect health or interact with other medications \[[@B44]\]. The respondents in this survey reported 85 cases of interactions between St John\'s wort and prescription medications. Dietetic professionals and physicians need to consider evaluating NVNM supplement use when assessing overall nutrient and medication use by clients or students. In summary, our results in this sample of Turkish students indicate that the prevalence of NVNM supplement use is still relatively modest, and neither associated with a healthier lifestyle nor related to a better nutritional knowledge. The need for further research on the relationships between dietary supplements and health has also been highlighted in the 1997 report by the United States Commission on Dietary Supplement Labels \[[@B45]\]. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AU conceived of the study, performed its design and coordination, sequence alignment, collected the data, performed the statistical analyses and drafted the manuscript, SN and SO collected the data and entered the data to SPSS packet program, All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/47/prepub> Acknowledgements ================ We thank all the students from the University of Osmangazi, Eskisehir, who participated in this survey, and the lecturers of the schools, who helped in the collection of data.	PubMed Central	2024-06-05T03:55:58.409569	2005-5-16	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173114/", "journal": "BMC Public Health. 2005 May 16; 5:47", "authors": [ { "first": "Unal", "last": "Ayranci" }, { "first": "Nazan", "last": "Son" }, { "first": "Osman", "last": "Son" } ] }
PMC1173115	Background ========== Tobacco is one of the leading causes of the global burden of disease, \[[@B1]\] demanding effective intervention strategies \[[@B2]\]. Smoking cessation treatment is generally considered to be among the most cost-effective life saving interventions in the health system \[[@B3]\] and telephone helplines for smoking cessation (quitlines) have proved to be both effective and cost-effective \[[@B4]-[@B6]\]. In evaluation of smoking cessation programs including surveys and clinical trials the tradition has been to treat non-responders as smokers \[[@B7]-[@B12]\]. However, bias due to failure to respond to follow-up is seldom assessed \[[@B13]\] for the different types of interventions and presently no reliable data exists supporting the pre-judgement that non-responders are more likely than responders to be present smokers. The existing empirical data on difference in smoking between responders and non-responders is often based on public health surveys \[[@B10]\]. However, the possibility exists that non-responders in smoking cessation programs using a telephone quitline may differ from non-responders in general health surveys. Thus, to categorically classify non-responders as smokers may underestimate the true effectiveness of smoking cessation treatment \[[@B9]\]. To our knowledge no previous studies investigating telephone treatment for smoking cessation have assessed the effect of non-response on reported abstinence rates. In the present study we compare abstinence rates and population characteristics in responders and non-responders included in an evaluation of the Swedish quitline and assess reasons for not responding to the original follow-up questionnaire. Methods ======= Study population ---------------- The original study base included 1606 individuals in different stages of change who 12 months earlier had contacted the Swedish quitline for help with smoking cessation. All had accepted to receive a follow-up questionnaire 12 months after first contact with the quitline and participated in the original follow-up. First contacts were made from February 2000 to November 2001 and the follow-up questionnaires were mailed out February 2001 to November 2002. After two reminders answers were collected from 1131 individuals (70%) leaving 475 non-responders. A sample of 84 (18 % of the total non-response) was included in the present assessment of non-responders (Fig [1](#F1){ref-type="fig"}). In the beginning of November 2002 we identified all 84 patients who had not responded to the follow-up questionnaire during the previous four months. The relative narrow time window was to minimise recall bias. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Graphic description of subjects in the study base. ::: ![](1471-2458-5-52-1) ::: Data collection --------------- The non-responders were contacted by a telephone interview during the autumn 2002. Each interview took approximately five minutes to perform and was conducted by only one person, to minimise interviewer bias. Attempts were made to phone first three times during daytime and then twice during evenings/weekends. Since all subjects had completed a questionnaire earlier (following their first call to the quitline) all background information, such as gender, age, language, tobacco habits and telephone number, was already available. In the original follow-up questionnaire abstinence was defined as \"not a single puff of smoke during the previous seven days\", which is a commonly used definition of point prevalence abstinence in larger surveys \[[@B12],[@B14],[@B15]\]. Two questions were used to assess present smoking behaviour: Have you smoked at least one single puff during the previous seven days? Has your smoking behaviour changed since you received the follow-up questionnaire? An additional retrospective question was asked to assess smoking behaviour at the time of the original 12 month follow-up: Have you smoked (one puff or more) in the week previous to receiving the follow-up questionnaire? The date when the original questionnaire was sent out was specified for all subjects. The interview was completed with a question on why he/she had not returned the original questionnaire. Ethical approval was obtained by Karolinska Institutet (Dnr 00-367). Statistical methods and presentation of data -------------------------------------------- We compared the prevalence of abstinence in the original study population (responders) with the 46 non-responders participating in the present study. In the table we start by comparing abstinence in the two groups at the time of the original follow-up and then at the time of the telephone interview. Reasons stated by the subjects for not returning the original follow-up questionnaire are presented in a table as proportion of those 46 participating in the telephone interview. Reasons identified by the interviewer for not participating in the telephone interview are presented in the text as proportion of those 38 not participating. Non-responders (at 12 month follow-up) participating in the telephone interview were compared with responders in the original study population in terms of population characteristics (sex, age, smoke-free at first call, and using nicotine at first call). These data are presented in table form. In the table we also present background data on all 84 non-responders selected for the study and the 38 non-responders not participating in the telephone interview. In the table we dichotomised age as a two category variable (≤40, ≥41), nicotine use at baseline as \"yes\" or \"no\" and stages of change as \"still smoking at the time of first contact\" and \"smoke free at the time of first contact\". For statistical comparison between groups we calculated Fisher\'s exact test, odds ratios (OR) and 95% confidence intervals (CI) on proportions and OR. In table [1](#T1){ref-type="table"} we used one- sided CI on the proportions since our main focus was on the lower limits. Statistical analyses were carried out using the Statistical Package for the Social Sciences (SPSS 11.5). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Percentage and proportions of abstinence in the original study population (responders) and the present study population (non-responders) at 12 months follow-up, and at the time of the telephone interview. ::: Original study population The present study population participating in the telephone interview ----------- ------------------------------- --------------------------------------------------------------------------- ------------------------------------------------- Abstinent at 12 months Abstinent at 12 months Abstinent at the time of the tel. Interview % (n/N) One-sided 95%CI % (n/N) One-sided 95%CI % (n/N) One-sided 95%CI Men 30 (69/226) 63 (5/8) 38 (3/8) ≥25 ≥29 ≥11 Women 31(285/905) 34 (13/38) 26 (10/38) ≥29 ≥22 ≥15 Total 31 (354/1131) 39 (18/46) 28 (13/46) ≥29 ≥27 ≥18 ::: Results ======= Of the 84 subjects not responding to the original questionnaire at 12 month follow-up (non-responders) recruited for the study base 55% (46/84) participated. Of the 38 subjects not participating 61% (23/38) could not be reached, 29% (11/38) declined, and 10% (4/38) were either sick or dead (not in table). Abstinence ---------- Of the 46 subjects participating in the present study 39% reported to have been smoke-free at the time when they received the original follow-up questionnaire (abstinent at 12 months) compared with 31% of responders in the original study population (Table [1](#T1){ref-type="table"}). No significant difference in abstinence was noted between the present study population and the original study population (Table [1](#T1){ref-type="table"}). However, men in the present study population were somewhat more likely to report being abstinent at 12 months compared with the men in the original study population (Table [1](#T1){ref-type="table"}). The reported higher level of 12 months abstinence in men did not persist at the time of the telephone interview (Table [1](#T1){ref-type="table"}). One woman did not remember whether or not she was abstinent at twelve months and was treated as a smoker. Reasons for not returning the postal questionnaire -------------------------------------------------- The most common reason stated for not having returned the original questionnaire was claiming that they had returned it (Table [2](#T2){ref-type="table"}). Approximately one in ten stated that they had believed that abstinence was a prerequisite for answering and therefore had not returned the questionnaire since they were smoking at the time (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Stated reasons among 46 participants for not returning the postal questionnaire ::: % (n) ----------------------------------------------------- ------- --------- Claimed to have returned the questionnaire 35 (16) Had not received the questionnaire/moved 20 (9) Thought abstinence was a prerequisite for answering 13 (6) Do not know 13 (6) Forgot to return it 9 (4) Uninterested 6 (3) Had lost the questionnaire 4 (2) Total 100 (46) ::: Comparison with responders -------------------------- The non-responders comprising the study base in the present study were somewhat younger than the responders in the original study population (Table [3](#T3){ref-type="table"}). The mean ages being 47 for the responders and 42 for the non-responders (data not in table). Men and women were equally represented both among responders and non-responders. Non-responders tended to a higher degree to have been smoke-free when they first called the quitline (Table [3](#T3){ref-type="table"}). They were also significantly more likely to have been totally nicotine free at first call compared with the responders (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Population characteristics of responding and non-responding subjects. Comparing 46 non-responders participating in the non-response analysis with the 1131 responders in the original study population. ::: Total Total sample of non-responders Non-responders not participating in the telephone interview Non-responders participating in the telephone interview Responders in the original study Comparison ¶ ------------------------------------- ------------------------------------ ----------------------------------------------------------------- ------------------------------------------------------------- -------------------------------------- ------------------ ------------ Sex Male(Ref) 20 (17) 24 (9) 17 (8) 20 (226) Female 80 (67) 76 (29) 83 (38) 80 (905) 1.2 0.5 -- 2.6 Age distribution: ≥ 41(Ref) 58 (49) 61 (23) 57 (26) 67 (755) ≤ 40 42 (35) 39 (15) 43 (20) 33 (376) 1.5 0.9 -- 2.8 Smoke-free at first call: No(Ref) 73 (61) 76 (29) 70 (32) 77 (875) Yes 27 (23) 24 (9) 30 (14) 23 (256) 1.5 0.8 -- 2.8 *Using nicotine\ at first call: Yes*(Ref) 82 (69) 87 (33) 78 (36) 89 (1010) No 18 (15) 13 (5) 22 (10) 11 (121) 2.3 1.1 -- 4.8 \Total consumption of nicotine, including smoked and smoke-free tobacco and NRT. ¶Comparing non-responders participating in the telephone interview to responders ::: Discussion ========== Our results indicate that non-responders in follow-ups of large cohorts of smokers trying to quit with the aid of telephone quitlines, should not by definition be considered as treatment failures. If anything, the non-responders in the present study reported higher abstinence rates at the time when they were supposed to return the original follow-up questionnaire. The observed trend of overrepresentation of non-smokers among non-responders may be explained by the fact that they were more likely to be nicotine free at first call to the quitline. In the original follow-up we presented data separately for different stages of change. Those who were in the contemplation stage at first call reported 12-month abstinence in 19% of the cases compared with 22% for those in preparation and 53% for people who were in the action stage at first call \[[@B14]\]. Thus, being nicotine free at first call is a significant predictor for abstinence at 12 months. Our results indicate that classifying non-responders as smokers may underestimate the true treatment effect of quitlines in line with a recent Dutch study \[[@B17]\]. There is a tendency to view non-responders as a homogenous group with common characteristics but studies have not confirmed this to be the case \[[@B16]\]. A previous Swedish study comparing prevalence of smokers amongst non-responders to a general health survey with responders, showed a tendency for an overrepresentation of smokers amongst non-responders, especially in low income groups \[[@B10]\]. Contrary, a study from Holland did not find such differences \[[@B18]\]. However, caution is needed when comparing these studies to the present study since it is possible that non-responders in general health surveys may differ from non-responders in studies assessing abstinence rates after smoking cessation treatment. Based on a response rate of approximately 70% in the original study, our results suggest that non-responders in the assessment of quitlines are probably not more likely than responders to be smokers at the time of follow-up. However, the possibility exists that non-responders in studies with lower response rate may differ from the present study population and our results may only apply for studies with a similar or higher response rate. The present results are in line with a study based on a one-year follow-up of participants from a national \"Quit and Win\" contest where bias in smoking prevalence because of non-response was studied \[[@B19]\]. Approximately one in three (29%) of those who were reached by telephone refused to participate in the interviews. There is a high probability that those individuals may be present smokers. Methodological considerations include the relative low number of subjects compromising statistical power. The data indicate that non-responders are not more likely than responders to be smokers. However, owing to small numbers we are not able to conclude that the non-responders may be more likely to be smoke-free as the data indicates. Another problem is the retrospective assessment of smoking behaviour at the time the non-responders were supposed to have turned in the questionnaire. Obviously recall bias may have affected the answers. A more conservative way to interpret the data is to compare the responders in the original study with non-responders smoking behaviour at the time of the telephone interview (Table [1](#T1){ref-type="table"}, columns one and three). This more conservative comparison does not change the main results that using methods with no visual contact between counsellor and patient, non-responders may well fail to respond due to other reasons than active smoking. As in all questionnaire surveys on smoking cessation the possibility of underreporting of smoking may exist e.g. due to the impact of social desirability (the desire to appear good). However, since this would most probably affect all subjects equally it is not a major concern in the present study. Also, a self-reported smoking status appears to be a reliable indicator of actual smoking-status \[[@B20]\] and the effect of social desirability in smoking cessation studies is probably less than previously suggested \[[@B21]\]. Further, in a recent study assessing if subjects who decline cotinine tests are lying about their smoking behaviour it was found that failure to comply may result from external factors such as demands or random factors such as being too busy at the time \[[@B22]\]. In the present study, thirteen percent stated that they thought abstinence was a prerequisite for returning the questionnaire (Table [2](#T2){ref-type="table"}). This is a pedagogical problem that needs to be taken seriously since this may be a potential source of bias although not a major problem in the present study. Conclusion ========== The present study, one of the first of its kind, indicates that routinely treating non-responders as smokers in smoking cessation research may underestimate the true effect of cessation treatment. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Tanja Tomson have made substantial contributions to conception and design, analysis and interpretation of data. She has been writing all drafts of this paper. Catrine Björnström have contributed in the planning, design and analysis of this paper and have given final approval of the version to be published. Hans Gilljam participated in the planning of the study design and assisted in the writing of the paper. He have given final approval of the version to be published Asgeir Helgason have made substantial contributions to conception and design, analysis and interpretation of data. He has been involved in drafting the paper and have given final approval of the version to be published. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/52/prepub> Acknowledgements ================ The authors would like to thank Mats Toftgård (Tobacco Prevention) for statistical advice. All staff members at the Swedish quitline are also acknowledged. The Swedish quitline is funded by the Swedish Government through, The National Institute of Public Health, and the Swedish Cancer Society.	PubMed Central	2024-06-05T03:55:58.414220	2005-5-23	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173115/", "journal": "BMC Public Health. 2005 May 23; 5:52", "authors": [ { "first": "Tanja", "last": "Tomson" }, { "first": "Catrine", "last": "Björnström" }, { "first": "Hans", "last": "Gilljam" }, { "first": "Asgeir", "last": "Helgason" } ] }
PMC1173116	Background ========== In the early evening of October 4^th^, 1992, an El Al Boeing 747-F cargo aircraft lost two of its engines just after take off from Schiphol Airport and crashed into two apartment buildings in the Bijlmermeer, a densely populated suburb of Amsterdam (the Netherlands) \[[@B1]\]. The air disaster killed 43 people, and destroyed 266 apartments \[[@B1],[@B2]\]. Fire-fighters and police officers were called to the scene to extinguish fires, to search and rescue people, to assist in the identification of human remains and personal belongings, to secure the surroundings and to clean-up the devastated area. Many of them were faced with bewildered residents and extensive destruction, and some witnessed dead or injured victims. Within a few days the wreckage of the aircraft was transported to a hangar at Schiphol Airport, where employees (i.e. \'hangar workers\') sorted and inspected the wreckage. In the extensive aftermath of the disaster, rumors and questions arose about the cause of the accident, the contents of the cargo, potential exposure to hazardous materials, and health consequences \[[@B2],[@B3]\]. Every now and then the media highlighted stories of individual victims, as well as uncertainties about potential exposures during the disaster \[[@B4]\]. One of the major topics concerned exposure to depleted uranium from the aircraft\'s balance weights, particularly because some of the depleted uranium has never been recovered from the rubble \[[@B1]\]. However, the authors of a retrospective risk analysis \"considered it improbable that the missing uranium had indeed led to the reported health complaints\" \[[@B5]\]. Nonetheless, it appeared that a growing number of exposed workers and affected residents reported health complaints, which some of them attributed to the disaster \[[@B6]\]. Public and political unrest thus waxed and waned in the aftermath of the disaster \[[@B2],[@B3]\]. Eventually, a parliamentary inquiry, that was held in 1998, recommended an epidemiological study on the health effects of the disaster \[[@B1]\]. About the same time, in 1998, the employers of professional fire-fighters and police officers in Amsterdam decided to start an independent assessment of the health status of professional workers involved in the disaster. The mayor of Amsterdam assigned their occupational health service, the KLM Health Services, to organize this assessment. The employer of the hangar workers at Schiphol Airport joined this initiative, as did government representatives of the affected inhabitants and volunteer workers. It was decided to offer a medical examination to all people involved in the air disaster, residents as well as assistance workers, and that an epidemiological study would be performed simultaneously by the Institute for Research in Extramural Medicine (EMGO Institute). In this paper we report on the design of the epidemiological study among professional assistance workers: the Epidemiological Study Air Disaster in Amsterdam (ESADA). Unfortunately, the epidemiological study among residents had to be cancelled, due to low response rates. The ESADA is the first epidemiological study that has ever been conducted after a major technological disaster in the Netherlands. The aim of this study is to assess the long-term psychological and physical health effects of occupational exposure to the air disaster in Amsterdam on professional assistance workers, i.e. fire-fighters, police officers and hangar workers. Based on the scientific literature on the health effects of disasters, the main hypotheses of the ESADA concern unexplained physical symptoms \[[@B7]-[@B12]\], and post-traumatic stress symptoms and associated psychological symptoms \[[@B13]-[@B15]\]. Due to the fact that the ESADA originated partly from societal concerns, we considered it necessary to also include some additional outcomes that will answer questions for some of the affected people, which, in turn, might help to reassure them. These societal questions relate to depleted uranium, Mycoplasma species and carnitine levels in plasma. The first of these questions stems from the concerns about the depleted uranium from the aircraft\'s balance weights, described above. The other two questions are primarily based on an alleged resemblance between the symptoms of some of the people affected by the air disaster in Amsterdam and the symptoms of patients with chronic fatigue syndrome (CFS) and Gulf War (I) Syndrome (GWS). Although some authors may have suggested a link between these syndromes and Mycoplasma species \[[@B16]-[@B21]\] or carnitine deficiency \[[@B22]-[@B26]\], others have rejected the existence of such links \[[@B27]-[@B29]\]. In this paper we describe and discuss the design of the ESADA. More details on the (organization of the) ESADA can be found on its website \[[@B30]\]. Methods / Design ================ Design ------ The ESADA is designed as a historical cohort study, in which the health status of the professional fire-fighters, police officers and hangar workers who were occupationally exposed to the air disaster in Amsterdam is compared with the health status of reference groups of workers with the same jobs and employers at the time of the disaster, but who were not occupationally exposed to this disaster. Study population ---------------- The ESADA study population consisted of professional fire-fighters, police officers and hangar workers. Eligible subjects had to (1) sign informed consent; (2) have sufficient mastery of the Dutch language to fill in the questionnaires; and (3) belong to one of the following three occupational groups: 1\) All professional fire-fighters who were, according to company records, employed in the Amsterdam fire department at the time of the disaster. Additional professional fire-fighters who started working in this fire department after the disaster were also invited to participate in the study, as almost the entire fire department had been exposed to the disaster. 2\) All police officers (i.e.constables, warrant officers, sergeants and their supervisors) who were, according to company records, employed in the Amsterdam-Amstelland regional police force on the date of the disaster (October 4^th^, 1992), and were still employed there on the 1^st^of January 2000. 3\) All the hangar workers registered as working for one of the departments involved in the transport, security and sorting of the wreckage on the date of the disaster (October 4^th^, 1992), and who reported to have been involved in these activities; as well as a random sample, matched with their colleagues for age, sex, department and job title, who were also registered as working for these departments on 30^th^November 1992, but who did not report to have been involved in any disaster-related activities. Procedures and data-collection ------------------------------ The study design was approved by the two independent Medical Ethics Committees of the medical facilities involved in this project: the VU University Medical Center (VUmc) and the \'Onze Lieve Vrouwe Gasthuis\' (OLVG) in Amsterdam. Potential participants were initially informed about the study via announcements in staff magazines, after which they were approached via personal letters, and eventually by telephone. All participants signed informed consent and participated voluntarily. Data were collected at the Prinsengracht out-patient clinic of the OLVG from January 2000 to March 2002, i.e. on average 8.5 years after the disaster. In addition, data on about half of the hangar workers were collected at Schiphol Airport for logistic reasons. Trained medical research assistants checked that the questionnaires had been completed, measured body height and weight, drew blood samples, and assisted with the collection of urine and saliva samples. A team of administrative employees carried out the data-entry of the questionnaires. Data of each participant were entered twice by two of these employees independently, after which inconsistencies were reviewed and any mistakes rectified. All remaining problems in the interpretation of data, such as dubious handwriting, were consistently resolved by one of the authors (AH, PS or AW). Blood, saliva and urine samples were dealt with according to standard procedures for collection, transportation, storage and laboratory analysis. Laboratory technicians could have been aware that the samples were from the ESADA, but they were blinded for exposure and health status. The laboratories were all certified according to accredited Dutch standards. Occupational exposure to the disaster ------------------------------------- All participants were asked to fill in a questionnaire on occupational exposure to the air disaster. This questionnaire addressed several specific disaster-related tasks, and also the total time spent on these tasks and the location in which they were performed (e.g. on or near the disaster site, in the hangar where the wreckage was temporarily placed, or elsewhere). They were also asked to describe any other disaster-related task(s) that they had performed. Answers to the latter question were categorized (by PS and AW). The questionnaire also covered disaster-related psychosocial events in a number of items on personal experiences during the disaster (e.g. \"were you in life-threatening danger?\", \"did you see the disaster scene during the first hours after the crash?\", and \"were any of your family members injured?\"). These personal records of occupational exposure to the disaster were used to define \'exposed\' workers, i.e. those who reported at least one disaster-related task, and \'non-exposed\' workers, i.e. those who did not report any disaster-related tasks. In addition to comparing exposed and non-exposed workers, we examined exposure-response relationships among exposed workers, in which level of exposure is characterized by the type of tasks and psychosocial events and the duration of exposure. As an additional dimension of level of exposure, we took into account the differences in potential psychotraumatic impact of exposure items, based on criterion A1 of the diagnostic criteria for Post Traumatic Stress Disorder (PTSD; American Psychiatric Association \[APA\]; Diagnostic and Statistical Manual of Mental Disorders-IV-Text Revision \[DSM-IV-TR, 2000\]) \[[@B31]\]. This criterion states that \"the person has experienced, witnessed, or been confronted with an event or events that involve actual or threatened death or serious injury, or a threat to the physical integrity of oneself or others\". Five experts on PTSD from different universities and psychiatric hospitals independently rated the likelihood of potentially psychotraumatic disaster-related tasks and events to meet this criterion on a 4-point Likert Scale ranging from 1 = \'very unlikely\' to 4 = \'very likely\'. Subsequently, we assumed that items with a mean item score of three or higher met the A1 criterion for PTSD (i.e. A1 tasks and events), as opposed to items with a lower mean score (i.e. non-A1 tasks and events). Table [1](#T1){ref-type="table"} lists the disaster-related tasks and the psychosocial events according to their potential psychotraumatic impact. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Disaster-related tasks and psychosocial events according to their potential psychotraumatic impact ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *A1\ (traumatic) Non-A1\* (non-traumatic) ------------------------- ------------------------------------------------------------------------------------------------------------------ ---------------------------------------------------------------------------------------------------- Tasks 1\. Identification or recovery of victims from the rubble/transport or search for human remains\ 1\. Fire-extinguishing\ 2. Rescue people 2. Clean up of destructed area\ 3. Transport of injured victims\ 4. Provide first aid/support injured victims or workers\ 5. Security tasks (surveillance, prevent burglary, keep disaster area free of bystanders)\ 6. Other tasks (e.g. traffic management)\ 7. Sort wreckage in hangar (at Schiphol Airport)\ 8. Other tasks in hangar in the presence of the wreckage\ 9. Transport of wreckage\ 10. Burning of contaminated soil remnants (from disaster site) Psychosocial events** 1\. Having been in life-threatening danger during disaster\ 1\. Saw the aircraft crash / saw or heard the aircraft when it crashed\ 2. Personal injuries due to disaster\ 2. Felt or heard the impact of the crash\ 3. Witnessed dead or injured victims\ 3. Saw the fire\ 4. Having been in or near one of the destroyed buildings at the time of the disaster\ 4. Saw the disaster site during the first hours after the crash/when the wreckage was still there\ 5. Immediate family members (partner, children) died / in life-threatening danger / injured due to the disaster\ 5. Other family members in life-threatening danger or injured due to the disaster\ 6. Other family members died due to the disaster 6. Friends or acquaintances died, injured or in life-threatening danger due to the disaster\ 7. Apartment of other family members, friends, or acquaintances damaged due to the disaster\ 8. Lived in the affected suburb of Amsterdam (Bijlmermeer) at the time of the disaster\ 9. Visited the hangar where the wreckage was kept ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- \A1 and non-A1 = items with a mean score of ≥ 3 or \<3, respectively, on a 4 point Likert Scale indicating the likelihood for an item to meet criterion A1 for post-traumatic stress disorder (from \'very unlikely\' \[=1\] to \'very likely\' \[=4\]) (see Methods). ::: Main health outcomes -------------------- ### Self-reported health measures • Post-traumatic stress symptoms: (a) The Dutch 22-item Self-Rating Inventory for PTSD (SRIP) \[[@B32]-[@B34]\] and, among exposed subjects only, (b) The 15-item Dutch version of the Impact of Event Scale (IES), which addressed post-traumatic stress symptoms with explicit reference to the air disaster in Amsterdam \[[@B35]-[@B37]\]. • General mental health: (a) The 90-item Symptom Checklist (SCL-90) \[[@B38],[@B39]\]; (b) The 20-item General Health Questionnaire (GHQ-12) \[[@B40]\]. • Fatigue and associated symptoms: The 20-item Checklist Individual Strength (CIS) \[[@B41],[@B42]\]. • Health-related quality of Life: The MOS 36-item Short-Form Health Survey (SF-36) \[[@B43],[@B44]\]. • Chronic conditions: One questionnaire assessed the current presence and history of the following chronic conditions, which are considered to have a significant impact on well-being: diabetes; stroke, brain hemorrhage or infarction; heart attack; other heart problems (such as heart failure, or angina pectoris); cancer; chronic osteoarthritis (wear) of the hip or knee joints; hypertension; asthma, chronic bronchitis or lung emphysema (Chronic Obstructive Pulmonary Disease \[COPD\]); serious or persistent intestinal disorders (longer than 3 months); chronic stomach disorders, stomach or duodenal ulcers; serious or persistent back complaints (including hernias); chronic inflammation of the joints (chronic rheumatism, rheumatoid arthritis). Workers with these chronic conditions were subsequently asked in what year the onset was, to determine whether this was before the disaster took place. • Physical symptoms: Multiple questionnaires were used to assess the current presence of various physical symptoms, such as a number of respiratory, musculoskeletal, and skin symptoms. • Attribution of current problems to the air disaster in Amsterdam and its aftermath. Another questionnaire assessed the extent to which exposed workers related any of their current physical, psychological or practical/financial problems to the air disaster and its aftermath. Those who attributed physical symptoms to the disaster and its aftermath were asked to specify these symptoms. ### Laboratory outcomes General laboratory tests \[[@B1]\]: • Hematological and blood chemical outcomes: hemoglobin, leukocyte count, differential count, platelet count and mean corpuscular volume (Sysmex SE 9000, TOA medical electronics Co. ltd); potassium (Roche Modular ISE900, Roche Diagnostics); creatinine, alkaline phosphatase, gamma-glutamyl transferase, alanine aminotransferase, creatine kinase and C-reactive protein (Roche Modular P800, Roche Diagnostics); ferritin and thyroid stimulating hormone (Centauer, Bayer Diagnostics); β2-microglobuline (IMx Abbott). • Autoantibodies: nuclear antigen antibodies, anti-double stranded DNA antibodies \[[@B46]\], Immunoglobulin (IgM) rheumatoid factor \[[@B47]\], antineutrophil cytoplasmic antibodies \[[@B48],[@B49]\], and cardiolipin antibodies \[[@B50],[@B51]\]. • Urine outcomes: creatinine (Hitachi 747, Roche Diagnostics GmbH, Mannheim, Germany); micro-albumin (Beckman Array 360 system); and β2-microglobuline (IMx Abbott); screening for protein, glucose, pH, blood and leukocytes (teststrip Boehringer Mannheim B.V.), followed by microscopic evaluation of the urinary sediment if indicated. • Saliva outcome: cortisol concentration (Wizard 1470, Perkin Elmer). Additional laboratory tests with respect to the societal questions: • Uranium 238: concentration in urine (Inductively Coupled Plasma-Mass Spectrometry \[ICP-MS\] analyser, Finnigan Mat Element) and, at concentrations above 50 ng/l or above 50 ng/g creatinine, also the ratio of uranium 235/238 isotopes \[[@B52]\]. • Total and free carnitine: concentration in blood plasma (Mira Plus, Roche Diagnostics) \[[@B53],[@B54]\]. • DNA of any Mycoplasma species: presence in peripheral blood mononuclear cells (DNA-isolation, Magna Pure, Roche Diagnostics; real time PCR, Taqman, Applied Biosystems); positive samples were subsequently evaluated for the presence of DNA of Mycoplasma fermentans \[[@B55],[@B56]\]. Self-reported socio-demographic characteristics ----------------------------------------------- • Age: at time of assessment in years. • Sex:male or female. • Ethnicity:categorized into those who considered themselves as European (i.e. Dutch, British, Dutch/Irish, Dutch/Chinese, Dutch/Indonesian, Portuguese, Spanish, Dutch/ Spanish and European), and others (e.g.Moroccan, Turkish, Surinam). • Level of education: highest level of education completed, categorized as: high (higher vocational education, university); medium (intermediate vocational education, higher general secondary education, or pre-university education); and low (no education, elementary school, lower vocational education, or lower general secondary education). • Current executive function: yes (i.e. supervising one or more people) or no. • Level of physical activity: the total number of hours spent each week on physical activities such as physical exercise, gardening and housekeeping, classified into high, medium and low according to the 33^rd^and 66^th^percentiles. • Alcohol consumption: Usual and exceptional consumption of alcoholic beverages, classified into: none; light-moderate; and (extremely) excessive, i.e. consumption of (a) six or more glasses on 9--20 days a month and on 3--4 days in the last week, (b) four or more glasses on at least 21 days a month and on at least 5 days in the last week, and/or (c) more than six glasses a day, on a weekly basis. • Cigarette-smoking:categorized as: never, former smoker, and current smoker. • Negative life events:*the number of reported negative life events, based on a questionnaire which specified 13 such events and also included two open-ended questions in which other events could be described. Subjects were asked to indicate whether any of these events happened to them before or after the disaster. Role of funding sources ----------------------- The study was funded by the Dutch Ministry of Health, Welfare and Sports; the City of Amsterdam; the Amsterdam-Amstelland regional police force; and KLM Royal Dutch Airlines. The funding sources had no role in the collection, analysis, or interpretation of the data, or in the decision to submit a manuscript for publication. Discussion ========== In recent years there has been increasing scientific and societal interest in the health consequences of man-made, technological disasters, i.e. a collective stressful experience with a sudden onset due to technological failure. Technological disasters have had psychiatric consequences \[[@B13]-[@B15],[@B57]\], such as PTSD, as well as medical consequences, in particular those of toxic exposures \[[@B58]-[@B61]\]. In addition to direct toxic health effects, the mere suspicion and fear of exposure to hazardous materials can also take its toll on the quality of physical, psychological and social well-being in the community \[[@B62]-[@B64]\]. Technological disasters strike unexpectedly and suddenly, which puts time-pressure on researchers to develop study protocols, gather exposure data, call in multidisciplinary experts, and obtain financial resources for immediate epidemiological research. Disaster researchers may also have to deal with complicated socio-political and legal aspects. In addition, they have to face a number of methodological problems. These difficulties include: (a) defining the entire potentially \'affected\' population and appropriate reference groups; (b) contacting potential participants, particularly in the case of evacuation and hospitalization; low response rates; usually without data on non-respondents \[[@B65]\]; (c) collecting exposure data immediately after the event, which is actually also needed for long-term epidemiological studies. Probably due to these difficulties, evidence from large-scale epidemiological studies that have been carried out after technological disasters is rather scarce \[[@B66],[@B67]\]. Furthermore, before-after comparisons are rare and only possible by chance in ongoing research projects, due to the unexpected nature of technological disasters \[[@B68]-[@B70]\]. Most of the studies that have been conducted so far have relied on \'convenience samples\', which were mainly composed of those who were directly affected, such as victims and residents; were based on non-epidemiological study designs; and used group-level or retrospective, self-reported exposure data, which can be affected by recall and reporting bias \[[@B71]-[@B74]\]. ESADA approach -------------- The purpose of the ESADA is to assess long-term health effects of occupational exposure to the air disaster in Amsterdam on professional assistance workers. In view of the above-mentioned difficulties in epidemiological research on disasters, the ESADA has some strong methodological points. With respect to the study population, we have been able to identify the complete cohort of exposed and non-exposed workers accurately, based on company records of employment at the time of the disaster. Another strong point of the ESADA is that we included reference groups of colleagues, who had the same jobs and employers, but who were not occupationally exposed to the disaster. Hence, we are able to draw group-level conclusions on associations between health status and occupational exposure to the disaster. With respect to exposure assessment, we were able to collect individual data on occupational exposure to the air disaster. Moreover, this consisted of multiple aspects of self-reported occupational exposure, including the duration and location of various disaster-related tasks and the experience of potentially stressful events during these tasks. Finally, we also included various assessments of long-term health, such as laboratory tests and self-reported symptoms and health-related quality of life, to obtain an integral evaluation of health status. Notwithstanding these strong methodological qualities, some limitations of the ESADA design should also be mentioned. Firstly, although company records of employment were available, we still had to resolve a few difficulties regarding the definition of the study population. For the fire-fighters, this was due to the fact that almost the entire fire department of Amsterdam had been exposed to the disaster. Therefore, in order to achieve an adequate reference group, we decided to also include fire-fighters who joined this fire department after the disaster took place. With respect to the police officers, we were unable to trace those who had left the Amsterdam-Amstelland regional police force in the years after the disaster, due to administrative difficulties. Hence, it was necessary to restrict this group to those who were still working for this police force in 2000. A second methodological issue concerns the self-report nature of occupational exposure status, and the average time-lag of 8.5 years between the disaster and the assessment. Due to administrative deficiencies in the historic registration of the exposure status, we used our detailed questionnaire data to define exposure status for all workers. Strictly speaking, the ESADA is therefore not a historic cohort study, but a cross-sectional one. The time-lag between the disaster and the exposure assessment may have led to recall bias, especially concerning certain details of exposure to the disaster, such as the duration of activities. However, it seems reasonable to assume that the workers did recollect whether or not they performed any as opposed to no disaster-related tasks, which was used to define occupational exposure status. It is therefore very unlikely that recall bias has resulted in (non-)differential misclassification of exposed and non-exposed workers. Nevertheless, recall bias should be kept in mind with respect to exposure-response relationships. We included multiple aspects of level of exposure, such as the duration and the potential psycho-traumatic character of disaster-related tasks, as it is unknown which aspect of occupational exposure to disasters is relevant for long-term health. However, we may still have missed other potentially relevant aspects, such as exposure to disaster-related media reports in the aftermath of the disaster \[[@B4],[@B75]\]. Thirdly, we acknowledge the fact that, with the exception of the laboratory variables, we rely on self-reported health outcomes. However, most of the health questionnaires that we used have been validated and widely accepted, except for those used to assess the physical symptoms. Differential misclassification in self-report health measures could occur if exposed workers are more likely than non-exposed workers to interpret and report bodily sensations as symptoms. On the other hand, hypervigilance and hypochondria themselves could well be adverse health effects of (toxicological) disasters \[[@B76],[@B77]\]. In conclusion, to increase our knowledge of potential health consequences of (technological) disasters, it is important to be prepared for epidemiological disaster research. Incorporating basic multidisciplinary, epidemiological research protocols into disaster management plans will stimulate scientifically sound research on the health effects of disasters. The ESADA will provide additional scientific knowledge on the long-term health effects of technological disasters on professional workers. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors participated in the multidisciplinary ESADA project team of the EMGO Institute, provided comments on the draft versions and approved the final manuscript. PS drafted the manuscript and performed statistical analyses. AH coordinated the acquisition of data and supervised the first part of the ESADA study. JT contributed to the design and supervised the statistical analyses. AW coordinated the rating of the potential psychotraumatic impact of exposure items and performed statistical analyses. HP and IB participated in the conception and design of the ESADA with respect to psychological outcomes, and IB rated the potential psychotraumatic impact of exposure items. NS supervised the second part of the ESADA. As Medical Director JB coordinated the data-collection. LB contributed to the design of the ESADA. WM contributed to the design and is Vice-President of the ESADA project team. TS conceived the study, and participated in its design and coordination as President of the ESADA project team. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/54/prepub> Acknowledgements ================ We wish to thank everyone who contributed to the organization and data-collection as well as the funding sources of the ESADA.	PubMed Central	2024-06-05T03:55:58.416574	2005-5-30	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173116/", "journal": "BMC Public Health. 2005 May 30; 5:54", "authors": [ { "first": "Pauline", "last": "Slottje" }, { "first": "Anja C", "last": "Huizink" }, { "first": "Jos WR", "last": "Twisk" }, { "first": "Anke B", "last": "Witteveen" }, { "first": "Henk M", "last": "van der Ploeg" }, { "first": "Inge", "last": "Bramsen" }, { "first": "Nynke", "last": "Smidt" }, { "first": "Joost A", "last": "Bijlsma" }, { "first": "Lex M", "last": "Bouter" }, { "first": "Willem", "last": "van Mechelen" }, { "first": "Tjabe", "last": "Smid" } ] }
PMC1173117	Background ========== In many developing countries and countries with economies in transition, health care injections are overused and are frequently administered in an unsafe manner \[[@B1],[@B2]\]. The World Health Organization (WHO) estimates that 20 million new hepatitis B virus (HBV) infections, 2 million new hepatitis C virus (HCV) infections, and 260,000 new HIV infections are associated with unsafe injections each year worldwide \[[@B3]\]. Concerns have been expressed over the use of injections within the Kingdom of Cambodia (2003 United Nations population estimate 14.1 million). Anecdotal reports suggest that therapeutic injections are often unnecessary and administered in an unsafe manner, and healthcare waste is inappropriately disposed \[[@B4]-[@B6]\]. A high prevalence of HIV infection has been documented in Cambodia, with 2.6% of the general population infected \[[@B7]\]. Cambodia also has high levels of HBV and HCV infection endemicity. A 1991 community-based study indicated that 8.0% of the population was hepatitis B surface antigen positive and 6.5% was anti-HCV positive \[[@B8]\]. The proportion of bloodborne pathogen infections in Cambodia attributable to unsafe injections is not known. To estimate the magnitude and patterns of unsafe injection practices in Cambodia, a rapid assessment of injection practices was conducted. The objectives of the assessment were to describe healthcare injection practices, including measures of injection frequency and injection safety, and determine knowledge, attitudes and practices related to medical injections among the general population and healthcare providers. This assessment focused mainly on the unregulated private healthcare sector, in which 70% of Cambodians seek medical care \[[@B9]\]. Methods ======= Overview -------- The assessment was conducted using methods established by WHO for the rapid assessment of injection practices \[[@B10]\]. As detailed in the following sections, we interviewed members of the general population, healthcare workers responsible for prescribing injections (i.e., \"prescribers\"), and healthcare workers responsible for administering injections (i.e., \"injection providers\"). All data were collected during face-to-face interviews using standardized questionnaires and forms adapted from the WHO assessment guide \[[@B10]\]. The general population sample was drawn from Takeo province (1998 population census 790,168), a rural province located in southeastern Cambodia near the Vietnamese border. Prescribers and injection providers were sampled in Takeo province and in Phnom Penh (1998 population census 999,809), Cambodia\'s largest city \[[@B11]\]. General population ------------------ We surveyed 500 Takeo province residents in November and December 2002 using standard cluster sampling methodology. To obtain a total of 500 participants, we surveyed 25 participants each in the catchment area of 20 district health centers. The 20 district health centers were selected by probabilities proportional to the population size of their catchment areas \[[@B12]\]. For each health center selected, one village was chosen at random to be surveyed. Households were sampled at random until 25 participants were included. Verbal informed consent was obtained from all household members \>16 years old to participate in the survey. For persons ≤ 16 years old, informed consent was obtained from a parent or guardian. For each participant, basic demographic data were collected, as well as the number of injections received during the 6 month period beginning on April 13, 2002 (New Year\'s Day in Cambodia). Injections were categorized as therapeutic injections, immunizations, and intravenous infusions. Participants \>16 years of age were also interviewed regarding their knowledge and attitudes about injections. Prescribers and injection providers ----------------------------------- We surveyed prescribers and injection providers in Takeo province and Phnom Penh in November and December 2002. In Takeo province, a convenience sample of 30 prescribers and 30 injection providers was accessed through the province\'s five public hospitals. In Phnom Penh, 30 prescribers were randomly selected from a list of registered private outpatient clinics and 30 injection providers were selected from 11 of the city\'s 13 private hospitals. For all prescribers, regardless of where they were identified, the survey focused on their medical practices in their private outpatient clinics. We ascertained their knowledge, attitudes and practices regarding injections. Injection frequency was determined by asking prescribers to estimate the total number of prescriptions they wrote per week and the number that included an injection or infusion. Injection providers were interviewed regarding their knowledge of disease transmission risks through unsafe injections, their hepatitis B vaccination status and the frequency of needle stick injuries during the previous 12 months. In addition, we observed each injection provider administer one injection. During all hospital visits, information was recorded regarding the use of incinerators and the presence of used needles and syringes and other medical equipment on the grounds of the facility. Data analysis ------------- Proportions, means, and confidence intervals were calculated using EpiInfo 6.04d (CDC, Atlanta, Georgia) with adjustment for the cluster sampling design effect. Univariate and multivariate analyses of variables associated with receiving injections were performed using SAS software version 8 (SAS Institute, Cary, North Carolina). Results ======= General population ------------------ The overall injection rate was 5.9 injections per person-year (95% CI: 5.3 -- 6.7), with 40% of the participants reporting receipt of one or more injections during the previous 6 months (Table [1](#T1){ref-type="table"}). Therapeutic injections accounted for 74% of the total number of injections reported, followed by intravenous infusions (16%), and immunization (10%). Among individuals who had any exposure to injections, nearly one-half reported receiving ≥ 5 injections. This group represented only 18% of all participants but accounted for 69% of the total injections reported. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of study participants, Takeo province, Cambodia, 2002 ::: ------------------------------------------------------------ --------------------- -------------------------- --------------------------- ------------------------ Total (n = 500) *Children\ (n = 212) Adult women (n = 189) Adult men (n = 99) Age: median 22 9 38 37 range (years) 0--80 0--16 17--79 17--80 [Injection experiences (6-month period)]{.underline} Number of injections 1483 557 713 213 Overall injection rate (per person-year) 5.9 5.3 7.5 4.3 Proportions of subjects who received: ≥ 1 vaccine injection 15% 35% 0% 0% ≥ 1 therapeutic injection 32% 31% 39% 20% ≥ 1 intravenous infusion 16% 16% 18% 14% ≥ 1 injection (any type) 40% 46% 42% 24% ≥ 5 injections (any type) 18% 18% 22% 9% [Knowledge and attitudes\^]{.underline}** Trusted practitioners if no injections prescribed 90% ND 90% 89% Preferred injection for treatment of fever 32% ND 35% 26% Believed injections more powerful than oral medication 47% ND 50% 40% Aware that dirty syringes can transmit HIV 95% ND 96% 93% Aware that dirty syringes can transmit hepatitis 59% ND 60% 57% ------------------------------------------------------------ --------------------- -------------------------- --------------------------- ------------------------ ND = not determined \* participants aged ≤ 16 years \^ adults only (288 total respondents) ::: The pattern of injection use differed by age and sex. Among adults, women were more likely than men to report receiving any (42% vs. 28%, p \< 0.001) or frequent injections (i.e., ≥ 5 injections during the previous 6 months; 22% vs. 9%, p \< 0.001) (Table [1](#T1){ref-type="table"}). Women were also more likely than men to report having received ≥ 1 therapeutic injection (39% vs. 20%, p = 0.001). Similar proportions of men, women and children reported receiving intravenous infusions, ranging from 14 -- 18%. Immunizations were limited to children, with 35% of children reporting having received ≥ 1 immunization. Ninety-six adult participants (93% of those reporting one or more injection in the previous 6 months) could recall the details of their last injection. All reported the injection was administered with a newly opened disposable syringe and needle. The majority (85%) of these injections was administered by health care workers; at least 13% were administered by lay persons (Figure [1](#F1){ref-type="fig"}). Injections were most commonly administered in the patient\'s home (65%) or in a private clinic (20%). Only 13% of injections were administered at public hospitals or public health centers. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Distribution of injections by provider type and setting\, Takeo Province, Cambodia, 2002 (n = 96) \Data refer to the most recent injection received by adults reporting ≥ 1 injection in the previous 6 months ::: ![](1471-2458-5-56-1) ::: Of the 288 subjects interviewed, 90% reported they would trust a medical practitioner who did not prescribe an injection (Table [1](#T1){ref-type="table"}). However, 47% (134/288) believed that therapeutic injections are more powerful than oral medications, with 75% (100/134) of these persons attributing a faster healing effect to injections. In addition, 32% of the respondents reported they preferred injections over oral medications for the treatment of illnesses characterized by fever. Knowledge of the HIV transmission risk associated with \"dirty\" syringes and needles was very high (95%), though only 59% of respondents were aware that hepatitis could be transmitted in this manner. None of the knowledge or attitude variables or other characteristics (besides gender) were associated with the receipt of injections. This was true even when the analyses were restricted to comparisons of persons reporting no injections versus frequent injection use (i.e., ≥ 5 injections during the previous 6 months). Prescribers ----------- The 60 prescribers interviewed were registered medical doctors or medical assistants. Prescribers estimated that nearly half (47%) of the total prescriptions they wrote included either a therapeutic injection (34%) or an intravenous infusion (14%) (Table [2](#T2){ref-type="table"}). The overall prescription rates and distribution by type was similar in Phnom Penh and Takeo Province. The main reasons cited for prescribing injections were illness severity (44%) and perceived patient preferences for injectable medications (40%). Seventy-seven percent of prescribers believed that injection prescriptions resulted in greater reimbursement. Only one prescriber perceived himself as over-prescribing injections. Most (92%) prescribers were aware that HIV, HBV and HCV could be transmitted through unsafe injections (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Characteristics of injection prescribers and providers, Takeo province, Cambodia, 2002 ::: ------------------------------------------------------------------------------------------------ -------------------- ----------------------------- ------------------------- Total (n = 60) Takeo Province (n = 30) Phnom Penh (n = 30) *[Characteristics of prescribers^1^]{.underline}* Medication prescription rate (prescriptions/week) average 20 21 20 median 20 21 20 range 4--140 4--140 7--140 Prescriptions including an injection^2^ 47% 48% 45% therapeutic injection 34% 32% 35% intravenous infusion 14% 18% 10% Main reason for prescribing injections illness severity 44% 50% 37% patient preference 40% 40% 40% more effective than oral medications 12% 7% 17% reimbursement 9% 7% 10% Preferred injectable med for treatment of febrile illness 64% 47% 80% Believed patient trust requires injection prescription 42% 53% 30% Believed reimbursement is higher for patient visits that result in injection prescription 77% 87% 66% Perceived themselves as over-prescribing injections 2% 3% 0% Knew HIV, HBV and HCV can be transmitted through unsafe injections 92% 87% 97% *[Characteristics of injection providers]{.underline}* Completed hepatitis B vaccination series 20% 7% 33% Needlestick injury in last 12 months 53% 50% 57% Average number (and range) of needlesticks in past 12 months among those reporting one or more 1.7 (1--10) 1.4 (1--6) 2.0 (1--10) Use of single use needles and syringes^3^ 98% 97% 100% Safety box (i.e., sharp container) present in injection administration area^3^ 25% 37% 13% Reported having sufficient number of sharps boxes 85% 77% 93% Practiced two hand recapping of used needles^3^ 58% 53% 60% Left used sharps in preparation area^3^ 13% 23% 3% Knew HIV, HBV and HCV can be transmitted through unsafe injections 90% 87% 93% ------------------------------------------------------------------------------------------------ -------------------- ----------------------------- ------------------------- ^1^Prescribers\' responses pertain to their private outpatient practices ^2^Denominator is the total number of weekly prescriptions ^3^Based on observation of provider by the investigators ::: Injection providers ------------------- All 60 injection providers interviewed were registered nurses, only 20% of whom had received the complete hepatitis B vaccination series. During our observations, 59 (98%) providers used new single use syringes (i.e., traditional plastic disposable syringes) and needles and one administered the injection with sterilized reusable equipment. Over half (58%) of the providers were observed practicing two-handed recapping of used injection equipment and 53% reported they had sustained a needlestick injury during the past 12 months. For only 25% of providers, a safety box (i.e. sharps container) was observed in close proximity to the injection area, yet 85% reported having a sufficient number of safety boxes. We observed that 13% of providers left used sharps in the injection preparation area (23% in Takeo province and 3% in Phnom Penh city). Ninety percent of injection providers stated they were aware that HIV, HBV and HCV could be transmitted through unsafe injections (Table [2](#T2){ref-type="table"}). Waste disposal -------------- In the general population survey in Takeo province, 32% (23/72) of participants who reported receiving their last injection in their home reported that the used injection equipment was left behind by the injection provider. Used sharps were observed on the grounds outside three of the five public hospitals in Takeo province, but not on the grounds of any of the 11 private hospitals in Phnom Penh. Incinerators and sharps pits in which used sharps waste could be burned or buried were present at all of the Takeo hospitals. In Phnom Penh, sharps pits were available in two hospitals. In the remaining nine hospitals, sharps waste was collected along with non-medical waste for disposal at the city landfill. Discussion ========== The rapid assessment of injection practices described in this paper suggests that injections are prescribed and administered in Cambodia at excessive rates. Intravenous infusions were common among the general population surveyed and represented approximately one-third of all injections prescribed in private outpatient clinics. Using WHO standardized measures as a basis for comparison, our assessment indicates that Cambodia has one of the highest rates of overall therapeutic injection usage ever reported worldwide \[[@B1],[@B2]\]. Studies conducted in other countries in the region have documented rates of 2--4 injections/person/year; approximately one third lower than in Cambodia \[[@B2]\]. This assessment documented potentially harmful injection practices including inadequate handling and disposal of used injection equipment. Given the country\'s high prevalence of HIV, HBV, and HCV, the overuse and misuse of injections carry substantial risks. While reuse of injection equipment was uncommon in this survey, the high prevalences of bloodborne infections may reflect past practices of needle and syringe reuse. We did observe breaks in aseptic technique (e.g., used sharps left in the injection preparation area) that may facilitate cross-contamination during injection preparation \[[@B13],[@B14]\]. Other infection control breaks such as mishandling of sterile materials and improper use of multidose vials have been frequently implicated in the transmission of bloodborne viral and bacterial infections, but were not studied in this survey \[[@B13],[@B15]\]. Anecdotal reports suggest that used needles and syringes might be repackaged and resold as new. While there was no evidence to suggest this practice was occurring in Cambodia, we did not evaluate the sterility of needles and syringes. In addition, many of the injection providers in our survey engaged in high risk behaviors (e.g., two-handed recapping needle), highlighting the need for training and other interventions aimed at reducing occupational exposures and infection risks. The patterns of attitudes, knowledge and practices regarding injections among the general population and health professionals observed in Cambodia were similar to those in many countries with high frequencies of injections and comparable healthcare systems such as Pakistan, India, and Indonesia \[[@B16]-[@B18]\]. In these countries as in Cambodia, patients and prescribers frequently believe that injections are more effective than oral treatments, and few health professionals perceive themselves as over-prescribing injections. This study and others have also shown that many healthcare providers believed that patients\' trust hinged on injection prescriptions, whereas only a minority of patients actually stated that they prefer injections for treatment of conditions that could be treated with oral medications. Both misconceptions may facilitate injection overuse. In this study, almost half of the prescriptions made by the private providers we surveyed contained at least one injection, compared with a previous finding that only 2% of prescriptions written in public sector outpatient clinics included injections \[[@B19]\]. Financial interests might lead some private health professionals to over-prescribe injectable medications \[[@B16]\]. Since oral and intravenous medications are readily available directly from pharmacies without prescriptions, it is possible that private healthcare providers provide and promote injections and infusions as a way to attract patients \[[@B20]\]. These factors might partly explain the popularity of the private sector compared with the public healthcare sector where the cost for clinic visits is lower and medications are available free of charge, but are dispensed according to standardized treatment protocols which generally specify oral formulations. High levels of awareness of infection risks associated with syringe and needle reuse, particularly regarding HIV were documented in this study. This is consistent with results of a recent national Demographic Health Survey \[[@B9]\] and likely reflects Cambodia\'s efforts during the late 1990\'s to educate the general public regarding risk factors for HIV infection. These education campaigns included information on the risk of transmission of HIV from re-used syringes. Following the campaigns, disposable needles and syringes became widely available in Cambodia at a relatively low cost. It is possible that the increased awareness of HIV infection risks associated with injections led to increased utilization of disposable needles and syringes rather than a shift toward alternative forms of treatment. This assessment had several limitations. First, the survey did not capture the medical indications that were associated with either prescription or receipt of injections. Second, resource constraints required the general population surveys to focus on one province instead of being conducted nationwide. Takeo province was selected because it is the third most populated province, its economic status is in the middle of the national spectrum, and like the general Cambodian population it is mainly rural \[[@B9]\]. Third, because only formal health professionals were surveyed, we could not assess practices in the informal health sector. We found that approximately one-in-seven injections in Takeo province were administered by untrained paramedical and lay health care workers and their practices may differ from health care workers in the formal sector. There is growing consensus that improvements in injection safety and overuse in developing countries can be achieved \[[@B21]\]. Such efforts require multidisciplinary engagement and behavioral changes on the parts of both patients and health professionals, including those from the private healthcare sector \[[@B22]\]. Although large-scale interventions to improve the safety of therapeutic injections are limited, several demonstration projects have proven to be effective and relatively long lasting. Studies in Indonesia demonstrated a sustained 19% reduction in injection rates following \"interactional group discussions,\" an intervention that brings health care providers and patients together to discuss their preferences regarding injections and oral medications \[[@B18],[@B23]\]. In Burkina Faso, interventions targeted at providing sufficient quantities of single-use syringes and needles resulted in \>90% reductions in unsafe injections that have been sustained for at least 5 years \[[@B24]\]. Comprehensive interventions aimed at reducing overuse of injections, promoting rational use of oral and injectable medications, and reinforcing safe injection and waste disposal practices are likely to be cost-effective in Cambodia \[[@B25],[@B26]\]. While initial efforts in injection safety have been largely focused on infant immunization, Cambodia is now expanding this initiative to include curative care, with an emphasis on the private sector. A national plan of action has been developed to decrease injection use and promote injection safety in accordance with WHO guidelines. The plan aims to involve multidisciplinary partners and improve coordination among relevant departments in the Ministry of Health. This plan also includes demonstration projects that address specific issues related to injection use in the private sector. It is hoped that Cambodia\'s efforts to address its injection safety and overuse problems will provide an example for other countries that are beginning to confront this important public health issue. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= SV, SS, SS and JT adapted the WHO assessment design to Cambodia. SV, SS, and SS conducted the survey, supervised different aspects of its implementation and collected data. SV and JP synthesized analyses and led the writing. SG helped interpret findings and review drafts of the manuscript. YH conceptualized the assessment and reviewed drafts of the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/56/prepub> Acknowledgements ================ We are grateful to Drs. Hok Phalla and Thou Chourn of the Cambodian National Institute for Public Health, and Mrs. Mam Boravann of the Cambodian Department of Drugs and Foods for their collaboration in conducting the survey.	PubMed Central	2024-06-05T03:55:58.419656	2005-6-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173117/", "journal": "BMC Public Health. 2005 Jun 2; 5:56", "authors": [ { "first": "Sirenda", "last": "Vong" }, { "first": "Joseph F", "last": "Perz" }, { "first": "Srun", "last": "Sok" }, { "first": "Seiharath", "last": "Som" }, { "first": "Susan", "last": "Goldstein" }, { "first": "Yvan", "last": "Hutin" }, { "first": "James", "last": "Tulloch" } ] }
PMC1173118	Background ========== For years Israel has been considered to be endemic for viral hepatitis, with an incidence of reported cases five to ten times that observed in the United States \[[@B1]\]. Most cases of viral hepatitis in Israel are caused by the hepatitis A virus (HAV), and in the 1970\'s the Israeli Ministry of Health instituted a wide-scale program of post-exposure prophylaxis with immune serum globulin (ISG). In 1999 universal immunization with inactivated virus (Havrix; Smith Klein Beecham) was instituted for infants, the first dose being given at 18 months and the second at 24 -- 30 months of age \[[@B2]\]. The Israel Defense Forces (IDF) have instituted vaccination programs as well, first with ISG and, more recently, with the inactivated virus \[[@B3]\]. Vaccination programs, as well socioeconomic changes, have led to a significant reduction in the prevalence of HAV infection in Israel today \[[@B4],[@B5]\]. In the United States, the U.S. Advisory Committee on Immunization Practices has given clear recommendations for vaccinating communities with high rates of infection (\>20 cases per 100,000) or populations at risk for liver failure \[[@B6]\]. However, no guidelines exist for routine testing for HAV serology prior to vaccination, a policy which could potentially prevent unnecessary immunizations (both active and passive) in seropositive patients as well as target those who are seronegative for active immunization. The following study examined the current prevalence of anti-HAV serology in a selected Israeli population (patients presenting to a primary care clinic), evaluating the benefits of routine testing for populations at risk. Methods ======= The records of teenage and adult patients (aged 16--99 years; mean 33.9) from two primary care clinics in Jerusalem, Israel, were retrospectively searched for anti-HAV IgG serology. The clinics are affiliated with the Maccabi Health Care Services, one of four government-funded services available to all Israeli citizens. Patients had undergone general blood tests for a number of reasons, either for follow-up of existing illness (none of which were hepatic-related) or for routine health screening, including anti-HAV IgG serology. A total of 1,017 patient files were found which contained serological results of anti-HAV antibodies from blood analyzed in the health fund\'s central lab, which used a microparticle enzymatic assay (MEA; AxSYM System, Abbott Diagnostics). Patients who tested seropositive for HAV were then contacted (either in person or by phone) and asked regarding the following: demographic data (country of birth); whether or not they remembered being sick with hepatitis (i.e. clinical jaundice, with or without known serological evidence of active HAV infection), and when; and whether or not they had been immunized, either passively (ISG) or actively (inactivated virus), and when. Recall of past infection was based on patient reporting alone and not through review of medical or laboratory records. Data was compiled and analyzed using the Microsoft Excel program. Results ======= Prevalence of HAV seropositivity -------------------------------- As mentioned above, a total of 1,017 patient records (503 males and 514 females) were found to contain results of anti-HAV antibodies (from March 1998 to September 2003). The mean age of the study group was 33.9 years; 37.8 among males (range: 16--99) and 31.2 among females (range: 16--87). Among the recorded results, 692 (68.0%) were seropositive -- 354 of the male patients (70.4%) and 338 of the female patients (65.8%; p = 0.113). The rates of seropositivity increased with age (see Figure [1](#F1){ref-type="fig"}; p \< 0.005), and remained constant throughout the years of the study (p = 0.451). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Prevalence of positive HAV serology according to age (%) ::: ![](1471-2458-5-60-1) ::: Patients who were born in Middle Eastern countries other than Israel had the highest seropositive rate (91.3%), while those born in South America the lowest (44.1%; p \< 0.005. See Figure [2](#F2){ref-type="fig"}). Immigrants from the Eastern Bloc countries (former Soviet Bloc and Eastern Europe) had a significantly higher prevalence of anti-HAV antibodies than native Israelis (70.3% vs. 58.7%; p = 0.016). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Prevalence of positive HAV serology according to country of birth(%). \* Eastern Bloc= former USSR and Eastern European countries\\ Western Bloc= North America and Western Europe ::: ![](1471-2458-5-60-2) ::: Recall of past infection and immunization ----------------------------------------- Of the 658 patients with anti-HAV antibodies, 456 were interviewed and asked regarding past infection and vaccinations (231 males and 225 females). Only 91 had any recollection of prior illness (20.0%) -- 51 of the males interviewed (22.1%) and 40 of the females (17.8%; p = 0.249), all of them at least 5 years prior to undergoing the serological testing. A total of 103 respondents remembered receiving post-exposure prophylaxis (passive immunization) with ISG. Those who were unaware of past infection were more likely to have been vaccinated with ISG than those who were aware (26.3% vs. 7.7%; p \< 0.005). The reasons given for receiving ISG are listed in Table [1](#T1){ref-type="table"}. Eight patients (2 male, 6 female) reported receiving the inactivated virus vaccine, of which none were aware of prior illness. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Reasons for Passive Immunization\* in Seropositive Patients ::: [Reason]{.underline} [Male (n)]{.underline} [Female]{.underline} [Total]{.underline} ------------------------------- ---------------------------- -------------------------- ------------------------- Post-exposure prophylaxis 25 25 50 Travel to endemic country 10 3 13 Army Service 14 3 17 [Other]{.underline} [17]{.underline} [6]{.underline} [23]{.underline} Total 66 37 103 *\ passive immunization**= immune serum globulin ::: Discussion ========== Although Israel is considered endemic for HAV infection, the prevalence of anti -HAV antibodies in our study population (68%) is much higher than that found in other studies, especially among the under-20 age group (62.8%). In a 1996 study of teenagers undergoing pre-draft evaluation, only 38.4% were seropositive \[[@B5]\], while serum samples from the Israeli Center for Disease Control\'s national serum bank in 1989 were seropositive for HAV in only 27% of 16 year-olds \[[@B4]\]. The relatively high prevalence of anti-HAV antibodies found in our study may be due to the fact that the study was conducted in a primary care clinic, with higher morbidity rates than the general population, though not liver-related. Another factor is the fact that both clinics are located in Jerusalem, one of Israel\'s poorest and most densely populated cities. Many of the patients were ultra-orthodox Yeshiva students who are at greater risk for HAV infection \[[@B7]\] and are not drafted into the IDF. At the same time, the higher rate of seropositivity among immigrants from Eastern bloc countries is consistent with the findings of an earlier study of this population \[[@B8]\], though another study of pre-draft teenagers did not find a significant difference between these immigrants and native Israelis \[[@B9]\]. Most of the seropositive patients in our study (80%) had no prior recollection of HAV infection. For 70% of children under the age of 6 (in communities with high rates of hepatitis A, 30--40% of children acquire infection before the age of 5 6) HAV infection is a self-limited, subclinical disease \[[@B10]\]. Among older children and adults, infection is usually symptomatic, with jaundice occurring in more than 70% of patients \[[@B11]\]. Signs and symptoms usually last less than 2 months, although 10--15% of symptomatic persons have prolonged or relapsing disease lasting up to 6 months \[[@B12]\]. Case fatality rates also increase with age, rising from 1.5/1000 among children less than 5 years old to 27/1000 after age 50 \[[@B13]\], making vaccination of unimmunized adults that much more important. Both ISG and inactivated HAV vaccines are safe, though not without potential side effects both locally and systemically \[[@B6],[@B14]\]. Aside from having to undergo two injections within a span of 6 months to a year, adult patients in Israel are required to purchase the active vaccine themselves. Seropositive patients in our study who were unaware of their immune status were more likely (by a factor of 3.4) to have unnecessarily received ISG (unnecessarily since they most likely have lifetime immunity) than those who were aware of past infection. This is probably an underestimate, since both post-exposure and pre-induction vaccinations have been implemented in Israel for nearly thirty years, and it is likely that many have forgotten receiving immunizations in the distant past. For the eight patients who received the inactivated virus without prior serological testing, it is possible (and even probable) that they too were immunized unnecessarily. Now it is too late to know whether their current seropositive status is a result of past occult infection (with lifetime immunity) or active immunization. These patients will now have to receive periodic booster vaccinations, which may have been unnecessary if prior immune status had been checked. Testing for HAV serology prior to active immunization has been shown to be both cost-effective and economically valid for Israeli travelers, assuming a cost of \$130 for vaccination (which is subsidized but not free for the patient) and \$30 for the IgG test (free for the patient) \[[@B15]\]. This may be true for the general population as well. Conclusion ========== The recent adoption of a nationwide infant HAV immunization policy in Israel has also been found to be both medically and economically justifiable \[[@B16]\]. However, this program is aimed at the infant population, for the purpose of preventing early childhood infection. Screening the adult population in endemic countries like Israel would enable seronegative adults to receive the inactivated virus, thus preventing morbidity and mortality resulting from infection. For the seropositive adult population, unnecessary immunizations could be avoided. Asking patients whether or not they need to be vaccinated is, at best, unreliable, since most cases of HAV infection occur in childhood and are asymptomatic, and even in cases where jaundice is present the cause may not necessarily be HAV. The health establishment in Israel needs to investigate the benefits of routing testing for anti-HAV serology, especially among the high-risk adult population who were not actively immunized during childhood or military service. Such routine testing is feasible, especially in an era where it is accepted practice to conduct routine screening tests for a number of preventive measures. Competing interests =================== The author(s) declare that they have no competing interest. Authors\' contributions ======================= The design and implementation of the study, as well as statistical analysis of the data, were all done by the author (NS). Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/60/prepub>	PubMed Central	2024-06-05T03:55:58.422936	2005-6-6	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173118/", "journal": "BMC Public Health. 2005 Jun 6; 5:60", "authors": [ { "first": "Noah", "last": "Samuels" } ] }
PMC1173119	Background ========== Ethiopia stands among the world\'s top 22 tuberculosis (TB) high-burden countries, with an estimated annual incidence of 250 TB cases/10^5^population \[[@B1],[@B2]\]. In the Southern Ethiopia Regional State (SNNPRS), TB is among the leading causes for sickness and death \[[@B3]\]. As in many other resource-constrained settings, treatment outcomes for tuberculosis have not been satisfactory \[[@B4]\], mainly due to poor treatment compliance and low coverage of short course chemotherapy (SCC). Delays in the diagnosis and treatment initiation, the devastating HIV/AIDS epidemic and the potential threat of anti-tuberculosis drug resistance represent serious threats to the TB control effort in the region. The HIV co-infection among TB patients in the region is estimated at 19% \[[@B5]\]. Directly Observed Treatment, Short-course (DOTS) was introduced in the region in 1996. However, the impact of DOTS and its effectiveness in the regional context has not been assessed yet. Global reviews \[[@B6]\] and reports from Asian, African and East European countries \[[@B7]-[@B9]\] have favourable implications for the DOTS strategy. On the contrary, some randomised controlled studies have failed to establish the superiority of the most notable component of this strategy, direct observation of treatment, over the conventional non-observed treatment in improving treatment outcomes \[[@B10],[@B11]\]. Nevertheless, DOTS is widely accepted and practised, with some TB high-burden countries achieving almost full coverage of their populations \[[@B2]\]. DOTS is equally effective in curing TB patients with HIV co-infection \[[@B9],[@B12]\] and hence, its importance in the era of the HIV/AIDS epidemic. With this study, we aimed to assess the results and trends in the expansion of DOTS in Hadiya Zone, Southern Ethiopia. Methods ======= Study setting ------------- We did the study in Hadiya zone, Southern Ethiopia (see map in Figure [1](#F1){ref-type="fig"}), with a population of 1.2 million \[[@B13],[@B14]\]. In addition to fifteen diagnostic centres, five health centres and 21 health stations provide treatment for TB patients referred or transferred from the diagnostic centres, thus making the number of treatment centres 41. All treatment units have standard Unit Registers from the National Tuberculosis and Leprosy Control Programme (TLCP). TB drugs used in a combination of two or more in the region included isoniazid (H), rifampicin (R), pyrazinamide (Z), ethambutol (E), streptomycin (S) and thioacetazone (T). The TB and Leprosy control programs were run as two separate vertical programs funded and managed mainly by the All African Leprosy and Rehabilitation Training Centre (ALERT) until the end of 1996 when the National and Regional TLCP took over the responsibility. The zonal TLCP gets administrative and technical support from the regional TLCP, and monitors the activities in the diagnostic and treatment centres through its seven woreda (district) offices. The woreda health offices have standard district TB registers from the National TLCP, and are responsible for the distribution of drugs and reagents, supervision of day to day activities in the treatment units, collection of slides for quality control and record keeping. Quarterly reports on cases detected and treatment outcomes from the woredas are gathered at the zonal level and sent to the regional TLCP. In 1994, only Hossana Hospital provided unobserved SCC for critically sick smear-positive pulmonary TB (PTB+), miliary TB and tuberculosis meningitis cases. DOTS was initiated in 1996 in two health facilities and gradually expanded, first to the health centres in all woredas and then to the health stations. Details of treatment regimens for various categories of patients under DOTS are given in the NTLCP manual \[[@B15]\]. In brief, all new patients were treated with SCC (two months on RHZ +/- E or S, followed by 6 months on EH or RH), and were required to take their medications under direct supervision by the health workers at least during the intensive phase of treatment. During this phase, only critically ill patients were hospitalised, while the rest received treatment on ambulatory basis. Re-treatment cases were treated with SERHZ for 2 months, ERHZ for 1 month and ERH for 5 months. In non-DOTS areas treatment with SCC was limited to critically sick PTB+, TB meningitis and miliary TB cases while others received LCC (two months on EH +/-S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). Design and data collection -------------------------- This is a retrospective trend analysis. The Unit Registers reviewed contain basic information such as patient\'s age, sex, address, category, TB type, drug regimen, date treatment started, treatment follow-up, follow-up sputum result and treatment outcomes. We visited all 41 TB treatment centres during July-November 2002. We checked when DOTS was initiated in each health institution and looked for the availability of TB drugs and reagents during the time of visit. We then reviewed the Unit Registers and entered the data to the computer. At three peripheral treatment centres where the unit registers could not be found, the respective district TB registers were reviewed. For each TB case, we copied the data from the registers to a computer program, SPSS for windows \[[@B16]\], according to the standard definitions of the National TLCP \[[@B15]\] on disease classification, patient categories, treatment regimens and treatment outcomes. To ensure the quality of data entered into the computer database, two people independently cross-checked each entry. Data analysis and Statistics ---------------------------- The data were analysed using SPSS for Windows version 11.0 \[[@B16]\]. For categorical data, we used proportions with 95% confidence intervals, Odds ratio and Chi-square test to compare different groups. Multivariate analysis using logistic regression model was used to analyse the association between treatment outcomes and potential predictor variables. We set the level of statistical significance at 5%. Treatment outcomes were analysed for the years 1994--2000 because a considerable proportion of patients registered during 2001 were still on treatment during the time of data collection. (The Ethiopian fiscal year goes from July to June. Fiscal year 2001, for example, goes from July 2001 through June 2002, and our study was commenced in July 2002). Patients with unrecorded treatment outcome were analysed as defaulters. Results ======= Patient registration and case notification ------------------------------------------ A total of 19971 tuberculosis patients, 11138 (55.8%) males and 8819 (44.2%) females were registered between 1994 and 2001 with the mean (SD) age of 25.6 (13.2) years. Forty-six percent (n = 9232) of the patients were PTB+ cases. Overall, 18687 (93.6%) patients were registered as new cases, 558 (2.8%) as transferred in, 273 (1.4%) as return after default, 142 (0.7%) as failure and 139 (0.7%) as relapse cases whereas patient category was not recorded for 172 (0.9%) cases. Table [1](#T1){ref-type="table"} shows the general characteristics of the patients. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### General Characteristics of the study subjects (n = 19971), 1994--2001 ::: Characteristics Number Percent --------------------------------------- ------------ ------------- *Age group (years)* 0--14 3356 16.8 15--24 6262 31.4 25--34 5366 26.9 35--44 2706 13.5 45--54 1248 6.2 55--64 501 2.5 ≥ 65 220 1.1 Unknown 312 1.6 *Sex* Male 11138 55.8 Female 8819 44.2 Not mentioned 14 0.1 *Patient Category* New 18687 93.6 Transferred-in 558 2.8 Return after default 273 1.4 Failure 142 0.7 Relapse 139 0.6 Unknown 172 0.9 *TB Classification* Pulmonary positive 9232 46.2 Pulmonary negative 4225 21.2 Extra-pulmonary 6453 32.3 Unknown 61 0.3 *Treatment Centre* Hossana Hospital 5362 26.8 Lemmo district health facilities 4011 20.1 Shashogo district health facilities 1208 6.0 Misha district health facilities 2660 13.3 Gibe district health facilities 1984 9.9 Soro district health facilities 2313 11.6 Duna district health facilities 216 1.1 Badewacho district health facilities 2217 11.1 ::: The proportion of women among TB patients registered for treatment remained at the range of 40--45% across the years and at the range of 43--47% across the age groups below 45. However, among patients older than 45 years, the proportion of women was significantly lower (31%; 95%CI 29--33) compared to those below 45 (46%; 95%CI 45--47). Among new cases, 46% (n = 8557) were PTB+ patients. Except for the years 1997 (38%; 95%CI 36--40) and 2000 (56%; 95%CI 54--58), this proportion showed little variations in the range of 40--50%. Table [2](#T2){ref-type="table"} shows a trend in the case notification of PTB+ over the study period. The proportion of expected PTB+ incident cases notified increased from 45% in 1994 to 116% in 1999, and declined to 67% in 2001. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Trends in case notification of smear-positive pulmonary TB, Hadiya Zone, 1994--2001 ::: Year Zonal Population\* New PTB+ cases Reported (n = 8558) Case Notification/ 10^5^persons/year Proportion of estimated PTB+ incident cases notified, %\\ --------- -------------------- ------------------------------------ -------------------------------------- ------------------------------------------------------------- 1994 1004000 512 49 45.2 1995 1070160 1025 96 87.9 1996 1101195 1162 106 96.8 1997 1133129 946 84 76.6 1998 1165990 1277 110 100.5 1999 1199804 1510 126 115.5 2000 1234598 1267 103 94.2 2001^§^ 1174118 858 73 67.0 \Projected from the 1994 Population and Housing Census, CSA \[13\] \\Estimated Incidence of PTB+ = 109/10^5^persons/year based on the WHO estimates of the Global Burden of Tuberculosis for 1997 \[1\] ^§^Population adjusted as an area with about 70--80 thousand population was taken away from Hadiya to the neighbouring Silti zone between years 2000 and 2001 PTB+: smear-positive pulmonary tuberculosis ::: Treatment given --------------- Forty percent (n = 7919) of patients received SCC. Treatment regimen was not recorded for 981 (5%) cases. Table [3](#T3){ref-type="table"} presents trend in treatment regimens across the years by TB classification. The proportion of patients treated with SCC increased from 7% in 1994 to 58% in 1999, and 97% in 2001 (χ^2^~trend~, p \< 0.001). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Trend in treatment regimens for the different categories of patients (n = 19970).\ ::: Category 1994 n (%) 1995 n (%) 1996 n (%) 1997 n (%) 1998 n (%) 1999 n (%) 2000 n (%) 2001 n (%) -------------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- All TB cases 1106 2201 2895 2742 3388 3253 2435 1950 SCC 73 (7) 311 (14) 288(10) 471 (17) 927 (27) 1881 (58) 2085 (86) 1883 (97) LCC 1011 (91) 1684 (77) 2360 (82) 2097 (77) 2292 (68) 1302 (40) 304 (12) 20 (1) No record 22 (2) 206 (9) 247 (8) 174 (6) 169 (5) 70 (2) 46 (2) 47 (2) New PTB+ 512 1025 1162 946 1277 1510 1267 858 SCC 9 (2) 111 (11) 94 (8) 232 (24) 537 (42) 1170 (78) 1187 (94) 837 (97) LCC 496 (97) 780 (76) 938 (81) 611 (65) 660 (52) 321 (21) 68 (5) 7 (1) No record 7 (1) 134 (13) 130 (11) 103 (11) 80 (6) 19(1) 12 (1) 14 (2) New PTB- 258 385 402 382 807 772 487 393 SCC 12 (5) 66 (17) 90 (22) 109 (28) 225 (28) 395 (51) 402 (83) 379 (96) LCC 242 (94) 286 (74) 248 (62) 251 (66) 543 (67) 362 (47) 73 (15) \- No record 4 (1) 33 (9) 64 (16) 22 (6) 39 (5) 15 (2) 12 (2) 14 (4) New EPTB 257 668 1177 1176 1124 721 528 570 SCC 19 (8) 73 (11) 57 (5) 70 (6) 97 (9) 197 (27) 385 (73) 559 (98) LCC 232 (90) 565 (85) 1081 (92) 1067 (91) 993 (88) 513 (71) 138 (26) 6 (1) No record 6 (2) 30 (4) 39 (3) 39 (3) 34 (3) 11 (2) 5 (1) 5 (1) Relapse 8 13 21 29 15 19 19 15 SCC 2 (25) 10 (77) 12 (57) 21 (72) 13 (87) 17 (90) 19 (100) 14 (93) LCC 4 (50) 2 (15) 9 (43) 6 (21) 2 (13) 2 (10) \- \- No record 2 (2) 1 (8) \- 2 (7) \- \- \- 1 (7) After default 35 56 27 45 33 25 21 31 SCC 5 (14) 14 (25) 15 (56) 20 (45) 15 (46) 21 (84) 17 (81) 30 (97) LCC 30 (86) 39 (70) 11(41) 23 (51) 18 (54) 4 (16) 4 (19) \- No record \- 3 (5) 1 (3) 2 (4) \- \- \- 1 (3) After failure 28 38 17 17 13 19 8 2 SCC 26 (93) 35 (92) 15 (88) 13 (77) 11 (85) 14 (74) 8 (100) 2 (100) LCC 2 (7) 3 (8) 2 (12) 4 (23) 2 (15) 5 (26) \- \- No record \- \- \- \- \- \- \- \- Transferred in 6 3 65 131 89 120 86 57 SCC \- \- 3 (5) 3 (2) 22 (25) 43 (36) 53 (62) 38 (67) LCC 5 (83) 2 (67) 55 (84) 124 (95) 58 (65) 57 (47) 19 (22) 7 (12) No record 1 (17) 1 (33) 7 (11) 4 (3) 9 (10) 20 (17) 14 (16) 12 (21) Missing info 2 13 17 10 20 46 13 29 SCC \- 2 (15) 1 (5) 3 (30) 3 (15) 11 (24) 9 (69) 29 (100) LCC \- 7 (54) 12 (71) 6 (60) 11 (55) 30 (65) 2 (16) \- No record 2 4 (31) 4 (24) 1 (10) 6 (30) 5 (11) 2 (15) \- \* Percentages are out of the total cases registered in the same category during the year. PTB+ = smear-positive pulmonary tuberculosis; PTB- = smear-negative pulmonary tuberculosis; EPTB = extra-pulmonary tuberculosis SCC = short course chemotherapy (2 months on RHZ+/- E or S, followed by 6 months on EH or RH) LCC = long course chemotherapy (2 months on EH +/- S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). R = Rifampicin,; H = Isoniazid; Z = Pyrazinamide; E = Ethambutol; S = Streptomycin; T = Thioacetazone. ::: During 1994--1997, more PTB- patents received SCC compared to PTB+ patients (19% vs. 12%; p \< 0.05) and EPTB patients (19% vs. 7%; p \< 0.05) among new cases of TB (Table [3](#T3){ref-type="table"}). This trend was reversed after 1997 when more PTB+ patients (76%) were put on SCC, followed by PTB- (57%) and EPTB (42%). There was no significant difference in the treatment regimens across the age groups and between the two sexes. Follow-up --------- Of the 8557 new PTB+ cases, sputum examination was repeated at the end of two months treatment for 60% (n = 5112), and 7.8% (n = 401) remained positive for AFB. At the end of five months treatment, 2747 (32%) patients had their sputum examined for AFB, and 78 (2.8%) remained smear-positive. 1823 (21%) patients had their sputum examined for AFB at (a month prior to) treatment completion, and 1.4% (n = 26) remained smear-positive. There has been a continuous and significant decline over time in the sputum positivity at two months irrespective of treatment regimen (Table [4](#T4){ref-type="table"}). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Trends in the follow-up smear results for new smear positive pulmonary tuberculosis ::: 2^nd^month 5^th^month End of treatment --------------------- ---------------- ---------------- ------------------ ---------- ---------- ---------- Cases on SCC 1994 6 (67) 0 5 (56) 0 3 (33) 0 1995 79 (71) 18 (23%) 53 (48) 4 (8%) 33 (30) 0 1996 68 (72) 15 (22%) 38 (40) 2 (5%) 25 (27) 0 1997 155 (67) 14 (9%) 94 (41) 0 74 (32) 2 (3%) 1998 342 (64) 24 (7%) 183 (34) 0 140 (26) 0 1999 824 (70) 52 (6%) 411 (35) 7 (2%) 365 (31) 2 (0.5%) 2000 817 (69) 43 (5%) 439 (37) 9(2%) 316 (27) 2 (1%) 2001^§^ 527 (63) 13 (3%) 171 (20) 1 (0.6%) 121 (14) 0 Cases on LCC 1994 330 (66) 45 (14%) 154 (31) 19 (12%) 86 (18) 5 (6%) 1995 508 (65) 75 (15%) 354 (45) 15 (4%) 179 (23) 4 (2%) 1996 564 (60) 59 (11%) 385 (41) 7 (2%) 223 (24) 4 (2%) 1997 302 (49) 15 (5%) 188 (31) 5 (3%) 86 (14) 3 (4%) 1998 263 (40) 19 (7%) 127 (19) 7 (6%) 80 (12) 2 (3%) 1999 182 (57) 7 (4%) 86 (27) 1 (1%) 68 (21) 2 (3%) 2000 46 (68) 1 (2%) 9 (13) 0 4 (6) 0 2001^§^ 2 (29) 1 2 (29) 0 1 (14) 0 Unknown regimen 1994 0 0 2 (29) 1 2 (29) 0 1995 9 (7) 0 4 (3) 0 3 (2) 0 1996 25 (19) 0 9 (7) 0 3 (2) 0 1997 23 (22) 0 11 (11) 0 3 (3) 0 1998 27 (34) 0 19 (24) 0 5 (6) 0 1999 4 (21) 0 1 (5) 0 1 (5) 0 2000 4 (33) 0 1 (8) 0 0 0 2001^§^ 5 (36) 0 17) 0 2 (14) 0 sm+ = sputum smear positive for acid-fast bacilli out of examined. SCC = short course chemotherapy; LCC = long course chemotherapy; AFB = acid-fast bacilli. ^§^Follow-up smears for the 5^th^month and end of treatment were not complete for the year 2001 as a considerable number of patients were still on treatment during the time of data collection. ::: Sputum positivity at the completion of fifth month treatment remained at the range of 0--2% for those on SCC, with the exception of 1995 and 1996 that exhibited 8% and 5% respectively. For those on LCC, smear positivity at 5^th^month decreased from 12% in 1994 to 1% in 1999. Overall, patients on LCC contributed more to the 5^th^month and end of treatment smear-positivity (treatment failure), than those on SCC (p \< 0.05; table [4](#T4){ref-type="table"}). Treatment results ----------------- We analysed treatment outcomes for the years 1994--2000 and, as a result, we evaluated 16943 (85% of the total) cases. Of these, 8268 (49%) successfully completed treatment, 3151 (18.6%) defaulted, 110(0.6%) had treatment failure, 446 (2.6%) died and 2000 (10%) were transferred out, while 2968 (17.5%) had no record of treatment outcomes. Assuming that patients with unrecorded treatment outcome had all defaulted from treatment, a total of 6119 (36.1%) patients defaulted. Among new PTB+ (and all TB) patients, treatment success was higher among patients on SCC, while default and failure rates were higher among those on LCC (Table [5](#T5){ref-type="table"}). Death in all categories was higher among those on SCC. Among new cases on SCC, PTB+ cases had 1.8 (65% vs.37%) and 1.4 (65% vs. 48%) times higher rate of treatment success compared to PTB- and EPTB cases, respectively (p \< 0.001). Female PTB+ patients had significantly higher treatment success (58% vs. 54%; p = 0.001) and lower defaulter rate (26% vs. 30%; p \< 0.001) than males. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Treatment outcomes for different categories of patients on short and long course regimens (n = 16943) ::: New cases --------------------- --------------- ---------- ----------- --------- --------- --------- ---------- Patients on SCC Treatment success 2166 (65) 475 (37) 436 (48) 62 (66) 49 (46) 65 (53) 79 (64) Default\* 726 (22) 483 (37) 320 (36) 13 (14) 45 (42) 31 (25) 33 (26) Failure 25 (0.7) \- \- 2 (2) 1 (1) 6 (5) 3 (2) Death 117 (3.5) 72 (5) 24 (3) 11 (12) 5 (5) 2 (1) 2 (2) Transferred out 306 (9) 269 (21) 118 (13) 6 (6) 7 (7) 18 (15) 7 (6) Patients on LCC Treatment success 1986 (51) 772 (39) 1932 (42) 14 (56) 32 (25) 11 (61) 189 (59) Default\* 1365 (35) 972 (48) 1922 (42) 8 (33) 79 (61) 2 (11) 24 (8) Failure 66 (2) \- \- 1 (4) 2 (2) 3 (17) 1 (0.3) Death 107 (3) 57 (3) 37 (1) 2 (8) 2 (2) 1 (5) 7 (2) Transferred out 350 (9) 204 (10) 697 (15) \- 14 (11) 1 (6) 3 (1) \* Patients with missing treatment outcome information were assumed to have defaulted and analysed as defaulters. PTB+ = smear positive pulmonary tuberculosis; PTB- = smear negative pulmonary tuberculosis; EPTB = extra-pulmonary tuberculosis SCC = short course chemotherapy (2 months on RHZ+/- E or S, followed by 6 months on EH or RH) LCC = long course chemotherapy (2 months on EH +/-S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). R = Rifampicin,; H = Isoniazid; Z = Pyrazinamide; E = Ethambutol; S = Streptomycin; T = Thioacetazone. ::: Patients registered as \"return after default\", had significantly lower treatment success compared to new cases, both for all TB (34% vs. 49%; p \< 0.001), and for PTB+ (34% vs. 58%; p \< 0.001). Furthermore, return after default cases were more likely to default again compared to new cases among PTB+ patients (53% vs. 29%; p \< 0.001) and all TB patients (53% vs. 36%; p \< 0.001). Treatment failure was five times higher among cases that received re-treatment for previous failure compared to new cases of PTB+ (6.4% vs. 1.2%; p \< 0.001); however, both groups had comparable treatment success, mainly due to a lower defaulter rate among failure cases (24% vs. 29%). Patients for whom follow-up sputum smear remained positive at the second month during treatment had significantly lower treatment success than those with negative smear result (53% vs. 74%; p \< 0.001), mainly as a result of higher defaulter (24% vs. 18%) and failure (10% vs. 1%) rates among the former group. Table [6](#T6){ref-type="table"} presents treatment outcomes for new cases of TB across the years. A steady rise in treatment success and decline in defaulter rate was noticed particularly among patients on SCC. Treatment failure showed a remarkable decline in both SCC and LCC groups. Death rate did not show much variation across the years. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Treatment outcomes across the years for new cases of tuberculosis ::: 1994 n (%) 1995 n (%) 1996 n (%) 1997 n (%) 1998 n (%) 1999 n (%) 2000 n (%) -------------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- PTB+ on SCC Treatment success 5 (42) 40 (34) 45 (38) 122 (47) 338 (60) 811 (69) 805 (74) Default\* 5 (42) 61 (51) 50 (42) 77 (30) 114 (20) 225 (19) 194 (17) Failure \- 3 (3) 2 (2) 2 (1) \- 8 (1) 10 (1) Death \- 4 (3) 2 (2) 10 (4) 24 (4) 44 (4) 33 (3) Transferred out 2 (17) 11 (9) 20 (17) 46 (18) 89 (16) 88 (7) 50 (5) PTB+ on LCC Treatment success 188 (38) 384 (48) 564 (56) 271 (45) 308 (52) 231 (72) 40 (68) Default\* 190 (38) 298 (38) 304 (30) 246 (40) 240 (40) 72 (22) 15 (25) Failure 23 (5) 18 (2) 8 (1) 8 (1) 6 (1) 3 (1) \- Death 17 (3) 21 (3) 18 (2) 20 (3) 19 (3) 9 (3) 3 (5) Transferred out 78 (16) 71 (9) 107 (11) 63 (10) 24 (4) 6 (2) 1 (2) PTB- on SCC Treatment success 1 (8) 15 (23) 18 (20) 32 (29) 67 (30) 153 (39) 189 (47) Default\* 8 (67) 40 (60) 67 (74) 55 (50) 85 (38) 116 (29) 112 (28) Death 1 (8) 6 (9) 2 (2) 10 (9) 8 (4) 25 (6) 20 (5) Transferred out 2 (17) 5 (8) 3 (3) 12 (11) 65 (29) 101 (26) 81 (20) PTB- on LCC Treatment success 107 (44) 88 (31) 84 (34) 78 (31) 199 (37) 182 (50) 34 (47) Default\* 101 (42) 163 (57) 127 (51) 141 (57) 280 (51) 135 (37) 25 (34) Death 4 (2) 5 (2) 6 (2) 5 (2) 16 (3) 20 (6) 1 (1) Transferred out 30 (12) 30 (11) 31 (13) 27 (11) 48 (9) 25 (7) 13 (18) EPTB on SCC Treatment success 4 (21) 19 (26) 25 (44) 29 (32) 43 (44) 101 (51) 215 (56) Default\* 11 (58) 43 (59) 29 (51) 30 (33) 29 (30) 68 (35) 110 (28) Death 2 (11) 6 (8) 1 (2) 3 (4) 1 (1) 4 (2) 7 (2) Transferred out 2 (11) 5 (7) 2 (4) 28(31) 24 (25) 24 (12) 53 (14) EPTB on LCC Treatment success 80 (35) 274 (49) 559 (52) 387 (36) 341 (34) 222 (43) 69 (50) Default\* 122 (52) 203 (36) 321 (30) 412 (39) 553 (56) 253 (49) 58 (42) Death 2 (1) 3 (1) 3 (0.3) 7 (1) 8 (1) 14 (3) \- Transferred out 28 (12) 85 (15) 198 (18) 260 (24) 91 (9) 24 (5) 11 (8) \* Patients with missing information on treatment outcome were assumed to have defaulted and analysed as defaulters. PTB+ = smear positive pulmonary tuberculosis; PTB- = smear negative pulmonary tuberculosis; EPTB = extra-pulmonary tuberculosis SCC = short course chemotherapy (2 months on RHZ+/- E or S, followed by 6 months on EH or RH) LCC = long course chemotherapy (2 months on EH +/-S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). R = Rifampicin,; H = Isoniazid; Z = Pyrazinamide; E = Ethambutol; S = Streptomycin; T = Thioacetazone. ::: When we adjusted the outcome measures for treatment results by various potentially confounding variables, significantly higher treatment success was exhibited among female patients, those 15--24 years, patients treated with SCC, those on re-treatment for relapse, PTB+ cases, those treated during 2000 and those treated in the Lemmo district health facilities (Table [7](#T7){ref-type="table"}). Meanwhile, male patients, patients aged 45--54, PTB- cases, those on LCC, return after default cases and those treated in Duna district health facilities exhibited significantly higher default rate. ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Adjusted odds ratios for various factors that might affect treatment outcomes among registered tuberculosis patients ::: Treatment success (all TB) Default (all TB) --------------------------- -------------------------------- ---------------------- --------- ------ ------------------- --------- Sex Male 45.5 1.00 37.5 1.00 Female 49.4 1.15(1.08--1.23) \<0.001 34.0 0.88(0.82--0.94) \<0.001 Age group (years) 0--14 42.6 0.83 (0.75--0.92) \<0.001 41.2 1.17 (1.06--1.30) 0.002 15--24 (reference group) 51.1 1.00 33.4 1.00 25--34 47.5 0.84(0.77--0.91) \<0.001 34.4 1.07 (0.98--1.17) 0.14 35--44 45.7 0.76 (0.68--0.84) \<0.001 36.1 1.14 (1.03--1.28) 0.02 45--54 44.9 0.69 (0.60--0.79) \<0.001 37.7 1.26 (1.09--1.46) 0.002 55--64 51.4 1.04 (0.84--1.29) 0.71 35.4 1.04 (0.81--1.27) 0.90 ≥ 65 37.4 0.67 (0.48--0.93) 0.02 43.7 1.27 (0.92--1.726 0.15 Patient category New 47.0 1.00 35.9 1.00 Transferred-in 59.5 1.09 (0.89--1.36) 0.40 35.7 1.19 (0.95--1.48) 0.13 Return after default 33.9 0.68 (0.51--0.92) 0.01 52.9 1.57 (1.19--2.07) 0.002 Treatment failure 54.3 1.30 (0.91--1.86) 0.15 23.6 0.72 (0.48--1.09) 0.12 Relapse 62.9 1.51(1.01--2.26) 0.05 16.9 0.50 (0.30--0.82) 0.006 TB type Pulmonary positive 55.3 1.00 28.7 1.00 Pulmonary negative 37.5 0.55 (0.50--0.60) \<0.001 43.7 2.06 (1.87--2.26) \<0.001 Extra-pulmonary 42.4 0.51 (0.47--0.55) \<0.001 41.1 1.94 (1.78--2.11) \<0.001 Treatment regimen Short course chemotherapy 55.5 1.00 27.7 1.00 Long course chemotherapy 45.0 0.68 (0.62--0.75) \<0.001 40.9 1.46 (1.33--1.61) \<0.001 *Treatment Centre\\* Hossana Hospital 27.0 1.00 40.0 1.00 Lemmo district HF 60.4 5.14 (4.62--5.72) \<0.001 20.6 0.28 (0.25--0.31) \<0.001 Shashogo district HF 50.8 2.81 (2.41--3.28) \<0.001 44.1 1.18 (1.01--1.37) 0.04 Misha district HF 60.3 4.78 (4.23--5.39) \<0.001 33.0 0.58 (0.51--0.65) \<0.001 Gibe district HF 60.7 4.78 (4.20--5.45) \<0.001 26.2 0.51 (0.45--0.59) \<0.001 Soro district HF 50.5 2.70 (2.39--3.06) \<0.001 42.5 1.11 (0.98--1.26) 0.09 Duna district HF 30.2 0.77 (0.55--1.08) 0.12 67.3 5.18 (3.71--7.24) \<0.001 Badewacho district HF 41.1 1.64 (1.44--1.86) \<0.001 54.4 2.11 (1.86--2.40) \<0.001 Year of treatment** 1994 38.4 1.00 42.7 1.00 1995 40.6 1.18 (0.99--1.39) 0.06 39.9 1.17 (1.00--1.37) 0.05 1996 48.7 1.29 (1.10--1.51) 0.002 33.3 0.83 (0.71--0.97) 0.02 1997 38.9 0.72 (0.61--0.85) \<0.001 39.5 1.08 (0.93--1.27) 0.32 1998 41.9 0.74 (0.63--0.87) \<0.001 43.8 1.20 (1.03--1.40) 0.02 1999 56.7 1.37 (1.16--1.62) \<0.001 30.0 0.59 (0.5--0.70) \<0.001 2000 59.8 1.41 (1.18--1.69) \<0.001 25.8 0.54 (0.45--0.65) \<0.001 OR = odds ratio; CI = confidence intervals; HF = Health facilities \* All the variables in the table are included in the model. \\ All treatment centres in a district grouped together as a unit. ^1^Percentage out of total registered for treatment. Patients with missing outcome record were assumed to have defaulted and analysed as defaulters. ::: Trend over time --------------- DOTS was initiated in 1996 in a hospital and one health centre, with potential population coverage of 25% (defined as population living within 2 hours walking distance from a health facility; estimated at 250,000 for the hospital and 25000 for the health centre). The number of health facilities providing DOTS increased to 10 in 1997, 30 in 1999 and 41 in 2001, making the population coverage by DOTS 31%, 58% and 75% respectively. The proportion of patients treated with SCC increased from 7% in 1994 to 27% in 1998, 58% in 1999, and 97% in 2001. 95% (39/41) of the treatment centres had at least isoniazid, rifampicin, pyrazinamide and ethambutol at the time of visit for data collection. Reagents for Acid-fast stain were available in nine of the 13 diagnostic centres functioning at the time. Simultaneously, treatment success for new PTB+ patients (on SCC and LCC together) increased from 38% in 1994 to 56% in 1998, 70% in 1999 and 73% in 2000 (χ^2^~trend~, p \< 0.001). Defaulting among new PTB+ patients declined from 38% in 1994 to 30% in 1998, 20% in 1999 and 18% in 2000 (χ^2^~trend~, p \< 0.001). Treatment failure decreased from 5% in 1994 to 1% in 2000. The proportion of reported deaths remained unchanged over years with some variations in the range of 2--5%. Treatment at small and large centres ------------------------------------ Thirty-one out of 40 (77.5%) peripheral treatment centres had significantly higher treatment success, and 28 (70%) of them had significantly lower defaulter rate compared to the zonal hospital (details not shown). Further comparison by districts revealed that six of the seven rural districts had significantly better treatment success while three out of seven had lower defaulter rate than the zonal hospital (Table [7](#T7){ref-type="table"}). Discussion ========== The results of our study show that there has been a continuous increase in treatment success and eventual decline in defaulter and failure rates in parallel to the expansion and decentralisation of DOTS to the lower treatment units. Treatment with SCC increased from a very low coverage in 1994 to almost full coverage in 2001 partly due to the expansion of DOTS (with SCC as one of its elements), and partly due to the regional TLCP commitment to SCC both in DOTS and non-DOTS areas. However, trends in case detection and notification showed inconsistencies over years. In the hierarchy of study designs, observational studies come after the randomised controlled trials \[[@B17]\]. Nevertheless, in the real field condition, where randomised trials are not logistically feasible, such observational study designs may provide important and valid information. Our study design does not allow comparison of treatment outcomes between the DOTS and non-DOTS areas. However, this study gives us a valuable information on the program performance before and along the course of expanding DOTS in this resource-constrained rural setting. Four years after the introduction of DOTS in the zone, it has been possible to achieve treatment success of 73% in 2000 for new PTB+ patients. Faring well towards the WHO/IUATLD recommended 85%, this confirms the finding of other studies \[[@B7],[@B18]\] that the DOTS strategy works well in resource-constrained settings with low overall health coverage. Increased coverage by SCC, improved access to care through decentralisation of the service and improved patient follow-up with the introduction of DOTS have most likely played a significant role in improving the treatment outcomes. Higher treatment success and lower default among female patients confirms previous study findings \[[@B11]\]. However, the proportion of females among patients registered for TB treatment was found to be consistently low, and it was exceptionally lower among patients older than 45 years. This may reflect a genuine gender difference in the TB epidemiology \[[@B19]-[@B21]\]. However, the possibility of this being a reflection of gender differentials in access to health care within the society may need to be ruled out by further studies. Patients treated at the peripheral treatment centres exhibited better treatment outcomes compared to those treated in the zonal hospital, which implies better follow-up of cases or better access to the TB care services. Despite significant decline in defaulter rate over time, patients that received re-treatment as \"return after default\", were much more likely to default again compared to new patients. This group of defaulters seems refractory to the conventional approach of treatment supervision; social and cultural factors that might play a role need to be explored. More death among patients on SCC was an indication of the policy that severely sick patients and those co-infected with HIV be a priority for treatment by SCC and these groups were more likely than others to die. Though lower among patients on SCC, the high proportion of failure among patients retreated for previous failure might signal the emergence of multi-drug resistant TB (MDR-TB). Earlier studies have shown that MDR has so far been below 1% \[[@B22],[@B23]\]. The proportion of PTB+ was fairly stable in the range of 40--50% across the years, which indicates that the diagnosticians uniformly observed the diagnostic algorithms recommended by the NTLCP. The case detection rate (CDR), as estimated by the proportion of expected incident cases notified, showed an encouraging increase from 45% in 1994 to 115% in 1999, well above the CDR recommended by the WHO. The most plausible explanation for this might be that the vigorous implementation of the program during the earlier years of introduction of DOTS might have enabled detection of a big pool of prevalent cases of TB. But the pace in case detection could not be maintained during 2000 and 2001 when the CDR fell to 94% and 67% respectively. Reduction in the incidence of TB could be a possible explanation, but one cannot justify such a fast decline in a very short time. This calls for some case detection improvement initiatives to be in place as the proportion of TB cases detected and cured under DOTS is one of the health-related indicators of the millennium development goals \[[@B24]\]. Quality of the laboratory services and the possibility of under-reporting are among factors to be explored further. Two major areas of program weakness need to be addressed. First, 94% of the patients were registered as new cases, which suggests that some re-treatment cases had been wrongly classified as new. Further, treatment outcomes have not been recorded for about one-fifth of the patients. Improving the record keeping system seems to be an immediate priority. Second, the proportion of PTB+ patients who had follow-up sputa examined for AFB was low. The threat of anti-TB drug resistance is imminent, and it is important to strengthen patient follow-up with sputum examinations. About one-third of the peripheral diagnostic facilities lacked reagents for Ziehl-Neelsen stain during the time of visit, and this might partly explain the low performance. Conclusion ========== The introduction and expansion of DOTS in Hadiya has led to a significant increase in treatment success and decrease in default and failure rates. The smaller institutions exhibited better treatment outcomes compared to the larger ones including the zonal hospital. The high number of patients with missing information in the unit registers is perhaps an issue that needs to be addressed as urgently as possible. Further studies are recommended to see the impact of the programme on the prevalence and incidence of tuberculosis. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= EBS was the principal investigator and participated in the design of the study, conducted the study, performed data entry and analysis, and wrote the manuscript. BL was the project co-ordinator and participated in the design, data analysis and write-up of the manuscript. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Map of Ethiopia with the study area highlighted. Ethiopia is administratively divided into nine regional states and two city administrations, and the Southern Nations, Nationalities and Peoples\' Regional State (SNNPRS) accounts for one-fifth (13 million) of the total population of the country. The study area (Hadiya zone) has got a population of 1.2 million. ::: ![](1471-2458-5-62-1) ::: Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/62/prepub> Acknowledgements ================ We would like to thank the Southern Nations, Nationalities and Peoples\' Regional Health Bureau, the regional and zonal TLCP and the staff of respective health facilities involved in the study for their genuine support during the field data collection. The Centre for International Health, University of Bergen funded this study.	PubMed Central	2024-06-05T03:55:58.424237	2005-6-6	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173119/", "journal": "BMC Public Health. 2005 Jun 6; 5:62", "authors": [ { "first": "Estifanos B", "last": "Shargie" }, { "first": "Bernt", "last": "Lindtjørn" } ] }
PMC1173120	Background ========== Almost 21% of the female population experience major depression in their life \[[@B1]\]. Twice as many females experience some form of depression when compared to male \[[@B2]\]. Female depression has higher risk on first onset, can last longer, and often recur \[[@B3],[@B4]\]. There is established relationship between life stress and depression \[[@B5]\]. Infertility is a stressful event in life of human being. In a comparison to patients with other medical conditions, psychological symptoms associated with infertility are similar to those related to cancer, hypertension, and cardiac rehabilitation \[[@B6]\]. Infertile women, in comparison with control group, showed higher scores on the depression and anxiety scales \[[@B7],[@B8]\]. Evidentially, depression has some neurobiological and hormonal mechanisms. Changes in ovarian hormones are likely necessary, but not sufficient for inducing depression \[[@B9]\]. The risk of relapse after treatment is 2.5 times higher in women; on the other hand, non-remission rate is similar in both sexes \[[@B10]\]. It appears that the sex difference cannot explain the sex-ratio relapse rate. As a result, there should be multiple factors including sex for inducing depression overall and in infertile subjects \[[@B3]\]. Many studies have been considered to determine risk factors of depression in infertility. In some studies, risk factors include female gender, repeated treatment cycles, unsuccessful treatments, low socioeconomic state, foreign nationality, lack of partner\'s support for women, life events for female, previous depression, and 2 to 3 years history of infertility \[[@B8],[@B11],[@B12]\]. However, some of these factors such as short period of infertility and husband cooperation have not been confirmed in other studies \[[@B12],[@B13]\]. The above-mentioned studies were designed as cross-sectional studies. So far, there is no prospective cohort or before-after study to find predictors of depression in infertile patients after treatment except a single one \[[@B14]\]. Lok et al showed that severity of depression following a failed treatment was positively associated with the duration of infertility. However, the post-treatment BDI scores were not associated with type of infertility treatment received, age, education, cause of infertility and number of previous treatments received \[[@B14]\]. Screening depression is a difficult task. The reason being is that there are numerous questionnaires for interpretation of psychological symptoms. Beck Depression Inventory (BDI) was developed and revised by Beck et al \[[@B15],[@B16]\]. This 21-item self-report questionnaire was intended to assess the severity of current depressive symptomatology in the psychiatric population. It is written on a fifth or sixth-grade reading level. It requires minimal time and no special training to administer. The BDI has been used, extensively, in clinical diagnosis and research \[[@B15],[@B16]\]. Versions in other languages are confirmed and used, as well \[[@B17]-[@B19]\]. In our study, Persian version of BDI has been employed. We analyzed Beck Depression Inventory score at the beginning and at the end of treatment cycle to determine which factors may influence the score of BDI, after treatment of infertility. Methods ======= Between April 2003 and 2004, 350 infertile women, whom had been visited in the Infertility Ward of Shariati Hospital in Tehran, participated in a prospective before-after study. Exclusion criteria were out of Tehran residency, definite diagnosis of mood disorders (with or without current treatment), medications reportedly associated with depression (e.g. steroidal contraceptives), and medical conditions associated with depression such as thyroid dysfunction and diabetes mellitus \[[@B20]\]. All subjects signed informed consent. Demographic data collected included age, education, type of infertility, duration of infertility, cause of infertility, and history of previous treatment (expressed as number). Candidates were asked to complete BDI (BDI1) in their first visit, during waiting periods, at the outpatient clinic. The questionnaire was completed by an assistant for uneducated subjects. Accordingly, BDI score was calculated. Index score of ≤9 is considered to be within normal range, a score of 10 to 15 shows minimal depressive symptomatology, a score of 16--31 points toward mild depression, a score of 32--47 is in favor of moderate depression and a score of \>47 indicates severe depression \[[@B16]\]. Due to rules of ethics, 10 patients who had the score of 47 or above were informed about their condition. Eight of these 10 patients were excluded from study since it was essential for them to seek psychiatric consultation as soon as possible. We continued the study with 342 patients. The treatment consisted of either intrauterine insemination (IUI) or assisted reproductive technology (ART). IUI candidates were excluded from the study. Therefore, the sample size was 257. At day 16 after embryo transfer, a positive result of treatment was defined as beta hCG (β-hCG) ≥200 IU/ml. One to Three weeks after receiving result of β-hCG, patients were asked to complete the second BDI (BDI2). This study was approved by ethics committee of Tehran University of Medical Sciences. The statistical analysis was performed using the Statistical Package for Social Sciences, version 10.0 (SPSS Inc., Chicago, IL). The primary outcome measure was the score of BDI after treatment. Data was expressed as means +/-SD and percentile of total. The tests being used were Wilcoxon Signed Ranks test, Mann-Whitney test, one way ANOVA, multiple regression analysis, and chi-square test. Results ======= A total number of 257 women were entered into the study after completing BDI1. Six (2.3%) patients refused answering BDI2. As a result, there were 251 paired complete questionnaires. Demographic characteristics of patients are given in Table [1](#T1){ref-type="table"}. We divided the patients into four groups based on cause of infertility. The prevalence was lowest in unexplained infertility (10%) and highest in female infertility (43%). The education level was divided into six groups. A large number of participants were at intermediate education level (35.1%) and a few of them were uneducated (2.4%). The number of previous attempts for treatment of infertility (IUI and/or ART) defers from 0 (24.7%) to 11 times (0.4%). Positive result of β-hCG was achieved in 61 patients (24.3%) (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of subjects ::: Characteristic No. Mean(± SD) or % ----------------------------- ------------------------ ----- ----------------- Age (y) 251 28.9 ± 5.5 Duration of infertility (y) 251 6.9 ± 4.5 No. of previous treatment 251 1.3 ± 1.4 Education No education 6 2.4 Primary school 36 14.3 Junior school 60 23.9 High school 12 4.8 Intermediate education 88 35.1 University 49 19.5 Type of infertility Primary 230 91.6 Secondary 21 8.4 Cause of infertility Male 86 34.3 Female 108 43 Combined 32 12.7 Unexplained 25 10 Result of β-hCG Positive 61 24.3 Negative 190 75.7 ::: The differences of somatic variables were analyzed between pregnant and non-pregnant subgroups. Pregnant subjects were younger (P\< 0.000), and had shorter duration of infertility (P= 0.005). There were no statistically significant difference in terms of type of infertility, education, cause of infertility, mean number of previous treatment, and mean score of BDI1 (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of demographic characteristics between pregnant and non-pregnant subjects ::: Pregnant (n = 61) Non-pregnant (n = 190) P ----------------------------- ----------------------- ------------------- ------------------------ -------- Mean age (y) 26.6 ± 4.2 29.6 ± 5.7 \<.000 Duration of infertility (y) 5.5 ± 3.7 7.4 ± 4.6 .003 Primary infertility (%) 58 (95.1) 172 (90.5) NS Cause of infertility (%) Male 22 (36.1) 64 (33.7) NS Female 27 (44.3) 81 (42.6) Both 7 (11.5) 25 (13.2) Unknown 5 (8.2) 20 (10.5) Education (%) No education 0 6 (3.2) NS Primary school 6 (9.8) 30 (15.8) Guiding school 11 (18) 49 (25.8) High school 3 (4.9) 9 (4.7) Completed high school 29 (47.5) 59 (31.1) University 12 (19.7) 37 (19.5) Previous treatment 1.2 ± 1.4 1.3 ± 1.4 NS BDI1 score 14.6 ± 11.3 14.5 ± 9.8 NS ::: The BDI score at the beginning of the study (BDI1) was within normal range (0--15) in 61% of patients. In 31% the score was 16--31, suggestive of mild depression, and only 8% of the women had BDI score higher than cutoff score (≥32), suggestive of moderate and severe depressive symptoms. After knowing the results of β-hCG, BDI score changed. After treatment, 47% of the whole sample was within normal range. In pregnant and non-pregnant subjects, normal score of BDI2 was found 95.1% and 31.6%, respectively. The difference in percentile between pregnant and non-pregnant patients were statistically significant (P \<.000). Our main outcome was evaluation of factors that can influence BDI2 score. When BDI2 score was considered as dependent variable, in a multiple regression analysis, it was positively correlated with BDI1 score (r = 0.61, P\< .000) and duration of infertility (r = 0.15, P \< .000). Negative correlation was found between BDI2 score and result of therapy (r = - .51, P= .000). Eighty six percent of variation of BDI2 score could be predicted by the above three variables. There was no correlation between BDI2 score and age, education, number of previous treatment, and type of infertility. The mean score of BDI2 was evaluated in relation to cause of infertility. The mean score in male factor was less than other groups. The difference between groups was statistically significant (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Mean score of BDI1 and BDI2 in relation to cause of infertility ::: Cause of infertility Mean of BDI1 Mean of BDI2 ---------------------- --------------- --------------- -------------- ------------ Male 11.2 ± 8.9 14.6 ± 11.3 P\< .000\* Z = -4.168 Female 15.8 ± 11 21.1 ± 13.2 P\< .000\* Z = -4.745 Combined 16.7 ± 9.2 21.9 ± 14.3 P= .006\* Z = -2.735 Unknown 15.2 ± 9.5 19.9 ± 10.5 P= .02\* Z = -2.303 P= .006\\ P= .002\\ Total 14.3 ± 10.2 18.9 ± 12.8 P\< .000 Z = -7.196 \: difference between variables in row \\: difference between variables in columns ::: Mean of BDI2 score rose after unsuccessful treatment and dropped after a successful treatment (Table [4](#T4){ref-type="table"}). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Comparison of mean BDI1 and BDI2 in pregnant and non-pregnant subjects ::: pregnant Non-pregnant P* ------------------ --------------------- ---------------------- --------------------- BDI1 (mean ± SD) 13.6 ± 11.3 14.5 ± 9.8 .28 (Z = -1.079) BDI2 (mean ± SD) 6.2 ± 5.4 22.9 ± 11.8 \<.000 (Z = -9.180) P \<.000 (Z = -5.845) \<.000 (Z = -11.481) ::: We observed the factors that affect BDI2 score in pregnant and non-pregnant groups, separately. In pregnant subjects, BDI2 score was positively correlated with BDI1 score (r = 0.82, P\< .000). In non-pregnant patients, BDI2 score was positively correlated with BDI1 score (r = 0.8, P\< .000), and duration of infertility (r = 0.17,P= .001). We studied BDI1 score, as well. In a multiple regression analysis, BDI1 score was negatively correlated with education (r = -0.26, P\< 0.000). There was no association between BDI1 score and age, duration of infertility, previous fertility, and number of previous cycles of treatment. Only 26% of variation of BDI1 could be predicted by the factor of education. In subgroups of infertility cause, the mean score of BDI1 in male factor was less than other groups. The difference between the groups was statistically significant (Table [3](#T3){ref-type="table"}). Discussion ========== The purpose of this study was to describe the subgroups of infertile women at risk of depression after treatment of infertility. This study can be valuable because it is using a self-reported inventory that differs in both cost and time from psychiatric structured interview. Nevertheless the shortcomings of this inventory, it has been widely used in research. Sensitivity and specificity of Beck Depression Inventory is not high but is reasonable. Beck et al suggested that a score greater than 9 points to depression symptomatology \[[@B16]\]. BDI score were also categorized in subgroups. The score of 9 and less shows normal range, a score of 10--15 indicates at least mild to moderate depressive symptoms, and a score of 16 and above indicates clinical depression \[[@B21]\]. We chose the second version since it screens subgroups of depression more precisely. In this study, the prevalence of depressive symptomatology as assessed by BDI1 score ≥ 16 was 39%, while the prevalence of moderate to severe depression(BDI ≥ 32) In an interesting study, the questionnaires were sent through internet to estimate degree of some psychological characteristics such as depression. Based on this study, authors found out that more than one quarter of patients could be considered moderately or severely depressed \[[@B22]\]. In screening depression in pregnancy, researchers chose a score greater than 9 and 43% of their population scored above this cutoff. As they designated BDI score greater than 15 as abnormal, nevertheless, the score exceeded the cutoff value in 19% of patients. In that study, a cutoff value greater than 15 yields a sensitivity of 0.83, a specificity of 0.89, a positive predictive value of 0.50, and a negative predictive value of 0.98 \[[@B21]\]. One of the best ways to determine the prevalence of a disease or symptom in relation to a risk factor is to compare a case group with a well-chosen control group. Domar et al showed that the prevalence of depression was 25.8% in infertile women compared with 13.2% in women who were waiting for a routine gynecologic examination \[[@B8]\]. These percentiles were reported to be 36.7% compared to 18.4% in another study \[[@B23]\]. Therefore, it is important that self-report questionnaires used for screening a disease should have a cutoff point reaffirmed either by a gold standard test or control group. The main outcome of this study was determining factors that may influence BDI2. The score after treatment (BDI2) was correlated with the score of BDI before treatment and duration of infertility and with the result of therapy. Hence, Table [4](#T4){ref-type="table"} shows no statistically significant difference between mean score of BDI1 in the two groups, we assume that the factors affecting BDI1 in the pregnant group and those in the non-pregnant group are the same. As BDI1 had a strongest affect on BDI2, a regression analysis considering BDI1 as a dependent variable was performed. It was found out that BDI1 score decreases as the level of education increases. In a stepwise regression, we failed to show any other factors to have relation with BDI1 score. There have been studies in which correlation between BDI score and demographic characteristics were evaluated. Demyttenaere et al showed that age, duration of infertility, and numbers of previous IVF attempts are factors affecting BDI score, but there was no data about level of education \[[@B24]\]. On the other hand, Beutel et al reported the correlation between education and depression (Spearsman correlation = -0.15 P\< .05) \[[@B11]\]. This inverse relationship between education and BDI1 score was considerable in our study (r = -0.26). This relationship can be explained by cultural views. It seems that high-educated people have objectives other than fertility to focus on. Data about age and risk of depression is not conclusive. Some studies have reported associations been them \[[@B11],[@B24],[@B25]\], while other studies could not show any relationships between these two \[[@B14],[@B26]\]. We suggested that other variables could act as intermediate variables between age and BDI1. Because of the strength of theses variables, the effect of age was hidden in multivariate analysis. The second important factor was the result of therapy. Apparently, subjects who are at risk of depression are more prone to develop depression at the time of exposure to life events. The third factor that had influence on BDI2 was duration of infertility. The association of BDI score after treatment with duration of infertility in subgroup with failed treatment was showed in another study \[[@B14]\]. In our study, BDI2 is associated positively with duration of infertility. However, the strength of this association is low (r = 0.15). It seems that such a low influence cannot play a significant role in practice. The assumption on whether the association between duration of infertility and post-treatment BDI score is a straight association or depends on intermediate variables such as number of treatment failure experiences needs more studies \[[@B27]\]. The perception of woman about the cause of infertility may change her BDI score. Pressure on infertile women with female factor infertility is high. The risk of depressive symptomatology is lower when a woman thinks that the problem is male factor. Combined factor indicates a severe problem, so it is reasonable that BDI score rises by this diagnosis (Table [3](#T3){ref-type="table"}). This type of cultural view has been showed in countries with family-based societies \[[@B28]\]. We emphasize that a number of meaningful relationships were introduced in other studies which were not mentioned here. Thus, in prospective studies other variables should be included and other more sensitive tools should be used. The other limitation of this study was using BDI for uneducated subjects that might induce a bias. It is clear that designing studies using interviews instead of questionnaires will be more valuable. Conclusion ========== Our findings suggest a plan for screening women who are prone to depression after infertility treatment. It is obvious that there is no significant difference among them before treatment (Table [2](#T2){ref-type="table"}). Answering the following two simple questions can conduct the physician to screen these women. 1\. Is my patient at risk of depression (due to BDI1 score) at the beginning of the treatment? 2\. What will be the result of therapy? Therefore, the physician can predict subjects who may need psychological supportive programs at the beginning of treatment cycles. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AK, AA and MA drafted the manuscript. AK performed the statistical analysis. AK, AAA and FR participated in the study design and coordination. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-244X/5/25/prepub> Acknowledgements ================ We gratefully acknowledge the participation of Dr. Ehsan Akbari Hamed and the infertility ward crew of Shariati Hospital: Nahid Abbassi, Roghaeh Amrollahi, and Mehrnoosh Amini. This study was supported by Vali-e-Asr Reproductive Health Research Center, Tehran University of Medical Sciences.	PubMed Central	2024-06-05T03:55:58.430849	2005-5-24	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173120/", "journal": "BMC Psychiatry. 2005 May 24; 5:25", "authors": [ { "first": "Afsaneh", "last": "Khademi" }, { "first": "Ashraf", "last": "Alleyassin" }, { "first": "Marzieh", "last": "Aghahosseini" }, { "first": "Fatemeh", "last": "Ramezanzadeh" }, { "first": "Ali Ahmadi", "last": "Abhari" } ] }
PMC1173121	Background ========== Duodenal ulcer perforations are a common cause of peritonitis. The classic, pedicled omental patch that is performed for the \'plugging\' of these perforations was first described by Cellan-Jones in 1929 \[[@B1]\], although it is commonly, and wrongly attributed to Graham, who described the use of a free graft of the omentum to repair the perforation in 1937 \[[@B2]\]. In this, a strand of omentum is drawn over the perforation and held in place by full thickness sutures placed on either side of the perforation, and this procedure has become the \"gold standard\" for the treatment of such perforations. However, occasionally, large perforations of the duodenum may be encountered in which there exists the threat of post-operative leakage following closure by this simple method \[[@B3],[@B4]\]. Here, other surgical options such as partial gastrectomy, jejunal serosal patch, jejunal pedicled graft, free omental plug, suturing of the omentum to the nasogastric tube, proximal gastrojejunostomy, or, even, gastric disconnection may be deemed necessary for adequate closure \[[@B3]-[@B8]\]. Very little data is available in literature regarding the definition, incidence, and the management of large perforations of duodenal ulcers. This paper represents our experience with the management of this subset of duodenal ulcer perforations over a period of three years from January 2001 to December 2003. Methods ======= A total of 162 patients underwent emergency surgery for duodenal ulcer perforations at our hospital over a period of three years (January 2001 to December 2003). The case files of all these patients were analyzed, and the patients were sorted into four groups according to the size of the perforation noted intra-operatively -- Group 1 (less than1 cm perforation); Group 2 (1 cm to 2 cm); Group 3 (2 cms to 3 cms); and, Group 4 (more than 3 cms perforation). No cases of anterior and posterior ulcers, or multiple perforations were encountered while reviewing the operative notes. The technique of omentopexy was essentially the same in all the cases -- a total of three sutures were placed onto the normal, healthy duodenum on either side of the perforation, a strand of omentum was placed directly onto the perforation, and the sutures were knotted above this. No attempt was made to close the perforation prior to placing the omentum as a graft. The case files of all the patients were retrospectively analyzed for patient particulars, intra-operative findings, surgery performed, post-operative stay, morbidity and mortality. The groups were then compared with each other in terms of age, leak rates, hospital stay, morbidity, mortality and the surgery performed. Statistical analysis was done using the chi-squareand the t- testby an independent comparison of each group singly against another by a statistician who was blinded to the study. A pvalue of \< 0.05 was taken as significant. It was found that the perforations between 1 cm and 3 cm in size (Groups 2 and 3 as above) behaved in a similar manner statistically, and therefore, the patients these two groups were combined to give a single group. We finally ended up with 3 groups of perforations -- Group A (less than1 cm perforations), Group B (perforations between 1 cm and 3 cms in size), and Group C (more than 3 cms perforation). Results ======= Of the total of 162 patients that underwent emergency surgery for duodenal ulcer perforations at our hospital over three years, there were 148 males (91.36 %) and 14 female (8.64 %) patients, giving a male to female ratio of 10.57 : 1. The average age of the patients was 40.63 years (range 15 -- 82 years), with an almost equal age of occurrence for males (40.52 years) and females (41.78 years). All the patients were divided into three groups as explained above. Group A was deemed to be the small perforation group, Group B was called \'large\' perforations, and Group C, \'giant\' perforations. The majority of patients came under the \'small\' perforation group, but there were 38 patients (23.46 %) with large perforations as per our definition. These patients had a higher age of presentation (47.18 years) than the patients with smaller perforations (39.46 years). Giant perforations, or perforations greater than 3 cms in size were seen only 2 cases, accounting for a small percentage (1.23 %) of all cases seen. When the small perforation group was compared with the larger perforations, it was found that the large perforations had a higher morbidity (x^2^= 37.4503, p\< 0.05), leak rate (x^2^= 4.9117, p\< 0.05), and hospital stay (t value5.117, p\< 0.001) and that this difference was statistically significant. This therefore, lends support to the popular opinion that large perforations have a worse outcome. Overall, the commonest surgery performed was the Cellan-Jones omental patching -- in 119 of the 122 cases in Group A; 30 of the 38 patients in Group B. When the results of omental patch were compared between the two groups, no significant difference was found in the leak rates (x^2^= 2.8698; p\> 0.5) and mortality (x^2^= 1.4732; p\> 0.1), thereby implying that this was an equally effective method for the closure of larger perforations also. Jejunal serosal patch using a loop of the jejunum, and antrectomy (4 cases each) were the other surgeries performed in Group B, when closure with the omentum was thought to be unsafe by the operating surgeon. Five (12.5 %) patients of this group had a leak following closure of the perforation; 3 following omental patch and 1 each after performance of jejunal serosal patch and antrectomy, whereas only 3 cases developed leak in Group A (2 after omental patch and 1 after truncal vagotomy and pyloroplasty). Two cases had \'giant\' perforations extending onto the pylorus -- in one, resection and Billroth II reconstruction was performed, and in the other, jejunal serosal patch. The patient who underwent resection had presented late, and succumbed to septicaemia on the very first post-operative day. The other remained well and was discharged on the 11^th^post-operative day. Overall, the patients with large perforations (Group B) had significantly increased hospital stay, leak rates, and morbidity (Table [1](#T1){ref-type="table"}). The hospital stay was almost double for these patients (13.65 days versus 6.93 days). Although the overall morbidity was 48.76 %, it was much higher in the larger perforations (groups B and C). The common morbidity encountered was chest infections (39 cases), but wound infection (12 cases), biliary leak (08 cases), intra-abdominal abscesses (06 cases), burst abdomen (06 cases), renal failure (02 cases), DIC (04 cases), jaundice and upper gastrointestinal bleeding (01 case each) were also recorded. The mortality in this series was 8.64 % (14 cases), and again, it was significantly higher in perforations more than 1 cm in size (x^2^= 3.8940; p\< 0.05). Table [1](#T1){ref-type="table"} gives the details of all the three groups. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient data ::: Group A -- \'Small\' (Less than 1 cm) Group B -- \'Large\' (1 cm -- 3 cm) Group C -- \'Giant\' (More than 3 cm) ---------------------------------- ------------------------------------------- ---------------------------------------------------------------------- -------------------------------------------------------- Number of cases 122 (75.31 %) 38 (23.46 %) 02 (1.23 %) Average age 39.46 years 47.18 years 37.50 years Male/Female 109 : 13 37 : 1 2 : 0 Average Duration of Symptoms 2.5 days 3.18 days 3.50 days Surgery Performed Omental Patch 119 \\ Pyloroplasty 03 \* Omental Patch 30 \\\* Jejunal Serosal Patch 04 \* Antrectomy 04 \* Antrectomy and Billroth II 01 Jejunal Serosal Patch 01 Post-operative Leak 03 (2.46 %) 05 (13.16 %) \- Morbidity 41 37 01 Post-operative Hospital Stay 6.93 days 13.65 days 6.00 days Mortality 07 (5.74 %) 06 (15.79 %) 01 (50 %) Each \* indicates one post-operative leak ::: Discussion ========== Duodenal ulcer perforation is a common surgical emergency in our part of the world. The overall reported mortality rate varies between 1.3 to nearly 20 % \[[@B9]-[@B11]\] in different series, and recent studies have shown it to be around 10 % \[[@B11]\]. Factors such as advancing age, concomitant disease, preoperative shock, size of the perforation, delay in presentation and operation, have all been defined by various authors to be risk factors for mortality in such a situation \[[@B9]-[@B11]\]. Although the size of a perforation is an important measure in determining the outcome, a review of literature failed to reveal, any accepted definition of either small or giant perforations of duodenal ulcers. Neither could we come across any specific recommendations regarding the management of giant / large perforations, which are said to be \"difficult\" to manage and have anecdotally been associated with high leak rates and mortality. This is in contrast to the well accepted and documented definition of giant duodenal ulcers (more than 2 cms in size), which may or may not perforate, but are usually considered to be an indication for definitive, elective ulcer surgery \[[@B8],[@B12]\]. Commonly, duodenal ulcer perforations are less than 1 cm in greatest diameter, and as such, are amenable to closure by omentopexy \[[@B3]\]. Our experience does seem to validate this, and this subset of \'small\' perforations does seem to have the best outcome. It is the perforations that are larger that have been the cause of much confusion in their definition and management. The size of such \'giant\' sized perforations has arbitrarily been defined by various authors as being greater than 0.5 cms \[[@B7]\], 1 cm \[[@B3],[@B4]\], or 2.5 cms \[[@B6]\] in greatest diameter, but we failed to uncover any specific size in available English language literature beyond which to label these perforations as \"giant\". These perforations are considered particularly hazardous because of the extensive duodenal tissue loss and surrounding tissue inflammation, which are said to preclude simple closure using omental patch, often resulting into post-operative leak or gastric outlet obstruction \[[@B3],[@B4]\]. The tendency to leak may further be aggravated by the high intraluminal pressures, extrusion of the duodenal mucosa through the closure, and, autodigestion by the pancreatic enzymes and bile, thereby further compromising an already sick patient \[[@B13]\]. Our data seems to suggest that based on the size, duodenal perforations can be classified into three main groups (1) smallperforations that are less than 1 cm in size, and have the best outcome; (2) largeperforations, that have a size between 1 cm and 3 cms; and, (3) giantperforations that exceed 3 cm size. The usage of the word \'giant\' for a duodenal perforation should be restricted to such large defects, where omentopexy may be deemed unsafe, and other options may be thought to be necessary. In the absence of any specific definition and guidelines regarding the management of such large / giant perforations in literature, different authors have recommended varied surgical options from time to time, based on their experience and research. These have included resection of the perforation bearing duodenum and the gastric antrum in the form of a partial gastrectomy, with reconstruction as either a Billroth I or II anastomosis, or the more morbid procedure of gastric disconnection in which vagectomy, antrectomy, gastrostomy, lateral duodenostomy and feeding jejunostomy are performed, with restoration of intestinal continuity electively after 4 weeks of discharge \[[@B8]\]. Others have recommended conversion of the perforation into a pyloroplasty, or, closure of the perforation using a serosal patch or a pedicled graft of the jejunum, or, the use of a free omental plug to patch the defect, and even, suturing of the omentum to the nasogastric tube \[[@B3]-[@B8]\]. Proximal gastrojejunostomy and / or vagotomy may be added to these procedures to provide diversion and a definitive acid reducing procedure respectively \[[@B8]\]. However, as can be appreciated, each of these procedures not only prolongs the operating time, but also requires a level of surgical expertise that may not be available in the emergency \[[@B6]\]. In addition, each of these procedures has it own morbidity that may add up significantly to alter the final outcome of the patient, and more importantly, none of them is immune to the risk of leak in the post-operative period, which has been the main concern against performing the omental patch in larger perforations \[[@B3],[@B4]\]. The results of omentopexy in small and large sized perforations in the present series give statistically similar results. The leak rates and mortality of the two groups after omentopexy remain comparable, thereby suggesting that this may be considered as the procedure of choice in all perforations upto a size of 3 cms. The procedure is simple and easy to master, and, avoids the performance of a major resection in a patient who is already compromised. In fact, Sharma et al also reported the success of the omental plug in perforations of duodenal ulcers more than 2.5 cms in size; only, they preferred using a free graft of the omentum rather than a pedicled one \[[@B6]\]. We feel that mobilization of the omentum on its pedicle from the colon, and placement of sutures into the normal duodenum away from the perforation makes the performance of omental patch safe even in the presence of large sized perforations. In the present series, only 2 cases were defined to be \'giant\' according to the size (more than 3 cm) that we have defined -- one underwent antrectomy and Billroth II reconstruction, the other, a jejunal serosal patch. The first patient (antrectomy) succumbed to the ongoing septicaemia on the very first post-operative day, but the other patient survived. This is the group of patients with truly giant perforations who need to be analyzed further to determine the best course of action i.e. resectional versus non-resectional surgery. However, the less number of patients in this group did not allow us to reach any definite conclusion regarding their ideal management. Further study is needed to optimize our efforts to this target group. Conclusion ========== Duodenal perforations should be classified as small, large or giant according to their size encountered at laparotomy. In the emergency setting, such patients are often seriously ill and it is not advisable to perform major surgical procedures on them. The Cellan-Jones omental patch is simple, can be performed in a relatively short time, and remains dependable even for the closure of large sized perforations (i.e. perforations upto 3 cms in size). The addition of a feeding jejunostomy and placement of a tube drain in the Morrison\'s space may offer a further sense of \'\'security\'\' to the operating surgeon, keeping by open the option of maintaining the nutrition of the patient as well as creating a controlled duodenal fistula in case of a post-operative leak. The word \'\'giant\'\' should be reserved only for perforations that exceed 3 cm in diameter. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= SGcarried out acquisition, statistical analysis and interpretation of the data and drafting of the initial manuscript RKconceptualized the paper, and was involved in the interpretation of data, drafting of the manuscript, and revising it critically for the intellectual content RScarried out drafting of the manuscript, and revised it critically for the intellectual content till the final version was reached AKAhelped in the revisions of the intellectual content and gave final approval of the version to be published. All authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2482/5/15/prepub>	PubMed Central	2024-06-05T03:55:58.434016	2005-6-25	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173121/", "journal": "BMC Surg. 2005 Jun 25; 5:15", "authors": [ { "first": "Sanjay", "last": "Gupta" }, { "first": "Robin", "last": "Kaushik" }, { "first": "Rajeev", "last": "Sharma" }, { "first": "Ashok", "last": "Attri" } ] }
PMC1173122	Background ========== The most important phosphate-containing calculi involved in urinary stone disease are carbonate apatite, brushite, and struvite. Overall, phosphate stones account for 12--20% of all urinary stones and rank first in the list of recurrent calculi \[[@B1]\]. The most definitive therapy involving any type of phosphate stone is calculus removal by shock-wave lithotripsy, percutaneous stone removal, or open surgery (especially in children). However, while technologies have successfully treated patients with phosphate calculi, they are associated with injury to the urinary tissue and retained stone fragments have resulted in recurrences. Stone dissolution through chemolysis remains an option for urologists even in the advent of more sophisticated modalities in the treatment of urolithiasis. Chemolysis via acidification of the urine and antibiotic therapy (especially for infection stones) are important adjuvant modalities for phosphate calculi therapy. Hitherto, a many solutions have been tested in the quest for more effective dissolution agents. The most effective solution among them is Renacidin (R), which is a buffer consisting mainly of citrate and gluconate \[[@B2]-[@B4]\]. These solutions are delivered directly into the kidney by ureteric catheter or percutaneous nephrostomy, and are successful in the dissolution of struvite. But the injury to urothelium restrains their wide use. So, solutions with more dissolution efficiency and less side effects are needed urgently and may be used widely. Our in vitro study was designed to evaluate the effectiveness of two new complex solutions (using D-gluconic acid-lactone, D-gluconic-acid and other ingredients which makes them different from R) in the chemolysis of calcium phosphate stones. Material and methods ==================== Stones specimen preparation --------------------------- Eighteen stones composed mainly of calcium phosphate were taken from 18 patients who underwent kidney stone surgery. These stones were large enough (weight range from 0.514--0.928 g) to be fragmented and matched equally into six groups. Each stone was analyzed using the Merckognost at Urinary Stone analysis Kit. The chemical composition of these stones was varied but there was a predominance of calcium phosphate in almost all the stones and calcium oxalate was less than 10% (Table [1](#T1){ref-type="table"}, Table [2](#T2){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Chemical analysis of 18 stones for stone dissolution. There was a predominance of calcium phosphate in almost all the stones and calcium oxalate was less than 10%. ::: Stone% Ca Oxalate NH~4~ PO~4~ Mg Uric Acid Tri-calcium phosphate Calcium oxalate Struvite -------- ---- --------- ------- ------- ---- ----------- ----------------------- ----------------- ---------- 1. 45 2 2 57 7 2. 60 45 10 9 3. 60 10 4 55 10 4. 45 15 45 5 4 5. 35 5 35 6. 80 11 3 65 6 7. 50 10 60 6 8. 40 2 7 25 5 7 9. 5 5 45 15 20 8 10. 15 20 15 40 6 11. 45 5 3 45 10 12. 35 20 2 35 13. 50 20 5 25 14. 40 10 35 15 15. 45 15 35 9 16. 40 15 0.5 35 5 17. 75 5 15 18. 15 5 45 10 15 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Mean chemical composition of 18 stones, showing that there was a predominance of calcium phosphate in almost all the stones and calcium oxalate was less than 10%. ::: Composition ± s(%) --------------- ------------- Ca^++^ 42.5 ± 20.7 Oxalate 9.6 ± 6.2 NH~4~ 1.5 ± 2.1 PO~4~ 40.1 ± 13.9 Mg 1.7 ± 4.9 Uric acid 3.3 ± 4.9 Cystine 0.0 ± 0.0 Ca^++^oxalate 3.4 ± 3.7 ::: Preparation of solvent ---------------------- Six different solvents were used for chemolysis of phosphate stones. The molecular formula and the molecular weights of some drugs used in these solvents are provided in Table [3](#T3){ref-type="table"}. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Molecular formulae and molecular weights of chemical drugs used in our experiment ::: Experimental drug Molecular formula Molecular weight -------------------------------- -------------------------------- ------------------ Citric acid C~6~H~6~O~7~·H~2~O 210.14 Sodium citrate Na~3~C~6~H~6~O~7~·2H~2~O 294.11 D-Gluconic Acid Lactone C~6~H~10~O~6~ 178.1 Magnesium citrate Mg~3~(C~6~H~5~O~7~)~2~·14H~2~O 703.4 Magnesium Carbonate (MgCO~3~)~4~·Mg(OH)~2~·5H~2~O 485.8 Disodium EDTA C~10~H~14~O~8~N~2~Na~2~·2H~2~O 372.24 Potassium citrate, tribasic K~3~C~6~H~5~O~7~·H~2~O 324.34 Potassium phosphate, monobasic KH~2~PO~4~ 136.09 ::: 1\. Physiologic sodium chloride solution (Phys). 2\. Artificial urine (Art, pH5.7, according to Griffith et al \[[@B5]\]) consisting of urea (25 g/L), sodium chloride (4.6 g/L), potassium-dihydrogen-phosphate (2.8 g/L), sodium sulfate (2.3 g/L), potassium chloride (1.6 g/L), ammonium chloride (1.0 g/L), calcium chloride dehydrate (0.1 g/L), sodium oxalate (0.05 g/L), and sodium citrate (0.01 g/L), dissolved in 1000 ml. distilled water. 3\. 0.03 M disodium ethylenediaminetetraacetic acid buffered to pH 8.5 with triethanolamine (EDTA, pH 8.5) 4\. 10% Renacidin (R, pH 3.9), consisting of citric acid (28.2 g), gluconic acid (5.0 g), calcium carbonate (1.0 g), magnesium bicarbonate (14.5 g), citrate magnesium (2.5 g), dissolved in 1000 ml. distilled water. 5\. Test solution 2 (S2, pH 4.0), prepared using citric acid (18.0 g), citrate magnesium (1.0 g), calcium carbonate (0.5 g), magnesium carbonate (7.5 g), D-gluconic acid-lactone (3.0 g), dissolved in 100 ml. distilled water. The solution was kept at 37? for 3 days, then diluted with 140 ml. distilled water. The concentrations of these ingredients are shown in Table [4](#T4){ref-type="table"}. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Concentrations of chemical drugs contained in S1 and S2 ::: Chemical drug Concentration (mmol/L) ----------------------------- ------------------------ Citric acid 357.1 Magnesium citrate 5.9 Magnesium Carbonate 64.3 D-Gluconic Acid Lactone 70.2 Calcium carbonate 20.8 D-Gluconic Acid 63.8 Sodium citrate 302.7 Potassium citrate, tribasic 302.7 Calcium chloride, anhydrous 454.5 Calcium carbonate is difficult to be dissolved in water, but in acidic solutions of S1 or S2, it can be dissolved thoroughly. So we give the concentration of Calcium as above. In diluted solutions these cations were diluted correspondingly, such that their pH did not change. ::: 6\. Test solution 1 (S1, pH 3.9), prepared using citric acid (18.0 g), citrate magnesium (1.0 g), calcium carbonate (0.5 g), magnesium carbonate (7.5 g), D-gluconic-acid (3.0 g), dissolved in 100 ml. distilled water, and processed as described for S2. The concentrations of these ingredients are shown in Table [4](#T4){ref-type="table"}). We also altered the pH of these solutions to 4, 5, 7, and 8.5 to test their effectiveness and added different cations (Na^+^, by adding 302.7 mmol/L citrate sodium 10 ml. into 100 ml. of the initial solution; K^+^, by adding 302.7 mmol/L citrate potassium 10 ml. into 100 ml. of the initial solution; Ca^2+^, by adding 454.5 mmol/L calcium chloride anhydrous 10 ml. into 100 ml. of the initial solution). S1 and S2 were formulated according to different dilutions (1:1, 1:2, 1:3, 1:4). The pH of these changed solutions had no detectable variation. After stones were immersed in these solvents for 24, 48 and 72 h, reduction in weight, percentage weight loss, and dissolution rate were determined. Results were analyzed using the Multiple Comparisons Test with the alpha set at 0.05. Stones remaining after the dissolution period were weighed (gm), and final weights were applied to the following formula: ![](1471-2490-5-9-i2.gif) The formula used to determine the dissolution rate was: ![](1471-2490-5-9-i3.gif) ### Chemolysis Every stone was placed in a vessel containing solvents, which were held at a constant temperature of 37? using a thermostat. Solvents were exchanged 2 times per day at 6 ml/100 mg per stone each time. The pH values were constantly controlled by a pH meter (Mettler Toledo 320-S, sensitivity 0.01). The weight of the calculus was measured continuously by Electronic Semi-micro-, Analytical and Precision Balances (Sartorius BP211D-OCE, sensitivity 0.01 mg) and documented online with a computer system (PC windows 486, Software Wedge for window, version 1.1). The analysis represents six stones for each experiment, repeated three independent times. Results ======= Urological calcium phosphate calculi were poorly dissolved by Phys and Art, and they had a low dissolution rate in EDTA at pH 8.5. The most effective solutions were R, S1 and S2 with 24 h mean dissolution rates: 5.05 ± 0.15 mg/hr (S1), 4.52 ± 0.64 mg/hr (S2), 4.53 ± 0.46 mg/hr (R); 72 h mean dissolution rates: 5.75 ± 0.44 mg/hr (S1), 5.2 ± 0.63 mg/hr (S2) and 4.55 ± 0.46 mg/hr (R) (![](1471-2490-5-9-i1.gif) ± s, P\< 0.01, R, S1 and S2 vs Phys, Art and EDTA; P\< 0.05, S1 vs R, LSD-test). The mean percentage weight loss at 72 h was: 40.5 ± 3.67 % (R), 52.1 ± 15.75 % (S1) and 44.4 ± 7.37 % (S2) (![](1471-2490-5-9-i1.gif) ± s, p\< 0.01 R, S1 and S2 vs Phys, Art and EDTA, LSD-test). Figure [1](#F1){ref-type="fig"} shows the average dissolution rates of the 6 different solutions at the end of 72 h. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Dissolution rates of phosphate calculi in vitro at 24 and 72 h using different solutions.Phys = physiologic sodium chloride solution, pH 7.0. Art = artificial urine, pH 5.7. EDTA = 0.03 M disodium EDTA+TEA, pH 8.5. R = renacidin, pH 4.0. S1 = test solution 1, pH 4.0. S2 = test solution 2, pH 3.9. ^a^P\< 0.01 vs Phys and Art. ^b^P\> 0.05 vs EDTA, Phys and Art. ^c^P\< 0.05 vs R·![](1471-2490-5-9-i1.gif) ± s·n = 18 for each group. For statistics, see results section. ::: ![](1471-2490-5-9-1) ::: Table [5](#T5){ref-type="table"} Shows mean weight loss and mean percentage weight loss of stones dissolved by the 6 solutions at the end of 24 and 72 h. (![](1471-2490-5-9-i1.gif) ± s, ^a^P\< 0.01 Phys, Art and EDTA vs R, S1 and S2; ^b^P\< 0.05 EDTA vs S1, S2 and R.^c^P\< 0.05 R vs S1). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Mean weight and percentage weight decrease after 24 and 72 h between groups dissolved with 6 different solutions. Phys = physiologic sodium chloride solution, pH 7.0. Art = artificial urine, pH 5.7. EDTA = 0.03 M disodium EDTA+TEA, pH 8.5. R = renacidin, pH 4.0. S1 = citrate complex 1, pH 4.0. S2 = citrate complex 2, pH 3.9. ± s, ^a^P\< 0.01 vs R, S1 and S2.^b^P\< 0.05 vs S1, S2.^c^P\< 0.05 vs S1. n = 18 for each group. For statistics, see results section. ::: Groups Weight loss (mg) Percentage weight loss (%) -------- ------------------ ---------------------------- ----------------- ------------------ Phys 3.52 ± 0.73^a^ 18.15 ± 13.15^a^ 0.00 ± 0.00^a^ 1.33 ± 1.53^a^ Art 2.48 ± 0.37^a^ 7.44 ± 1.9^a^ 0.00 ± 0.00^a^ 0.00 ± 0.00^a^ EDTA 43.78 ± 6.23^ab^ 153.6 ± 20.84^ab^ 4.67 ± 2.08^ab^ 19.00 ± 5.19^ab^ R 114.4 ± 8.92^c^ 346.14 ± 26.56^c^ 13 ± 1.35^c^ 40.5 ± 3.67^c^ S2 127 ± 23.09 408.11 ± 62.94 13.5 ± 2.76 44.4 ± 7.37 S1 138.5 ± 31.09 439.5 ± 102.00 5.77 ± 1.29 52.1 ± 15.75 ::: The dissolution rates were pH dependent. Even Phys and Art were effective to some extent at pH 4. The addition of 1 mol/L sodium hydroxide 10 ml into S1 or S2 100 ml, though not enough to lead to a detectable pH rise, would make the solutions cloudy and greatly reduce their effectiveness. At pH 5.0 they were nearly ineffective, as also observed at pH 7.0, 8.5. However, the dissolution rate of EDTA increased at an elevated pH value. At pH 8.5, EDTA approached a maximum dissolution rate with 72 h mean 1.56 ± 0.05 mg/hr (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Effectiveness of six solvents at different pH (mg/hr).Phys = physiologic sodium chloride solution, Art = artificial urine, EDTA = 0.03 M disodium EDTA+TEA, R = renacidin, S1 = test solution 1, S2 = test solution 2. ::: ![](1471-2490-5-9-2) ::: The addition of Na^+^, K^+^or Ca^2+^(302.7 mmol/L Sodium citrate 10 ml, 302.7 mmol /L Potassium citrate 10 ml or 454.5 mmol/L Calcium chloride dihydrate 10 ml into initial solution 100 ml, respectively) caused no significant change in dissolution rate. Diluted solutions demonstrated an interesting result. Diluted twice, S1 and S2 were more effective than their initial solutions with 72 h mean dissolution rates: 6.04 ± 1.36 mg/hr (S1), 5.60 ± 1.23 mg/hr (S2) (![](1471-2490-5-9-i1.gif) ± s, P\< 0.05, S1 and S2 vs R). The effectiveness of three times diluted S1 or S2 was the same as the initial solution. When diluted further, the effectiveness was reduced gradually until nearly 0 after solutions were diluted five times (Figure [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Concentration and effective of S1 and S2 ::: ![](1471-2490-5-9-3) ::: Discussion ========== Chemolysis is useful for eliminating cystine stones as well as for cases in which lithotripsy or endourology is considered to be difficult or risky. It proved to be a useful method for reducing staghorn stones before performing lithotripsy \[[@B6]\]. The type of chemolytic solution is dependent upon the composition of the stone and must be regarded as an effective adjuvant treatment \[[@B7]\]. Calcium phosphate can be dissolved with Suby or R, but the treatment is often tedious and time consuming. Calcium oxalate, the major urinary stone component, cannot be dissolved by these solutions. EDTA and other strong calcium chelators cannot be used because of their local toxicity. Certain enzymes can digest the organic matrix of the stone \[[@B8]\]. The first attempts to dissolve calcium stones were done by Hellstrom and Albright in 1930. They used citric acid for phosphate calculi, but this proved irritating to tissues. Suby and Albright modified the solution by adding magnesium oxide and sodium carbonate (Suby\'s G solution) \[[@B9],[@B10]\] to reduce injury to the rabbit bladder mucosa even though adding this cation reduced the speed of dissolution of struvite Mg (NH~4~) PO~4~. A contradictory conclusion was been made by a Dutch study \[[@B11]\] that magnesium in R promotes stone dissolution by cation exchange with calcium in apatite \[[@B12]\]. Our S1 and S2 solutions were similar to R although actual formulas were different. First, we added D-gluconic acid-lactone or D-gluconic-acid as new chelators which bind with calcium well by their special trait. Second, we used new ingredients and ingredients at different weights. The higher density of our solutions than R may be one of the factors which enhanced stone dissolution rates, but it may prevented further dissolution. When diluted twice, the solutions may have gained more space to accommodate cations than the initial solutions and achieve better effectiveness (P\< 0.05, S1 and S2 vs R). The further diluted solutions had a lower concentration suggesting that less ingredient took part in the reaction thus accounting for a lower dissolution rate. So, we think twice diluted S1 and S2 may be more useful than their initial solutions. The dissolution of the majority of the stones in this study can be attributed on the basis of their reaction equation: 3Ca^+2^+ 2 Citrate→CaCitrate→Ca^2+^+ CaCitrate The citrate in the solution binds with the calcium component of the stone producing calcium citrate, thus preventing crystallization. The dissolution process may be brought about by the combination of the above reaction. As the solubility of calcium phosphate is very pH dependent, the acidification of the urine by the incorporation of citric acid produce a pH between 3.0--4.0, the ideal pH for dissolution as described by Albright et al \[[@B9],[@B10]\]. When the pH is elevated, the citrate will react with cations thus losing its effectiveness. On the other hand, excessive alkalization may lead to the formation of calcium phosphate calculi. Added minor quantities of cations (Na^+^, K^+^or Ca^2+^) in S1 and S2 apparently causes no significant change to their dissolution rate. Conclusion ========== The study has shown that solutions S1 and S2 can dissolve calcium phosphate stones effectively in vitro at a precise dilution of their chemical components. Based on these findings, it is suggested that S1 and S2 may become useful complements to modern techniques of stone fragmentation such as extracorporeal shock wave lithotripsy and percutaneous surgery. Their roles may be suited for the treatment of infectious stones with a CaOx content of less than 10%. However, their safety profiles should be further investigated in order to support their use in subsequent human trials. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Xiang-bo Zhang participated in the design of the study, in the sequence alignment, carried out the experiment, performed the statistical analysis and drafted the manuscript. Zhi-ping Wang conceived of the study, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Jian-min Duan, Jian-zhong Lu, Bao-liang Ma, participated in the sequence alignment. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2490/5/9/prepub> Acknowledgements ================ We are indebted to Zhi-ping Wang, Professor and director of the institute of urinary, the 2nd Hospital of LanZhou University, for his help in conducting the study and for reviewing the manuscript. We thank Jian-min Duan, Jian-zhong Lu, Bao-liang Ma, staff at the institute of urinary, the 2nd Hospital of Lanzhou University, for generously supplying the room for our experiment.	PubMed Central	2024-06-05T03:55:58.435674	2005-5-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173122/", "journal": "BMC Urol. 2005 May 22; 5:9", "authors": [ { "first": "Zhang", "last": "Xiang-bo" }, { "first": "Wang", "last": "Zhi-ping" }, { "first": "Duan", "last": "Jian-min" }, { "first": "Lu", "last": "Jian-zhong" }, { "first": "Ma", "last": "Bao-liang" } ] }
PMC1173123	Background ========== Aseptic loosening is the most common cause for long-term failure of total hip replacement (THR). The success rate after revision surgery is much lower than that after the primary operation. Wear, which contributes to aseptic loosening \[[@B1]-[@B4]\], has been shown to correlate with the contact pressure of the UHMWPE acetabular cup \[[@B5],[@B6]\]. An increase of the abduction angle of the acetabular cup has been shown to increase contact pressure and wear \[[@B7],[@B8]\]. Different geometrical cup designs, polyethylene thicknesses and clearances between the cup and head have been demonstrated to modulate implant survival \[[@B2],[@B7],[@B9]-[@B15]\]. However, the combined effect of these parameters on polyethylene stress is still unclear. For instance, the different geometries of the acetabular cups may modify the contact pressure in different ways depending on the abduction angle. Parameters reducing contact pressure reduce plastic deformation and wear, and thus may diminish the risk for aseptic loosening. Pre-clinical testing of hip implants prior to marketing is known to be important \[[@B16]\], since failure to perform proper testing may lead to unsatisfactory clinical results. Therefore, the geometries and orientations of acetabular cups should be optimized. Finite element (FE) modeling enables the evaluation of stress distribution throughout the THR prostheses \[[@B8],[@B17]\] and an assessment of geometries and material properties which would be difficult or time-consuming to test experimentally. In this study, two UHMWPE acetabular cups with different geometries and good long term clinical records, according to Scandinavian hip registers, were studied \[[@B18],[@B19]\]. However, the cups belong to the same hip prosthesis system and their individual survivorships are not described separately in the registers. FE modeling and cyclic testing of hip implants in a simulator were used to test the hypotheses that the geometry, thickness and abduction angle of the cup, as well as the clearance between the femoral head and the cup could affect acetabular cup stress upon loading. Information from this study can be used for the future development of THR implants prior to their surgical installation. Methods ======= Laboratory tests ---------------- Two UHMWPE acetabular cups\* with different designs (Waldemar LINK GmbH & Co., Hamburg, Germany: IP, n = 4 and Lubinus eccentric (with a snap-fit primary locking mechanism), n = 8) were tested experimentally under a cyclic load \[[@B20]\]. Axial loading was used to investigate the plastic deformations occurring in the cups and to minimize wear. Before the tests, the cups were fixed in stainless steel holders with bone cement (Palacos^®^R-40 cum Gentamicin, Schering-Plough Europe, Brussels, Belgium). To speed up the experimental tests, two combinations of head, cup and cement were tested at the same time which led overall to six tests (Table [1](#T1){ref-type="table"}, Fig. [1](#F1){ref-type="fig"}). Testings were carried out using a dynamic tester (Instron 8874, Instron Corporation, Canton, MA) at a 5 Hz frequency for 5 million cycles. The load profile was a Paul gait curve with a peak load of 3 kN, and abduction angles of 45° and 60° were used. Diluted bovine serum supplemented with EDTA and antibacterial agents was used as lubricant \[[@B21]\]. The serum was filtered through a 0.2 μm filter and had a total protein content of 25 mg/ml. Temperature (37°C) and pH (7) were regularly recorded. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of the parameters obtained from experimental tests and numerical simulations. ::: Experimental tests Finite element analyses -------------------------------------- ----- -------------------- ------------------------- ----------- ----------- ----------- ----------- Compression (μm) 45° Mean: 115 (n = 1) Mean: 129 (n = 2) 98 132 \- \- Range: 125--132 60° Mean: 123 (n = 1) Mean: 140 (n = 2) 107 133 \- \- Range: 129--150 Penetration rate (μm/million cycles) 45° Mean: 11.8 (n = 2) Mean: 9.5 (n = 4) \- \- \- \- Range: 11.6--11.9 Range: 8.9--10.2 60° Mean: 14.7 (n = 2) Mean: 11.2 (n = 4) \- \- \- \- Range: 14.2--15.2 Range: 10.0--11.9 Von Mises stress 45° \- \- Fig. 3 Fig. 3 Fig. 3 Fig. 3 60° \- \- \" \" \" \" Contact pressure and/or area 45° Fig. 6 Fig. 6 Figs. 4-8 Figs. 4-8 Figs. 4-8 Figs. 4-8 60° \- \- \" \" \" \" 45° and 60° refer to the abduction angle. Experimental maximum compressions were analyzed from the initial loading cycles for two cup-cement combinations (12 acetabular cups, 6 experimental tests) at the same time (Fig. 1). Penetration rates of the cups, i.e. plastic deformations, were calculated after 5 million loading cycles. ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Hip implant simulator and tested acetabular cups. Two types of acetabular cups (up right, IP and Lubinus eccentric) were tested in a simulator (left) under cyclic axial loading. Two cups, fixed in a metal backing with bone cement, were tested at the same time. ::: ![](1475-925X-4-32-1) ::: During the tests, compressions of whole measurement systems, i.e. compressions of femoral heads into the cup-cement combinations during each cycle, were recorded. Maximum compression values during the initial cycle were used for the validation of the finite element models. As the compression was applied according to Fig. [1](#F1){ref-type="fig"}, the final value for one cup-cement combination was approximated to be half of the recorded value, and consequently six compression values were obtained (Table [1](#T1){ref-type="table"}). After 5 million cycles, the penetration rates (μm / million cycles) of the acetabular cups, i.e. plastic deformations, were analyzed using a coordinate tester Dea Global C 091508 (Dea, Tourin, Italy). The penetration rates were evaluated as the displacement of the center of the spherical surface of the cups, which were mathematically estimated after the cyclic tests. Finite element modeling ----------------------- The femoral head-acetabular cup-cement complexes, used in experimental tests, were processed for finite element analyses. In addition, two new designs were reconstructed to investigate the effect of clearance and polyethylene thickness on the contact pressure. Thus, altogether four 3-D meshes consisting of 15834 -- 16072 hexahedral elements (Fig. [2](#F2){ref-type="fig"}) were created: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Finite element meshes of the present study. Four acetabular cups, cement mantle and femoral head were reconstructed by using hexahedral elastic elements. ::: ![](1475-925X-4-32-2) ::: 1\) Lubinus eccentric (inner diameter of acetabular cup = 28.5 mm, experimentally tested), 2\) Lubinus concentric, (inner diameter of acetabular cup = 28.5 mm, virtual design), 3\) IP1 (inner diameter of acetabular cup = 28 mm, experimentally tested), 4\) IP2 (inner diameter of acetabular cup = 28.5 mm, virtual design). The outer diameter of all acetabular cups was 52 mm and the diameter of the femoral head was 28 mm. Lubinus concentric was reconstructed to analyse differences between eccentric and concentric cups and, consequently, the effect of polyethylene thickness. The effect of clearance was studied by increasing the inner diameter of the IP cup (IP1 → IP2) to correspond to the diameter of Lubinus eccentric/concentric. Young\'s moduli (E) for femoral head, acetabular cups and bone cement, as determined experimentally, were 193 GPa, 0.69 GPa and 2.74 GPa, respectively. A friction coefficient of 0.05 was assumed for the contact between acetabular cup and femoral head \[[@B22]\]. The cup-cement interface was fixed and the boundary constaint was applied for the outer surface of the cement layer to simulate the metal backing used in the experimental tests. First, the compressions of the femoral heads into the acetabular cup -- cement combinations were analyzed using 3 kN loading force and compared with corresponding experimental findings. Second, the effect of abduction angle on the stresses of acetabular cups was investigated with different loading forces in order to simulate different body weights performing regular daily activities and exceptionally high impact loads. In addition to the experimentally used abduction angles of 45° and 60°, simulations were also conducted using an angle of 80°. Von Mises stress distributions in the cups and contact pressures both at cup-head and cup-cement interfaces were studied. Abaqus code (v6.3, Hibbitt, Karlsson & Sorenssen, Inc., Pawtucket, RI, USA) was used for the FE simulations. Results ======= Experimentally, the maximum compressions of the femoral heads during the initial loading cycle using the IP and Lubinus eccentric acetabular cups were 115 μm and 129 μm, respectively, at the 45° abduction angle (Table [1](#T1){ref-type="table"}). The corresponding values obtained in the FE simulations were 98 μm and 132 μm. At the abduction angle of 60°, experimental maximum compression values during the initial cycle using the IP and Lubinus eccentric cups were 123 μm and 140 μm, respectively. The corresponding values obtained in the FE simulations were 107 μm and 133 μm. The measured penetration rates of the cups were 11.8 μm (11.6 -- 11.9) and 9.5 μm (8.9 -- 10.2) per million cycles for IP and Lubinus eccentric, respectively, at the abduction angle of 45° (Table [1](#T1){ref-type="table"}). The corresponding values at the 60° angle, 14.7 μm (14.2 -- 15.2) and 11.2 μm (10.0 -- 11.9), were significantly higher (p \< 0.05, Wilcoxon signed ranks test, n = 6 + 6). The penetration rates were significantly (p \< 0.05, Mann-Whitney U-test, n = 4 + 8) higher for IP than for Lubinus eccentric. FE simulations disclosed that there were different stress distributions (von Mises stress) in Lubinus eccentric and IP cups (Fig. [3](#F3){ref-type="fig"}). Due to the varying geometries, the direction of the stress was different. The decrease of the thickness of the Lubinus acetabular cup (eccentric versusconcentric) did not affect the stress distribution, but the increase in the gap between the head and the IP cup had a major influence on the von Mises stresses. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Stress distribution in the acetabular cups. Von Mises stresses of a) Lubinus eccentric, b) IP1, c) Lubinus concentric and d) IP2 models at 45°, 60° abduction angles. ::: ![](1475-925X-4-32-3) ::: FE simulations showed that the abduction angle (45° -- 80°) had only a minor influence on the contact pressure between the head and the Lubinus eccentric cup (Figs. [4a](#F4){ref-type="fig"} and [5a](#F5){ref-type="fig"}). In contrast, the contact pressure on the surface of IP cup increased notably as a function of the abduction angle (Figs. [4b](#F4){ref-type="fig"} and [5a](#F5){ref-type="fig"}). However, the larger contact area between the IP cup and head induced lower contact pressure at low loading forces on the surface of IP compared to the corresponding situation with Lubinus eccentric (Fig. [6](#F6){ref-type="fig"}). At high loads, the edge of the IP acetabular cup experienced very high contact pressures, especially at high abduction angles (60° and 80°). The difference in the contact pressure between the eccentric and concentric Lubinus models at the inner surface of the acetabular cup was not significant (Figs. [4a, c](#F4){ref-type="fig"} and [5](#F5){ref-type="fig"}). As the increase of the radius of the IP acetabular cup from 28 mm to 28.5 mm reduced the contact area, the peak contact pressure increased (Figs. [4b, d](#F4){ref-type="fig"} and [5](#F5){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Contact pressure at the inner surface of acetabular cups at 45° 60° and 80° abduction angles.a) Lubinus eccentric, b) IP1, c) Lubinus concentric and d) IP2. ::: ![](1475-925X-4-32-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Peak contact pressure at the cup-head interface as a function of loading force and abduction angle. a) Lubinus eccentric and IP1, b) Lubinus concentric and IP2. ::: ![](1475-925X-4-32-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Contact area between humeral head and acetabular cup. Experimentally analyzed (a) and numerically simulated (b) contact area between Lubinus eccentric and IP acetabular cup and femoral head. After cyclic tests of hip implants in a simulator, the surfaces of UHMWPE cups were covered with a thin layer of diluted colourant (Tuschierpaste blau, Emil Alberts GmbH, Gevelsberg, Germany), and femoral heads were positioned in contact with the acetabular cups in the loading direction. After removing the femoral heads, the presence of blue color indicated the contact area. The contact area in the FE-simulations was analyzed under 3 kN axial load which was the maximum load in the experimental cyclic tests. ::: ![](1475-925X-4-32-6) ::: At the interface between the acetabular cup and the cement, the contact pressure increased as a function of abduction angle in all models (Figs. [7](#F7){ref-type="fig"} and [8](#F8){ref-type="fig"}). Due to the larger contact area between cup and head and the smaller deformation of the cup, the contact pressure remained lower on the outer surface of the IP cup at all abduction angles (Figs. [7a, b](#F7){ref-type="fig"} and [8a](#F8){ref-type="fig"}). A decreased thickness of the acetabular cup (Lubinus eccentric → concentric) (Figs. [7a,c](#F7){ref-type="fig"} and [8](#F8){ref-type="fig"}) and the increased clearance between IP cup and femoral head (Figs. [7b,d](#F7){ref-type="fig"} and [8](#F8){ref-type="fig"}) elevated contact pressures on the outer surfaces of the cups. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Contact pressure at the outer surface of acetabular cups at 45°, 60° and 80° abduction angles. a) Lubinus eccentric, b) IP1, c) Lubinus concentric and d) IP2 ::: ![](1475-925X-4-32-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Peak contact pressure at cup-cement interface as a function of loading force and abduction angle. a) Lubinus eccentric and IP1, b) Lubinus concentric and IP2. ::: ![](1475-925X-4-32-8) ::: Discussion ========== In the present study, 3-D finite element analyses and experimental cyclic tests were used for the estimation of the combined effects of acetabular cup geometry and orientation on the stresses in cemented UHMWPE. It was found that IP and Lubinus eccentric acetabular cups, manufactured by Link^®^, were subjected to considerably different stress distributions and contact pressures both on the inner and outer surfaces of the cups. The abduction angle and thickness of the cup, as well as the clearance between the femoral head and the cup, contributed significantly to stress experienced by the acetabular cup. Experimental and theoretical compressions during one loading cycle were consistent with each other. The slight discrepancies (2 -- 15%) were probably due to the fact that the processing of UHMWPE induces uncertainties in the diameters of the cups, whereas in the finite element simulations, optimal diameters, supplied by the manufacturer, were used. On the other hand, the initial fit of the head-cup-cement complex may not have been perfect in all of the experimental configurations, and this may have induced higher compression values. The experimentally determined penetration rates of the present study supported the conclusions based on the FE analyses. The experimental tests showed that the penetration rate increased significantly as a function of the abduction angle. This was consistent with the FE analyses which suggested that the contact pressure increased as a function of the abduction angle, even though this increase with the Lubinus model was relatively low. Even though the peak contact pressure remained lower on the inner surface of the IP acetabular cup compared to that on the Lubinus eccentric when simulating loading forces of less than 3 kN, the experimentally determined penetration rate was faster for the IP cup. The contact area between the IP cup and the femoral head was larger than that between Lubinus eccentric and the head (Fig. [6](#F6){ref-type="fig"}), which probably induced this discrepancy. The minor clearance and consequently high contact area have been shown to increase also volumetric wear \[[@B10],[@B15]\]. Previously, in simulator tests, wear rates of Lubinus eccentric cups were lower than those of IP cups and some other models \[[@B23]\]. This advantageous effect might be due to the formation of a protective protein layer on the cup surface as a result of the more closed lubrication environment in the Lubinus eccentric design. The behavior of contact pressure and penetration rate of the present study were consistent with earlier studies \[[@B8],[@B9],[@B12]\]. Jin et al. (1999) investigated contact pressure and area in eight combinations of femoral head and UHMWPE cup, and found similar behavior of contact pressure as a function of load as observed in the present study \[[@B9]\], although their model was 2-D axisymmetric and was not affected by the edge of the cup. Patil et al. (2003) studied the effect of abduction and anteversion angle of acetabular cup on the contact stress and wear of UHMWPE \[[@B8]\]. FE simulations of their study showed that peak contact stresses increased as a function of abduction angle, as was also found in the present study. Wear rates after testing in the hip simulator and linear head penetrations of the clinical study \[[@B8]\] as well as plastic deformations of the cups in the present study supported those findings. Some studies have suggested that the cup orientation plays minor role on the polyethylene wear, but more important is to determine the shape of the wear distribution \[[@B10],[@B24]\]. In the present study, the distribution of von Mises stress and contact pressure were highly influenced by the cup geometry. In general, slight differences between the previous studies and the present study are probably due to differences in the acetabular cup geometries, material properties and the loading protocols used in the experimental tests and analytical or numerical analyses \[[@B8],[@B25],[@B26]\]. Abduction angle clearly differentiated the acetabular cups. Lubinus eccentric seemed to be relatively forgiving in dealing with the variations in the abduction angle, whereas at 60° and 80° abduction angles, the IP cup experienced high peak contact pressures, especially at high loading forces. Evidently, geometrical factors were behind this kind of behaviour. The snap-fit primary locking mechanism in Lubinus cup eliminated the effect of the abduction angle. In this model the contact area between the ball and cup did not change as a function of the abduction angle. This minimizes the variations in the contact pressure and indicates that the survival of Lubinus acetabular cup may not be so operation sensitive. On the other hand, even though the geometry of IP cup permits a large range of motion, the contact area between IP cup and head changed as a function of the abduction angle, modifying the contact pressure. Furthermore, the edge of the IP cup was the most critical area in elevating the contact pressure, pointing to an increased risk for the cup failure \[[@B27]\]. The thickness of the acetabular cup and the clearance between cup and femoral head may have significant clinical relevance \[[@B9]-[@B12],[@B14],[@B15]\]. For instance, in the present study, a 13% decrease of the thickness at the upper part of the acetabular cup (Lubinus eccentric versusconcentric) increased the peak contact pressure at the cup-cement interface at a 45° abduction angle by 20% (3 kN load). Consequently, eccentricity may decrease the risk for cup rotation. The increase of the clearance between IP acetabular cup and femoral head from 0 to 0.25 mm elevated peak contact pressure at cup-head interface by 27% and at the cup-cement interface by 65% (3 kN load). Also earlier studies have shown that the increased clearance and decreased thickness raise peak contact pressure \[[@B9],[@B12]\]. These percentual values suggest that rather minor changes in the geometrical parameters of UHMWPE acetabular cups may have significant consequences (e.g. rotation, wear, loosening) in long-term use. In the present study, all of the materials and boundaries were kept similar to permit a comparative evaluation of acetabular cups with different geometrical designs. In the clinical setting, however, the bone support and joint capsule will affect the stresses to which the acetabular cups are subjected \[[@B28]\]. Also, the contact area between the ball and the cup increases during the so-called running-in period, which typically lasts less than a year \[[@B29]\]. The bearing surface of the polyethylene accommodates to the head radius through deformation, wear and creep, so that the effective clearance approaches zero. Therefore, the contact pressure and fracture risk decrease substantially. After the first year, the penetration rate of the head into the cup is fairly constant \[[@B29],[@B30]\]. In long-term clinical use, this phenomenon may modify the behavior of the acetabular cups tested in the present study. Finally, cement fixation was assumed to extend up to the edge of the acetabular cups. However, in the clinical situation, the upper edge of the cup is not always fully supported by the cement mantle, especially at low abduction angles (\<45°), inducing variations in the stress and wear concentrations of the cups. A previous finite element model with elasto-plastic polyethylene cup proposed a perfect plastic yield after 8 MPa von Mises stress \[[@B15]\], which is lower than the maximum von Mises stress observed in the present study. However, the radiation-crosslinked UHMWPE tested in this study has a yield strength of 20 MPa or more (ASTM F648) \[[@B31]\] and the von Mises stress of IP cup at the abduction angles of 45° and 60° did not exceed 20 MPa under any load. In this study the cup was assumed to be elastic to enable analysis of potential peak stresses also at very high impact loads, not only during normal walking. Under low loads (\< 3 MPa), the von Mises stress did not exceed 20 MPa in any cup design. In addition, if von Mises stresses are calculated for the IP acetabular cup at the 3 kN load, which was the maximum value used in the experimental tests of the present study, only approximately 8 MPa peak von Mises stress is reached at an abduction angle of 45°. Plasticity, nonlinear elasticity or viscoelasticity \[[@B8]\] of UHMWPE would have changed absolute pressure values, but not the conclusions of the present study. FE modeling provides a rapid and inexpensive estimate of implant-related factors (e.g. geometry of the acetabular cup) and surgery-related factors (e.g. abduction angle of the implant), but the final judgement is based on laboratory tests and clinical data. The hip arthroplasty registers \[[@B18],[@B19]\], used to obtain clinical reference data, do not differentiate between Lubinus eccentric and IP acetabular cups, but the present study points to differences in the performances of these cups. The results of this study indicate that it would be beneficial if cups of different designs were recorded separately in the registers, rather than combining cups from a single hip system. All experimental and numerical data presented in this study address the importance of optimizing the geometry and orientation of the acetabular cup before and during the operation, respectively. These are factors which may affect significantly failure of THR. Conclusion ========== FE analyses of the present study showed that the geometrical design, thickness and abduction angle of acetabular cup, as well as the clearance between cup and head modify the mechanical stresses experienced by UHMWPE. These factors should be taken into account in the future development of THR and the FE technique is a powerful tool for this purpose. Authors\' contributions ======================= RKK was involved in the design of the study, developed the FE models and calculated numerical predictions, as well as drafted the manuscript. AK participated in the design of the study, experimental tests and data analyses, as well as in the statistical analyses. YTK participated in the design of the study and coordination as well as helped in the manuscript preparation. SSS was involved in conceiving the study, coordination and manuscript preparation. RL participated in conceiving the study, coordination, experimental tests, FE model development and manuscript preparation. All authors have read and approved the final manuscript. Note ==== \*Partially cross-linked polyethylene cups are processed according to ISO 5834-II and ASTM F648 standards. Cups are compression molded and gamma irradiated in a vacuum to a level of 27 kGy. Acknowledgements ================ Financial support from the Academy of Finland, Sigrid Jusélius Foundation, Helsinki, Finland, and the Invalid Foundation, Helsinki, Finland, is acknowledged. The authors thank MSc Petteri Väänänen for conducting the experimental measurements and Esa Miettinen for technical support.	PubMed Central	2024-06-05T03:55:58.438381	2005-5-17	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173123/", "journal": "Biomed Eng Online. 2005 May 17; 4:32", "authors": [ { "first": "Rami K", "last": "Korhonen" }, { "first": "Arto", "last": "Koistinen" }, { "first": "Yrjö T", "last": "Konttinen" }, { "first": "Seppo S", "last": "Santavirta" }, { "first": "Reijo", "last": "Lappalainen" } ] }
PMC1173124	Background ========== High Frequency Chest Compression (HFCC) \[[@B6],[@B9],[@B19],[@B29]\] is a useful and popular therapy for clearing bronchial airways of excessive or thick mucus since it does not require patients to do any directed efforts for respiration while therapy is given, unlike other airway clearance techniques such as active cycle of breathing or autogenic drainage \[[@B17]\]. A HFCC machine pumps air into an inflatable jacket worn by patient. By means of a surrogate piston, sine waveform compression pulses with frequencies ranging from \~5 Hz to \~21 Hz are supplied to the thorax of a patient through the jacket. These pulses squeeze and vibrate the patient\'s thorax at prescribed frequencies. These actions help in the evacuation of mucus through changing the rheological property of the mucus and airflow oscillation. King et al. \[[@B13]\], Krumpe et al. \[[@B15]\], and Tomkiewicz at al. \[[@B28]\], showed that HFCC pulses decrease the viscosity of mucus and helps evacuation. A more important consequence is the respiratory airflow oscillated during use of HFCC, which results from variation of intrapleural pressure. Lapin \[[@B17]\] and Warwick \[[@B30]\] pointed out that airflow is the most important factor for mucus transport since airflow produces the shear stresses for evacuation of mucus. Although several models have been developed to simulate respiration by other researchers and their models successfully worked for their own purposes, these models are not appropriate for HFCC simulation because these models were either linearized \[[@B7],[@B8]\] or because they assumed that respiration was driven by a mechanical ventilator at the mouth \[[@B5],[@B20]\]. The model presented in this paper provides reliable simulation results on fast change of intrapleural pressure (P~pl~) altered by HFCC because this model is described with nonlinear equations and P~pl~is selected as the driving force of respiration. The nonlinear characteristics of the respiratory system are not easily noticed during quiet tidal breathing, but the airflow signal oscillation measured at the mouth during use of HFCC is a strong indicator of the nonlinear characteristics of the respiratory system. In this study, we simulated the respiratory system with a computational model that carefully reflected its nonlinear characteristics. The model is an electrical circuit analogue, in which nonlinear resistors (R\'s), capacitors (C\'s), and inductors (L\'s) represent airway resistance, lung capacitance, and inertance of air, respectively. Just like in a living organ, the driving force of this model is P~pl~which is a superposition of HFCC pulses on the spontaneous breathing effort. We compared the simulation data to the in vivodata to demonstrate and understand the characteristics of airflow signal oscillation. Methods ======= Conceptual model of the respiratory system ------------------------------------------ Our modelling and simulation for reproducing airflow signals required simplification of airway structure in the lung. The geometrical and dimensional structures of the airways were proposed by several researchers, among which Horsefield et al. \[[@B10]\] and Weibel \[[@B31]\] dissected and measured the human lungs and airways, and more recently, and Tawhai et al. \[[@B11]\] and Kitaoka et al. \[[@B14]\] proposed algorithmic approaches for reconstructing the branching structure. We employed Weibel\'s morphometry of the lung, which provides the geometries and the dimensions based on the symmetric dichotomous structure of airway branching when the lung volume is assumed to be 75% of the total lung capacity (TLC). According to his morphometry, the trachea is defined as airway generation 0 and it is separated into two geometrically identical daughter branches. Each daughter is repeatedly branched up to 22 times, thus the lung is considered to have 24 (0--23) airway generations and the number of airway branches total \~1.7 × 10^7^. In each airway generation, there are 2^z^(z is a generation number) identical branches whose lengths and radii are provided and the dimensions of airway branches in each generation differ from generation to generation. He also suggested that airway generations 0--16 comprise the conducting zone, whereas airway generations 17 -- 23 make up the respiratory zone in which gases are exchanged. Based on Weibel\'s morphometry of the lung, we simplified the geometry of the airways. For airway generation Zin conducting zone, a bundle of 2^z^identical airway branches are considered as the big tube whose cross sectional area equals to 2^z^times the cross sectional area of a single airway branch. The tubes for each airway generation are represented by RCL T-networks shown in fig. [1 (b)](#F1){ref-type="fig"}, and evaluation of the R\'s, C\'s, and L\'s in the RCL T-network are explained in equations (4) -- (6). Meanwhile, the respiratory zone is considered as a big lump, the alveolar space, since alveolar ducts and sacs are scattered throughout the respiratory zone \[[@B8]\]. Although the upper airway is not presented in Weibel\'s morphometry, it is one of the chief sites for airway resistance. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### \(a) The conceptual model based on Weibel\'s morphometry of the lung \[31\]. The lung and airways are comprised of five regions, namely, the alveolar space, the conducting airway zone, the pleural cavity, the thorax outside the lung, and outside the thorax. (b) RCL T-network that represents a region (airway generation) in the conducting airway zone. (c) Electrical circuit analogue converted from (a). For acronyms in the figure, see list of abbreviations. ::: ![](1475-925X-4-34-1) ::: Our conceptual model as a summary of the simplification, \'the upper airway + 17 conducting airway generations + the alveolar space\', is demonstrated in fig. [1 (a)](#F1){ref-type="fig"}, and circuit analogue converted from fig. [1 (a)](#F1){ref-type="fig"} is shown in fig. [1 (c)](#F1){ref-type="fig"}. Dimensions of airway branches ----------------------------- Since the dimensions of every single airway branch in the lung vary during respiration, it is crucial to track them to extract proper values of R\'s, C\'s, and L\'s. The radii of the airway branches in airway generation Zare determined by transmural pressure (P~tmZ~), pressure difference between the pleural cavity and inside airway generation Z. In order to determine P~tmZ~, we need to rely on the study about transpulmonary pressure (P~tp~), pressure difference between the pleural cavity and alveolar space. According to Salizar et al. \[[@B26]\], P~tp~is a function of lung volume (LV) and TLC. With given LV, P~tp~is obtained by, P~tp~= -log(1 - LV/ TLC) × 7.22. (1) If LVis functional residual capacity (FRC) and no airflow exists in the airways, the lung is at rest. In this situation, air pressure at any site is the same and atmospheric (zero), therefore P~tmZ~and P~tp~are the same. Our simulation begins with assuming that the lung was at rest. The equation of Lambert et al., or Lambert\'s tube law \[[@B16]\], is a function of P~tmZ~, and gives the ratio of the cross sectional area (A~Z~) to the maximum cross sectional area of an airway branch (A~maxZ~) in airway generation Z. That is, A~Z~/ A~max\ Z~= 1.0 - (1.0 - α~0~)(1.0 - P~tmZ~/ P~0~)^-N^ (2) where α~0~, α~0~\' and Nare constants for each airway generation given from Lambert\'s tube law, and P~0~= (α~0~-1)N/α~0~\'. As mentioned earlier, the dimensions of the airway branches provided by Weibel\'s morphometry are based on 75 % of TLC. Let A~Z75~be Weibel\'s cross sectional areas in airway generation Z. By using ![](1475-925X-4-34-i1.gif) A~maxZ~for each airway generation can be found. Since A~maxZ~does not change whether or not the lung is at rest, the equation is valid to find out A~Z~during respiration. The length of an airway branch is assumed to vary in away that; l~z~/ l~FRC~= r~z~/ r~FRC~ (3) where l~z~and r~z~are the length and the radius of an airway branch in airway generation Z, respectively. l~FRC~and r~FRC~are the length and the radius on FRC, respectively. Equivalent circuit elements modelling ------------------------------------- The RCL T-network consists of two resistors (R~gZ~), two inductors (L~gZ~), a capacitor (C~gZ~), and a DC voltage source (P~tm\_ini~). R~gZ~and L~gZ~are the airway resistance and the inertance of air in the entire airway generation Z, respectively. C~gZ~, the airway capacitance of the entire airway generation Z, represents inflation and deflation of the non-rigid airway wall during respiration. P~tm\_ini~is the initial value of P~tmZ~. This initial pressure counterbalances P~pl~, which is negative when the lung is at rest. The values of the circuit elements in the RCL T-network of airway generation Zcan be obtained by totalling the 2^z^airway branches. Therefore, R~gZ~is given by ![](1475-925X-4-34-i2.gif) Similarly, L~g~zis given by ![](1475-925X-4-34-i3.gif) and C~gZ~is C~gZ~= C~sZ~× N~z~\[ml/cmH~2~O\], (6) where R~sZ~, L~sZ~, and C~sZ~indicate the values of a single airway branch in airway generation Z. N~z~is 2^z^, the number of airway branches in generation Z. R~sZ~depends on the classic Poiseuille equation and the Zetacorrection factor proposed by Pedley et al. \[[@B21]\]. The Zetacorrection factor is given by ![](1475-925X-4-34-i4.gif) and ![](1475-925X-4-34-i5.gif) R~e~is Reynolds number ![](1475-925X-4-34-i6.gif), Vis air velocity, ρis the density of air, and ηis the viscosity of air. Because electric current flows at the speed of light in an electrical circuit, it is necessary to compensate for the incomparably slower behaviour of airflow in the airways by placing inductors. Inertance of air in a single airway branch (L~sZ~) depends on the dimension of the airway branch \[[@B5]\] and it is given by ![](1475-925X-4-34-i7.gif) To compute the values for C~sZ~in equation (2), relation between A~Z~and P~tmZ~in airway generation Z, is used again. That is, ![](1475-925X-4-34-i8.gif) ![](1475-925X-4-34-i9.gif) The alveolar space is represented by a capacitor in the circuit analogue. Alveolar capacitance (C~as~) is the ratio of the alveolar volume (ΔAV) to ΔP~tp~. Lung volume (LV) consists of airway volume and alveolar volume, and airway volume is negligible compared to alveolar volume. Therefore, ![](1475-925X-4-34-i10.gif) which implies that ![](1475-925X-4-34-i11.gif) represents C~as~. It can be obtained by equation (1); however, the equation did not consider the hysteresis of P~tp~-volume curves that is caused by several proposed reasons \[[@B1]\]. To overcome this, we defined C~as~as ![](1475-925X-4-34-i11.gif) \*k(k\<0), and kwas continuously changed over the time course of the simulation. The values of kwere empirically determined to achieve the acceptable shape of the hysteresis, which is shown in fig. [2](#F2){ref-type="fig"}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### P~tp~-volume curves of the lung model during use of HFCC. Although the initial LV(FRC) is 2700 ml, LVduring use of HFCC is smaller. The curves became jagged due to HFCC pulses. ::: ![](1475-925X-4-34-2) ::: A resistor and an inductor characterize the upper airway, from the nasal/oral cavity to larynx. The equations for the resistance of the upper airway (R~ua~) that Jackson et al.\[[@B12]\] validated are described below: ![](1475-925X-4-34-i12.gif) where F~ua~is the airflow rate in the upper airway. As Marchal et al. \[[@B18]\] estimated, the value of the inductor that represents inertance of air is 0.00003 \[cmH~2~O·s^2^/ml\]. Numerical methods for nonlinear circuit analogue ------------------------------------------------ Although the number of elements is manageable and the structure of circuit is fairly simple, analysis of the circuit analogue in fig. [1 (c)](#F1){ref-type="fig"} is not trivial since the values of all the energy-storing (C\'s and L\'s) elements as well as resistive elements (R\'s) change at every sequence of the simulation time-step. To exemplify a general idea of numerical methods for the nonlinear circuit analogue, analysis of a simple nonlinear second-order system in fig. [3](#F3){ref-type="fig"} was demonstrated. The system of fig. [3](#F3){ref-type="fig"} is described by an equation: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A nonlinear second-order system in the form of an electrical circuit. iis electrical current, and v~C~is voltage difference between each end of C. The value of the resistor (R) in this figure is dependant on ias well as v~C~whereas the capacitor (C) and the inductor (L) are dependent only on v~C~. V~in~(t) is a voltage source for this simple circuit system. Nonlinear circuit elements in this circuit imply the nonlinearity of fig. 1 (c). ::: ![](1475-925X-4-34-3) ::: ![](1475-925X-4-34-i13.gif) where v~C~is the voltage difference between each end of the capacitor. Note that R(•) is a function of iand v~C~, and L(•) and C(•) are functions of v~C~just like in fig. [1 (c)](#F1){ref-type="fig"}. Using backward Euler approximation \[[@B22]\], equation (14) is converted to a difference equation: ![](1475-925X-4-34-i14.gif) where Δt is time difference between sequence \[n\] and \[n-1\]. Equation (15) is the same as ![](1475-925X-4-34-i15.gif) By the definition of backward Euler approximation, the current of the circuit ![](1475-925X-4-34-i16.gif). Then equation (16) can be restated as an equation; ![](1475-925X-4-34-i17.gif) In equation (17), the right hand side consists of all known values, and the left hand side is a function of v~C~\[n\]. Suppose that v~C~\[n\] is x, equation (17) can be expressed as f(x) = c, (18) where c is a constant. Equation (18) can be easily solved using an iteration method \[[@B24]\]. To do this computation, MATLAB (Mathworks, Natick, MA) codes were written. Protocols for measurement airflow signals ----------------------------------------- In this study, The Vest™ (Advanced Respiratory, St.Paul, MN, USA; now named Hill-Rom Co.,Inc.), which delivers sine waveform compression pulses, was used for application of HFCC to a subject for measuring and recording the airflow signals at the mouth. The general usage of HFCC device was early described in the review of Hansen et al. \[[@B9]\] and the typical respiratory airflow during use of HFCC is shown in fig. [4](#F4){ref-type="fig"}. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### \(a) The airflow and (b) the oesophageal pressure measured by Fink et al. \[6\] (modified with the permission), which shows typical respiratory airflow during use of HFCC. The positive numbers of airflow rate represent inspiration and the negative numbers represent expiratory airflow. Oesophageal pressure reflects the intrapleural pressure (P~pl~) that is altered by HFCC pulses (tiny peaks). ::: ![](1475-925X-4-34-4) ::: The subject sat upright on a chair for measuring and recording the airflow signals at the mouth with an in-house built electronic spirometer. The subject worn a nose clip and breathed through a mouthpiece. After the HFCC device is properly set up, the compression pulses were applied and then the subject made several slow and large, but not to TLC, breaths. During the breaths the subject hold his glottis open until data collection was completed. This protocol was followed for the low (5 Hz), high (21 Hz), and medium frequencies (15 Hz) of HFCC pulses. Before each frequency recording of airflow signals, the subject rested for one minute. To ensure that the subject had adapted to HFCC pulses and had reached a steady state, only the last ten seconds of the one- minute breathing were recorded and analyzed for our study. Results ======= For the simulation, the parameters of the lung were determined to represent a normal healthy lung. The TLC, FRC, and RVof the model lung are 6000 ml, 2700 ml, and 1000 ml, respectively. The ambient atmospheric pressure is assumed to be zero, and the air density and viscosity are 0.00113 g/cm^3^and 0.00019 g/cm·s, respectively. It was assumed that there would be no airflow in the airways and that initially the LVwould be the same as the FRC. Change of P~pl~is the primary driving force of respiration and is initially -4.32 cmH~2~O, which is the counterbalance to initial P~tp~given by equation (1). Fig. [4](#F4){ref-type="fig"} is the airflow the oesophageal pressure measured by Fink et al. \[[@B6]\], which shows typical respiratory airflow during use of HFCC. Fig. [5](#F5){ref-type="fig"} shows the subject\'s airflow signals during ten seconds at the three different frequencies. In this figure, each airflow signal is also viewed as the low-pass filtered and the high-pass filtered curves. The high-pass filtered curve indicates the fast airflow signal oscillation generated by HFCC pulses (HFCC airflow signal) whereas the low-pass filtered curve is the airflow during the spontaneous breathing effort (spontaneous airflow signal) of the subject. We divided one cycle of the respiration into four phases. Phase I is the portion of the inspiration phase when the spontaneous airflow signal is greater than the amplitudes of HFCC airflow signal. In this phase LVincreases. During phase II, the spontaneous airflow signal stays within the amplitudes of the HFCC airflow signal. Phase II is the pause before expiration begins. The amplitudes of HFCC airflow signals change considerably as the phase moves from I to II. To emphasize the difference of HFCC airflow signals, phase II were consciously prolonged and they are longer than phase II in fig. [6](#F6){ref-type="fig"}. In phase III, passive or active expiration begins and the amplitudes of HFCC airflow signals decrease to about phase I amplitudes. The low-pass curve gets greater to the negative direction curve than the high-pass curve. The next and last phase is phase IV. Phase IV is the resting period before inspiration, begins. During this phase the amplitudes of HFCC airflow signals again become greater than the spontaneous airflow signal. To reproduce similar airflow signals, a cycle of respiration from phase IV to phase IV was simulated using our computational model. The simulation results with the three frequencies are shown in fig. [6](#F6){ref-type="fig"}. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### The airflow signals measured at the mouth of the subject, the high-pass filtered (HFCC airflow signal) and the low-pass filtered (spontaneous airflow signal) curves. The subject was using HFCC with (a) 5 Hz, (b) 15 Hz, and (c) 21 Hz. Regardless of the frequencies, larger spontaneous airflow signals result in smaller HFCC airflow signals. Since it is difficult to breathe hard during 21 Hz, spontaneous airflow signals in (c) are smaller than in (a) and (b). Phase I is the portion of the inspiration phase when spontaneous airflow signals are greater than the amplitudes of HFCC airflow signals. During phase II, spontaneous airflow signals stay within the amplitudes of HFCC airflow signals. In phase III, expiration begins, and the amplitudes of airflow oscillation decrease to about the amplitudes during phase I. Spontaneous airflow signals get greater to the negative direction than HFCC airflow signals. Finally phase IV, the resting period before inspiration, begins, and the amplitudes of HFCC airflow signals again become greater than spontaneous airflow signals. To emphasize the difference of HFCC airflow signals, phase II were consciously prolonged and they are longer than phase II in fig. 6. ::: ![](1475-925X-4-34-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Simulated respiration, spontaneous airflow signals and HFCC airflow signals. The frequencies of HFCC were set (a) 5 Hz, (b) 15 Hz, and (c) 21 Hz. Definitions for Phases I, II, III, and IV in this figure are the same as those in the text and fig. 5. ::: ![](1475-925X-4-34-6) ::: From the measured values and the simulation of airflow signals at the mouth, we could observe two important phenomena during respiration pertaining to the airflow signal oscillation generated by HFCC. First, the amplitudes of HFCC airflow signals in phases I and III were smaller than those in phases II and IV. Second, the amplitudes in phases I and III became even smaller as the spontaneous airflow signal became greater. We used the simulation results to investigate how the nonlinearity of airway resistance, lung capacitance, and inertance of air characterized the respiratory airflow. The simulation was repeated after setting one of the three properties as a linear constant value. Fig. [7](#F7){ref-type="fig"} compares the linear values with the nonlinear values of the three properties. The linear values are means of the nonlinear values during the simulation. Fig. [8](#F8){ref-type="fig"} presents the simulation results of the lung model under the imaginary assumptions. Fig. [8 (a)](#F8){ref-type="fig"} is normal respiratory airflow at 15 Hz, and fig. [6 (b)](#F6){ref-type="fig"} and fig. [8 (a)](#F8){ref-type="fig"} are from the same simulation data. Fig. [8 (b)](#F8){ref-type="fig"} demonstrates the predicted airflow signals at the mouth with linear airway resistance, which indicates the amplitudes of HFCC airflow signals do not vary significantly. Fig. [8 (c)](#F8){ref-type="fig"} is the airflow when lung capacitance is set to a linear value. Just like fig. [8 (a)](#F8){ref-type="fig"}, HFCC airflow signals are the largest when spontaneous airflow signals are close to zero. Therefore, it can be presumed that the nonliearity of lung capacitance does not play a role in the non-uniformity of HFCC airflow signals. And neither is inertance of air. Fig. [8 (d)](#F8){ref-type="fig"}, which shows the simulation data with linear inertance of air, is almost identical to fig. [8 (a)](#F8){ref-type="fig"}. This indicates that the inertance of air is also not a factor in the non-uniformity of HFCC airflow signals. Fig. [9](#F9){ref-type="fig"} shows the HFCC airflow signals as a function of spontaneous airflow signals based on the same simulation data shown in fig. [8](#F8){ref-type="fig"}. Fig. [8 (a), (c)](#F8){ref-type="fig"}, and [8 (d)](#F8){ref-type="fig"} indicate that larger spontaneous airflow signals result in smaller oscillations of HFCC airflow signals whereas HFCC airflow signals do not seem to be related to spontaneous airflow signals in fig. [8 (b)](#F8){ref-type="fig"}. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### \(a) Airway resistance, (b) lung capacitance, and (c) inertance of air during the simulation used for fig. 8. The linear (fixed) values are the mean values of each nonlinear (varying) value. ::: ![](1475-925X-4-34-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Simulated spontaneous airflow signals and HFCC airflow signals, (a) when the values of airway resistance, lung capacitance, and inertance of air are nonlinear (varying), (b) when the value of airway resistance is linear (fixed), (c) when the value of lung capacitance is linear (fixed), and (d) when the value of inertance is linear (fixed). The frequencies of each plot are 15 Hz, so (a) is the same as fig. 6 (b). In (b), the amplitudes of HFCC airflow signals almost do not vary. ::: ![](1475-925X-4-34-8) ::: Discussion ========== The number of the airway branches is estimated to be about 17 million \[[@B31]\], and the dimensions of each airway branch vary while the lung is being inflated and deflated during respiration. To reduce the computational burden, models of the respiratory system are often simplified by ignoring the nonlinear natures of the respiratory system \[[@B7],[@B8]\]. Linearization entails considerable flexibility in numerical analysis since linear circuits are computationally much cheaper than nonlinear circuits, either on a time-domain or a frequency-domain basis. However, even during the slowest breathing, airway resistance, lung capacitance, and inertance of air change due to the variation of airway dimensions as well as turbulence. In particular, increase of the airway resistance due to turbulence is so drastic that inaccurate evaluation of airway resistance may result in a misleading simulation result. Our simulation results demonstrated that such errors are more likely when HFCC intervention is applied. Evidence for this was that the result from a linear model was significantly different from that from a nonlinear model (fig. [8](#F8){ref-type="fig"} and [9](#F9){ref-type="fig"}). As parts of the respiratory system, lung capacitance and inertance of air are also nonlinear although it was observed that their nonlinearity did not cause the non-uniform amplitudes of HFCC airflow signals. However, this does not imply that the nonlinear characteristics of lung capacitance and inertance of air have no role in the simulation of the respiratory system. Dimensional changes of the airway branches in the lung are responsible for the nonlinearity of lung capacitance and inertance of air. When dimensions of the airways are involved for simulation and prediction, the nonlinearity is very important for accurate results. For example, in the study of Sohn et al. \[[@B27]\], our model is used for estimation of air velocity, which is airflow rate ÷ cross sectional area. Since the cross sectional areas of airway branches do not vary linearly, lung capacitance and inertance of air also should not be linear in order to avoid discrepancy. Ideally, studies about airflow in the airways would be best resolved by CFD (Computational Fluid Dynamics), however, using a CFD approach in this study presents several problems that cannot be overcome by modern technology. First, the turbulence mechanism is not completely known \[[@B23]\]. Even during the slowest breathing manoeuvre, turbulence exists in the proximal airways, and ignoring turbulence would not give true simulation results. Another obstacle is that the whole lung cannot be taken into consideration even with the latest supercomputing power. A CFD approach to simulate dynamics of airflow interacting in a huge number of airway branches in the lung requires extremely massive computation. There is no way to deal with it, if any, its computation time would be incredibly long. It should be also pointed out that CFD techniques for airflow in non-rigid wall tubes are not yet mature. Although not perfect, our circuit analogue model allows us to effectively simulate the nonlinear characteristics of the respiratory system. It is well known that respiratory system impedance consists of airway impedance and chest wall (tissue) impedance \[[@B2]-[@B4]\]. Since the driving force of airflow in our model is the intrapleural pressure altered by HFCC pulses transferred from the body surface, our model does not necessarily incorporate chest wall (tissue) impedance. However, it is not clear how effectively the chest wall transfers HFCC pulses to the pleural cavity. Other factors should also be considered for impedance between the jacket of HFCC and the pleural cavity -- such as clothes, posture, and tightness of the jacket. Currently, based on observations of Milla et al. \[[@B19]\], we assume that HFCC pulses on the body surface are transferred to the pleural cavity without any distortion of pulsation waveforms although some attenuation may exist. The HFCC device was initially developed only for cystic fibrosis patients who normally have healthy lungs \[[@B29]\]; consequently we modelled a healthy lung as the first step of the research. As this medical treatment becomes widely applicable to other lung diseases, we are planning to develop models that can be used to simulate HFCC on various lung diseases in the future. ::: {#F9 .fig} Figure 9 ::: {.caption} ###### HFCC airflow signals are demonstrated as a function of spontaneous airflow signals based on the same simulation data shown in fig. 8. Nonlinear airway resistance (a, c, and d) produces the non-uniform amplitudes of HFCC airflow signals. ::: ![](1475-925X-4-34-9) ::: Conclusion ========== In this study, the airflow signals measured at the mouth during use of HFCC are viewed as a composite of two causes: the spontaneous breathing effort and HFCC pulses. However, since the respiratory system is nonlinear, airflow signals at the mouth are not a mere superposition of the effects from these two causes. In laboratory measurements, the amplitudes of the airflow signal oscillation varied considerably despite the uniformity of the HFCC pulses. After confirming that the simulation results matched up with the observations, we analyzed the simulation data to explain the observed inconsistency in the HFCC airflow signal amplitudes. We found that the amplitudes of HFCC airflow signals behave as a function of spontaneous airflow signals. This is due to the nonlinearity of the respiratory system, particularly variations in airway resistance. The findings in this paper may not be immediately applicable for HFCC therapy, but they do lead to ways to better prescribe HFCC therapy. Most importantly, the usefulness of our computational simulation and the model-based approach as a tool to understand clinical observations of HFCC was well demonstrated in this paper. List of abbreviations ===================== HFCC: High Frequency Chest Compression FRC: Functional Residual Capacity TLC: Total Lung Capacity RV: Residual Volume LV: Lung volume AV: Alveolar volume R: resistor C: capacitor L: inductor A~Z~: Airway cross sectional area in airway generation Z A~maxZ~: Maximum Airway cross sectional area in airway generation Z A~Z75~: Airway cross sectional area in airway generation Zwhen LVis 75% of TLC C~as~: Alveolar capacitance C~gz~: Total airway capacitance in airway generation Z C~sZ~: Single airway capacitance in airway generation Z F~inZ~: Incoming airflow in airway generation Z F~outZ~: Outgoing airflow in airway generation Z F~ua~: Airflow in the upper airway L~gz~: Total inductance (inertance of air) in airway generation Z L~sZ~: Single inductance (inertance of air) in airway generation Z l~Z~: The length of an airway branch in airway generation Z N~Z~: Number of airway branches in airway generation Z P~alv~: Alveolar pressure P~awZ~: Airway pressure in airway generation Z P~pl~: Pleural pressure or intrapleural pressure P~tm\_ini~: The initial value of transmural pressure P~tmZ~: Transmural pressure in airway generation Z P~tp~: Transpulmonary pressure P~tp\_ini~: The initial value of transmural pressure R~gz~: Total airway resistance in airway generation Z R~sZ~: Single airway resistance in airway generation Z r~z~: The radius of an airway branch in airway generation Z Authors\' contributions ======================= KS designed all the necessary computational and experimental procedures, analysed data to reach the conclusion, and also prepared the texts and figures in the manuscript. WJW conceived this study, helped organizing the manuscript, and advised all aspects of physiology and medicine for this study. YWL and JL helped measuring the respiratory airflow. JEH advised all aspects of technology and engineering for this study.	PubMed Central	2024-06-05T03:55:58.440825	2005-5-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173124/", "journal": "Biomed Eng Online. 2005 May 19; 4:34", "authors": [ { "first": "Kiwon", "last": "Sohn" }, { "first": "Warren J", "last": "Warwick" }, { "first": "Yong W", "last": "Lee" }, { "first": "Jongwon", "last": "Lee" }, { "first": "James E", "last": "Holte" } ] }
PMC1173125	This book is intended for biologists, medical physicists, biomechanical engineers and clinicians who are involved in bone studies. For these practitioners, the book is a comprehensive review of physical measurement techniques applied to bone, with useful emphasis on practical aspects such as protocols of measurement, sample preparation and sources of errors. Chapters were written by leading scientists in the relevant fields, and overall, the outcome is a well-written, comprehensive and useful textbook that will definitely serve the biomedical community. The book contains 18 chapters classified into four major sections: Introduction, Invasive Techniques, Ionizing Radiation Techniques, and Non-Ionizing Techniques. As typically occurs with books that are collections of chapters from several authors, the quality is not uniform throughout. For example, while some figures are clear and illustrative, other figures suffer poor quality or were reproduced at a too low resolution. The important points of strength and weakness are summarized below. The introduction includes a comprehensive review of bone anatomy and physiology, in which each skeletal component -- bone cells, matrix and mineral -- is separately addressed. It would have been useful here to also refer to tissues that connect to bone (tendons, ligaments, cartilage) and produce mechanical loads which act on bone (muscles) and particularly, to the interfaces between bone and these soft tissues. A review of the pathophysiology, etiology, epidemiology, and economic impact of osteoporosis follows, but other, less common bone diseases are only briefly mentioned. A useful chapter on biology safety considerations, with focus on bone studies, is also included in the introduction. I found this chapter particularly helpful as a guide for students or laboratory technicians who become involved in preparing bone sections for histopathology or mechanical testing. Typically, the information on how to avoid accidents such as when dealing with bone saws, chemicals for demineralizing, fixing and staining bone samples, radiation, and biological waste, is posted in internal laboratory or institutional regulations. This is the first textbook chapter that integrates this information specifically for experiments with bone, and I will certainly make it an obligatory reading material for those involved in bone studies in my laboratory. The last chapter in the introduction concerns instrument evaluation. While appreciating the important information provided on measurement errors, precision, and accuracy, I found this chapter to be too general in nature, and was not convinced that it should have been included in a textbook that focuses on bone measurements. For example, the classical Bland and Altman paper in Lancet (1986) \[[@B1]\] which is discussed in detail as related to bone mineral density measurements, originally referred to flow and oxygen saturation measurements, and can be generalized for bone as for any other application. Accordingly, this chapter could have fit a textbook on biomedical instrumentation, but it does not contain contents that are specific enough for bone studies. The second section, on invasive techniques, puts heavy emphasis on the chapter which concerns mechanical testing of bone. Unfortunately, the contents of this chapter substantially overlap information already provided in the Bone Mechanics Handbook \[[@B2]\] which is a classic in this field. I was disappointed to find out that the tabulated data of mechanical properties for cancellous bone, cortical bone and individual trabeculae mostly summarize publications from the 1970\'s and 1980\'s (with few references to publications from the 1990\'s). Publications from 2000\'s (e.g. \[[@B3],[@B4]\]) are not covered. I would have expected this new book to cover a more recent literature than that covered in the Bone Mechanics Handbook published on 2001 \[[@B2]\] or in Prof. J.D. Currey\'s book \"Bones: structure and mechanics\" published on 2002 \[[@B5]\], particularly because of the growth in attention to bone mechanical properties as osteoporosis becomes a major epidemic. The issue of bone viscoelasticity, which is discussed in detail in \[[@B2]\], is only briefly mentioned here in terms of the dependence of the elastic modulus of cortical bone on the strain rate. On the contrary, I found the subsequent chapters on bone histomorphometry, microscopy and related techniques very useful, specifically because these chapters include detailed procedures for bone histological staining as well as for immunohistochemistry, immunofluorescence, confocal microscopy and scanning electron microscopy analyses. The third section concerns ionizing radiation techniques and specifically, absorptiometric methods to asses bone mineral density such as DXA, quantitative CT, peripheral QCT and micro-CT. Again, the parts which describe the fundamentals of radiation physics, interactions of x-rays with matter and radiological instrumentation may be too general for this type of book, but the applications to assessment of bone quality in humans and animal models are useful. A remarkable drawback in this section is that nearly all attention is given to bone mineral density, and very little information is provided on morphological parameters of bone that are demonstrated by the CT-based imaging modalities. Specifically, measurements of trabecular separation, trabecular thickness, bone volume fraction, trabecular number, trabecular connectivity etc., and accuracy of these measurements using clinical QCT and micro-CT are not addressed. Relations between these morphologic parameters, which are important characteristics of cancellous bone, are also not mentioned. Although the chapter on MRI that is included in the fourth section (non-ionizing techniques) partially concerns these issues, the reader is lacking information on morphological parameters that can be measured specifically by CT-based methods, and on the relation of accuracy to CT resolution for each morphological parameter. The last section is dedicated to non-ionizing techniques for bone studies. These include imaging methods such as MRI and ultrasound, as well as finite element and animal models. The chapter on ultrasound measurements of bone is particularly worth mentioning as a well-written review on the technological aspects, accuracy and precision and clinical applications. The chapter on finite element analysis covers up-to-date modeling approaches based on CT cross-sectional images at the whole bone level, or based on micro-CT or micro-MR images at the trabecular level. Geometrical model reconstructions from cross-sectional images and bone mechanical properties relevant to structural modeling are described in adequate detail, but meshing techniques and the effect of mesh density and element selection on the predicted stresses and deformations in bone were not addressed (qualitatively or quantitatively). This may disappoint a reader who was hoping to use this chapter as a practical, rather than a theoretical guide for model development. A second drawback relates to the discussion on models at the trabecular levels, where generic models of the trabecular architecture are ignored. Generic modeling of trabecular bone, which utilizes the theory of cellular solid mechanics, is a complementary modeling approach to specimen-specific (micro-CT or micro-MR image based) modeling. Generic models allow to study the mechanics of \'typical\' trabecular bone samples \[[@B6],[@B7]\], as opposed to specimen-specific models which may include abnormalities or anatomical variations that are specific to the studied specimens \[[@B7]\]. The concluding chapter in the last section concerns animal models to study bone, and particularly, osteoporosis. I found this chapter to be extremely useful for experimentalists who weigh alternative models for an experimental design of an animal study of osteoporosis. Having considered the points of strength and weakness in the different sections of this book as discussed above, I certainly recommend individual researchers and laboratories who deal with bone studies to hold a copy of this book. It should be noted that for those investigators focused on biomechanics of bone, it is not a substitute for the classics \"Bone Mechanics Handbook\" \[[@B2]\] and \"Bones\" \[[@B5]\], but rather, a complimentary book with particular added value in biology-oriented topics of safety considerations, microscopical techniques and animal model specifications. Bioengineering students, medical physics students and perhaps biology students at the graduate levels will also find this book useful for obtaining insight into the broad field of bone analysis. Abbreviations ============= DXA = dual x-ray absorptiometry CT = computed tomography QCT = quantitative computed tomography MRI = magnetic resonance imaging MR = magnetic resonance	PubMed Central	2024-06-05T03:55:58.444216	2005-6-14	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173125/", "journal": "Biomed Eng Online. 2005 Jun 14; 4:37", "authors": [ { "first": "Amit", "last": "Gefen" } ] }
PMC1173126	Background ========== An elevated immunoglobulin G (IgG) index and the presence of oligoclonal bands (OCB) in the cerebrospinal fluid (CSF) are a hallmark of multiple sclerosis (MS) \[[@B1],[@B2]\]. Although this finding is not specific for MS, 72 % percent of patients present with an elevated IgG index and even 98 % show an oligoclonal distribution of IgG bands exclusively in the CSF \[[@B3],[@B4]\]. Intrathecal IgG is thought to be the product of B lymphocytes residing in the brain of MS patients after they have crossed the blood brain barrier in an activated state with the help of various co-stimulatory signals \[[@B5]\]. Instead of undergoing apoptosis, the B cells expand clonally within the central nervous system (CNS) giving rise to a persistent antibody production \[[@B6]\]. Despite intense investigations, no single antigen against which the antibodies might be directed has been isolated so far. In contrast, the intrathecal antibody response covers a large number of CNS and non-CNS antigens as well as various pathogens \[[@B7]-[@B14]\], including the viral antigens such as measles, rubella and varicella zoster \[[@B15]\]. In up to 96 % of MS patients an intrathecal antibody production against at least one of the three antigens has been found \[[@B3],[@B16]\]. Little is known, however, about the clinical significance of these findings. Previous studies attempting to evaluate the long-term evolution of intrathecal viral antibodies were hampered by technical shortcomings such as few sensitive detection methods and an absence of correction for blood-CSF-barrier disturbance \[[@B17],[@B18]\]. Hence, the findings of these preliminary studies were partially contradictory with regard to the stability of the CSF antibody production \[[@B17],[@B18]\]. Furthermore, no correlation to the clinical course could be demonstrated \[[@B17],[@B18]\]. Here we report the results of a follow-up study on 70 MS patients from which at least two CSF analyses including cell count, IgG index, local IgG synthesis, antibody specific index and oligoclonal bands were available. The clinical implications of the immunological findings are discussed. Methods ======= Patients -------- The study used 70 consecutive patients with definite MS according to the criteria of Poser \[[@B19]\] and a primary relapsing course. All patients had at least two spinal taps, mostly during disease exacerbations. Three lumbar punctures (LP) were performed in 26 patients, four in 12 patients and five in 7 patients. Patients were characterized clinically by age, sex, disease duration, course of the disease and the expanded disability status scale (EDSS), documented at the time of the first LP. Furthermore, the progression index was calculated as the ratio of the EDSS and the disease duration for each patient. Patients with corticosteroid treatment in the last four weeks or with immunomodulatory or immunosuppressive therapy in the last 3 months prior to the first LP were excluded. Cerebrospinal fluid and serum sample pairs were analyzed for cell count in the CSF, oligoclonal bands in serum and CSF, local IgG synthesis, IgG index and antibody index for the following antigens: measles, rubella and varicella zoster virus. We focused only on these specific antigens because of the frequent detection of these antigens within CSF in the case of MS described in prior studies \[[@B3]\]. Cell count ---------- The CSF cell count was determined immediately after LP. For this purpose, 90 μl of CSF were stained with 10 μl dye containing 20% crystal violet solution, 20 % glacial acetic acid and 60% H~2~O. Cells were enumerated in a Fuchs-Rosenthal counting chamber. IgG index --------- The intrathecal IgG production was quantitated by the IgG index. For this purpose, albumin and IgG were measured in matched serum and CSF pairs by nephelometry according to the manufacturer\'s instructions using commercially available kits (antiserum against human albumin or IgG, respectively) and the BN 100 nephelometer (both Dade Behring GmbH, Marburg, Germany). The IgG index was calculated according to the following formula: (CSF IgG × serum albumin) / (serum IgG × CSF albumin). An IgG index above 0.7 was indicative of an intrathecal IgG synthesis \[[@B20]\]. Local IgG synthesis ------------------- We quantitated the local IgG synthesis within the CSF with the help of a method described by Reiber and colleagues \[[@B3]\]. Therefore, we used the following formula: Ig~loc~= \[Q~Ig~- Q~lim~(Ig)\] × Ig(serum) (Q = quotient CSF: serum). Details of the calculation have been described previously \[[@B21]\]. Oligoclonal bands (OCB) ----------------------- All samples were analyzed directly after performing the LP without freezing, to minimize post-sampling changes. The presence of oligoclonal bands in the CSF indicated an intrathecal IgG production. For detection of OCB, isoelectric focusing was performed on matched serum and CSF sample pairs. The serum and CSF samples were diluted to the same IgG concentration and run on polyacrylamide gel precoated with ampholytes (pH between 4.5 and 10.0) at increasing voltage according to the manufacturer\'s instructions (Servalyt Precotes, Serva Electrophoresis GmbH, Heidelberg, Germany). Subsequently, a silver staining with commercially available reagents was done as indicated by the manufacturer (Serva Electrophoresis GmbH). The patterns were interpreted qualitatively by comparing the presence or absence of OCB in the serum and CSF. Antibody specific index (AI) ---------------------------- During the whole study period, antibody production against measles, rubella and varicella zoster virus was analysed in matched serum and CSF pairs by commercially available, specific enzyme immune assays (Enzygnost by Dade Behring) on an ELISA II processor using the alpha-method (Dade Behring). The AI was calculated according to the following formula: AI = Q spec. / Q lim. with Q spec. = CSF antibodies / serum antibodies and Q lim. being calculated from the CSF / serum albumin ratio according to the Reiber formula \[[@B21]\]. AI larger than 1.4 indicate an intrathecal antibody synthesis against the given antigen. Statistical analyses -------------------- This study was carried out as retrospective case-control study. The mean values of cell count, IgG index, local IgG synthesis and AI were compared for the whole group at the different time points (LP 1 to 5) by non-parametric tests for dependent samples (Friedman test). The percentage of patients positive for OCB or with a positive AI against measles at the 1^st^and 2^nd^LP, respectively, was compared with the Fisher\'s Exact test. Different clinical subgroups, i.e. those with a high or low EDSS, were analyzed for their mean measles AI with the help of a median split. The mean measles AI in the subgroups was compared with a non-parametric test for independent samples (Mann Whitney U test). Results ======= Patient characteristics ----------------------- A total of 70 patients, 45 women and 25 men, were included in this study. The mean age of the patients was 38 years (range 21 to 67 years). All patients had a primary relapsing course of the disease. The mean disease duration was 4.5 years with a range of one to 37 years. The mean EDSS at study entry was 2.9 (standard deviation SD 1.33). The mean time from the first LP to the second was 15 months with a range from 1 to 81 months. The mean time from the first to the last LP was 22 months (range: 1 to 106 months) in all patients. In 30 patients the range between the first and second LP was at least 12 months. In 24 patients the time between these LPs was less than 6 months. At the occasion of the second LP, 22 patients received an immunomodulatory or immunosuppressive treatment (interferon beta, glatiramer acetate, intravenous immunoglobulin, azathioprine). It has to be stressed that the study population is inhomogeneous. If the LPs were sorted by the follow-up years in which they had been arranged, no new information could be observed. This might be due to the wide range of time between the different LPs in this patient cohort (1 to 106 months). The data are not shown. Mean IgG index, local IgG synthesis and AI ------------------------------------------ There was no significant change of the mean IgG index, local IgG synthesis or AI for any of the viral antigens investigated over time (p = 0.414 for IgG index, p = 0.578 for local IgG synthesis, p = 0.673 for measles AI, p = 0.984 for rubella AI, p = 0.941 for varicella zoster AI, Friedman test). The data are summarized in Table [1](#T1){ref-type="table"}. In addition, there was no significant change in the mean cell count (p = 0.291, Friedman test, data not shown). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Mean and SD of IgG index, local IgG synthesis (Ig~loc~) and antibody index (AI) for measles, rubella and varicella zoster virus at each successive lumbar puncture (LP). The number of patients in each category of serial LP is given. All included patients received at least two serial LPs (first and second). Additional LPs were only obtained in the mentioned number of patients. ::: LP no. No. of patients with the respective LP IgG index Ig~loc~(mg/L) AI measles AI rubella AI varicella zoster -------- ---------------------------------------- ------------ --------------- ------------ ------------- --------------------- 1. 70 1.3 ± 0.75 33 ± 5,0 3.4 ± 3.29 4.6 ± 7.11 3.6 ± 5.86 2. 70 1.3 ± 1.58 25 ± 4,1 3.5 ± 3.67 4.8 ± 7.67 3.6 ± 5.85 3. 26 1.4 ± 0.99 46 ± 1,1 3.9 ± 5.59 5.4 ± 9.28 4.7 ± 7.28 4. 12 1.2 ± 0.73 28 ± 0,9 2.8 ± 2.09 6.7 ± 14.78 4.4 ± 8.74 5. 7 1.0 ± 0.39 19 ± 4,8 2.1 ± 0.89 1.9 ± 1.28 1.9 ± 1.46 ::: OCB and AI positive patients ---------------------------- As expected, the presence of OCB in the CSF was more sensitive in the detection of an intrathecal IgG production than a positive IgG index \[[@B22],[@B23]\]. For instance, at the 1^st^LP 94% of patients had positive oligoclonal CSF bands whereas 88% of patients had a positive IgG index. Once positive the OCB bands remained positive during the follow-up. The percentage of patients who were positive for oligoclonal bands in the CSF increased after the 1^st^LP from 94 to 100 %. Likewise, the relative number of patients with an intrathecal antibody synthesis against measles virus increased significantly over time (Figure [1](#F1){ref-type="fig"}). No obvious change in rubella and varicella zoster antibody positive patients over time was observed. This significant increase in patients with an intrathecal antibody production against measles was seen in both patient subgroups irrespective of whether or not they were treated with any immunomodulatory treatment mentioned above at the 2^nd^LP or not (Table [2](#T2){ref-type="table"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Percentage of patients with an intrathecal antibody synthesis against measles, rubella, and varicella zoster virus at five successive spinal taps (LP, n = 70 for LP 1 and 2, n = 26 for LP 3, n = 12 for LP4 and n = 7 for LP5). There was a significant increase (p\<0.05) in patients positive for antibody synthesis against measles virus over time. There were no significant changes in rubella or varicella zoster antibody synthesis. ::: ![](1743-8454-2-3-1) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Percentage of patients who had oligoclonal bands in the CSF or a positive AI for measles at the respective lumbar puncture (LP). ::: 1^st^LP 2^nd^LP p value -------------------- ------------ --------- --------- ---------- Untreated patients OCB 96 100 not done measles AI 68 73 0.013 Treated patients OCB 91 100 not done Measles AI 55 68 0.001 All patients OCB 94 100 not done Measles AI 64 71 \<0.001 ::: Individual long-term follow-up studies -------------------------------------- In 7 patients, data from 5 consecutive spinal taps were available. Except for patient no. 57, all data were derived from patients with a secondary progressive course of the disease and collected over a period of 1 to 6 years. Although the AI for the various antigens was not identical, 4 of the patients were stable over time with regard to their status as antibody producers (patients no. 16, 21, 41, 52). Of these patients, one was positive for measles alone, one for measles and rubella, one for all three viral antigens and one patient was antibody negative. The introduction of an immunomodulatory or immunosuppressive treatment at the 2^nd^LP (interferon beta patients 16, 21 and 41, azathioprine patient no. 52) had no impact on the antibody production. Two additional patients experienced an expansion of their antibody response from one (rubella virus in both cases) to three antigens, despite treatment initiated at the 2^nd^LP (interferon beta in patient no. 5 and azathioprin in patient no. 34). Correlation with clinical features ---------------------------------- There was no significant difference in the mean AI against measles in patients with high or low disability (p = 0.995, Mann Whitney U test) at the 1^st^LP. High disability was defined as an EDSS score above the median of 3.0. Furthermore, there was no difference in the mean measles AI in patients with long and short disease duration (median 2 years, p = 0.304). The EDSS scale and disease duration was used to calculate the progression index for each patient. A progression index of more than 1 (median), indicating an increase of at least 1 point on the EDSS scale per year was regarded as high. The mean AI against measles was significantly higher in patients with a high progression index than in those with a low progression index (p = 0.038). In the subgroup with a high progression index the mean measles AI was 4.2, whereas in the low progression index group a mean measles AI of 2.6 was measured. In the low progression index subgroup 53% of patients had an intrathecal antibody production against measles, but almost 74% were positive for measles in the high progression index subgroup. There were no significant differences in the local IgG synthesis when comparing the groups with high or low disability at the 1^st^LP or the groups with a high or low progression index (p = 0.71, p = 0.962, respectively). In these respective groups, the IgG index was not changed (p = 0.165 high disability group; p = 0.828 high progression index). Discussion ========== There are three main findings of our study: i) Once an intrathecal antibody production is established it is maintained over time; ii) For the first time we provide evidence for an expansion of antibody reactivity in the CSF of MS patients; and iii) disease progression and B cell activation may be linked. Persistent IgG production ------------------------- In our study patients positive for oligoclonal IgG bands in the CSF remained positive at the follow-up. These findings are in accordance with previous data indicating a clonally stable IgG production in the CSF over long periods \[[@B24]\]. This was true despite the introduction of immunomodulatory or immunosuppressive treatment. In 132 MS patients serial CSF analysis before and after 2 years of interferon beta treatment or placebo failed to reveal differences in the IgG index or OCB in either treatment group \[[@B25]\]. Even after autologous hematopoietic stem cell transplantation identical OCB compared to pre-treatment CSF analysis were found \[[@B26],[@B27]\]. Similarly in our study, there was no effect of treatment on IgG production as documented by the IgG index or on the local IgG synthesis in the CSF. In addition to the well-known finding of a stable OCB production, we demonstrated for the first time that the intrathecal antibody production is stable over time as well. This means, once an intrathecal antibody response against measles, rubella or varizella zoster virus was established it persisted at the follow-up analyses for at least 4 years. Although persistent IgG and antibody production is a well-known phenomenon in multiple sclerosis, the mechanisms remain unknown. It has been demonstrated recently in an experimental model of B cell migration over the blood brain barrier that activated human B cells readily cross the blood brain barrier with the help of adhesion molecules and chemokines \[[@B5]\]. Longevity of B cells and plasma cells is a common feature \[[@B28]\], and the microenvironment of MS lesions promotes the persistence and activation of B cells \[[@B29]\]. Although repeated antigenic stimulation favors longevity of B cells, persistance of antigen is not required for the survival of B cells \[[@B30]\]. Instead, high expression of anti-apoptotic proteins may be involved in the survival of B cells. Interestingly, increased expression of the apoptosis-inhibitory proteins Bcl-2 and FLIP were detected in B cells of MS patients \[[@B6]\]. Alltogether, microenvironment and antiapoptotic signaling might explain in part the persisting IgG and antibody production in the CSF of MS patients even in the absence of persistent antigen exposure. Expansion of antigen response ----------------------------- More intriguing was the finding that the number of patients positive for intrathecal OCB increased over time from 94 to 100%. It has been assumed that in early disease activated B cells might not yet have become resident in the CNS. Progression from a monoclonal to an oligoclonal CSF pattern has been reported in multiple sclerosis patients \[[@B31]\]. An increase in CNS clones after the initial stage of the disease has been suggested as a possible explanation. Further evidence for an expanding B cell activity during the course of the disease comes from a CSF analysis in which the prevalence of mature plasma cells was found to be higher in patients with longer disease duration \[[@B32]\]. In our study, an increasing number of patients showed a newly established intrathecal antibody response against measles virus: whereas 64% of patients were positive for measles at the 1^st^LP, 71% became positive at the 2^nd^and 86% at the 5^th^LP. Furthermore, in two out of seven patients with repeated CSF analysis, there was an expansion of the B cell response from one to three viral antigens by the 5^th^LP. To our knowledge, this is the first follow-up report on intrathecal polyspecific antibody production in multiple sclerosis patients. This follow-up for the first time provides evidence for an expansion of antibody reactivity over time in multiple sclerosis patients. However, this expansion of antibody is in contrast to the stable local IgG synthesis over time. One possible explanation for a polyspecific intrathecal antibody production might be a bystander activation of B cells with a given antibody reactivity. This is possible in the presence or absence of T cells. If certain cytokines -namely tumor necrosis factor alpha- is present, IgG secretion by activated B cells may occur even in the absence of T cells \[[@B33]\]. Why preferentially measles, rubella, varicella zoster and to a lesser extent other virus reactive B cells become activated during the disease process remains unclear. Association between B cell activity and clinical aspects of MS -------------------------------------------------------------- Although the evidence of B cell activity in the CNS is a common feature in MS, its clinical significance remains to be elucidated. Early studies on fluctuations in the CSF antibody production failed to show an association between antibody titers and the clinical course, possibly due to methodological shortcomings \[[@B17],[@B18]\]. In contrast, a recent publication revealed an association between a very high IgG index and a rapidly progressing course of the disease \[[@B34]\]. Further evidence for an impact of B cell products on disease severity comes from a publication by Villar and colleague. They found an intrathecal IgM synthesis to be associated with a worse clinical outcome \[[@B35]\]. Accordingly, a predominance of B cells over T cells and macrophages in the CSF has been shown to be associated with a more rapid disease progression, but not with a longer disease duration or a higher disability score \[[@B36]\]. Interestingly, when our patients were divided into two subgroups according to the rate of disease progression, there were more measles positive patients in the rapidly progressive subgroup. Furthermore, the AI for measles was significantly higher in those with a rapid progression of disability. Conclusion ========== In summary, our data indicate that once an intrathecal antibody production has been established it is stable over time. Furthermore, an expansion in antibody specifity occurred in a proportion of patients. The presented data support the hypothesis that a strong B cell activation is associated with a worse clinical outcome in MS. List of abbreviations ===================== AI -- antibody specific index CNS -- central nervous system CSF -- cerebrospinal fluid EDSS -- expanded disability status scale Ig -- immunoglobulin LP -- lumar pucture MS -- multiple sclerosis OCB -- oligoclonal bands Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= H-FP gave the idea of this study and participated in the design of the study. The statistical analysis and the design of the study was performed by DR. Both authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:58.445531	2005-6-27	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173126/", "journal": "Cerebrospinal Fluid Res. 2005 Jun 27; 2:3", "authors": [ { "first": "Hela-Felicitas", "last": "Petereit" }, { "first": "Dirk", "last": "Reske" } ] }
PMC1173127	Background ========== Chronic inflammation and airway smooth muscle dysfunction are consistent features of asthma responsible for disease progression and airway remodeling \[[@B1]\]. The increase in bronchial smooth muscle, both hypertrophy \[[@B2]\] and hyperplasia \[[@B3]\], plays a critical role in the development of airway hyperreactivity (AHR), the hallmark of asthma. Airway smooth muscle cells (ASMC) may also play a secretory or immunomodulatory role by producing pro-inflammatory cytokines, chemokines, polypeptide growth factors, extracellular matrix proteins, cell adhesion receptors, and co-stimulatory molecules, which perpetuate submucosal inflammation \[[@B4],[@B5]\]. These mediators may act on the ASM itself in an autocrine manner as well to further contribute to the asthma phenotype \[[@B6]\]. Therefore, smooth muscle itself may be capable of initiating and maintaining airway inflammation. Also, ASMC have been shown to undergo cell migration, which could contribute to airway remodeling \[[@B7]\]. Thus, regulation of airway smooth muscle hypertrophy and migration may be a new target for treatment of asthma \[[@B7],[@B8]\]. It is well known that IgE plays a critical role in the pathogenesis of asthma in the early and late phases by interacting with its two receptors, the high affinity receptor (FcεRI) and the low affinity receptor (FcεRII) \[[@B9]\]. IgE plays a key role in bronchial hyperresponsiveness and smooth muscle hyperreactivity \[[@B8]\]. Crosslinking of the high affinity IgE receptor (FcεRI) on mast cells leads to cellular degranulation and the release of various proinflammatory mediators and cytokines contributing to bronchoconstriction. The low affinity IgE receptor (CD23) (FcεRII) has been identified on B cells, monocytes, follicular dendritic cells, Langerhan\'s cells, eosinophils, and platelets \[[@B10]\]. Upregulation of the CD23 receptor is thought to increase allergic responses in the bronchial mucosa through the enhancement of antigen uptake and presentation \[[@B8]\]. The receptor has two isoforms that differ only in their cytoplasmic domains \[[@B11]\]. CD23a is constitutively expressed on B cells and is associated with endocytosis of IgE coated particles, and CD23b is induced by IL-4 and is also found on non- B cells such as T cells, Langerhan\'s cells, monocytes, macrophages, platelets, and eosinophils \[[@B12],[@B13]\]. IL-4 causes CD23 induction on B cells through CD40 \[[@B12]\]. CD23b mediates phagocytosis of soluble IgE complexes. An autocatalytic process involving cleavage of membrane bound CD23 by a matrix metalloprotease yields a series of soluble elements (sCD23) which increase IgE production via the CD21 receptor on B cells \[[@B13],[@B14]\]. The CD23 receptor has been shown to be upregulated on monocytes and alveolar macrophages in a T helper cell type 2 (TH2) environment and may contribute to chronic inflammation in asthma through this mechanism \[[@B15]\]. It has been shown that IL-4 and GM-CSF induce CD23 expression on monocytes, and GM-CSF primes monocytes for cellular activation and secretion of IL-1 upon subsequent exposure to IgE-containing immune complexes \[[@B8]\]. CD23 is also involved in antigen presentation to B cells as well as cellular interactions between B and T cells \[[@B12]\]. In previous studies, Hakonarson et al. \[[@B16]\] demonstrated the expression of CD23 (FcεRII) messages and a low level of the protein in airway smooth muscle cells. In two patients who died of status asthmaticus, CD23 expression was also markedly upregulated on ASMC. The CD23 expression was inducible with human atopic sera or IgE immune complexes in naïve (control) ASMC, and this upregulation was blocked when pretreated with anti-CD23 blocking antibody. The authors concluded that IgE coupled activation of CD23 contributes largely to its upregulation \[[@B5],[@B14]\]. In a corresponding experiment with rabbit ASMC subjected to control and atopic serum, they were able to demonstrate, through Western blot analysis, a markedly enhanced expression of CD23 in the ASMC sensitized with atopic serum or IgE immune complexes. They were able to achieve significant inhibition of upregulation by pretreatment with anti-CD23 mAb. They hypothesized that IgE was responsible for the upregulation of the low affinity IgE receptor \[[@B17]\]. Hakonarson, et al. have also demonstrated that ASMC in vitro exposed to human atopic sera results in an initial increase in TH2 cytokines including IL-5 and GM-CSF followed hours later by production of IL-1 and TH1 cytokines \[[@B18],[@B19]\]. Recently, phase I trials have been completed on IDEC-152, an IgG1 anti-CD23 antibody, for patients with mild to moderate persistent asthma. The drug was well tolerated by participants, and a dose-dependent decrease in mean IgE values was reported \[[@B20]\]. T helper cell type 2 (TH2) mediated inflammatory cytokines, such as IL-4, IL-13, and IL-5, as well as other enzymes and chemokines are active in the asthmatic patient. GM-CSF has been shown to be involved in asthma pathogenesis and in vivo can induce TH2 differentiation independent of IL-4 \[[@B21]\]. Tryptase and prostagladin D2 (PGD2) were chosen as major mast cell mediators, and more recently the PGD2 receptor gene (PTGDR) has been shown to be an asthma susceptibility gene \[[@B22]-[@B24]\]. It is possible that many factors are responsible for the upregulation of the low affinity IgE receptor in addition to and independent of IgE. The purpose of this study was to identify specific mediators released in the asthmatic patient that are responsible for the upregulation of CD23 on human airway smooth muscle cells independent of IgE. Methods ======= Cell culture and flow cytometry ------------------------------- Alpha-smooth muscle isoactin positive Human ASMC (Cambrex, Walkersville, MD) in T-75 flasks were starved for 24 hours in 0.1% (vol/vol) fetal bovine serum (FBS) containing medium M199 (Cellgro, Herndon, VA) supplemented with 1% (vol/vol) antibiotic/antimycotic solution (Sigma Chemical Co., St Louis, MO). The cells were then stimulated with either vehicle (bovine serum albumin, BSA, vehicle for cytokines (1 mg/ml), in M199; ethanol (EtOH), vehicle for LTD4 (6% final concentration) and PGD2 (0.001--0.01% final concentration); M199, vehicle for tryptase), an individual mediator, or a mediator in combination with GM-CSF at their optimum concentrations for 24 hours. The doses of cytokines used were up to four time ED50 including: IL-4 (0.04--1 nM), GM-CSF (0.07--0.8 nM), IL-13 (0.4 nM), IL-5 (0.01--0.07 nM), IL-13 (0.3--2.2 nM), PGD2 (1--10 μM), LTD4 (1--10 μM), tryptase (30 nM, a concentration sufficient to induce ASMC proliferation) (Sigma). Dose ranging studies were performed to determine the optimum concentration of IL-4 and GM-CSF on the expression of CD23, and the doses chosen were IL-4 (0.5 nM) and GM-CSF (0.4 nM) (Figure [1](#F1){ref-type="fig"}). All cytokines were obtained from R & D Systems Inc. Minneapolis, MN except GM-CSF which was obtained from Sigma. The cells were then harvested with a soft rubber edged scraper, centrifuged for 5 minutes at 1000 rpm (200 g), washed and resuspended in 1% BSA in phosphate buffered saline (PBS) and fixed with 70% ETOH. After washing twice more, the cells were resuspended in 1% BSA in PBS. Finally, they were filtered through a 40 μm nylon mesh to obtain single cell suspension and stained with (20 μl) of PE (phycoerythrin)-CD23 (EBVCS-5, BD Biosciences, San Jose, CA) or PE-mouse IgG~1~for 15 minutes in the dark to facilitate staining for flow cytometry. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Upregulation of CD23 by IL-4 and GM-CSF. Dose-ranging studies were performed to determine the optimum concentrations of IL-4 and GM-CSF. Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml), IL-4 (0.125. 0.25, 0.5, or 1.0 nM) or GM-CSF (0.1, 0.2, 0.4, or 0.8 nM) for 24 h. The cell lysates in RIPA buffer were subjected to western blot analysis for CD23. Mouse anti-human CD23 monoclonal antibody (clone M-L233, BD Biosciences, 1 μg/5 ml) was used as the primary antibody and anti-mouse horseradish peroxidase linked antibody as the secondary antibody (Amersham). The immunoreactive protein bands were detected by enhanced chemiluminescence light (ECL) (Amersham). ::: ![](1476-7961-3-6-1) ::: Protein analysis ---------------- A commercially available bicinchoninic acid (BCA) kit (Pierce, Rockford, IL) was used for protein analysis according to the manufacturer\'s instructions. The optical densities were read using a Bio-Kinetics EL-312 Microplate reader. Indirect immunofluorescence --------------------------- Indirect immunofluorescence stainings were performed with anti-smooth muscle-α isoactin antibody (Sigma) and anti-human CD23 antibody (M-L233, 1 μg/ml, BD Biosciences), which are specific monoclonal antibodies and either a FITC or TRITC fluorochrome, conjugated second antibody. Fixed huASMC were incubated with the above antibodies diluted in PBS with 3% BSA for 60 min at room temperature. The cells were then washed three times with PBS for 10 minutes for each wash. Non-specific binding was blocked by incubating cells with 3% BSA in PBS for 60 minutes. The blocking solution was then removed and cells were incubated with FITC- or TRITC- fluorochrome conjugated antibody for 45 minutes in the dark to facilitate staining. Cells were then washed with PBS three times. Finally, one drop of Fluoromount-G (Southern Biotechnology Inc., Birmingham, AL) was added. Western blot ------------ Standard Western blot analyses were performed to detect anti-STAT6 (1:500, Calbiochem, San Diego, CA) polyclonal rabbit, anti-p-STAT-6 (1:500, Calbiochem) polyclonal rabbit. Human ASMC lysates in radio-immunoprecipitation assay (RIPA) buffer were transferred onto Hybond-ECL nitrocellulose membranes and were immunoblotted with monoclonal anti-human CD23 (1:500 dilution, clone M-L233, 1 μg/5 ml, BD Biosciences), polyclonal anti-IL-4Rα (1:500 dilution, Santa Cruz), monoclonal anti-IL-2Rγc (1:250 dilution, R&D Systems, Inc.). The nitrocellulose membranes were incubated with a 1:1,000 dilution of anti-rabbit or anti-mouse horseradish peroxidase linked whole antibody (Amersham, Piscataway, NJ) in PBS-T for 1 hour at room temperature. Paxillin monoclonal antibody (1:500 dilution, Transduction Laboratories) was used as a positive isotype control for CD23, and fibronectin polyclonal antibody (1:250, Sigma) was used as a positive control for the remaining antibodies. The immunoreactive protein bands were detected by enhanced chemiluminescence light (ECL) (Amersham). Statistical analysis -------------------- Data were analyzed with Prism 4 software (GraphPad, San Diego, CA). One-way analysis of variance (ANOVA) was used. Results are expressed as mean ± SEM. A P value less than 0.05 was considered statistically significant. Results ======= CD23 protein expression is upregulated in huASMC by IL-4, GM-CSF, or IL-4/GM-CSF -------------------------------------------------------------------------------- Previous studies have shown that IgE immune complexes in atopic serum caused an increase in CD23 expression in ASMC \[[@B16]\]. To determine if other humoral factors in atopic serum effect CD23 expression in human ASMC, we have tested the effect of the relevant cytokines, arachidonic acid metabolites, and the mast cell enzyme tryptase. Flow cytometry was performed to evaluate differences in cell populations after stimulation of the huASMC for 24 hours with either individual mediators IL-4 (0.5 nM), GM-CSF (0.4 nM), IL-13 (0.4 nM), IL-5 (0.4 nM), PGD2 (10 μM), LTD4 (10 μM), tryptase (30 nM) or a combination of IL-4, IL-5, and IL-13 each with GM-CSF. Within the huASMC stimulated by IL-4, GM-CSF or the combination of IL-4/GM-CSF, two populations of cells were detected distinguishable by cell size. While the smaller cells did not show a significant expression of CD23, many of the larger cells showed increased expression of CD23. In the example in Figure [2](#F2){ref-type="fig"}, 66% of the larger cells (gate D) showed an increase in cell expression of CD23 when compared to the controls. As stated previously, the functions of ASMC are heterogeneous including proliferation and synthesis. Previous studies have shown, on flow cytometry of ASMC stimulated in vitro with IL-1β and TNF-α, only 20--60% of ASMC produce GM-CSF. The ASMC producing GM-CSF include some which also have increased proliferative properties. This suggests that considerable heterogeneity exists in the phenotypic expression of the ASMC in culture \[[@B25]\]. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Upregulation of CD23 by IL-4, GM-CSF and IL-4/GM-CSF. Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with IL-4 (0.5 nM)/GM-CSF (0.4 nM) for 24 h. The FACS analysis showed that the smaller cells passed through Gate A and larger cells passed through Gate D. Background noise was eliminated using the BSA-stimulated control cells that were labeled with PE-anti-CD23 (EBVCS), represented by C in Gate A, and F in Gate D. The FACS results of a representative experiment showed 66% of the larger cells (Gate D) had an increase in cell expression of CD23 when compared to the controls. ::: ![](1476-7961-3-6-2) ::: In addition to the combination of IL-4/GM-CSF inducing increased expression of CD23, both IL-4 and GM-CSF alone independently increased the expression of CD23 in huASMC. The percentage of cells with increased fluorescence intensity above the control was 25.1 ± 4.2% (IL-4), 15.6 ± 2.7% (GM-CSF) and 32.9 ± 13.9% (IL-4/GM-CSF combination). On the other hand, IL-5, IL-13, cysteinyl leukotrienes, and tryptase did not induce CD23 expression (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Il-4, GM-CSF, and IL-4/GM-CSF Increase CD23 Expression on ASMC (n = 3) ::: ------------------------------------------------------------------------------ Cytokine (24 hours) \#Cells in Gate D\ \%Cells with Increased CD23\ (n = 3) Expression Above the Control -------------------------- -------------------- ------------------------------ BSA (1 μg/ml) 1,343 ± 122 0 IL-4 (0.5 nM) 1,413 ± 197 25.1 ± 4.2\* GM-CSF (0.4 nM) 1,346 ± 243 15.6 ± 2.7\* IL-4/GM-CSF (0.4/0.5 nM) 1,324 203 32.9 ± 13.9\* IL-5 (0.4 nM) 1,130 ± 251 0 IL-13 (0.4 nM) 1,316 ± 269 0 PGD2 (1 μM) 1,521 ± 123 0 PGD2 (10 μM) 1,159 ± 204 0 LTD4 (10 μM) 2,037 ± 375 0 Ethanol (6% vol/vol) 2,507 ± 200 0 Tryptase (10 μM) 2,385 ne ± 405 0 ------------------------------------------------------------------------------ Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml), IL-4 (0.5 nM), GM-CSF (0.4 nM), IL-4 (0.5 nM)/GM-CSF (0.4 nM), IL-13 (0.4 nM), IL-5 (0.4 nM), PGD2 (10 μM), LTD4 (10 μM), tryptase (30 nM) for 24 h. Results are mean ± SEM of the percentage of cells with an increased fluorescence intensity above the control (n = 3). Control values: BSA-stimulated, PE-anti-CD23 labeled, 10.9 ± 1.4 %; BSA-stimulated, PE-mouse IgG~1~, non-immune, 0.5 ± 0.1 % (n = 3). \denotes significant increase in CD23 expression above the BSA control value. ::: Expression of CD23 in response to IL-4, GM-CSF, IL-4/GM-CSF is accompanied by changes in huASMC morphology ---------------------------------------------------------------------------------------------------------- Western blot analysis of huASMC stimulated with IL-4, GM-CSF, or Il-4/GM-CSF for 24 h showed an increase in CD23 expression compared to BSA vehicle control (Figure [3](#F3){ref-type="fig"}). Indirect immunofluorescence was used also to identify any morphological changes associated with the cytokine stimulation and upregulation of CD23 (Figure [4A--D](#F4){ref-type="fig"}). Those cells stimulated with the combination of IL-4/GM-CSF demonstrated CD23 expression along with changes in cell morphology including depolymerization of isoactin fibers, cell spreading, and membrane ruffling (Figure [4B](#F4){ref-type="fig"}). These changes in phenotype are consistent with flow cytometry results in that the larger cells expressed CD23 (Figure [4D](#F4){ref-type="fig"}). In contrast, the control BSA stimulated population showed no changes in cell cytoskeletal structure and morphology (Figure [4A](#F4){ref-type="fig"}) or specific staining for CD23 (Figure [4C](#F4){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Western blot analysis of CD23 after stimulation of IL-4, GM-CSF, IL-4/GM-CSF. Alpha-smooth muscle isoactin positive huASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml) (vehicle control), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.5 nM/0.4 nM) for 24 h. The cell lysates in RIPA buffer were subjected to western blot analysis for CD23. Mouse anti-human CD23 monoclonal antibody (clone M-L233, BD Biosciences, 1 μg/5 ml) was used as the primary antibody and anti-mouse horseradish peroxidase linked antibody as the secondary antibody (Amersham). The immunoreactive protein bands were detected by enhanced chemiluminescence light (ECL) (Amersham). Paxillin mouse monoclonal IgG~1~(Transduction Laboratories) was used as an irrelevant isotype control. ::: ![](1476-7961-3-6-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Expression of CD23 in response to IL-4/GM-CSF is accompanied by changes in huASMC morphology. Alpha-smooth muscle isoactin positive huASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with either BSA or IL-4 (0.5 nM)/GM-CSF (0.4 nM) for 24 h, and stained with either anti-smooth muscle-isoactin (A & B) or anti-CD23 antibody (C & D). Those cells stimulated with the combination of IL-4/GM-CSF demonstrated CD23 expression (D) and changes in cell morphology including depolymerization of isoactin fibers, cell spreading, and membrane ruffling (B). Cells stimulated with BSA (vehicle for IL-4/GM-CSF) alone did not increase the expression of CD23 (C) nor changes in phenotype (A). These findings were confirmed by three independent observers. ::: ![](1476-7961-3-6-4) ::: To confirm activity of protein synthesis, the protein content of the control and the experimental groups of cells were compared using a BCA protein analysis kit. Human ASMC were starved for 24 hours in 0.1% FBS containing medium M199 and then stimulated with BSA (1 μg/mL), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4 (0.5 nM)/GM-CSF (0.4 nM) for 24 hours. The protein content was increased by 19% in the IL-4/GM-CSF treated cells above that of the control (Table [2](#T2){ref-type="table"}). The increase in protein concentration with IL-4 alone was not statistically significant. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### IL-4/GM-CSF Combination Increases Protein Content in huASMC. ::: Cytokine (nM) mg/10^6^cells (n = 3) ------------------------- ----------------------- BSA (vehicle) 1.17 ± 0.08 IL-4 (0.5) 1.18 ± 0.01 GM-CSF 1.12 ± 0.03 IL-4 (0.5)/GM-CSF (0.4) 1.39 ± 0.02 \ Alpha-smooth muscle isoactin positive huASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml)), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.4/0.5 nM) for 24 h. The cell lysates in RIPA buffer were analyzed for protein content using a commercially available BCA kit (Pierce). The optical density was read using a Bio-Kinetics EL-312 Microplate reader. Results are mean ± SEM (n = 3). \denotes value significantly different from the BSA vehicle treated control. ::: Stimulation of huASMC with IL-4 induces phosphorylation of STAT-6 and expression of IL-2Rγc ------------------------------------------------------------------------------------------- IL-4 binds the IL-4R with high affinity, and signaling through IL-4 causes enhanced expression of IL-4R \[[@B21]\]. The induction of these genes is mediated through signal transduction molecules including signal transducer activator of transcription (STAT-6). The binding of IL-4 to its receptor complex induces the formation of an IL-4 receptor complex which consists of IL-4Rα and the common gamma chain (γc) of the receptors for IL-2, IL-4, IL-7, IL-9, IL-15, and IL-21 \[[@B21]\]. It has not been previously reported that airway smooth muscle cells express the IL-2Rγc, the signaling unit of the IL-4 receptors. Western blot analysis of IL-4Rα and IL-2Rγc in huASMC lysates showed the presence of these receptor components on huASMC. Figure [5](#F5){ref-type="fig"} shows abundant expression of IL-4Rα and a low level expression of IL-2Rγc protein on huASMC. After stimulation of huASMC with IL-4 (0.4 nM) for 24 h, a two fold increase in γc expression was observed compared to the BSA vehicle control (Figure [6](#F6){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### IL-4Rα and IL-2Rγc Expression in huASMC. A: IL-4Rα and IL-2Rγc expression in huASMC at baseline. Unstarved huASMC lysates were subjected to western blot analysis for IL-4Rα and IL-2Rγc using polyclonal anti-IL-4Rα (1:500 dilution, Santa Cruz), or monoclonal anti-IL-2Rγc (1:125 dilution, R&D Systems, Inc.). The nitrocellulose membranes were incubated with a 1:1,000 dilution of anti-rabbit or anti-mouse horseradish peroxidase linked whole antibody (Amersham). The immunoreactive protein bands were detected by ECL (Amersham). IL-2Rγc is minimally expressed in huASMC while IL-4Rα is expressed abundantly in huASMC. ::: ![](1476-7961-3-6-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Upregulation of IL-2Rγc Expression in huASMC by IL-4 and IL-4/GM-CSF. Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then either stimulated with BSA (vehicle) (1 μg/ml), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.5 nM/0.4 nM) for 24 hours. The IL-4 and IL-4/GM-CSF stimulated cells had increased IL-2Rγc expression compared to the BSA (vehicle) group. Fibronectin polyclonal rabbit antibody (Sigma) (1:250) was used as an irrelevant isotype control. ::: ![](1476-7961-3-6-6) ::: To confirm that IL-4 was activating the IL-4Rα during the stimulation of huASMC, we examined the phosphorylation of downstream STAT-6 by western blot. Human ASMC were starved for 24 h and then stimulated with IL-4 (0.4 nM) for fifteen minutes. Results of Western blot revealed an approximately a four fold increase in intensity of the band for phosphorylated-STAT-6 in IL-4 stimulated cells when compared to BSA control (Figure [7](#F7){ref-type="fig"}). This supports the role of an IL-4 mediated signal transduction pathway involvement in CD23 upregulation in huASMC. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Phosphorylation of STAT-6 by IL-4 and IL-4/GM-CSF in huASMC*. Starved huASMC were incubated with either BSA vehicle control (1 μg/ml), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.5 nM/0.4 nM) for 15 minutes. Standard Western blot analyses were performed to detect STAT-6 and phosphorylated-STAT-6 (p-STAT-6) using a anti-STAT-6 polyclonal rabbit antibody (Calbiochem) and anti-p-STAT-6 polyclonal rabbit antibody (Calbiochem). Anti-rabbit horseradish peroxidase linked antibody was used as the secondary antibody. Protein bands were detected by ECL. STAT-6 was abundantly expressed by all four groups, while p-STAT-6 was only expressed in the IL-4 and IL-4/GM-CSF groups. Fibronectin polyclonal rabbit antibody (Sigma) was used as an irrelevant isotype control and was abundantly expressed in all four groups. ::: ![](1476-7961-3-6-7) ::: Discussion ========== In our study, we have demonstrated that CD23, the low affinity IgE receptor, is upregulated on human airway smooth muscle cells by the cytokines IL-4, GM-CSF, and the combination of IL-4/GM-CSF. This upregulation of CD23 by the combination of IL-4 and GM-CSF was accompanied by an increase in cell volume and protein content, cytoskeletal depolymerization, cell spreading and membrane ruffling. Because ASMC require a doubling time of 48 hours, the increase in protein content could not be attributed to an increase in cell number. Also, the increase in cell number in gate D seen with LTD4 was likely secondary to the effects of ethanol (Table [1](#T1){ref-type="table"}). Stimulation of huASMC by IL-4 caused an activation of STAT-6 and an increase in γc expression. Collectively, our findings suggest that CD23 expression can be stimulated by IL-4 and GM-CSF cytokines independent of IgE in huASMC and the upregulation of CD23 may play a role in cell migration and hypertrophy. Previous studies have demonstrated an increase in CD23 expression in alveolar monocytes after stimulation with IL-4 and GM-CSF \[[@B10]\]. In that study, the use of the individual mediators alone did not increase the CD23 levels to that of asthmatic patients suggesting a possible synergistic role between IL-4 and GM-CSF. Our findings are consistent with these in that the combination of IL-4 and GM-CSF was most effective in upregulating CD23 in huASMC. Not all TH2 cytokines are involved in this process; IL-5 (0.4 nM) and IL-13 (0.4 nM) had no effect on CD23 expression. Cysteinyl leukotriene LTD4 (10 μM), prostaglandin PGD2 (10 μM) and tryptase (30 nM) did not induce CD23 expression on huASMC \[[@B26]\]. We evaluated the effect of stimulation of huASMC with IL-4 on phosphorylation of STAT-6 via the IL-4R which would confirm the presence of the receptor in huASMC. STAT-6 is a critical mediator of IL-4 stimulated gene activation, and it is regulated by both tyrosine and serine kinases \[[@B27]\]. It has been shown in a mouse model that STAT-6 binds the CD23a murine promoter, and STAT -/- mice stimulated with IL-4 are unable to upregulate CD23. This suggests STAT-6 is a critical mediator for IL-4 induced upregulation of CD23 \[[@B28]\]. IL-4 along with CD40 mediated signals are responsible for upregulation of CD23 on B cells \[[@B14]\]. In this study, we have confirmed the expression of the IL-4Rα and a low level of common gamma chain in huASMC and that after stimulation of huASMC with IL-4, there was a two fold increase in γc chain expression (Figure [6](#F6){ref-type="fig"}). Phosphorylation of STAT-6 after stimulation with IL-4 for 15 min confirms that IL-4 has bound and activated the IL-4 receptor complex (Figure [7](#F7){ref-type="fig"}). It has been shown that the common gamma chain is a functional β chain of the IL-4 receptor complex in certain cells \[[@B27]\], and our data suggest that this is the case in huASMC. Interestingly, IL-13 (4 nM, a concentration sufficient to simulate huASMC proliferation, unpublished observation) did not upregulate CD23. For proliferation of ASMC by IL-13, IL-4Rα and IL-13Rα1 are required for signal transduction and downstream activation of p44/42 extracellular regulated kinases (ERK, unpublished data). Apparently, IL-4Rα and IL-13Rα1 engagement is not sufficient for CD23 expression, further supporting the role of γc chain in CD23 expression by IL-4. The signs of signal transduction in response to IL-4, and the increase in protein content of the cell in response to IL-4 and GM-CSF combination (Table [2](#T2){ref-type="table"}) represent activation of transcription and translation of CD23 in this case. Coupling of GM-CSF and its receptor complex is known to activate ERK that may have contributed to the synergistic effect of GM-CSF on CD23 expression. CD23 expression was associated with changes in cell morphology including depolymerization of isoactin fibers, cell spreading, and membrane ruffling (Figure [4B](#F4){ref-type="fig"} &[4D](#F4){ref-type="fig"}). Actin in ASMC is in a dynamic state and undergoes polymerization-depolymerization during the contraction-relaxation cycle \[[@B29],[@B30]\]. Membrane ruffling and cell migration involve signaling pathways including PI3-kinase, Rac and other Rho family G protein members in a variety of cell types, including vascular smooth muscle cells. Rac has an essential role in cell migration and regulation of the actin cytoskeleton \[[@B31],[@B32]\]. Moreover, ASMC are capable of switching their phenotypes from contractile to synthetic phenotype that is mediated by Rho kinases \[[@B32],[@B33]\]. In summary, we have demonstrated that the low affinity IgE receptor can be induced on huASMC by specific cytokines including IL-4, GM-CSF, and the combination of IL-4/GM-CSF. The combination of IL-4/GM-CSF also induced morphologic changes in the ASMC that may contribute to the synthetic function or migration. In addition, IL-4 and IL-4/GM-CSF stimulation of huASMC increased the protein content of the cell, suggesting hypertrophy. Conclusion ========== T helper type 2 cytokines including IL-4 have major role in asthma pathogenesis. GM-CSF is a hemopoetic growth factor, mostly released by activated monocytes and T cells. Additional sources of GM-CSF include epithelium of asthmatic airways \[[@B34]\] and human airway smooth muscle cells \[[@B6],[@B35]\]. Therefore, the effect of GM-CSF on CD23 expression can be both via paracrine and autocrine mechanisms. Previous studies by Hakonarson et al. \[[@B16],[@B17]\] have shown that upregulation of the CD23 receptor has been associated with proasthmatic changes in agonist-mediated ASM constrictor and relaxant responsiveness. Our study suggests that CD23 expression is associated with elements of hypertrophy (i.e. an increase in cell volume and protein content), thus consistent with their findings. Inhaled corticosteroids, the mainstay in treatment of asthma, effectively reduce inflammation and remodeling of the epithelium and basement membrane. However, no agents have been proven effective in reducing smooth muscle mass in asthmatic patients. Recent study results on anti-CD23 therapy showed decrease in serum IgE. Further studies to intervene the upregulation of CD23 expression by cytokines IL-4 and GM-CSF may open a new avenue to target smooth muscle hypertrophy, an important element of severe asthma \[[@B2]\]. Competing interests =================== The author(s) declare that they have no competative intrests. List of abbreviations ===================== huASMC: Human airway smooth muscle cells FceRI: High-affinity receptor for IgE FceRII (CD23): Low-affinity receptor for IgE AHR: Airway hyperreactivity PGD2: Prostaglandin D2 LTD4: Leukotriene D4 PE: Phycoerythrin FBS: Fetal bovine serum BSA: Bovine serum albumin PBS: Phosphate bufferred saline BCA: Bicinchoninic acid FITC: Fluorescein Isothiocyanate TRITC: Tetramethyl Rhodamine Iso-Thiocyanate RIPA: Radio-immunoprecipitation assay ECL: Enhanced chemiluminescence light Rα: Receptor alpha γc: Common gamma chain FACS: Fluorescent Activated Cell Sorter STAT: Signal Transducer Activator of Transcription TH2: T helper cell type 2 ERK: Extracellular regulated kinases Authors\' contributions ======================= JTB participated in designing experiments and performing CD23 analysis by flow cytometry. RKG carried out flow cytometry, immunofluorescence studies, and drafted the manuscript. HM carried out western blot analyses. DBL supervised all aspects of the project. All authors have read and approved the final manuscript. Acknowledgements ================ This work was supported in part by the grant from the Le Bonheur Children\'s Medical Center, NIH-HL56812, Children\'s Foundation Research Center, Molecular Resource Center, and American Academy of Allergy, Asthma and Immunology. The authors would like thank Jan Aldrich for her technical assistance.	PubMed Central	2024-06-05T03:55:58.447562	2005-5-20	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173127/", "journal": "Clin Mol Allergy. 2005 May 20; 3:6", "authors": [ { "first": "Joseph T", "last": "Belleau" }, { "first": "Radha K", "last": "Gandhi" }, { "first": "Holly M", "last": "McPherson" }, { "first": "D Betty", "last": "Lew" } ] }
PMC1173128	Background ========== Allergic asthma ranks in Germany among frequent occurred occupational diseases. Cow hair allergy in Finland is one of the most frequent reasons for the development of occupational asthma \[[@B1]\], but epidemiologically in Germany still insufficiently explored. The prevalence of respiratory complaints in Finnish farmers with cow husbandry is 27% for rhinitis respectively 12% for asthma in consequence of occupational exposure \[[@B2],[@B3]\]. Until now there is only limited insight into the risk factors with regard to this disease. The prevention of sensitization in stock farming to cow hair and epithelial allergens including lipocalins is often difficult due to intense exposure, with traditional milking still requiring close animal contact. The German Professionals insurance laws require a complete elimination of allergens for the recognition of an allergic airway disease as a compensable occupational disease. Complete avoidance of occupational exposure of cow hair allergens is only possible if stock farming is abandoned. This is, however, often not feasible in small dairy plants because of the resulting loss of income. In the present case report we discuss the efficiency of the strategy of exposure minimization to allergens by the use of a milking robot, i.e., an automatic milking system (AMS) as an alternative measure to total elimination of allergens for the prevention of occupational asthma caused by cow hair allergy. Case presentation and discussion ================================ A 37-year-old female farmer reported to suffer from daily asthmatic complaints by milking of cows with a conventional device. Exposure associated respiratory symptoms increased continuously during a period of 12 years. Allergic bronchial asthma, caused by sensitization against cow dust-derived allergens, was diagnosed by a positive skin prick test (Bencard^®^) and by the detection of IgE antibodies against Bos d 2at 74.9 kIU/l (Pharmacia, UniCAP^®^). Moreover, nonspecific bronchial hyperresponsiveness was verified by a positive histamine inhalation challenge test. The diagnosis of occupational asthma was confirmed by a separate specific inhalation challenge test using a 10% extract of cow hair allergens which led to a 330% increase of airway resistance (Raw). The German Professionals insurance laws require complete elimination of allergen exposures for the compensation of occupational allergic asthma as occupational disease \[[@B4]\]. Legal consequence would usually be to abandon dairy farming. However, in our case presented, dairy farming substantially contributed to family-income and the compensation offered would be insufficient for subsistence. Hyposensitization is the most effective measure in the prevention of allergic symptoms in patients with asthma. In that farmer the hyposensitization was not feasible due to permanent exposure to allergens. This farmer did not wear a helmet during the milking process because of claustrophobic panic attacks (fear of suffocation). Moreover an immunotherapy with recombinant lipocalins is currently not available. Respirator helmets have been predominantly used as effective measures for prevention of the inhalation of fumes and dusts \[[@B5]\]. Investigations about the effectiveness of respirator helmets against high-molecular allergens to prevent allergic asthma are rare. Taivainen et al. (1998) demonstrated that due to the use of a dust respirator helmet with a P2-class filter, dairy farmers with occupational asthma induced by cow dander showed less rhinitic symptoms and an improved peak flow rate \[[@B6]\]. However, workers using respirator helmets were not prevented against a progressive behavior of occupational asthma \[[@B7],[@B8]\]. Thus it is to doubt whether respirator helmets are efficient in farmers with occupational asthma to provide a complete protection against allergens \[[@B9]\]. To enable further dairy farming, the farmer installed a milking robot in 1999, i.e., an automatic milking system (AMS). At present circa 1300 AMS are installed, predominantly in the Netherlands, followed by Germany. AMS were developed in the course of increasing automation of the agricultural production. AMS consist of a milking box with the milking equipment and an electronic controlling system, an automatically operating dosing unit of a specific food-concentrate as well as a software controlled management system. The cows are decoyed from a specific food-concentrate to enter the milking box. A transponder, fixed on the cow neck, initiates milking operations automatically controlled by software and laser. The milking robot operates in the cowshed, which is air-shielded from the control-centre. After the installation of AMS the farmer was involved only in terms of monitoring the milking process. Figure [1](#F1){ref-type="fig"} shows the job and the current exposure situation of the farmer. It is apparent that the novel milking technique reduced exposure significantly by some 90% (from over 2 hours to approx. 10 min.) without a decline of milk production. Beyond allergen exposure reduction, the physical workload was lessened and the time available for activities to improve coping was increased -- these factors are increasingly recognized as relevant for the prognosis of asthma \[[@B10]\]. Table [1](#T1){ref-type="table"} shows the influence of the AMS technique on the current occupational exposure of the farmer. According to German Professionals insurance laws all conceivable measures must be taken into account even AMS is an expensive technical device to prevent occupational asthma. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Milking process monitoring by an air-shielded control-centre. ::: ![](1476-7961-3-8-1) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Parameters influenced by automatic milking system (AMS) ::: -------------------------------------------------------------------------------------------------------- Parameter Power of parameter Influence of AMS ----------------------- ------------------------ ------------------------------------------------------- Daily exposure period ↓↓↓ Drop of more than 2 hours to approx. 10 min. (\> 90%) Exposure intensity ↓↓↓ Increasing space to cow Physical workload ↓↓ Monitoring tasks Timing flexibility ↑↑ • No commitment to fixed milking times\ • Free working plan Quality of life ↑ More time available for other activities -------------------------------------------------------------------------------------------------------- Explanation: ↓ = reduction of power, ↑ = improvement of power ::: After installation of the milking robot the farmer reported significantly less frequent allergic and asthmatic symptoms. Asthma medication was reduced and required only once daily in the evening. Improvement was noted also in terms of lung-function and decreased total serum IgE: serum IgE decreased from 1332 IU/ml before milking robot installation to 572 and 474 IU/ml two and three years after installation, respectively. FEV~1~improved by 20% and vital capacity by 11%, comparing values one year before with two years after installation. A CAP-RAST score \"5\" versus Bos d 2was determined after the intervention. Preceding EAST-Test results vs. Bos d 2were not suitable for comparison. Current discussions address threshold limit values for allergens that may prevent allergic symptoms \[[@B11]\]. These have been proposed also for the cow dust-derived allergen Bos d 2\[[@B12]\]. Table [2](#T2){ref-type="table"} shows examples for successful measures of exposure minimization at workplaces to reduce sensitizations. Recent studies have already demonstrated that allergen exposure reduction at workplaces is successful in reducing sensitizations, e.g., to latex proteins \[[@B13],[@B14]\], laboratory animals \[[@B15]\] or acid anhydrides \[[@B16]\]. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Examples for successful measures of exposure minimization at workplaces in reducing of sensitizations ::: Exposure areas Allergen Measure of exposure minimization Study ------------------------------ --------------------------------- ------------------------------------------------------------------ ----------------------------------------- Hospitals Latex proteins Replacement of powdered gloves through powder-free gloves Edelstam et al. 2002, Tarlo et al. 1994 Animal research laboratories Allergens to laboratory animals Technical improvement of housing cabinets for animals Thulin et al. 2002 Electrical industry Acid anhydrides Installation of a closed system in the production of epoxy resin Drexler et al. 1999 ::: Conclusion ========== The case presented and the evidence obtained from the literature indicates that the strategy of exposure minimization to allergens at workplaces can be an effective alternative to total elimination, which may not be feasible for economical reasons. In farmers with cow dust allergy a milking robot is an appropriate technical measure to minimize allergen-exposure particularly with regard to the reduced animal contact. Independently from this, such milking robots also increasingly used to enhance productivity (usually in herds comprising more than 50 animals). The efficacy of milking robots in terms of primary and secondary prevention should be evaluated also epidemiologically in a suitable cohort study, preferably in countries with intensive dairy farming. List of abbreviation ==================== AMS; Automatic milking system Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= GK was the principal investigator. GK examined the patient, performed the investigations and drafted the manuscript. HCB revised the manuscript. WU revised the manuscript and evaluated the epidemiological aspects. HD conceived the case report. All authors read and approved the final manuscript.	PubMed Central	2024-06-05T03:55:58.449923	2005-6-22	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173128/", "journal": "Clin Mol Allergy. 2005 Jun 22; 3:8", "authors": [ { "first": "Gintautas", "last": "Korinth" }, { "first": "Horst Christoph", "last": "Broding" }, { "first": "Wolfgang", "last": "Uter" }, { "first": "Hans", "last": "Drexler" } ] }
PMC1173129	Background ========== The average annual incidence rate of hepatitis A virus (HAV) infection varies significantly worldwide ranging from 9.1 per 100,000 in the United States and Europe to 500 per 100,000 in the Central Asian Republics \[[@B1],[@B2]\]. According to the latest available data, the incidence of HAV infection remains as high as 150 cases per 100,000 population in the Central Asian Republics and Kazakhstan in 2001 \[[@B3],[@B4]\]. Although HAV infection incidence has decreased slightly compared to the figures for the last decade, the persistence of high incidence in Kazakhstan is mainly due to problems related to poor supplies of good-quality drinking water, sanitation and hygiene practices \[[@B1],[@B5],[@B6]\]. It has been documented that the incidence of HAV infection is closely linked to poor socio-economic living conditions \[[@B3]\]. HAV infection is usually a mild illness, but can be severe if it occurs concurrently with other diseases or comorbid conditions \[[@B7]-[@B10]\]. In Britain, and North America, a patient with HAV infection is hospitalized only if (a) (s)he is too sick with diarrhea, nausea, and vomiting to be managed appropriately with antiemetics and oral rehydration at home, (b) (s)he has hepatic and/or renal failure (sometimes seen with HAV-infection in persons with underlying alcoholic liver disease or with chronic hepatitis caused by other viral agents) \[[@B11]\], or (c) (s)he has a clinical presentation that is truly anomalous (e.g., meningoencephalitis has been reported in association with HAV infection but is exceedingly rare \[[@B12]\]). However, in many republics of the former Soviet Union, HAV infection has been a cause for routine hospitalization, as mandated in administrative directives. The premise on which many such patients are hospitalized should be reconsidered because relatively few patients with HAV infection become seriously ill, no treatment to influence the underlying course of the disease exists, and neither admission to hospital nor isolation from others in the home lessens the spread of the infection \[[@B13]\]. Because HAV infection is neither a severe illness nor one with important sequelae, it has been observed that the rates of infection of family members of patients with mild HAV managed at home or in hospital were similar \[[@B13]-[@B15]\]. Previous studies using data from the region reported that the hospitalization of patients after the appearance of jaundice was unjustified, and that the quality of treatment between inpatient and outpatient settings would be similar \[[@B13],[@B14]\]. However, in severe cases (\<10% of HAV infections), patients are more likely to be managed in a hospital and to receive more intensive treatment, including ultrasound and computerized tomography exams, as well as liver biopsies \[[@B7],[@B8],[@B10]\]. In such cases, the assumption of similar quality of inpatient and outpatient treatment does not hold. Using data gathered from patients with mild HAV infection in Almaty, Kazakhstan, we conducted a cost-comparison analysis of the policy of mandatory hospitalization and an alternative policy of outpatient management for patients with mild HAV infection. The purposes of this study were (a) to determine the mean treatment costs associated with managing cases of mild HAV infection in an inpatient setting under a policy of mandatory hospitalization, (b) to compare those costs with costs of treating patients with mild HAV infection using outpatient management, assuming a similar quality of treatment under inpatient and outpatient conditions, and (c) to determine total direct medical and nonmedical costs as well as productivity losses associated with mild HAV infection and then compare these costs both from the perspective of the resource-allocating authority (i.e., the Ministry of Health) and from a broader societal perspective. Materials and methods ===================== Data sources ------------ Data were collected using survey questionnaires that involved interviews with patients and medical practitioners between November 2000 and February 2001. Information was collected on patient\'s age, sex, principal diagnosis, duration of illness, use of services, and cost data. Patients were interviewed to collect information regarding out-of-pocket and direct nonmedical expenses, such as expenses paid for transportation to outpatient clinics and indirect costs associated with the number of days absent from work. Interviews with medical personnel and hospital administrators were conducted to determine the costs and quantify the amount of health and nonhealth staff time spent per day in caring for patients with mild HAV infection. A sample size of 451 subjects (200 inpatients and 251 outpatients) was used to conduct a cost-comparison analysis. We enrolled patients presenting to either of two hospitals in Almaty, Kazakhstan, in a geographical area where mandatory hospitalization for mild HAV infection was still practiced. We also enrolled patients in 19 outpatient clinics in geographical areas where mandatory hospitalization was no longer practiced. Patients selected for the study included members of both sexes, all age groups, and varied in ethnicity. Case definition --------------- A case of mild HAV infection was defined as a clinical presentation with manifestations of acute hepatitis and serologic confirmation of HAV infection in a person who reported to one of the selected hospitals or outpatient clinic sites. The clinical manifestations of HAV infection included sudden onset of fever, anorexia, fatigue, and/or abdominal discomfort followed within a few days by jaundice and hepatomegaly \[[@B16]-[@B18]\]. Serologic confirmation of mild HAV infection was determined by elevated alanine serum transferase (AST) not in excess of 5 times the upper limit of normal of the reference value, and by presence of detectable IgM anti-HAV Criteria for inclusion in this analysis were (a) having mild HAV infection as a primary or sole diagnosis confirmed by laboratory; (b) being between 1 and 65 years of age and being either hospitalized or managed at one of the study sites by a health care worker during the study period; and (c) providing voluntarily signed consent that included permission to review medical records. Persons who had multiple concurrent diseases or comorbid conditions and those who had not been diagnosed and tested positive for mild HAV infection more than 30 days before the study period were excluded from the study. Cost description ---------------- Costs for direct medical and non-medical services provided for patients with mild HAV infection were calculated based on interviews with medical providers, hospital administrators, health officials and patients. Units of direct medical services recorded included the number of days spent in hospital, health staff time allocated to treatment, and costs for those services specifically identified as part of the management of patients with mild HAV infection in hospital and outpatient clinics. Time spent by health care personnel was obtained based on reviews of medical records. Costs were based on estimates of costs for time spent by hospital- or clinic-personnel per day in caring for patients with mild HAV infection: in hospital, physicians spent approximately 54 minutes (min.), nurses approximately 89 min., and nonhealth staff about 40 min.; in the outpatient clinics, physicians spent approximately 26.4 min., nurses approximately 18 min., and nonhealth staff a negligible amount of time. In this analysis, we are considering only the base salary for the health care personnel; physicians are paid US\$15.84 per 8-hour workday, US\$4.8 per day for nurses, and US\$2.4 per day for nonhealth staff obtained from the budget of each health facility and based on the national contract for health personnel. Although physicians earn more than their base salary through various bonus payments and through informal payments from patients, consistent estimates of these payments could not be captured and were not included in this analysis \[[@B6],[@B19],[@B20]\]. Since an estimated 80% to 90% of physician income is obtained informally \[[@B20]\], it is difficult to track the flow of incomes for doctors, and we were able to use only the base salary for our estimates. The cost per patient-hospital day included the costs of purchase and repair of medical equipment and buildings, administrative and operational costs in the hospitals allocated to the treatment of mild HAV patients. To cover the full capital and recurrent cost, cost per patient-hospital day was determined based on average national estimates provided by hospital administrators \[[@B21]-[@B23]\]. Although hospital costs cover capital and recurrent costs such as the cost of building and medical equipment, laboratory services, health and nonhealth personnel, and hospital administrative costs, we estimated separately the costs for laboratory and personnel services for managing patients in hospital based on medical records and interviews with medical practitioners. Costs for hospital meals and prescription drugs were obtained based on patient self-reported out-of-pocket expenses. The cost of hospital meals was estimated for the respective number of inpatient hospital days. The number of hospital days was determined based on patient medical records. Because all patients paid for hospital meals and prescription drug charges where incurred, we calculated the cost of hospital meals and prescription drugs separately. Total treatment costs were calculated as the sum of direct medical and non-medical costs which included hospital costs, costs for time spent with patients by physicians, nurses, and other staff, laboratory and prescription drug costs, as well as non-medical costs such as hospital meals and transport. Because there are no long-term sequelae of mild HAV infection \[[@B7],[@B10]\], no home care was needed for such patients; therefore, the cost of home care was not calculated. Productivity losses were calculated as indirect costs associated with lost workdays due to illnesses of HAV infection. The number of lost workdays were determined based on self-reports of missed work. The exchange rate used was US\$1 = 148 Kazakhstan Tenge (KZT), a rate that has not varied more than 15% since January 2001. As of May 5, 2005, the exchange rate was US\$1 = 131 KZT. Treatment outcome ----------------- Having measured the direct medical and nonmedical costs, as well as productivity losses of managing mild HAV infection, we defined effective treatment or management for a mild HAV infection as an episode of care (which included all services provided) that led to a complete recovery. As has been observed epidemiologically elsewhere, we assumed that the rates of infection of family members of patients with mild HAV infection managed at home or in hospital were similar \[[@B13],[@B15]\]. Statistical analyses -------------------- We completed our analyses in five stages: First, we estimated the mean annual treatment cost per case of mild HAV infection when managed in inpatient and outpatient settings. Using SAS statistical programs, comparison between inpatient and outpatient groups were made using PROC TTEST of the null hypothesis of equal group variances, and we determined the difference in mean treatment cost per case using t tests of differences in means. The confidence interval (CI) for the difference between two sample means was obtained using a \"pooled\" estimate of the standard deviation \[[@B24]\]. Second, we stratified the samples by four different age groups, given significant differences in mean age between the two groups. We categorized patients into four age groups: ≤ 18, 19 -- 26, 27 -- 34, and ≥ 35 years to estimate the mean annual treatment cost per case of mild HAV infection by age group. We used PROC GLM to conduct one-way analysis of variance (ANOVA) using Bonferroni t tests of pair of means within age groups to test the null hypothesis of equality of the group means among different age groups. Third, we estimated the number of lost workdays for both inpatient and outpatient group. We calculated the productivity losses by multiplying the number of lost workdays by employed patients aged 18 -- 64 years or by adult parents or guardians of children with mild HAV infection, and by daily urban wage rates. In this analysis, we did not consider replacement costs for unemployed patients. Fourth, we estimated the total annual cost of mild HAV infection as the sum of the annual treatment costs consisting of direct medical and nonmedical costs as well as productivity losses resulting from lost workdays. Finally, we conducted sensitivity analyses to determine which parameters exert the greatest influence on the results of this analysis and establish the range of total annual cost of mild HAV infection. A p-value ≤ 0.05 was considered to be statistically significant. Epi Info software \[[@B25]\] was used for data entry. Data from questionnaire forms were entered twice into a computer database to reduce the chances of data-entry errors and were analyzed using SAS statistical programs \[[@B26]\]. Results ======= Table [1](#T1){ref-type="table"} presents sociodemographic characteristics of the inpatient and outpatient groups. The mean age of the inpatient group was 22 years; the mean age of the outpatient group was 12 years. The inpatient group reported a higher annual household income of US\$898.2 (2001 US\$) compared with US\$502.3 for the outpatient group. Several other comparisons reflected disparities in age of the two sample groups: 26.5% of the inpatient group was employed, compared with 6.7% of the outpatient group; 21% of the inpatient group were married, compared with 8.9% of the outpatient group; 32% of the inpatient group smoked cigarettes, compared with 9.9% of the outpatient group. The average wage rate per patient in urban areas was 14,617 KZT (US\$98.76) per month compared to 5000 KZT (US\$33.78) per patient in rural areas. There were 27.21 lost workdays for employed inpatients and 6.19 lost workdays for family members caring for minor patients in hospital; employed outpatients had an average of 15.73 lost workdays (lost workdays: 33.40 vs. 15.73). The average length of hospital stay for a patient with mild HAV infection was 9.7 days, compared with 3.4 visits per patient in the outpatient group. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Sociodemographic characteristics of patients with mild HAV infection, Almaty, Kazakhstan, 2001 ::: Variables All patients (n = 451) Inpatients (n = 200) Outpatients (n = 251) -------------------------------------- ------------------------ ---------------------- ----------------------- Gender (%) Male 50.9 50.0 51.4 Female 49.1 50.0 48.6 Marital status (%) Married 17.8 21.0 8.9 Unmarried 85.2 79.0 91.1 Education level (%) Never attended or kindergarten 13.2 0.5 23.4 School 1--8 years 55.8 72.0 42.7 Grades 9--11 years 16.3 27.5 17.3 High school graduate 1.1 \- 2.0 College 1--3 years or graduate 13.6 \- 14.5 Ethnicity (%) Kazakhs 40.6 36.0 44.2 Russians 49.2 54.5 45.0 Uzbek 2.4 2.0 2.8 Others 7.8 7.5 8.0 Employment status (%) Employed 16.4 26.5 6.7 Unemployed 83.6 73.5 93.3 Tobacco use (%) Smoker 20.3 32.0 9.9 Nonsmoker 79.7 68.0 90.1 Urban status (%) Urban 99.8 100.0 99.6 Rural 0.2 \- 0.4 Home ownership (%) Owned 98.2 99.5 97.2 Rented 1.8 0.5 2.8 Age group (%) ≤ 18 66.3 37.5 89.2 19 -- 26 22.6 42.5 6.8 27 -- 34 6.7 12.5 2.0 ≥ 35 4.4 7.5 2.0 Mean Age (years) (SE) ^¶^ 17 (0.41) 22 (0.49) 12 (0.43) Mean household income (US\$) (SE)^¶^ 886.6 (60.03) 898.2 (61.65) 502.3 (28.38) ^¶^Standard error (SE) are values in parenthesis. ::: Table [2](#T2){ref-type="table"} presents the mean treatment costs per patient with mild HAV infection in inpatient and outpatient settings. The mean total direct medical and nonmedical cost of managing mild HAV infection was estimated at US\$157.4 (2001 US\$) per inpatient (95% CI = US\$151 -- 163), compared with US\$22.5 per outpatient (95% CI = US\$21 -- 23). The hospital costs (US\$81) were calculated by taking the product of the average number of hospital days (9.7 days) and the cost per HAV patient-hospital day (US\$8.4) obtained from the survey. Of the US\$81 of hospital costs, as much as 40% (US\$32) of the hospital costs was allocated to the costs for the purchase and repair of medical equipment and buildings and 60% (US\$49) was spent on recurrent hospital administrative and operational costs including overhead and utilities. Laboratory costs accounted for 6.9% (US\$11.0) of the mean treatment costs for inpatients compared with 35.6% (US\$8.0) of those costs for outpatients. We estimated that about 19% (US\$30) of the mean treatment costs were allocated to hospital meals. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Mean treatment costs of managing cases of mild HAV infection for inpatient and outpatient settings, Almaty, Kazakhstan (2001 US\$) ::: Inpatient Outpatient ---------------------------- ----------- ------------ ---------------- ------ ------ --------------- Hospital cost 81.0 1.78 (89.0, 96.0) \- \- \- Physicians\' time 18.0 0.49 (17.0, 19.0) 3.0 0.08 (2.80, 3.20) Nurses\' time 9.0 0.17 (8.7, 9.30) 1.6 0.07 (1.46, 1.70) Non-health staff 2.0 0.04 (1.92, 2.10) \- \- \- Laboratory 11.0 0.35 (10.0, 12.0) 8.0 0.11 (7.8, 8.2) Prescription drugs 6.4 0.21 (6.19, 7.01) 5.9 0.33 (5.25, 6.55) Hospital meals 30.0 0.57 (29.0, 31.0) \- \- \- Transport \- \- \- 4.0 0.20 (3.60, 4.40) Total inpatient/outpatient 157.4 3.04 (151.0, 163.0) 22.5 0.42 (21.0, 23.0) Productivity losses^¶^ 150.9 2.12 (146.7,155.1) 71.5 4.75 (66.8, 76.3) Total societal cost^§^ 308.3 4.90 (298.7, 317.9) 94.0 4.28 (85.6, 102.4) Note: Data may add exactly due to rounding ^†^Standard error (SE) ^‡^95% Confidence interval (95% CI) ^¶^Productivity losses calculated based on the number of lost workdays reported and average urban wage rate. ^§^Total societal costs include treatment costs and productivity losses ::: From the perspective of the Ministry of Health, the mean direct medical costs of managing mild HAV infection was estimated at US\$127.4 per patient in hospital, compared with US\$18.5 in an outpatient setting. From the societal perspective, the mean total cost of direct medical and nonmedical services as well as of productivity losses was estimated at US\$308.3 per patient in hospital compared with US\$94.0 per patient in an outpatient setting. Of the US\$157.4 total for direct medical and nonmedical services, a patient in hospital paid approximately US\$36.4 out of pocket costs for drugs and hospital meals, compared with US\$9.9 for such costs paid by a patient in an outpatient setting. Results from the PROC TTEST indicated that the difference between the mean treatment costs for management of mild HAV infection among the inpatient and outpatient groups was US\$134.6 (2001 US\$), 95% CI = (US\$ 128.0 -- 140.0) and statistically significant at (p≤ 0.0001). When we controlled for the effect of age on medical cost by stratifying the samples by age group, we found that the mean treatment costs in the outpatient setting was still considerably lower than those in the inpatient setting. For example, the mean treatment costs of managing cases of mild HAV infection for outpatients aged 19 to 26 years was estimated at 13.9% of that for the inpatients of the same age. The one-way ANOVA analysis indicated that the mean treatment costs per inpatient decreased as age increased (Figure [1](#F1){ref-type="fig"}). The mean treatment cost per inpatient for the age group ≥ 35 years was estimated at US\$144 compared with US\$163 for the age group ≤ 18 years (F-value = 1.40; p= 0.25). Results from this analysis indicated that although there were differences in population means among different age groups, these differences were not statistically significant. We also noted that the mean treatment cost per outpatient in the age group ≥ 35 years was estimated at US\$28 versus US\$22 for the age group ≤ 18 years (F-value = 2.05; p= 0.10). Again, results from this analysis indicated that although there were differences in population means, these differences were not statistically significant. Mean treatment costs as well as productivity losses per outpatient increase slightly as age increases (Table [3](#T3){ref-type="table"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Mean treatment costs per mild HAV inpatient by age group ::: ![](1478-7547-3-4-1) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Mean, median total cost and standard error of managing mild HAV infection, by age group and treatment setting (inpatient vs. outpatient), Almaty, Kazakhstan (2001 US\$) ::: Total cost\* per inpatient Total cost\* per outpatient ---------- ---------------------------- ----------------------------- ----- ----- ------------- ----- ≤ 18 75 320 (32.42) 319 224 92 (8.64) 97 19 -- 26 85 303 (6.72) 304 17 96 (4.75) 97 27 -- 34 25 302 (11.25) 304 5 111 (11.60) 111 ≥ 35 15 290 (12.88) 292 5 93 (12.95) 101 All 200 308.3 (4.90) 304 251 94 (4.28) 98 \* Note the total cost includes the treatment costs as well as productivity losses ^§^Sample size ^†^Standard error (SE) ::: Nationally, we estimated about 21,600 cases of mild HAV infection occurred during 2001. Of these cases, an estimated 19,008 employed patients with HAV infection would report lost workdays due to illness considering the 12% unemployment rate in Kazakhstan \[[@B6]\]. Assuming no replacement costs of unemployed sick workers, we calculated the productivity losses by multiplying the average lost wages of US\$150.9 per inpatient by the cases of employed inpatients. We estimated the productivity losses at US\$2.87 million (95% CI = US\$2.79 million -- US\$2.95 million). Using 21,600 cases of mild HAV infection, an average of 9.7 hospital days and 33.4 lost workdays per inpatient, we calculated a total of 209,520 hospital days and 721,440 lost workdays. Annual mild HAV infection treatment costs for inpatients were estimated at US\$3.39 million (95% CI = US\$3.26 million -- US\$3.52 million). These costs could be lowered by managing patients with mild HAV infection in an outpatient setting for as low as US\$486,000 (95% CI = US\$453,600 -- US\$496,800). From the societal perspective, the annual total cost of mild HAV infection for patients treated in hospital was estimated at US\$6.26 million, 95% CI = (US\$6.05 -- US\$6.47). The treatment costs represent 51% of the total costs whereas productivity losses account for 49% of the total costs. The total cost of managing mild HAV infection in an outpatient was estimated at US\$1.85 million, 95% CI= (US\$1.72 -- US\$1.95). The annual total cost of mild HAV infection could be reduced by almost 70% from US\$6.26 million to US\$1.85 million by encouraging patients with mild HAV infection to be treated in outpatient rather than in inpatient settings. Using an average saving (US\$214) per inpatient and 21,600 cases of mild HAV infection, the potential savings at the national level from changing the treatment setting from inpatient to outpatient would be about US\$4.62 million (2001 US\$) per year. Discussion ========== Our findings suggest that the mean treatment costs per patient managed in hospital were almost seven times the costs per patient in an outpatient setting. Our results were consistent with previous reports that inpatient care is more expensive than outpatient care settings \[[@B6]\]. Even though no other study has directly estimated the costs of mild HAV infection in Kazakhstan, we observed how expensive the inpatient setting in the region was, relative to the outpatient setting. Health systems in the Central Asian Republics, including Kazakhstan, have not been designed and managed with attention to the criteria of technical efficiency(the production of a defined set of services at the lowest cost within the health system; also known as operational efficiency) or allocative efficiency(the distribution of resources across varied range of services in an optimum manner). These health systems have tended to overinvest services provided in inpatient settings; hospitals utilize approximately 75% of the Kazakhstan health budget compared to 10% of the budget for outpatient care \[[@B6]\]. Many disease entities managed with outpatient treatments in European Union countries (including viral hepatitis and tuberculosis) are managed on an inpatient basis in Kazakhstan \[[@B6]\]. In order to provide optimal care within the budget limits imposed by the allocation of hospital resources to various emerging infectious diseases like human immunodeficiency virus (HIV) infection, health authorities should consider choosing health care practices known to be beneficial on the basis of evidence of cost-effectiveness \[[@B27]\]. Inefficiency in the use of public funds has resulted in the increasing practice of informal or \'under-the-table\' payments in the health system \[[@B6]\]. Informal out-of-pocket expenses are increasingly required due to declining funds for the health sector. In the late 1990s, a 65% reduction in government revenues resulted in insufficient health sector funding that led to delays in wage payments and shortage of medical supplies \[[@B20]\]. Although the benefit package in Kazakhstan has covered inpatient drug costs, many hospitals in practice cannot afford to supply them. Therefore, patients often pay for food and drugs while they are in hospital \[[@B6]\]. These expenditures are informal payments that patients are asked to pay out-of-pocket. Our results were consistent with the findings of Ensor and Savelyeva, that estimates of patient per capita payment for prescription drugs in hospital range between US\$ 8.5 and US\$10.4 \[[@B20],[@B28],[@B29]\]. In Kazakhstan, private practice has been permitted since 1991, and more physicians are becoming semi-autonomous practitioners in group practices funded through a patient capitation fund. Many local health authorities permit health facilities to establish charge departments in order to obtain additional revenue \[[@B20]\]. User charges for medical services by public sector health organizations were legalized in 1995. Now, health care facilities charge for services, and as a result, outpatients make official co-payments for the treatment \[[@B6]\]. In addition to these official charges, recent surveys indicated that informal payments to health care practitioners continue to exist, ranging from US\$5 for consultations to thousands for an operation \[[@B6],[@B20],[@B30],[@B31]\]. In this paper, estimates of out-of-pocket expenses were obtained by interviewing patients. Higher costs in inpatient settings may reflect unnecessary use of hospital resources. Other factors that contributed to higher costs in inpatient services may be due to the decrease in government revenues that used to subsidize the hospital services, low official salaries of health care staff, and growth in private health care services \[[@B6]\]. Because of these latter factors, medical practitioners seek augmentation of income via the informal payments \[[@B20]\]. User fees for those who can afford them or cost-sharing strategies could address the problems of under-funding in health systems, stabilize these informal payments, and free up public funds. By balancing the mix of resources and increasing the proportion of spending on services managed in outpatient settings, Ministries of Health in Central Asia might be able to achieve optimal allocation of resources in the health care system. Our findings may be conservative for three reasons. First, in estimating mean treatment costs, we did not consider informal payments to health care staff. Second, our estimates of productivity losses were based on 264 working days per year (22 working days per month) and did not include the value of housekeeping productivity losses. Third, we used lower estimates of the number of patients with mild HAV as 150 per 100,000; the upper estimates of this figure could be as high as 500 per 100,000. This study has some limitations. First, we did not evaluate post-treatment health outcomes after hospital discharge, including completion of outpatient visits. However, there is a wealth of literature that has demonstrated a lack of long-term health outcome differences between persons restricted to bedrest or institutionalized and persons forced to exercise or allowed to engage in activities ad libitumwhen symptomatic with acute HAV infection \[[@B32],[@B33]\]. Second, because of lack of reliable data, we were unable to obtain the capital costs of outpatient facilities including the cost of space, maintenance for facility and medical equipment, overhead, and utilities associated with the treatment of mild HAV infection; therefore, our cost estimates for the outpatient might be biased toward the lower end. Despite these potential concerns, these limitations would not invalidate the findings of this study. In sensitivity analyses, we varied some parameters to establish the range of annual cost estimates of mild HAV infection. We observed that annual cost estimates are sensitive to the following four variables: value of medical services, number of lost workdays, average wage rates, and incidence of HAV infection cases. In these analyses, the annual cost estimates of mild HAV infection varied from US\$4.37 to US\$24.66 million (Table [4](#T4){ref-type="table"}). Considering patients with multiple concurrent diseases or patients diagnosed with severe HAV, cost of medical services can vary significantly. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Total annual cost estimates of mild HAV infection for inpatients in sensitivity analysis, Kazakhstan, (2001 US\$) in millions ::: Analysis Treatment cost Productivity losses Total cost ------------------------------------------------------ ---------------- --------------------- ------------ Base case assumption 3.39 2.87 6.26 Value of medical services ^†^ 4.89 2.87 7.76 Rural wage rate (KZT 5000 per month) ^§^ 3.39 0.98 4.37 Greater adjustment of hepatitis A under-reporting^¶^ 12.59 12.07 24.66 Note the total cost includes the treatment costs as well as productivity losses ^†^an estimated informal payment is included. When 80--90% informal payment is taken into account, the treatment costs were estimated at US\$226.2 per inpatient. ^§^reported wage rate based on the survey results ^¶^annual incidence of hepatitis A as high as 500 per 100,000 (reported in the Central Asia Republics in 1998) ::: Future research should consider conducting cost-effectiveness analyses (CEA) taking into account the quality of post-treatment outcomes rather than assuming similar treatment outcomes among inpatients and outpatients. In conducting these analyses, researchers should consider two possible scenarios: (a) the treatment outcome for the inpatient setting is better than that for the outpatient setting (in this case, the inpatient setting is more costly and more effective than the outpatient setting, so an incremental cost-effectiveness analysis should be conducted); and (b) the treatment outcome for the outpatient setting is as good as or better than that for the inpatient setting (thus, because the cost of managing mild HAV infection for outpatients is lower than for inpatients, cost savings should be calculated). Conclusion ========== From a societal perspective, the annual total treatment costs and productivity losses associated with mild HAV infection ranged from US\$6.05 million to US\$6.47 million for inpatients compared to a range of US\$1.72 million to US\$1.95 million for patients in the outpatient setting. The results of this study show that compliance with a change in the policy of hospitalization for patients with mild HAV infection could produce savings of US\$4.62 million at the national level in Kazakhstan. Appendix ======== Before beginning this study, we obtained institutional review board (IRB) approval (Protocol No. 2708) from the Centers for Disease Control and Prevention (CDC) Assurances-Human Subjects Office. In addition to CDC IRB approval, we also obtained an approval letter from the Ethics Committee Review from the Ministry of Health in Kazakhstan. Prior to administering questionnaires and interviews with patients, we explained the purposes of the study and obtained informed consent forms signed by all adult patients with mild HAV infection. Parents or guardians of all child participants with mild HAV infection under 18 years of age signed the informed consent forms for children. This study was conducted in collaboration with the Data for Decision Making and Policy Branch of the Division of International Health (DIH), Epidemiology Program Office (EPO) of CDC, the CDC Central Asian Regional Office (CDC/CAR), and the Almaty City Sanitary and Epidemiological Stations of the Ministry of Health in Almaty, Kazakhstan. Acknowledgements ================ The authors thank colleagues at CDC, the CDC Central Asian Regional Office, and the Ministry of Health and health care practitioners in the Republic of Kazakhstan for their support in collecting the data. The study was conducted entirely at the Division of International Health, Epidemiology Program Office, Centers for Disease Control and Prevention. The opinions expressed in this article do not necessarily represent those of the agencies of the United States Government	PubMed Central	2024-06-05T03:55:58.451351	2005-5-12	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173129/", "journal": "Cost Eff Resour Alloc. 2005 May 12; 3:4", "authors": [ { "first": "Abdiaziz S", "last": "Yassin" }, { "first": "Michael", "last": "Favorov" }, { "first": "Edmond", "last": "Maes" }, { "first": "Ramses", "last": "Sadek" }, { "first": "Aliya", "last": "Jumagulova" }, { "first": "Victor", "last": "Merker" }, { "first": "Tatiana", "last": "Surdina" }, { "first": "Terence", "last": "Chorba" } ] }
PMC1173130	Introduction ============ In spite of continuous improvements in revascularization techniques, there remain patients ineligible for coronary artery bypass graft surgery (CABG) and coronary angioplasty. Those patients frequently suffer from severe angina, CCS (Canadian Cardiovascular Society) class III to IV; their complaints may not be managed even with optimized drug treatment. An estimated 0.5--1% of catheterized patients belongs to this category. Taking into account the above numbers of catheterizations, in Germany alone there are up to 6000 patients per year suffering from angina-related pain without an option for revascularization. Those patients master their daily lives only by an excessive use of short-acting nitrates and avoiding physical activity. Regular use of morphine is not uncommon. A vicious circle is established: The pain limits physical activity thus inducing further decrease of physical fitness. Frequent hospitalizations, visits to the attending physician and the need for intensive care (when a myocardial infarction is suspected) cause immense costs for the health care system. Based on the \"gate control theory\" \[[@B1]\] Wall and Sweet \[[@B2]\] were the first to use electric stimulation of spinal nerves or dorsal roots in treating chronic pain. This new technique called spinal cord stimulation (SCS) caused a significant decrease in ischemia-related pain and an increased perfusion of capillaries in skeletal muscle and skin through stimulating different types of neurons and fibers in the spinal cord. In recent years the number of studies showing decreasing symptoms, improved quality of life and less signs of myocardial ischemia (mostly measured by ECG) in SCS-treated patients rose continuously \[[@B3]-[@B13]\]. However, the exact mode of action and the influence on myocardial perfusion are yet unknown. Very recently Jessurun and co-workers could demonstrate an improved myocardial perfusion in TENS (transcutaneous electric neurostimulation) treated patients with syndrome X after 4 weeks \[[@B14]\]. Besides a study with 9 patients and a 6 week follow up \[[@B15]\], there exist no long-term data about myocardial ischemia under SCS in patients with coronary artery disease. In our trial, we therefore investigated not only the development of symptoms, quality of life and exercise capacity but also the myocardial perfusion measured by MIBI-SPECT (single photon emission computer tomography) in a one year follow up of 31 patients with CAD and intractable angina. Materials and methods ===================== Patients -------- Between March 1999 and May 2003 31 patients were included in the study. Criteria for accepting a patient was a refractory angina CCS III-IV in spite of optimized medical therapy (ACE inhibitors, aspirin, β-blockers, statins, calcium channel blockers, nitrates). All patients had a coronary angiogram in the last 3 months before screening and were discussed in detail for possibility of CABG and/or PCI (percutaneus coronary intervention) by a specialized team of cardiologists and heart surgeons. Only if none of these revascularization techniques was considered as suitable was the patient included in the trial. Since relieving of pain is the main aim of SCS treatment we only included patients with angina CCS III-IV without relevant symptoms of heart failure. Hence, patients with a left ventricular ejection fraction less than 40% or a NYHA class greater than II were excluded from the study. All inclusion and exclusion criteria are listed in table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Inclusion and exclusion criteria ::: ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Inclusion criteria Chronic coronary artery disease (CAD) Evidence of ischemia in MIBI-SPECT examination Coronary angiogram in the last 3 months Angina pectoris, CCS III-IV Optimal medical treatment (Beta-blocker, ACE-inhibitor or AT1-antagonist, Ca^2+^-antagonist, ASS, CSE-inhibitor, long-acting nitrate \[LAN\]) Not suitable for percutaneous coronary intervention (PCI) or aortocoronary bypass surgery (CABG) after discussion with a team of specialists (cardiologists and heart surgeons) Exclusion criteria Myocardial infarction or unstable in the last 3 months Pacemaker with unipolar electrode Relevant valvular heart disease Symptomatic heart failure \> NYHA II Left ventricular ejection fraction (LVEF) \< 40% ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: Implantation of the electrode ----------------------------- The electrode was placed in the epidural space at the level of Th4-5 through a small skin incision under local anesthesia, and then it was moved up to C7-Th1 subcutaneously. Correct placement was defined by test-stimulation and interviewing the patient. Patients should experience paraesthesia in the same dermatome normally affected by angina. After final placement, the electrode was connected to an external stimulator, which allowed the patient to test the effects of SCS for 3--4 days. Afterwards the final stimulator (Medtronic Itrel III) was implanted subcutaneous in the lateral abdomen and connected to the electrode. The stimulator was controlled through an external device by the patient allowing to change the amplitude of the stimulus or to switch it on/off. During the course of the study, examinations were scheduled after four weeks, three months (with MIBI-SPECT and SAQ), twelve month (with MIBI-SPECT and SAQ) and whenever wished by the patients. Exercise capacity ----------------- The bicycle ergometry protocol began at a workload of 25 Watt and increases in 25 Watt increments every 2 minutes. Exercise was terminated by the patient, when angina pectoris occurred or he was exhausted. Criteria for early termination of exercise were ischemic ST-segment depression \> 2 mm; ventricular or supraventricular tachycardia; and a decrease or abnormal elevation in systolic blood pressure. In our study in none of the patients termination was necessary because of tachycardia or abnormal blood pressure. The six-minute walk test was performed using an internal hallway with a marked distance of 50 meters. Patients were instructed to walk the distance at their own pace but to cover as much ground as possible. They were allowed to stop and rest during the test if needed. Nitrate intake and Seattle Angina Questionnaire (SAQ) ----------------------------------------------------- The SAQ is a self-administered, disease-specific measure for patients with CAD that has previously been demonstrated to be valid, reproducible, and sensitive to clinical change \[[@B16]-[@B18]\]. The SAQ measures five dimensions of coronary artery disease: patients\' physical limitations caused by angina, the frequency of and recent changes in their symptoms (angina frequency and angina stability), their satisfaction with treatment and the degree to which they perceive their disease to affect their quality of life. Each scale is transformed to a score of 0 to 100, where higher scores indicate better function (e.g. less physical limitation, less angina, and better quality of life). The amount of short acting nitrates (SAN) during one week was reported by the patients at every visit in the clinic. For our analysis we used average SAN consumption after 3 and 12 months. Myocardial scintigraphy (MIBI-SPECT) ------------------------------------ SPECT- examinations were run on two days, one day under resting conditions, the second day after stress, using 740 MBq 99-Technetium-MIBI. Stress was induced either by physical exercise (bicycle ergometry) or by dobutamine injection (10, 20, 30, 40 μg per /kg/ min). Scanning was run one hour after the injection in the stress examinations and two hours after resting-conditions. To compare conditions before SCS implantation and during follow up stress scans were made at a comparable rate pressure product. The camera was a three-head gamma camera with a LEHR- (low-energy-high-resolution) -collimator (prism 3000, Picker/Philips; matrix 128 × 128, 40 steps with 3° each, 20 seconds per step). For generating tomograms in short and long axis views, a filtered backprojection technique with additional semiquantitative polarmap analysis was used. SPECT examinations took place prior to the implantation of the SCS system as well as three and twelve months after. Myocardial scintigraphy is usually used to access regional perfusion, e.g. after a heart attack or PCI. However, if the total myocardial perfusion in one patient is to be examined over a course of time, one frequently encounters the reduction of ischemia in one area combined with increased ischemia in another, e.g. improved perfusion in the anterior myocardium with reduced perfusion in the posterior myocardium. This makes it difficult to judge whether conditions have improved or actually worsened. To meet this problem, results were interpreted with the „Emory Cardiac Toolbox" (Emory Medical Device Group). This software compares the perfusion found in the patient to a virtual normal perfusion, setting the normal nuclide distribution- that is the normal perfusion- as 1. If a patient has a score of 0.5 (50% of normal radionuclide distribution and myocardial perfusion) in the initial scan and reaches a score of 0.8 (80% of normal radionuclide distribution and myocardial perfusion) in follow-up examinations this means total perfusion has improved, even if perfusion of some areas has decreased. The score is not determined in direct comparison of two SPECT-scans, but by comparing each scan to virtual normal results. This software makes it possible to monitor the development of myocardial perfusion over a course of time. In spite of this method, accessing the total perfusion remains more complex than judging regional perfusion (e.g. after PCI or heart attack). We therefore decided to limit our interpretation of data on the development of myocardial perfusion in the course of the study as follows: A score improved at least 0.2 was judged an improvement; a score lowered by 0.2 was regarded as a worsened state. A score of ± = 0.1 was defined as unchanged. We left out any semi-quantitative scores with standard deviation as we felt they would over-interpret the data. Statistical analysis -------------------- Data are presented as mean ± 95 % confidence interval. Differences among groups were compared by an unpaired t-test. Significance was assigned to a value of p \< 0.05. Results ======= Patients -------- The mean age was 65 years (± 11). As angina is more common in men, there were more men (n = 26) than women (n = 5) enrolled. 26 Patients suffered from a 3-vessel coronary artery disease, 5 from a 2-vessel disease. All patients had undergone bypass surgery (n = 24) and/or PCI (stent) (n = 28); five patients already underwent transmyocardial laser revascularization. Detailed patient characteristics are presented in table [2](#T2){ref-type="table"}. Drug therapy had been optimized; apart from morphine intake it was not changed significantly during the trial\'s course (see table [3](#T3){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Patient characteristics at baseline ::: ---------------------------------------------- -------------------- Gender: male/female 26 / 5 Mean Age (range), years 65 (35--79) Diagnosis 3-vessel coronary artery disease 26 (84%) 2-vessel coronary artery disease 5 (16%) History, number of patients and percentage Myocardial infarction 23 (74%) Heart Failure \>NYHA II 0 (0%) Previous coronary intervention (PTCA/Stent) 28 (90%) Previous CABG 24 (77%) TMLR 5 (16%) Cerebrovascular Disease (stroke and TIA) 5 (16%) Symptomatic peripheral vascular disease 4 13%) Hypertension 16 (52%) Diabetes mellitus 7 (23%) Hyperlipoproteinemia 23 (74%) Adipositas (BMI \>30) 8 (26%) Nicotine abuse past/current 24 / 6 (77% / 19%) ---------------------------------------------- -------------------- NYHA = New York Heart Association, PTCA = percutaneous transluminal coronary angioplasty, CABG = coronary artery bypass graft, TMLR = transmyocardial laser revascularization, TIA = transitoric ischemic attack, BMI = body mass index ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Medication of the patients at baseline and one year follow up ::: Baseline (n = 31) 1 Year (n = 27) ----------------------------------- ----------------------- --------------------- Beta-blocker 31 (100%) 27 (100%) ACE-Inhibior/AT~1~-Antagonist 29 (94%) 26 (96%) Ca^++^-Antagonist 26 (84%) 24 (89%) CSE-inhibitor 31 (100%) 27 (100%) Long-acting nitrate 31 (100%) 25 (93%) Morphine derivate 12 (39%) 1 (4%) Medication of the patients at baseline (n = 31) and one year follow up (n = 27). Before one year follow up SCS device was explanted in 3 patients (2 during test phase, one after 8 months). One patient died before last visit. ::: 2 patients did not feel a positive effect of SCS on their symptoms in the test period (after implantation of the electrodes and using the external stimulator) and therefore dropped out during the initial phase of the study. SCS device was explanted in another patient after 8 months because of a lack of anti-anginal effect, too. Unfortunately, after explantation of the SCS device none of the 3 patients joined further follow up visits or MIBI-SPECT examinations. One patient died because of a liver cell cancer one month before one year visit. Cancer was not known at time of inclusion in the study. Exercise capacity ----------------- The physical fitness of patients improved as measured by six-minute walk test and bicycle ergometry already after 3 months (142.74 ± 22.13 vs. 224.83 (± 23.96) meters and 67.50 ± 7.11 vs. 96.43 ± 11.74 watt) and maintained at this level after one year (250.56 ± 25.24 meters; 98.08 ± 8.12 watt, see table [4](#T4){ref-type="table"}). The 3 months follow up was finished by 29, the one year follow up by 27 patients (see above). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### SAN consumption per week and results of bicycle ergometry and 6-minute walk test at baseline and follow up ::: Baseline 3 Months 1 Year -------------------------------- ------------------ -------------------- -------------------- SAN Consumptionamount/week 12.35 (± 1.6) 3.38 (± 0.96)\* 2.78 (± 0.90)\* Bicycle Ergometrywatt 67.50 (± 7.11) 96.43 (± 11.74)\* 98.08 (± 8.12)\* 6-Minute Walkmeters 142.74 (± 22.13) 224.83 (± 23.96)\* 250.56 (± 25.24)\* \* = p \< 0.05 compared to baseline. Results are listed in mean ± 95% confidence interval. ::: Symptoms, nitrate intake and Seattle Angina Questionnaire (SAQ) --------------------------------------------------------------- Apart from the three patients who decided to have the electrodes explanted as they felt no relieve in symptoms, all other patients felt a significant decrease of symptoms due to treatment with SCS (28 out of 31). Eleven out of twelve patients treated with morphine at baseline no longer needed morphine at all; one patient was at least able to reduce the dose from 60 mg morphine sulphate (MST^®^) to 20 mg per day. Two patients no longer needed long-acting nitrates (see table [3](#T3){ref-type="table"}). The use of short-acting nitrates (SAN) was lowered significantly, as well (12.35 ± 1.6 vs. 3.38 ± 0.96 and 2.78 ± 0.90 times per week, see table [4](#T4){ref-type="table"}). The Seattle Angina Questionnaire showed a significant increase in quality of Life after 3 months under SCS treatment (SAQ score 39 at baseline vs. 73). After one year the improved quality of life maintained (score 75). Scores for physical limitation 33 vs. 47 \[3 months\] and 48 \[1 year\]), angina stability (30 vs. 51 and 53) and frequency (26 vs. 53 and 54) improved significantly, as well (see figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Results from the Seattle Angina Questionnaire (SAQ). \* = p \< 0.05 compared to baseline. Results are in mean ± 95% confidence interval (error indicator). ::: ![](1468-6708-6-7-1) ::: Myocardial perfusion (MIBI-SPECT) --------------------------------- As described above, data were interpreted with the „Emory Cardiac Toolbox" software. The examinations 3 months after SCS implantation (n = 29) showed unchanged SPECT results in most patients (n = 16). Only 6 patients showed a better myocardial perfusion while in 7 ischemia worsened (figure [2](#F2){ref-type="fig"}). Interestingly, a shift towards an improved myocardial perfusion could be found after one year. Of 27 examined patients a worsened situation compared to baseline was found only in one patient. 16 patients showed clearly less myocardial ischemia. Results of the MIBI-SPECT analysis with the \"Emory Cardiac Toolbox\" software are shown in figure [2](#F2){ref-type="fig"}. Black arrows indicate the shift of the patients between the three groups (worse, equal or improved myocardial perfusion) during 3 and 12 months follow up. An exemplary scintigram prior to and twelve months after implantation is shown in figure [3](#F3){ref-type="fig"}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Results of the MIBI-SPECT analysis with the \"Emory Cardiac Toolbox\" software. Black arrows indicate the shift of the patients between the three groups (worse, equal or improved myocardial perfusion) during 3 and 12 months follow up. ::: ![](1468-6708-6-7-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Scintigram of one patient prior to (left) and twelve months after (right) implantation. Each first line shows a stress scan, the second line the appendant rest scan. The third line shows again a stress scan etc. After one year clearly reduced hibernation resp. prolonged ischemia of the anterior LV wall (arrows vertical long axis views). Less pronounced reduction of the ischemia of the lateral LV wall (arrow heads short axis views). ::: ![](1468-6708-6-7-3) ::: Complications ------------- No serious complications were caused by the implantation of the neurostimulators. Neither injuries of the spinal cord nor infections occurred. The major problem in the beginning were a dislocation of the electrodes (n = 4). This could be remedied without any difficulties by surgical re-intervention. As the operating surgeons became more experienced, electrode dislocations requiring surgical intervention did not happen any longer. Small dislocations could be handled easily by re-programming the stimulation parameters. Discussion ========== This is the first clinical report with a long term follow up of myocardial perfusion measured by MIBI-SPECT in patients with multivessel coronary artery disease under spinal cord stimulation (SCS). Like in prior investigations we additionally measured symptoms, quality of life and exercise capacity of the patients, as well. There was an early reduction of angina related symptoms under SCS treatment. Patients showed not only a significant reduction of SAN consumption (table [4](#T4){ref-type="table"}) but also less use of morphine derivates already in the first weeks under neurostimulation therapy (table [3](#T3){ref-type="table"}). The analysis of the self-administered Seattle Angina questionnaire (SAQ) resulted in improved scores for physical limitation, angina frequency, angina stability and quality of life after 3 months with unchanged results after one year (figure [1](#F1){ref-type="fig"}). In opposite to the early functional improvement of the patients (symptoms, physical activity, quality of life), myocardial ischemia measured by MIBI-SPECT was not decreased after 3 months. 22 patients showed unchanged or even more ischemia compared to baseline before SCS implantation (figure [2](#F2){ref-type="fig"}). After one year -- while functional parameters (exercise capacity, quality of life etc.) did not show a further change -- myocardial perfusion was remarkable improved in many patients (16 of 27). Only one patient still had more cardiac ischemia than in baseline examination (figure [2](#F2){ref-type="fig"}). Interestingly, the score for physical limitation (SAQ) of this patient did not improve very much, as well (55 \[baseline), 62 \[3 months\], 58 \[1 year\]). Since a direct effect of SCS on myocardial perfusion should be established early after beginning of the neurostimulation, one would have expected to see alterations in the MIBI-SPECT after 3 months. The improved myocardial perfusion found after one year might therefore be due to the development of collateral vessels associated with increased physical fitness. Two recent studies reported alterations in myocardial blood flow already after 4 or 6 weeks \[[@B22],[@B23]\]. Both found a homogenization of myocardial perfusion measured by positron emission tomography (PET) in patients with multivessel disease or syndrome X. They suggested that the change in myocardial blood flow can be denoted as a steal phenomenon at the microcirculatory level eventually mediated by a direct effect of SCS on coronary vasomotion. However, they could not find an improved resting myocardial blood flow in their studies, as well. Hautvast and co-workers even found a decreased flow reserve under dipyridamole stress \[[@B22]\]. Thus, our study is the first to show that myocardial perfusion can improve under neurostimulation in patients with coronary multivessel disease. Since we applied a different method to measure myocardial ischemia, we cannot provide data regarding a possible homogenization of cardiac blood flow in the early phase of spinal cord stimulation. But our data suggest that the main effect of SCS treatment is anti-symptomatic and changes in myocardial ischemia are rather secondary and might be due to increased physical activity. This hypothesis is supported by data of Belardinelli and co-workers, who demonstrated that moderate exercise training improved myocardial perfusion measured by thallium scintigraphy in patients with chronic coronary artery disease \[[@B19]\]. If this decrease of myocardial ischemia does also improve the prognosis of the patients cannot be answered yet. Nevertheless, the impressive improvement of angina-related symptoms, the exercise ability and last but not least the increase in quality of life of 90% (28/31) of our patients in a one year follow up supports the data of other studies which treated patients with CAD and refractory angina with spinal cord stimulation. In a recent study, which included 32 patients with a follow up of 65 months, SCS relieved angina effectively also after long-term treatment without development of tolerance. SAQ scores still were improved after years \[[@B20]\]. Maintenance of increased exercise ability for years might be an ongoing stimulus for coronary collateralization. In conclusion electrical spinal cord stimulation is a valuable therapeutic option for the treatment of refractory angina pectoris in patients not suitable for coronary revascularization procedures. Additionally, the symptomtic relief may allow more physical activity which might induce coronary collateralization resulting in an improved myocardial perfusion. Study limitations ----------------- Since it is difficult to conduct placebo-controlled trials with electrical neuromodulation, because there is no alternative for the paraesthesia induced by the neurostimulators, our study has -like others- no placebo group. However, measurement of myocardial perfusion with MIBI-SPECT is an objective parameter not impressionable by the patient \[[@B21],[@B22]\]. As placebo effect is generally believed to decrease with time \[[@B23]-[@B26]\] the sustained relief of symptoms over one year in this study seem to contradict the estimation that placebo effects are the major mode of action of SCS treatment. To finally prove whether the altered myocardial perfusion is due to direct SCS effects or increased exercise ability with improved collateralization a control group without SCS but with exercise training would be necessary.	PubMed Central	2024-06-05T03:55:58.455574	2005-5-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173130/", "journal": "Curr Control Trials Cardiovasc Med. 2005 May 19; 6(1):7", "authors": [ { "first": "Holger", "last": "Diedrichs" }, { "first": "Carsten", "last": "Zobel" }, { "first": "Peter", "last": "Theissen" }, { "first": "Michael", "last": "Weber" }, { "first": "Athanassios", "last": "Koulousakis" }, { "first": "Harald", "last": "Schicha" }, { "first": "Robert HG", "last": "Schwinger" } ] }
PMC1173131	I see no important disagreement between me and the commentators, but disagreement does exist on these topics. I would classify opposing views into two major categories. On one side, some writers assert that causal inference can and should be done without counterfactuals. In my view, Dawid \[[@B1],[@B2]\] is a moderate on that side, as he deploys devices that are isomorphic to counterfactuals, so the difference seems mostly one of labeling and emphasis (which is not always unimportant). More radical counterfactual deniers include Shafer \[[@B3]\], who appears mostly upset because counterfactual models continue to weave into the foundation of statistics, econometrics, sociology, and health sciences, while his approach \[[@B4]\] appears destined for the dustbin, along with other truly noncounterfactual theories of causation. Pearl \[[@B5]\] gives a succinct account of the failings of these theories (reference 5, section 7.5), noting (as do Greenland and Brumback \[[@B6]\]) that the causal models entering into scientific teaching and application (including causal graphs, causal \"pies,\" and structural equations) have mappings into counterfactual formalisms. The target of my article was the other side, those who take counterfactuals uncritically or superficially, without paying enough attention to what these hypothetical quantities are supposed to mean. My article originated as a chapter in a WHO volume \[[@B7]\]. This volume arose from a conference which seemed a festival of counterfactual abuse, rife with talk of cause-of-death removal as if it were an intervention. It is disheartening if not frightening to witness discussion of global health policy framed in such terms. In this context, the concerns expressed by Dawid \[[@B1],[@B2]\] about counterfactuals seem reserved. Counterfactual abuse can be diminished by connecting potential outcomes to interventions. Susser and Schwartz \[[@B8]\] point out that this connection is needed for lifestyle risk factors (smoking, physical inactivity, etc.) just as for social factors. I agree; risk-factor epidemiology could better serve public health if it addressed what could be done, rather than estimating effects of the unattainable (like removal of all tobacco exposure), without regard to how change is brought about. Becoming more realistic involves more than just operationalizing the exposure (right-hand) side of the structural equation; one also needs to expand the left-hand side to consider the full spectrum of intervention effects, such as all effects of smoking cessation (e.g., weight gain, depression). The traditional narrow focus on a few prominent endpoints (like cancer and cardiovascular disease), encouraged by the case-control viewpoint, has discouraged grappling with the multivariate complexity of outcomes as well as exposures. Worse, in the smoking context, there may be a bias against acknowledging that one of our most damaging population exposures (tobacco) may bring worthwhile benefits to a non-negligible portion of the population -- not just medical benefits like Parkinsonism prevention, but also psychologic benefits like enhanced sense of well-being, which are hard to measure and weigh against costs. If we really believe in informed consent, then we must inform the public about how lifestyle choices are not just about lifespan maximization, but are also choices of how to live and die. This view will not sit well with those for whom good sensations are evil if the sensations do not come from sanctified sources like religious faith, licensed entertainment, or prescription drugs.	PubMed Central	2024-06-05T03:55:58.457904	2005-5-19	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173131/", "journal": "Emerg Themes Epidemiol. 2005 May 19; 2:2", "authors": [ { "first": "Sander", "last": "Greenland" } ] }
PMC1173132	Background ========== Diethylcarbamazine citrate (DEC) has been used in the treatment and control of lymphatic filariasis (caused by the nematodes Wuchereria bancrofti, Brugia malayiand B. timori) since 1947 and it continues to play an important role, being one of the drugs used in the Global Programme for the Elimination of Lymphatic Filariasis \[[@B1]\]. However, despite this long period of use, DEC\'s mode of action is still poorly understood. Particularly intriguing is the marked contrast between its rapid action in vivoand the lack of significant activity in vitro. In vivo, the response is rapid: within a few minutes of treatment, peripheral blood microfilariae counts drop dramatically \[[@B2]\]. The poor in vitroactivity indicates that DEC probably requires some host factor for its activity, and previous work has highlighted the role of the innate immune system and leucocytes independent of T cells and complement in the activity of DEC \[[@B3],[@B4]\]. DEC also has anti-inflammatory properties, as a result of its interference with arachidonic acid metabolism \[[@B4]\]. The products of the arachidonic acid metabolic pathway, eicosanoids, have a number of biological effects, including inhibition of platelet aggregation; regulation of leucocyte activation and adherence; mediation of granulocyte chemotaxis and degranulation; and promotion of vasodilatation \[[@B5]\]. It is well known that DEC inhibits enzymes of the 5-lipoxygenase pathway, leukotriene synthases \[[@B6],[@B7]\]. Additionally, in vitro, DEC blocks endothelial cell production of the cyclooxygenase (COX) pathway products prostaglandin (PG) E~2~, prostacyclin (PGI~2~) and thromboxane A~2~but has no effect on platelet prostanoid production \[[@B8]\]. In addition, the drug increases the rate and degree of microfilariae adherence to granulocytes, with eosinophil adhesion in particular being augmented \[[@B9]-[@B11]\]. Nevertheless, a role for some of these activities has yet to be demonstrated in vivoand so we have used a mouse model to identify the host factors responsible for the rapid efficacy of DEC. The arachidonic acid pathway includes lipoxygenase and cyclooxygenase enzymes. The COX pathway has similarities with the nitric oxide (NO) pathway, since both have constitutive and inducible isoforms of their enzymes and are key regulators of inflammatory responses \[[@B12],[@B13]\]. The COX and NO pathways are known to interact with each other, with there being \'cross-talk\' between NO/PGE~2~and iNOS/COX which is generally stimulatory but may also be inhibitory \[[@B14],[@B15]\]. Therefore, we have used a combination of pharmacological inhibitors and gene-knockout technology to elucidate the role of these two pathways in DEC\'s activity in vivo. Materials and methods ===================== Parasites and mice ------------------ Microfilariae of Brugia malayiwere obtained from TRS Laboratories (Georgia, USA), suspended in RPMI 1640 with 5% FCS, and 300000 parasites in a volume of 200 μl were injected intravenously into mice. Systemic parasitaemia was allowed to equilibrate for 24 hours, then heparinised blood samples were taken by tail bleeding and parasitaemia was measured. Mice were allocated into age- and size-matched groups and treated as described below. All animals were kept in the Biological Services Unit of the University of Liverpool in accordance with Home Office regulations and were fed and watered ad libitum. BALB/c mice were kept under standard conditions, and the 129/SV and targeted knockout of the iNOS gene (iNOS^-/-^, kindly provided by Prof. F.Y. Liew, University of Glasgow) strains in filter-top cages. Action of DEC against microfilariae in vivoin mice ---------------------------------------------------- Three BALB/c mice infected with B. malayimicrofilariae were treated with a single, oral dose of DEC 100 mg/kg \[[@B3]\] (Sigma, U.K) in distilled water and the parasitaemia monitored from five minutes to two weeks post treatment. To investigate the role of the arachidonic acid metabolic pathway in the mode of action of DEC, indomethacin (10 mg/kg in 1% ethanol), water-soluble dexamethasone (3 mg/kg in water, both obtained from Sigma, U.K.), or vehicle was given by intra-peritoneal (i.p.) injection to microfilaraemic male BALB/c mice 30 minutes before oral DEC administration (100 mg/kg, three mice per treatment group). One animal was kept as an untreated control. Heparinised blood samples were taken at intervals post treatment for measurement of parasitaemia. Experiments were repeated three times. The requirement for inducible NO in DEC\'s efficacy was determined in iNOS^-/-^mice. DEC (100 mg/kg) or vehicle were administered orally to three female iNOS^-/-^mice or their background strain, 129/SV. Mice were tail-bled at regular intervals post-treatment for evaluation of parasitaemia. To test the efficacy of another anti-filarial drug, ivermectin, in these mice, ivermectin phosphate (1 mg/kg in 1% DMSO) was administered by i.p. injection. This experiment was repeated three times. Expression of COX-1, COX-2 and iNOS in DEC-exposed peritoneal exudate cells --------------------------------------------------------------------------- Male 129/SV and iNOS^-/-^mice were injected i.p. with 10 mg/kg DEC in endotoxin-free water or 100 μl of endotoxin-free water (three mice in each group). After 30 minutes, peritoneal exudate cells were collected in sterile PBS with 1 g/L glucose, 1% bovine serum albumin and 1 U/ml heparin. The cells were pelleted and lysed in 1 ml TRI reagent (Sigma, U.K.) then protein extracted according to the supplied protocol. For Western blot analysis, 10 μg of each protein were separated on a 7.5 % denaturing SDS polyacrylamide gel and blotted on to 0.45 μM pore size PVDF membrane (Immobilon P, Micropore, U.K.). After blocking overnight at 4°C in block buffer (1% casein in PBS/0.1% Tween) and washing in PBS/0.1% Tween, membranes were incubated for 1 hour in rabbit anti-mouse COX-1, COX-2 or iNOS polyclonal IgG (Cayman Chemical Co., Alexis Corporation, U.K.) diluted to 1 in 5000 in block buffer. The anti-COX antibodies showed no cross-reactivity with the opposite isoform, whilst the anti-iNOS antibody showed only 5% cross-reactivity against nNOS and none against eNOS. Membranes were then washed and incubated for 1 hour in goat anti-rabbit IgG conjugated to horse radish peroxidase (Nordic, The Netherlands) diluted to between 1 in 20000 and 1 in 100000, depending on the primary antibody, followed by further washing. The electrochemiluminescent reagent SuperSignal West (Pierce Perbio, U.K.) was used to visualise the bands on X-ray films. Statistical analysis -------------------- Parasitaemia data were expressed as mean percentage of pretreatment microfilariae or as a percentage of untreated control microfilaraemias per 100 μl of blood and were analysed by the two-tailed Student\'s t-test. P values of \< 0.05 were considered to be significant. Results ======= Action of DEC against microfilariae in vivoin mice ---------------------------------------------------- In BALB/c mice treated with DEC alone, microfilaraemia levels were reduced by five minutes with a sustained reduction for at least 60 minutes post-treatment (Fig. [1](#F1){ref-type="fig"}). However, by 24 hours after treatment, microfilarial levels had partially recovered and two weeks later they had returned to levels approaching those pre-treatment (Fig. [1](#F1){ref-type="fig"}). Subsequent experiments focused on the rapid activity of DEC over the first one to two hours. Neither vehicle, indomethacin nor dexamethasone by itself had any effect on microfilaraemia in BALB/c mice (data not shown). However, in mice pre-treated with indomethacin or dexamethasone, microfilaraemias were reduced by only 11% (dexamethasone) or 44% (indomethacin) of untreated controls at 60 minutes post DEC administration (Fig. [2](#F2){ref-type="fig"}). The differences from the DEC-only group were statistically significant for all time points for indomethacin (P \< 0.004) and for 15 and 30 minutes post-treatment for dexamethasone (P \< 0.017) pre-treatments. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### *DEC causes rapid sequestration of B. malayimicrofilariae in BALB/c mice.BALB/c mice intravenously injected with B. malayimicrofilariae were dosed orally with 100 mg/kg DEC and microfilaraemia monitored from 5 to 60 minutes post treatment, then at 24 hours and two weeks. ::: ![](1475-2883-4-4-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Indomethacin or dexamethasone pre-treatment reduces efficacy of DEC in BALB/c mice infected with B. malayimicrofilariae.Indomethacin (10 mg/kg), dexamethasone (3 mg/kg) or vehicle was administered 30 minutes before oral dosing with DEC (100 mg/kg). Symbols are means of three mice for the DEC plus dexamethasone group (triangles), seven mice for the DEC plus indomethacin group (white circles) and four mice for the DEC-only group (black circles). Significantly different results from the DEC-only group are denoted by \ (P \< 0.017), \\ (P = 0.001) or \\\* (P = 0.000). ::: ![](1475-2883-4-4-2) ::: DEC administration also rapidly reduced microfilaraemias in 129/SV mice but, in contrast, had no effect on microfilariae levels in iNOS^-/-^mice, in which microfilaraemia was maintained at pre-treatment levels for at least 2 hours (Fig. [3a](#F3){ref-type="fig"}), with no significant differences from untreated iNOS^-/-^controls (P \> 0.887 for all time points). In contrast, ivermectin was effective in both 129/SV and iNOS^-/-^mice (Fig. [3b](#F3){ref-type="fig"}), although it had a slower onset of action than DEC. However, by 24 hours no microfilariae were detected in either strain of mouse given ivermectin. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### *DEC is ineffective against B. malayimicrofilariae in the absence of iNOS.Efficacy of (a) a single, oral dose of DEC (100 mg/kg) or (b) a single, i.p. dose of ivermectin phosphate (1 mg/kg) in 129/SV and iNOS^-/-^mice infected with B. malayimicrofilariae. Black symbols represent 129/SV mice, white symbols iNOS^-/-^. Squares indicate DEC administration, triangles ivermectin administration and circles untreated controls. Symbols represent mean results from at least three or four mice, except in the case of those treated with ivermectin (two animals) from two combined experiments which were representative of a further repeat. Significantly different microfilaraemias between 129/SV and iNOS^-/-^mice after DEC administration are denoted by \ (P = 0.001) or \\ (P = 0.000). ::: ![](1475-2883-4-4-3) ::: Expression of COX-1, COX-2 and iNOS in DEC-exposed peritoneal exudate cells --------------------------------------------------------------------------- Thirty minutes after administration of endotoxin-free water to 129/SV and iNOS^-/-^mice, peritoneal exudate cells were expressing COX-1 protein, whereas those from DEC-exposed animals contained markedly less COX-1 (Fig. [4](#F4){ref-type="fig"}). Interestingly, there seemed to be a higher level of COX-1 remaining in the iNOS^-/-^than the 129/SV macrophages after DEC treatment. Neither COX-2 nor iNOS protein was detected in any of the 129/SV or iNOS^-/-^groups (not shown). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Western blot detection of COX-1 protein from peritoneal exudate cells.COX-1 protein was detected in 129/SV and iNOS^-/-^peritoneal exudate cells thirty minutes after i.p. injection of endotoxin-free water (control) or DEC (10 mg/kg). Proteins (10 μg) were separated on a 7.5% denaturing SDS polyacrylamide gel, transferred to PVDF membrane, incubated with rabbit anti-mouse COX-1, then goat anti-rabbit IgG-horse radish peroxidase conjugate and detected by chemiluminescence. ::: ![](1475-2883-4-4-4) ::: Discussion ========== Here we have used a murine model to elucidate the processes within the mammalian host that contribute to DEC\'s rapid in vivoaction. The involvement of two interacting pathways, the cyclooxygenase and inducible nitric oxide pathways, were shown to mediate the activity of DEC in vivo. Treatment of mice with DEC resulted in a rapid reduction in microfilaraemia. This reduction, however, was transient and microfilaraemia began to recover 24 hours after treatment, with almost full restoration to pre-treatment levels two weeks after treatment. This has been previously observed in other models \[[@B16],[@B17]\] and suggests that the disappearance of the microfilariae from the peripheral circulation and their sequestration in the central vascular system occur independently of parasite killing. A prolonged course of DEC treatment of B. malayi-infected mice led to sustained reductions in circulating microfilariae for at least 30 days \[[@B18]\]. Our results confirm previous findings showing that an important target for DEC is the arachidonic acid metabolic pathway. Inhibition at the first stage in the pathway by dexamethasone, which inhibits phospholipase A2, almost completely abolished the activity of DEC, whereas inhibition of the cyclooxygenase enzymes COX-1 and COX-2 by indomethacin reduced its efficacy by 56%, indicating that in addition to its well documented inhibition of the 5-lipoxygenase pathway \[[@B6],[@B7]\], DEC acts on the cyclooxygenase pathway. We have shown that at least one way it does this in vivois by the loss of COX-1 protein within 30 minutes of administration. The lack of activity of DEC in mice deficient in iNOS identifies a novel enzyme system involved in the in vivoactivity of DEC. Previously we have shown that B. malayimicrofilariae are susceptible to nitric oxide in vitro\[[@B19]\]. However, we found no evidence that DEC itself up-regulated iNOS activity either in vitro(not shown) or in vivo, in agreement with Rajan et al. \[[@B20]\], who did not find any induction of NO release from murine macrophages or rat endothelial cells treated with DEC. It therefore seems probable that iNOS exerts an effect on DEC activity via its interaction with cyclooxygenases, an idea supported by the reduced loss of COX-1 protein in peritoneal exudate cells derived from iNOS^-/-^mice. Several studies have shown that NO and iNOS interact with COX enzymes to cause an increase in enzymatic activity \[[@B21]\] and consequently increased prostaglandin synthesis \[[@B22]-[@B25]\], although large amounts of endogenous NO inhibited COX expression and activity in murine macrophages \[[@B26]\]. One explanation of the differential effects of NO on COX activity may relate to effects on different COX isoforms. For example, NO can activate COX-1 in fibroblasts but inhibit COX-2 in the same cell \[[@B15]\]. Although our studies do not distinguish between the role of COX-1 and COX-2 in DEC\'s activity, the rapid activity of DEC sequestration and the depletion of COX-1 protein suggest a role for COX-1. COX-1 but not COX-2 is essential for the early production of prostaglandins from macrophages and mast cells \[[@B27],[@B28]\]. Further studies on mice deficient in COX isoforms or the use of isoform-specific pharmacological inhibitors could address this question. Several polymorphisms in the human iNOS gene have been described that are associated with a variety of diseases, including malaria \[[@B29]-[@B31]\] and hypertension \[[@B32]\]; it would be interesting to know if these or other polymorphisms affected responsiveness to DEC therapy. Our findings could help expand our understanding of the mechanisms involved in the cellular processes leading to sequestration and the subsequent killing of parasites. In addition to the elevation of granulocyte adherence, platelets have also been shown to bind to and kill microfilariae \[[@B33]\]. In view of the well know effects of NO and prostaglandins on platelet function and evidence to suggest the presence of inducible NO in human platelets \[[@B34],[@B35]\], the role of platelets in parasite sequestration and killing should be re-evaluated in vivo. Filarial parasites also produce and release prostanoids, including PGE~2~, PGI~2~and PGD~2~\[[@B36]-[@B41]\], which result in inhibition of platelet aggregation \[[@B40]\], vasodilatation of the blood vessels and immune suppression, and may contribute to the long persistence of these parasites in their natural hosts \[[@B41]\]. This prostanoid production is also inhibited by DEC \[[@B8]\]. Significantly, they do not produce thromboxane A~2~\[[@B36]\]. In contrast to mammalian systems, in which eicosanoid formation is often in response to agonist-induced stimulation, microfilariae produce prostanoids constitutively \[[@B36]\], but the mechanisms by which they do so have not yet been described in detail, although a glutathione S-transferase of O. volvulussynthesizes PGD~2~from PGH~2~\[[@B39]\]. It is not clear if DEC acts predominantly against the prostanoids of the worm or of the host. The lack of any direct effect of dexamethasone and indomethacin on microfilaraemia suggests that these drugs either do not influence parasite prostaglandins in vivoor that if they do, they are not involved in DEC-mediated sequestration. Further studies that involve inhibition of the key parasite enzymes would be required to determine the role of parasite-derived prostanoids in DEC activity. Recent studies have reported a direct activity of DEC against Wuchereria bancroftimicrofilariae that results in exsheathment, organelle damage and cytolysis \[[@B42]\], which occur both in vitroand in vivoand suggest that DEC may have a direct effect on worms in addition to its interaction with host-derived pathways as reported here. Much remains to be discovered of the mode of action of DEC. What mechanisms lead to parasite killing following sequestration in the central vasculature; and how does this relate to the paradoxical appearance of microfilariae in the peripheral circulation following the \'DEC provocative test\'? What is the role of host immunity and effects on adult worms in the long-term efficacy of DEC? This model should be a powerful tool to address these questions and others to further unravel the mysteries of this elusive drug. Conclusion ========== Inducible nitric oxide synthase and the cyclooxygenase pathway were found to be essential for DEC\'s activity in vivo. Along with its well-documented activity on the lipoxygenase pathway, DEC administered in vivoreduced the amount of the host\'s COX-1. Further elucidation of DEC\'s mechanism of action with this murine model could provide a clearer understanding of the interaction of the nitric oxide and cyclooxygenase pathways and the cellular and molecular events at the site of sequestration. List of abbreviations ===================== DEC, diethylcarbamazine citrate; COX, cyclooxygenase; i.p., intra-peritoneal; PG, prostaglandin; PGI~2~, prostacyclin; NO, nitric oxide; iNOS, inducible nitric-oxide synthase. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= HFM assisted with the in vivoexperiments, performed the Western blot detection, analysed and interpreted the results, conducted statistical analysis and wrote the manuscript. LDP collected the parasitaemia data and assisted with the in vivoexperiments. MJT conceived the study, performed the in vivoexperiments, interpreted the results and advised on the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ MJT is supported by a Senior Wellcome Trust Fellowship in Basic Biomedical Science. LDP was funded by the University of Liverpool Research Development Fund. We thank Prof. F.Y. Liew (University of Glasgow) for providing the iNOS^-/-^mice.	PubMed Central	2024-06-05T03:55:58.458351	2005-6-2	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173132/", "journal": "Filaria J. 2005 Jun 2; 4:4", "authors": [ { "first": "Helen F", "last": "McGarry" }, { "first": "Leigh D", "last": "Plant" }, { "first": "Mark J", "last": "Taylor" } ] }
PMC1173133	Background ========== Protease inhibitor based antiretroviral therapy (PI-ART), defined as the combination of at least two nucleoside analogues with at least one protease inhibitor (PI) \[[@B1]\], was introduced in 1996 and has greatly reduced the incidence of HIV-related morbidity and mortality in the industrialised world \[[@B2],[@B3]\]. PI-ART would thus be expected to have a positive effect on health-related quality of life (HRQL). On the other hand, HRQL might be negatively affected by strict adherence requirements as well as by short and long-term adverse effects \[[@B4],[@B5]\]. Several studies (cross-sectional or longitudinal) have focused on the HRQL of HIV-positive individuals in different stages of the HIV infection and under different treatment regimes. The results have varied but in general HIV infection affects several physical, psychological and social dimensions of HRQL and patients with symptomatic disease and/or an AIDS-defining complication are more severely affected than those with other comparable chronic diseases \[[@B6]-[@B8]\]. HRQL has been shown to be related to the CD4 value, viral load and symptoms, so that patients with a more advanced state of HIV infection reported poorer HRQL \[[@B9]-[@B15]\]. Moreover, symptoms, physical function, role function and sexual function deteriorated over time, while emotional domains were unchanged or improved \[[@B9],[@B10],[@B16]\]. Only a few studies on the influence of long-term (\>1 year) PI-ART on the quality of life have yet been published. Nieuwkerk et al. \[[@B17]\] compared the development of HRQL during three different PI based regimes and concluded that in terms of HRQL, patients with higher CD4 values at start experienced less benefit from the treatment. Burgoyne et al.\[[@B18],[@B19]\] followed HRQL in 41 patients with different treatment status over a period of four years and found no overall change of HRQL and that HRQL was less sensitive to CD4 changes than to symptom changes as well as that change in HRQL was somewhat related to change in social support. The main aim of the present study was to investigate HRQL before and after two years of first generation PI-ART. The following research questions were addressed: (a) Does HRQL change during two years of PI-ART? (b) Do viral response, adherence, subjective experience of adverse effects and initial CD4 count predict changes in HRQL during two years of PI-ART? Methods ======= Data collection --------------- ### Subjects The study was performed at the Department of Venhälsan at South Stockholm General Hospital, Sweden. A convenient sample of 72 subjects (70 men and 2 women) in an advanced state of HIV infection and who were among the first patients to receive PI-ART in Sweden responded to the HRQL instrument described below before the start of PI-ART (pre PI-ART). HIV infection was documented by at least two laboratory tests (two repeated ELISA tests or one ELISA test and one Western Blot). The patients were treated according to best clinical practice and the participant\'s physician chose the drug combination of the PI-ART (at least 2 nucleoside analogue reverse transcriptase inhibitors and either indinavir (n = 55) or ritonavir (n = 17) at start). Approximately two years after the introduction of PI-ART (follow-up; mean 25.1 months, standard deviation (SD) 2.8 months, post initiation), 54 of the 72 subjects (75 %) completed the follow-upmeasure. Thirteen patients had died, four had changed clinic and one did not receive the follow-upquestionnaire. ### The Swedish Health-Related Quality of Life Questionnaire (SWED-QUAL) The patients completed the SWED-QUAL at the pre PI-ARTand two-year follow-upvisits. SWED-QUAL was developed by Brorsson et al.\[[@B20]\] from the measures used in the US Medical Outcomes Study (MOS) \[[@B21]-[@B24]\]. The questionnaire, which is designed to measure HRQL, consists of 70 items, of which 63 forms two single-item and 11 multi-item dimension scales of Likert type: physical functioning (7 items), mobility (1 item), satisfaction with physical ability (1 item), role limitations due to physical health (3 items), pain (6 items), emotional well-being: positive affect (i.e. positive feelings; 6 items), emotional well-being: negative affect (i.e. negative feelings; 6 items), role limitations due to emotional health (3 items), sleep problems (7 items), satisfaction with family life (relations with parents, siblings, children etc.; 4 items), relation to partner (6 items), sexual functioning (4 items) and general health perception (9 items). In the present study, the relation to partner section was slightly modified to make it suitable for the investigated group (i.e. the word \"spouse\" was replaced by \"partner\"). Each scale is transformed into a 0--100 index; the higher the score, the better the perceived HRQL. In a general population sample, the internal-consistency reliability coefficients (Cronbach\'s α) ranged from 0.79 to 0.89 \[[@B20]\]. Brorsson et al.have also reportedpreliminary support for the construct validity \[[@B20]\]. The instrument has been used in our previous study of HIV-positive subjects in Sweden \[[@B25]\]. A factor analysis (see below) was performed on a merged material from three groups of patients: (a) the group described above at pre PI-ART(n = 72), (b) a previously described group of protease inhibitor naïve HIV patients (n = 73) \[[@B25]\] and (c) an unpublished material of 164 HIV-negative men who have sex with men visiting our information and screening clinic for sexually transmitted diseases (own unpublished data). ### Adherence/adverse effects At the follow-up, the patients\' adherence and drug-related adverse effects were examined. Adherence was assessed by asking the following question, modified from Morisky et al.\[[@B26]\]: \"How many times during the last month did you skip doses of your HIV medication for the following reasons: because you felt so good that it did not matter, because you felt worse due to the medication or because you forgot?\" For each of the three given reasons the patients were asked to rate the number of missed doses on a scale with five alternatives (1 = 0 doses, 2 = 1--2 doses, 3 = 3--5 doses, 4 = 6--9 doses or 5 = \>9 doses). Patients were classified as non-adherent if they indicated \>9 missed doses in at least one of the three mentioned categories (i.e. indicating 90% adherence or less) or if they indicated missed doses in all three categories. The subjective experience of medication-related adverse effects was addressed in one question, which asked the patients to rate their global experience of adverse effects during the past month on a 10 cm visual analogue scale (VAS) ranging from \"none at all\" to \"severe\". ### Laboratory measures Lymphocyte subsets were determined, using routine flow cytometry \[[@B27]\], at least every 3--4 months. The level of HIV-1 RNA in plasma was quantified at the Swedish Institute for Infectious Disease Control by a commercially available reverse transcriptase polymerase chain reaction (PCR)-based assay (HIV Monitor; Roche Diagnostic Systems, Branchburg New Jersey, USA). The level of quantification at the time of the study was 500 viral copies ml^-1^. ### Viral outcome On the basis of the participants\' long-term virologic outcome to PI-ART, they were subdivided into viral responders and viral non-responders. Viral responders were defined as participants who, after the first 3 months of PI-ART, had either HIV RNA less than 500 copies ml^-1^in more than 75% of the analysed samples or a continuous decrease in HIV RNA to below 500 copies ml^-1^before 18 months. All other participants were regarded as viral non-responders \[[@B28]\]. ### Demographic and medical data Demographic and medical data were obtained from specific items in the questionnaires and from the medical files. Statistical methods ------------------- Statistical calculations were performed with the assistance of the personal-computer program SPSS for Windows, version 11.0.0. The criterion for statistical significance was p\< 0.05. Since some parameters did not fulfil the assumptions of a normal distribution, the Wilcoxon signed ranks test \[[@B29]\] was used to compare differences between two related groups (i.e. pre PI-ARTversus follow-up). In order to achieve data reduction to create composite scores for use in multiple regression analyses of the material, a rotated component matrix analysis \[[@B30]\] was performed on the scores from ten of the 13 SWED-QUAL scales (i.e. those measuring physical and emotional health). The following scales were used in the factor analysis: physical functioning, mobility, satisfaction with physical ability, role limitations due to physical health, pain, emotional well-being: positive affect, emotional well-being: negative affect, role limitations due to emotional health, sleep problems and sexual functioning. The analysis resulted in two factors: (a) physical HRQL composite score, i.e. PCS, comprising the averaged scores of the SWED-QUAL scales physical functioning, mobility, satisfaction with physical ability, pain, role limitations due to physical health and sexual functioning, and (b) emotional HRQL composite score, i.e. ECS, comprising the averaged scores of the SWED-QUAL scales role limitations due to emotional health, positive affect, negative affect and sleep problems (Table [1](#T1){ref-type="table"}). These two factors explained 65 % of the total variance. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Rotated factor loading (varimax) of the principal component analysis (n = 309) ::: Rotated factor loadings ------------------------------------------ ----------------------------- ------ SWED-QUAL Scale Physical functioning 0.92 0.07 Mobility 0.83 0.04 Satisfaction with physical ability 0.79 0.24 Pain 0.56 0.45 Role limitations due to physical health 0.66 0.46 Sexual functioning 0.63 0.32 Role limitations due to emotional health 0.33 0.75 Emotional well-being, positive affect 0.08 0.81 Emotional well-being, negative affect 0.08 0.85 Sleep problems 0.32 0.70 ^PCS^Physical HRQL composite score; ^ECS^Emotional HRQL composite score ::: A set of hierarchical multiple linear regression analyses \[[@B31]\] were performed to investigate whether the initial CD4 value, viral response, adherence and subjective experience of adverse effects predicted the change in HRQL. The two analyses (one for PCS and one for ECS) were conducted in two forced steps with the follow-upPCS and ECS as dependent variables. The pre PI-ARTPCS and ECS, respectively, were entered in the first step, while the three variables viral response, adherence and subjective experience of adverse effects were entered in the second step together with the initial CD4 value. Ethical approval ---------------- The study was approved by the Local Ethical Committee at Huddinge University Hospital. Information about the study was given to the subjects in connection with an ordinary, scheduled visit to the clinic. Results ======= Demographic and medical data ---------------------------- At pre PI-ART, the mean age of the 72 patients was 41 (SD 9, range 23 -- 65) years, 59 were of Swedish origin, 61 had prior antiretroviral treatment with only nucleoside analogue reverse transcriptase inhibitors, 32 had a CD4 count of \<200 × 10^6^cells l^-1^and 28 had an AIDS diagnosis. Further demographic and medical data are shown in Table [2](#T2){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Demographic and medical data regarding a Swedish sample of HIV-infected persons before the start of PI-ART (n= 72) ::: No. of subjects Percentage ----------------------------------------------------------------- -------- --------------------- ---------------- ------------ Mode of transmission Male to male 69 96 Female to male/male to female 3 4 Working status Full or part time work/studies 36 50 Sick leave or disability pension 31 43 Not stated 5 7 Education Compulsory school 13 18 Upper secondary school/high school 27 38 University studies 27 38 Not stated 5 7 Having a partner: Yes/no 30/41 42/57 not stated 1 1 CD4 count, cells × 10^6^l^-1^: median 127 range 1--660 HIV RNA, log~10~copies ml^-1^: median 4.87 range 2.70--6.18 Time since first start of antiretroviral treatment, months: mean 33 SD 31.6 ^PI-ART^Protease inhibitor based antiretroviral therapy; ^SD^Standard deviation ::: Changes from pre PI-ARTto follow-up --------------------------------------- From the pre PI-ARTto the follow-up, the CD4 count of the surviving patients increased from a median (Md) of 150 (range 1 -- 660) × 10^6^cells l^-1^to 325 (range 5 -- 840) × 10^6^cells l^-1^(p \< 0.001) and the HIV-1 RNA level decreased from 4.90 (range 2.70 -- 6.00) log~10~copies ml^-1^to 2.70 (range 2.70 -- 6.31) log~10~copies ml^-1^(p \< 0.001). Thirty patients had switched PI during the first two years of PI-ART. A total of 36 subjects were classified as viral responders and 18 as viral non-responders. ### The patients\' emotional well-being deteriorated during the first two years of PI-ART The pre PI-ARTand follow-upresults of the composite scores and the single SWED-QUAL scales are shown in Table [3](#T3){ref-type="table"}. The ECS decreased from pre PI-ARTto follow-up(Md 71.2, interquartile range (IQR) 47.1 -- 86.9 versus Md 63.6, IQR 41.0 -- 77.3; p \< 0.01;). A comparison of the single SWED-QUAL scales at pre PI-ARTwith the same scales at follow-uprevealed statistically significant decreases in the SWED-QUAL scales role limitations due to emotional health (pre PI-ARTMd 88.9, IQR 66.7 -- 100 versus follow-upMd 77.8, IQR 50.0 -- 100; p \< 0.05) and emotional well-being: negative affect (pre PI-ARTMd 62.5, IQR 37.5 -- 94.8 versus follow-upMd 50.0, IQR 25.0 -- 75.0; p \< 0.01; Table [3](#T3){ref-type="table"}). The effect sizes of the change were 0.34 for the PCS, 0.36 for role limitations due to emotional health and 0.37 for emotional well-being: negative affect. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The Swedish Health-Related Quality of Life Questionnaire (SWED-QUAL) results before and after 2 years of protease inhibitor based antiretroviral therapy (pre PI-ARTand follow-uprespectively). The higher the score the better the health-related quality of life (n = 54) ::: Score ------------------------------------------ ------------------- ------------------- -------- Physical HRQL composite score (PCS) 79.3 (64.5--92.5) 77.2 (58.4--92.9) NS Emotional HRQL composite score (ECS) 71.2 (47.1--86.9) 63.6 (41.0--77.3) \<0.01 Physical functioning 95.2 (81.0--95.2) 95.2 (84.5--100) NS Mobility 100 (66.7--100) 100 (66.7--100) NS Satisfaction with physical ability 66.7 (33.3--83.3) 66.7 (33.3--100) NS Pain 100 (53.8--100) 100 (84.7--100) NS Role limitations due to physical health 66.7 (55.6--100) 88.9 (50.0--100) NS Sexual functioning 75.0 (41.7--91.7) 58.3 (41.7--91.7) NS Role limitations due to emotional health 88.9 (66.7--100) 77.8 (50.0--100) \<0.05 Emotional well-being, positive affect 72.9 (47.9--83.3) 62.5 (37.5--83.3) NS Emotional well-being, negative affect 62.5 (37.5--94.8) 50.0 (25.0--75.0) \<0.01 Sleep problems 67.9 (39.3--87.5) 57.1 (36.6--75.0) NS Satisfaction with family life 72.9 (49.5--90.6) 72.9 (55.7--85.4) NS Relation to partner 87.5 (75.0--100) 79.2 (52.1--95.8) NS General health perception 55.6 (36.1--75.0) 63.9 (38.9--80.6) NS ^a^Wilcoxon signed ranks test, change from pre PI-ARTto follow-up; ^NS^Non significant; ^IQR^Interquartile range; ^HRQL^Health-related quality of life ::: ### Subjective experience of adverse effects contributed to deteriorated emotional HRQL at follow-up Forty-seven (87 %) of the 54 patients were classified as adherent and seven (13 %) as non-adherent. The mean VAS score for subjective experience of adverse effects at follow-upwas 3.39 (SD 2.95, Md 2.57, IQR 0.79 -- 5.84) cm on a 10 cm scale. Two hierarchical multiple regression analyses were performed to investigate whether the change in physical and emotional HRQL composite scores could be predicted by initial CD4 count, viral response, adherence and subjective experience of adverse effects (Table [4](#T4){ref-type="table"}). Only the ECS model showed a statistically significant R^2^change; that is, the subjective experience of adverse effects predicted a decrease in the emotional HRQL. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Hierarchical multiple linear analyses of the influence of initial CD4 value and the variables viral response, subjective experience of adverse effects and adherence on the change in PCS and ECS from before start (pre PI-ART) to after two year of treatment (follow-up) ::: **Follow-upPCS Follow-upECS** ---------------------------- ------------------------- -------------------- ------------------------- -- Pre PI-ARTPCS 0.722\\\* \- Pre PI-ARTECS \- 0.681\\\* R^2^ 0.605\\\* 0.456\\\* F 78.26 42.72 d.f. (regression;residual) 1;51 1;51 Step 2 Partial correlation Partial correlation Viral response 0.202 0.076 Adherence 0.167 0.235 Adverse effects 0.091 -0.366\* Pre PI-ARTCD4 value -0.043 -0.024 R^2^-change 0.028 0.128\* F-change 0.91 3.63 d.f. (regression;residual) 4;47 4;47 ^PCS^Physical HRQL composite score; ^ECS^Emotional HRQL composite score; ^PI-ART^Protease inhibitor based antiretroviral therapy; ^a^partial correlation of the final model (step 2); \p \< 0.05; \\\p \< 0.001 ::: Internal consistency of the SWED-QUAL scales -------------------------------------------- Cronbach\'s α reliability estimates for the 11 SWED-QUAL multi-item scales for the pre PI-ARTand follow-upranged between 0.74 and 0.92 indicating good internal consistency. The internal consistency for the two HRQL composite scores was for the PCS α 0.88 and 0.85 (pre PI-ARTand follow-up, respectively) and for the ECS α 0.91 and 0.89 (pre PI-ARTand follow-up, respectively). Discussion ========== In this study we used the SWED-QUAL instrument to investigate the HRQL of 54 relatively advanced HIV patients before and after two years of first generation PI-ART in a setting where protease inhibitors were introduced. We also studied the patients\' viral outcome, self reported adherence and subjective experience of adverse effects and the relationship between these variables and the development in HRQL. To minimise multiple comparisons, a set of linear regression models were used. To increase the power to detect relations as a result of these models, we created two SWED-QUAL composite scores (PCS and ECS) through a data reduction procedure. The main findings from the present study were that the physical HRQL remained stable while the emotional HRQL deteriorated for two years of first generation PI-ART and that subjective experience of adverse effects was the strongest predictor of the deterioration in emotional HRQL ratings. The present study thus suggests that first generation PI-ART interrupted the progressive negative development of the physical domain of HRQL that has been reported in studies performed before the introduction of PI-ART \[[@B9],[@B10],[@B16],[@B32],[@B33]\]. Our findings are consistent with those of Goujard et al.\[[@B34]\] and Burgoyne et al.\[[@B18]\] who also failed to detect any changes in HRQL after one and a half and four years, respectively, when monitoring patients in a period after PI-ART had become available. The somewhat surprising decrease in emotional health found in the present study has, to our knowledge, not been reported in other longitudinal studies of PI-ART. Longitudinal studies performed before the introduction of PI-ART have, in general, showed stable or improved mental/emotional health over time \[[@B9],[@B10],[@B16],[@B35]\]. We also found stable scores in the emotional domain of HRQL in a previous pre PI-ART era study, where patients with no or only single drug antiretroviral therapy were followed for two years (own unpublished data). However, a study investigating the HRQL of patients receiving didanosine monotherapy or in combination with delavirdine (not approved in Europe due to side effects) showed slightly declining mental health scores for up to two years after the start of the trial \[[@B33]\]. After the introduction of PI-ART, improvements in depressive symptoms and mental health was reported in a shorter time period, i.e. after up to one year following the addition of a PI to existing antiretroviral therapy \[[@B36],[@B37]\]. Similarly, Rabkin et al.\[[@B38]\] found a reduction in psychological distress and clinical depression during a two-year period when PI-ART became widely available. In the latter study, however, PI-ART was introduced to the cohort continuously during the follow-up, resulting in a mean time with PI-ART shorter than two years (i.e. the analysis did not evaluate patients before and after start of PI-ART in a consistent manner). From a longer term perspective, however, emotional HRQL was found to be stable over a four-year period \[[@B18]\]. When we further investigated the relation between the change in HRQL and the variables viral response, adherence, subjective experience of adverse effects and baseline CD4 value, our study showed that it was the subjective experience of adverse effects that contributed most to the deterioration in the emotional HRQL. It should be stressed that our measure of adverse effects was a global single item, where the patients were asked to rate their total perception of adverse effects during the previous month and that we did not sub-analyse this experience further into different symptoms. However, the negative impact of perceived side effects/symptoms on HRQL, is confirmed in a great number of quantitative investigations \[[@B12],[@B18],[@B37]-[@B42]\]. Also, the results from our quantitative study agreed with those of a qualitative study that aimed to elicit the patients\' real-life descriptions of their experience of combination therapy. Erlen & Mellors \[[@B43]\] found that side effects were one of the major problems associated with the therapy. Furthermore, symptoms or adverse events have been shown to be related to medication adherence \[[@B44],[@B45]\]. In order to improve HRQL and adherence, it is therefore crucial to find treatment combinations and strategies that minimise these negative effects and to individualize treatment. The simplified and less toxic treatment regimes available today, with antiretroviral therapy based on three nucleoside analogue reverse transcriptase inhibitors or nucleoside analogues combined with non-nucleoside analogue reverse transcriptase inhibitors or ritonavirboosted PIs once or twice daily \[[@B1]\], may be less liable to have a negative impact on HRQL. This was indicated in a randomised study comparing two nucleoside analogue reverse transcriptase inhibitors combined with either a PI or efavirenz, over a one year period and where the combination with efavirenz had a better influence on HRQL than the combination with the PI \[[@B46]\]. Similarly, Carrieri et al.\[[@B37]\] found a positive development of HRQL in patients switching to a non-PI-containing therapy, as compared to patients continuing stable PI-ART. However, these aspects need to be further investigated and there is a need for further study of the effect of different treatment regimes on HRQL, in experienced as well as naïve patients. Certain study limitations should be emphasized when interpreting the present results. Firstly, the small sample may mean that the study failed to detect further relations between the investigated variables. Secondly, a large proportion of the investigated patients were in an advanced state of their HIV infection and many had previous experience of antiretroviral monotherapy. The inferences drawn from this study may not be applicable to ART naïve patients starting antiretroviral therapy in a less advanced state of the disease. Today treatment is started when certain laboratory criteria are fulfilled and before symptoms, including AIDS-defining disease, are at a high risk of appearing. Thirdly, our study population may not be totally representative of all patients receiving PI-ART. The majority of our patients were well-educated males with a homosexual route of transmission. The influence on HRQL may be different for female patients or patients with other educational status or routes of transmission. It should be noted that we chose to consider viral outcome of the treatment in a longitudinal context, i.e. viral response defined by repeated HIV RNA measures over up to 18 months. The impact on the development of HRQL might have been different if only one single measure of viral load at follow-uphad been taken into account. Conclusion ========== In this sample of mainly advanced patients, the emotional dimension of HRQL had deteriorated for two years after the start of first generation PI-ART, and the subjective perception of adverse effects made a major contribution to this decrease. The results of the present study show the importance of studying HRQL in a situation where there is a desperate need for life-saving new therapies. Therefore, consistently taking HRQL into account when treating HIV patients is of the utmost importance. Finding treatment combinations and strategies with the least negative long-term influence on HRQL is essential at a time when those having access to antiretroviral combination therapy have a dramatically increased life span. Furthermore, considering the current numerous treatment options together with the fluctuations in HRQL over time, further short and long-term investigations of HRQL in patients receiving antiretroviral therapy are of the utmost importance. Authors\' contributions ======================= LEE, ES and GN conceived of the study. All authors made substantial contributions to conception, planning and design. LEE participated in the coordination, carried out the statistical analysis and interpretation of data and drafted the manuscript. GAB participated the coordination and acquisition of the study and helped draft the manuscript. ES participated in the acquisition of the study and helped draft the manuscript. GN participated in the interpretation of data and helped draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We would like to thank the nurses and physicians at the Department of Venhälsan, South Stockholm General Hospital for their help in administering the instruments to the participants, Eva-Lena Fredriksson, Department of Venhälsan, South Stockholm General Hospital for her great assistance in the co-ordination of the study and J. Petter Gustavsson, Department of Nursing, Karolinska Institutet for helpful discussions. We would also like to acknowledge the assistance of the participants, who gave time and effort to respond to the questionnaires. This study was supported by grants from the Board of Research for Health and Caring Sciences at Karolinska Institutet.	PubMed Central	2024-06-05T03:55:58.460198	2005-5-4	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173133/", "journal": "Health Qual Life Outcomes. 2005 May 4; 3:32", "authors": [ { "first": "Lars E", "last": "Eriksson" }, { "first": "Göran A", "last": "Bratt" }, { "first": "Eric", "last": "Sandström" }, { "first": "Gun", "last": "Nordström" } ] }
PMC1173134	Background ========== The question of whether disease cases are clustered in space has received considerable attention in the literature \[[@B1]-[@B4]\]. Although many statistical tests for disease clusters have been proposed, most tests suffer from multiple testing problems due to one or two unknown parameters that must be set prior to their applications. For example, Cuzick and Edwards\'s procedure \[[@B5]\] has an unknown number kof nearest-neighbours and Besag and Newell\'s method \[[@B6]\] has an unknown number of cases kfor the size of the cluster. As far as we know, the spatial scan statistic proposed by Kulldorff \[[@B7],[@B8]\] and Tango\'s maximized excess events test \[[@B9],[@B10]\] are exceptions and take multiple testing into account in the sense that we have only to specify the maximum possible cluster size. Especially, Kulldorff\'s circular spatial scan statistic has been applied to a wide variety of epidemiological studies for cluster detection (for example, see \[[@B11]-[@B13]\]). In recent power comparisons of disease clustering tests, his scan statistic has been shown to be the most powerful for detecting localized clusters \[[@B14],[@B15]\]. It should be noted, however, that the power estimates provided reflect the \"power to reject the null hypothesis for whatever reason\" and that the probability of both rejecting the null hypothesis and detecting the true cluster correctly is a different matter. As the circular spatial scan statistic uses a \"circular window\" with variable size to define the potential cluster area, it is difficult to correctly detect noncircular clusters such as those along a river. Most geographical areas are noncircular. Furthermore, in our experience in applying SaTScan program \[[@B16]\] to various data, even if the null hypothesis is rejected, the circular spatial scan statistic tends to detect a larger cluster than the true cluster by absorbing surrounding regions where there is no elevated risk. It should be noted that although Kulldorff originally made no assumptions about the shape of the scanning window in his paper \[[@B8]\], a circular scanning window has been used in almost all purely spatial applications especially for the availability of software and computational speed. Recently, Patil and Taillie \[[@B17]\] and Duczmal and Assunção \[[@B18]\] proposed non-circular spatial scan statistics based on the likelihood ratio test formulated in the same way as in the circular spatial scan statistic. To avoid undertaking computationally infeasible searches, they considered different approaches. Patil and Taillie \[[@B17]\] used the notion of \"upper level set\" to reduce the size of windows to be scanned and proposed \"upper level set scan statistic\". However, they do not discuss how to select the level gwhich defines the upper level set and do not provide any illustrations of their method nor any results of comparison with the circular scan statistic. Duczmal and Assunção \[[@B18]\], on the other hand, have applied a simulated annealing method in which they try to examine only the most promising windows using a graph-based algorithm to obtain the local maxima of a certain likelihood function over a subset of the collection of all the connected regions. Their method seems to be very complicated but they do not show any programmable procedure of their method. In our experience using their program (personal communication to Professor Duczmal via email) which is executable with the Borland C++ Builder 6, their scan statistic, in most cases, detected a cluster of peculiar shape that was much larger than the true cluster by absorbing not only surrounding regions with non-elevated risk but also faraway regions with non-elevated risk. An example of such properties of Duczmal and Assunção\'s procedure is shown later in comparison with the circular spatial scan statistic and the proposed flexible spatial scan statistic. That is why we did not include both the Patil and Taille method and Duczmal and Assunção\'s procedure in our simulation for comparison. In this paper, we propose an alternative flexibly shaped spatial scan statistic(\'flexible spatial scan statistic\' hereafter) in which the detected cluster is allowed to be flexible in shape while at the same time the cluster is confined within relatively small neighborhoods of each region. The performance of the flexible spatial scan statistic is compared with that of the circular spatial scan statistic using Monte Carlo simulation. In comparing performance we examined not only the usual power but also the newly introduced bivariate power distribution classified by the number of regions detected as the most likely cluster and the number of hot-spot regions included in the most likely cluster. The proposed flexible spatial scan statistic is illustrated with some simulated disease maps for the Tokyo Metropolitan area. Methods ======= Consider the situation where an entire study area is divided into mregions (for example, county, enumeration districts, etcetera). The number of cases in the region iis denoted by the random variable N~i~with observed value n~i~, i= 1, \..., m. Under the null hypothesis H~0~of no clustering, the N~i~are independent Poisson variables such that H~0~: E(N~i~) = ξ~i~, N~i~\~ Pois(ξ~i~), i= 1, \..., m (1) where Pois(e) denotes Poisson distribution with mean eand the ξ~i~are the null expected number of cases in the region i. To specify the geographical position of each region, we will use the coordinates of the administrative population centroid. Under this situation, the circular spatial scan statistic imposes a circular window Zon each centroid. For any of those centroids, the radius of the circle varies from zero to a pre-set maximum distance dor a pre-set maximum number of regions Kto be included in the cluster. If the window contains the centroid of a region, then that whole region is included in the window. In total, a very large number of different but overlapping circular windows are created, each with a different location and size, and each being a potential cluster. Let Z~ik~, k= 1,\..., K, denote the window composed by the (k- 1)-nearest neighbours to region i. Then, all the windows to be scanned by the circular spatial scan statistic are included in the set Z~1~= {Z~ik~\\| 1 ≤ i≤ m, 1 ≤ k≤ K} (2) A flexible scan statistic we propose, on the other hand, imposes an irregularly shapedwindow Zon each region by connecting its adjacent regions. For any given region i, we create the set of irregularly shaped windows with length kconsisting of kconnected regions including iand let kmoves from 1 to the pre-set maximum K. To avoid detecting a cluster of unlikely peculiar shape, the connected regions are restricted as the subsets of the set of regions iand (K- 1)-nearest neighbours to the region iwhere Kis a pre-specified maximum length of cluster. In total, as in the circular spatial scan statistic, a very large number of different but overlapping arbitrarily shaped windows are created. Let Z~ik(j)~, j= 1,\..., j~ik~denote the j-th window which is a set of kregions connected starting from the region i, where j~ik~is the number of jsatisfying Z~ik(j)~⊆ Z~ik~for k= 1,\..., K. Then, all the windows to be scanned are included in the set Z~2~= {Z~ik(j)~\\| 1 ≤ i≤ m, 1 ≤ k≤ K, 1 ≤ j≤ j~ik~} (3) In other words, for any given region i, the circular spatial scan statistic consider Kconcentric circles, whereas the flexible scan statistic consider Kconcentric circles plus all the sets of connected regions (including the single region i) whose centroids are located within the K-th largest concentric circle. So, the size of Z~2~is far larger than that of Z~1~which is at most mK. Details of the algorithm that we adopted to find all these arbitrarily shaped windows within a pre-specified maximum length Kare given in the Appendix. Under the alternative hypothesis, there is at least one window Zfor which the underlying risk is higher inside the window when compared with outside. In other words, we are considering the following hypothesis: H~0~: E(N(Z)) = ξ(Z), for all Z, H~1~: E(N(Z)) \> ξ(Z), for some Z (4) where N() and ξ() denote the random number of cases and the null expected number of cases within the specified window, respectively. For each window, it is possible to compute the likelihood to observe the observed number of cases within and outside the window, respectively. Under the Poisson assumption, the test statistic, which was constructed with the likelihood ratio test \[[@B8]\], is given by ![](1476-072X-4-11-i1.gif) where Z^c^indicates all the regions outside the window Z, and n() denotes the observed number of cases within the specified window and I() is the indicator function. The window Z\* that attains the maximum likelihood is defined as the most likely cluster(MLC). To find the distribution of the test statistic under the null hypothesis, Monte Carlo hypothesis testing \[[@B19]\] is required. In this paper, p-value of the test is based upon the null distribution of likelihood ratio test statistic with a large number (we used 999) of Monte Carlo replications of the data set generated under the null hypothesis. It should be noted that, in the same manner as the circular spatial scan statistic, the flexible spatial scan statistic is also able to locate secondary clusters that do not overlap the most likely cluster but are still statistically significant. Results ======= Illustrations and powers ------------------------ In this section, we will compare the flexible spatial scan statistic with the circular spatial scan statistic. As an entire study population, we will use m= 113 regions comprising the wards, cities and villages in the area of Tokyo Metropolis and Kanagawa prefecture in Japan (Figure [1](#F1){ref-type="fig"}). The variability of regional populations for m= 113 regions is: 25 percentile = 56, 704, median = 142, 320 and 75 percentile = 200, 936. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### An entire study population for simulation studies. The 113 regions comprising wards, cities and villages in the area of Tokyo Metropolis and Kanagawa prefecture in Japan. The region number used in the text is shown. Especially, The region numbers of four hot-spot clusters A-Dare A= {14, 15, 20}, B= {14, 15, 20, 26}, C= {14, 15, 26, 27}, and D= {73, 74, 75, 76, 78}, respectively. ::: ![](1476-072X-4-11-1) ::: ### Hot-spot clusters We will consider the following four hot-spot clusters where the expected total number of cases ![](1476-072X-4-11-i2.gif) is set to be 200 under the null hypothesis. 1\. Cluster A= {14, 15, 20} 2\. Cluster B= {14, 15, 20, 26} 3\. Cluster C= {14, 15, 26, 27} 4\. Cluster D= {73, 74, 75, 76, 78} where the region included in a hot-spot cluster is called a \"hot-spot region\" (hot-spot region numbers are shown in Figure [1](#F1){ref-type="fig"}). The relative risk within any cluster Ris set to three, i.e., H~1~: N(R) \~ Pois(θξ(R)), θ= 3.0 (6) The cluster Ais considered here as an example of a circular cluster that can be in the set of the circular windows and is expected to be identified by the circular spatial scan statistic more often than by the flexible spatial scan statistic. The other clusters are examples of noncircular clusters that are not in the set of the circular windows and thus cannot be identified correctly by the circular spatial scan statistics. For example, consider the region i~0~= 15 as the starting region and the set of (K- 1)-nearest neighbours to the region 15, which is listed as follows in the ascending order of distance from the region 15: 15, 14, 20, 12, 4, 26, 13, 27, 16, 40, 19, 42, 10,\..., In this case, circular windows are {15}, {15, 14}, {15, 14, 20}, {15, 14, 20, 12}, \... When the starting region is 14 or 20, the corresponding set of (K- 1)-nearest neighbours is 14, 15, 20, 4, 16, 13, 19, 12, 5, 1, 17, 10, 26, 3, 27,\..., and 20, 14, 15, 19, 16, 4, 17, 26, 40, 13, 5, 12, 1, 27,\..., respectively. In both cases, cluster Band Care easily found to be not in the set of circular windows. The cluster Dis considered as an example of a long and narrow cluster as is shown in Figure [1](#F1){ref-type="fig"}. ### Illustrative example As an illustration, we will apply the circular spatial scan statistic, the flexible spatial scan statistic and Duczmal and Assunção\'s spatial scan statistic to the disease map shown in Figure [2](#F2){ref-type="fig"} which is a random sample of n= 235 cases assuming the cluster model C. Circles are drawn only for the regions whose observed-expected ratio (standardized risk ratio) is statistically significantly larger than 1 at α= 0.05. The radius of the circles is set inversely proportional to the upper tail p-value. The number shown in Figure [2](#F2){ref-type="fig"} indicates the region number. Figure [2](#F2){ref-type="fig"} obviously suggests the clusters occurring in the area including regions {14, 15, 26, 27, 33}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### A random sample from cluster model C. Dots describe the centroids of regions with some cases. Circles are drawn only for the regions whose standardized risk ratios are statistically significantly larger than 1 at α= 0.05 and the region number is placed in stead of dot. The radius is set inversely proportional to the tail probability. ::: ![](1476-072X-4-11-2) ::: Before applying the three spatial scan statistics, we have to specify a common maximum length Kfor the most likely cluster. This makes comparisons to a certain extent fair. In this example, we chose two kinds of maximum length K= 15 and K= 20 since it is not unreasonable to assume that an actual cluster size will be less than one third or one fourth of the size of the whole study area. Irrespective of the value of K, the circular spatial scan statistic detected the regions {14, 15} as MLC with log likelihood ratio = 20.1, p= 1/(999 + 1) = 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 3.47. This is shown in Figure [3(a)](#F3){ref-type="fig"}. The flexible spatial scan statistic, regardless of the value K, detected the regions {14, 15, 26, 27, 33} as MLC with log likelihood ratio = 29.7, p= 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 3.41. This is shown in Figure [3(b)](#F3){ref-type="fig"}. Duczmal and Assunção\'s method, on the other hand, detected a cluster of peculiar shape that is much larger than the true cluster. In the case of K= 15, their scan statistic detected an area consisting of K= 15 connected regions {14, 15, 24, 26, 27, 31, 32, 33, 48, 54, 69, 77, 78, 90, 110 } as MLC with log likelihood ratio = 31.8, p= 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 2.40. This is shown in Figure [4(a)](#F4){ref-type="fig"}. Figure [4(b)](#F4){ref-type="fig"} shows the most likely cluster {14, 15, 26, 27, 31, 32, 33, 48, 60, 61, 62, 67, 69, 77, 78, 80, 89, 90, 108, 110 } detected by Duczmal and Assunção\'s scan statistic for K= 20 where the length of MLC is also the same as K= 20 and log likelihood ratio = 36.0, p= 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 2.26. In the case of K= 15, the results of the three scan statistics are summarized in Table [1](#T1){ref-type="table"}. Although the most likely cluster detected by Duczmal and Assunção\'s scan statistic has the largest log likelihood ratio among three scan statistics, it has detected MLC surprisingly larger than the true cluster. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The most likely cluster detected by the circular and the flexible spatial scan statistic. (a) Detected by the circular spatial scan statistic for both K= 15 and K= 20 and (b) by the flexible spatial scan statistic for both K= 15 and K= 20, when applied to a random sample from the cluster model C= {14, 15, 26, 27}. ::: ![](1476-072X-4-11-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The most likely cluster detected by the Duczmal and Assunção\'s scan statistic. (a) Detected for K= 15 and (b) for K= 20, when applied to a random sample from the cluster model C= {14, 15, 26, 27}. ::: ![](1476-072X-4-11-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Regions detected as the most likely cluster by three procedures. Regions detected as the most likely cluster by the circular scan, the flexible scan and Duczmal and Assunção\'s scan, with the maximum length of cluster set to be K= 15 for the simulated random sample from the cluster model C where the hot spot cluster is assumed to be the set of connected four regions {14, 15, 26, 27} with the assumed relative risk θ= 3.0. For details, see text. ::: region no. population observed no. cases expected no. cases relative risk estimated (true) Log likelihood ratio (LLR) and estimated relative risk for the most likely cluster ------------ ------------ -------------------- -------------------- -------------------------------- ------------------------------------------------------------------------------------ ------------ ------------ 14 319,687 14 3.794 3.69 (3.0) \* \* \* 15 529,485 21 6.283 3.34 (3.0) \* \* \* LLR = 20.1 = 3.47 26 139,077 6 1.650 3.64 (3.0) \* \* 27 165,564 6 1.964 3.05 (3.0) \* \* 33 105,899 4 1.257 3.18 (1.0) \* \* LLR = 29.7 = 3.41 24 466,347 8 5.534 1.44 (1.0) \* 31 197,677 3 2.346 1.27 (1.0) \* 32 349,050 5 4.142 1.20 (1.0) \* 48 58,635 1 0.696 1.43 (1.0) \* 54 3,808 1 0.045 22.12(1.0) \* 69 119,575 3 1.419 2.11 (1.0) \* 77 177,742 5 2.109 2.37 (1.0) \* 78 125,127 2 1.485 1.34 (1.0) \* 90 194,866 5 2.312 2.16 (1.0) \* 110 21,535 1 0.256 3.91 (1.0) \* LLR = 31.8 = 2.41 ::: Using a PC(Windows XP, CPU pentium 4, 3.2 GHz), the execution time of the flexible spatial scan statistic in this example is 14 seconds for K= 15 and 379 seconds for K= 20 which is certainly greater than that for the circular spatial scan statistic (less than 1 second for both K= 15 and K= 20). ### Power comparison In the power comparison, we chose K= 15. To compare the power of the flexible spatial scan statistic with that of the circular spatial scan statistic based upon Monte Carlo simulation, we will introduce a new bivariate power distribution P(l, s) classified by the length lof the significant MLC and the number sof hot-spot regions included in the most likely cluster: ![](1476-072X-4-11-i4.gif) where l≥ 1 and s≥ 0. Based on P(l, s), we examined the following powers, 1\. the usual power, i.e., P(+,+) = ∑~l≥1~∑~s≥0~P(l, s), 2\. the joint power P(l, s), especially P(s\, s\) where s\* is the length of the hot-spot cluster assumed in the simulation. 3\. the marginal power distribution of s(≥ 0), P(+, s) = ∑~l≥1~P(l, s) and its conditional power P(+, s)/P(+,+), 4\. the marginal power distribution of l(≥ 1), P(l, +) = ∑~s≥0~P(l, s). The powers are calculated for tests of nominal αlevels of 0.05 and for the expected total number of cases 200 under the null hypothesis, which are based on Monte Carlo simulation using Poisson random numbers. For each simulation, 1,000 trials were carried out. The resultant power distribution P(l, s) × 1000 is shown in Tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"} for each of the four cluster models, respectively, in the form of cross table classified by l(\"length\"in tables) and s(\"include\"in tables). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model A. Comparison of bivariate power distribution P(l, s) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster A= {14, 15, 20}. Nominal α-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ------- --- --- --- ----- ----- Flexible (K= 15) Circular (K= 15) Length l Include s\ Total Length l Include s\ Total hot-spot regions hot-spot regions 0 1 2 3 0 1 2 3 1 0 0 0 1 0 0 0 2 0 0 0 0 2 1 0 0 1 3 0 0 0 142 142 3 0 0 0 738 738 4 0 0 0 116 116 4 0 0 0 134 134 5 0 0 0 137 137 5 0 0 0 39 39 6 0 0 0 149 149 6 0 0 0 12 12 7 0 0 0 165 165 7 0 0 0 9 9 8 0 0 0 131 131 8 0 0 0 1 1 9 0 0 0 84 84 9 0 0 2 3 5 10 0 0 0 27 27 10 0 0 0 2 2 11 0 0 0 11 11 11 0 0 0 4 4 12 0 0 0 2 2 12 0 0 0 12 12 13 0 0 0 0 0 13 0 0 0 14 14 14 0 0 0 0 0 14 0 0 0 3 3 15 0 0 0 0 0 15 0 0 0 6 6 Total 0 0 0 964 964 Total 1 0 2 977 980 usual power = 0.964 usual power = 0.980 -------------------- --------------------- ------- ------------ --------------------- ------- ------- --- --- --- ----- ----- ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model B. Comparison of bivariate power distribution P(l, s) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster B= {14, 15, 20, 26}. Nominal α-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- --- ----- ----- ----- Flexible (K= 15) Circular (K= 15) Length l Include s\ Total Length l Include s\ Total hot-spot regions hot-spot regions 0 1 2 3 4 0 1 2 3 4 1 0 0 0 1 0 0 0 2 0 0 0 0 2 0 0 0 0 3 0 0 0 0 0 3 0 0 0 523 523 4 0 0 0 0 127 127 4 0 0 0 65 0 65 5 1 0 0 0 157 158 5 0 0 0 23 0 23 6 0 0 0 0 205 205 6 0 0 0 7 66 73 7 0 0 0 2 198 200 7 0 0 0 0 15 15 8 0 0 0 1 151 152 8 0 0 0 0 32 32 9 0 0 0 5 85 90 9 0 0 0 1 15 16 10 0 0 0 1 24 25 10 0 0 0 0 7 7 11 0 0 0 0 17 17 11 0 0 0 2 3 5 12 0 0 0 0 5 5 12 0 0 0 2 63 65 13 0 0 0 0 0 0 13 0 0 0 0 96 96 14 0 0 0 0 0 0 14 0 0 0 0 30 30 15 0 0 0 0 0 0 15 0 0 0 0 22 22 Total 1 0 0 9 969 979 Total 0 0 0 623 349 972 usual power = 0.979 usual power = 0.972 -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- --- ----- ----- ----- ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model C. Comparison of bivariate power distribution P(l, s) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster C= {14, 15, 26, 27}. Nominal α-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- ----- ----- ----- ----- Flexible (K= 15) Circular (K= 15) Length l Include s\ Total Length l Include s\ Total hot-spot regions hot-spot regions 0 1 2 3 4 0 1 2 3 4 1 0 0 0 1 1 0 1 2 0 0 0 0 2 0 0 351 351 3 0 0 0 0 0 3 2 0 4 0 6 4 0 0 0 0 138 138 4 0 0 3 0 0 3 5 0 0 0 3 147 150 5 2 0 2 0 0 4 6 1 0 0 2 200 203 6 1 0 0 0 0 1 7 0 1 0 4 147 152 7 0 0 0 81 0 81 8 0 0 2 9 107 118 8 0 0 10 18 38 66 9 0 0 0 10 71 81 9 0 0 2 0 26 28 10 1 0 2 5 28 36 10 0 0 0 29 3 32 11 0 0 0 0 10 10 11 0 0 1 13 1 15 12 0 0 0 0 2 2 12 0 0 2 4 60 66 13 0 0 0 0 0 0 13 0 0 0 5 62 67 14 0 0 0 0 0 0 14 0 0 0 10 27 37 15 0 0 0 0 0 0 15 0 0 0 6 37 43 Total 2 1 4 33 850 890 Total 6 0 375 166 254 801 usual power = 0.890 usual power = 0.801 -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- ----- ----- ----- ----- ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model D. Comparison of bivariate power distribution P(l, s) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster D= {73, 74, 75, 76, 78}. Nominal α-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ----- ----- ------- ---- --- ---- ---- ----- ---- ----- Flexible (K= 15) Circular (K= 15) Length l Include s\ Total Length l Include s\ Total hot-spot regions hot-spot regions 0 1 2 3 4 5 0 1 2 3 4 5 1 0 0 0 1 6 0 6 2 1 0 0 1 2 3 5 0 8 3 0 0 0 0 0 3 0 0 0 14 14 4 1 0 0 1 0 2 4 1 0 4 5 0 10 5 0 1 0 3 1 242 247 5 0 0 2 1 0 0 3 6 1 0 0 1 2 162 166 6 1 0 0 1 363 0 365 7 2 3 0 5 5 93 108 7 0 0 1 0 56 0 57 8 1 2 1 6 7 53 70 8 0 0 2 2 28 0 32 9 0 2 0 1 5 38 46 9 0 0 2 2 10 0 14 10 0 2 0 1 1 18 22 10 1 0 0 3 3 0 7 11 0 0 0 2 2 5 9 11 0 0 0 0 3 11 14 12 0 0 1 0 0 1 2 12 0 0 0 2 3 8 13 13 0 0 0 0 0 0 0 13 0 0 0 1 1 16 18 14 0 0 0 0 0 0 0 14 0 0 1 0 0 5 6 15 0 0 0 0 0 0 0 15 0 1 0 0 1 7 9 Total 6 10 2 20 23 612 673 Total 12 6 12 31 468 47 576 usual power = 0.673 usual power = 0.576 -------------------- --------------------- ------- ------------ --------------------- ------- ----- ----- ------- ---- --- ---- ---- ----- ---- ----- ::: #### 1) Usual power Both tests have the same size 0.043 (distribution of length of significant MLC is omitted) and are shown to have high powers for the hot-spot clusters considered here. The flexible spatial scan statistic generally has higher power except for the model A(circular cluster) where, however, the difference is small. #### 2) Joint powers at (s\, s\) and at its neighbours Table [2](#T2){ref-type="table"} shows the good characteristics of the circular spatial scan statistic. Namely, the circle-based scan statistic could detect circular hot-spot cluster Awith length s\* = 3 considerably more accurately with power 738/1000 compared to 142/1000 of the flexible spatial scan statistic. Tables [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}, on the other hand, show that the power of the circular spatial scan statistic in detecting exactly noncircular hot-spot clusters is 0/1000 due to the circular window. However, the circular spatial scan statistic is seen to be able to include some of the hot-spot regions into MLC reasonably well. For example, when applied to the noncircular cluster Bwith length s\* = 4, three or four regions including three hot-spot regions can be detected as the most likely cluster with relatively high power (523 + 65)/1000 = 0.588 (Table [3](#T3){ref-type="table"}). When applied to the model Dwith length s\* = 5, the similar high power 363/1000 can be observed at (l, s) = (6, 4) (Table [5](#T5){ref-type="table"}). The flexible spatial scan statistic, on the other hand, has no such high power at a single point (l, s) near (s\, s\). However, the characteristic of the flexible spatial scan statistic is that the support of the power distribution is distributed in a relatively narrow range of / on the line s= s\,i.e, we have s\ ≤ l≤ 12 in the four cluster models considered here. #### 3) Marginal power P(+, s) and its conditional marginal power P(+, s)/P(+, +) Regarding the marginal power P(+, s\) at s= s\, the flexible spatial scan statistic is shown to have much higher power than the circular spatial scan statistic for the case of noncircular clusters (Tables [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}). Furthermore, the conditional marginal power P(+, s)/P(+, +) of the flexible spatial scan statistic is 964/964 = 1.000, 969/979 = 0.990, 850/890 = 0.955 and 612/673 = 0.909 for the cluster A-D, respectively. These results indicate that the identified MLC by the flexible spatial scan statistic includes the hot-spot cluster with quite high probability. For the noncircular clusters, the mode of P(+, s) of the circular spatial scan statistic is around s= s\* - 1 or s= s\* - 2. #### 4) Marginal power distribution P(l, +) For the flexible spatial scan statistic, the probability that the length of significant MLC is less than s= s\* is shown to be zero or quite small and the maximum length is around 10 to 12. the circular spatial scan statistic, on the other hand, tends to detect a much longer cluster than expected from the hot-spot cluster assumed in the simulation. For example, the probability that the length of MLC for the cluster B with length s\* = 4 is greater than or equal to 12 is 213/1000 = 0.213 compared with 5/1000 = 0.005 for the flexible spatial scan statistic. The probability that the length of MLC for the cluster Cwith length s\* = 4 is greater than or equal to 12 is 213/1000 = 0.213 compared with 2/1000 = 0.002 for the flexible spatial scan statistic. This tendency is shown even in the circular cluster Awhere the same probabilities are 0.035 vs. 0.002. ### Cost comparison Based upon the bivariate power function P(l, s), we can compute the following expected total cost incurred by incomplete identification of the true cluster: C= C~2~{rE(s\* - S) + E(L- S)}, r= C~1~/C~2~ (8) where C~1~and C~2~denote the average cost of missing one region in the true cluster and that of incorrectly detecting one region not in the true cluster, respectively. Land Sdenote the random variable of land s, respectively. Two expected numbers E(s\* - S) and E(L- S) for four kinds of clusters A-Dare shown in Table [6](#T6){ref-type="table"}. In general, we can assume r\> 1. For example, the ratio C/C~2~is shown for the case of r= 1 and r= 2, respectively, in Table [6](#T6){ref-type="table"}. However, in this example, irrespective of the value of r(\> 1), the circular spatial scan statistic is shown to have lower cost for detecting circular cluster Abut to have higher cost for detecting non-circular clusters B-D. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Cost comparison Expected number of undetected regions included in the true cluster E(s\* - S), expected number of detected regions not in the true cluster E(L- S) and the ratio of costs C/C~2~(r= 1, 2) incurred by incomplete identification of the true cluster. The spatial scan statistic with low values is better. ::: Hot-spot Cluster Scan statistic E(s\* - S) E(L- S) the ratio C/C~2~ ----------------------------- ------------------- ------------------ --------------- ---------------------- ------- A= {14, 15, 20} Flexible (K = 15) 0.108 2.951 3.059 3.167 Circular (K = 15) 0.065 0.722 0.787 0.852 B= {14, 15, 20, 26} Flexible (K = 15) 0.097 2.548 2.645 2.742 Circular (K = 15) 0.735 2.525 3.260 3.995 C= {14, 15, 26, 27} Flexible (K = 15) 0.492 2.243 2.735 3.227 Circular (K = 15) 1.736 3.153 4.889 6.625 D= {73, 74, 75, 76, 78} Flexible (K = 15) 1.774 1.088 2.862 4.636 Circular (K = 15) 2.770 1.709 4.479 7.249 ::: Limitations of current work --------------------------- Needless to say, the results derived here are based upon a small Monte Carlo simulations study and thus the characteristic observed in the current work could change a little bit depending on the cluster model adopted. We assumed here only one hot spot cluster and did not consider the case of several hot spot clusters. Therefore, we need a further simulation study to compare the performance of the two spatial scan statistics under several different clusters. Regarding the algorithm adopted for the flexible spatial scan statistic, we set the restriction that irregularly shaped windows Zwith length k(≤ K) are constructed from members of the (K- 1)-nearest neighbours to the starting region. It seems that this restriction plays an important role in preventing the flexible spatial scan statistic from reaching out for and absorbing faraway regions with non-elevated risk. However, to avoid undertaking computationally infeasible searches, the flexible spatial scan statistic has to be set with an upperbound for K. This depends on the disease map under study and the capability of the computer. The current practical upperbound is around K= 30 for the reason that the execution time of our current algorithm will take more than a week if K \>30 for the number of regions m= 200 \~ 300. However, it seems to be unlikely that the length of the true cluster would be larger than 10 \~ 15 percent of the total number of regions. So, we think that our current algorithm can be applied to many epidemiological studies with small to moderate cluster sizes. However, for larger cluster sizes, a more sophisticated algorithm to increase the upperbound for Kis needed. Regarding data type, the proposed spatial scan statistic can only be applied to regional count data whereas the circular spatial scan statistic can be applied to not only count data but also individual point data. However, at least in disease surveillance, most of the data that people analyze is aggregated, so the method covers most real-world situations. Finally, one of the reviewers commented that using small areas as basis for clustering without any attempt to incorporate heterogeneity in background rates is a fundamental flaw of all existing scanning methods. In general, we know that disease risks over study regions are heterogeneous to a certain extent and the null hypothesis of complete spatial randomness is not true. However, statistical hypothesis testing is based upon the null hypothesis which is not true. Likewise, we will use complete spatial randomness as the null hypothesis as indicated in equation (1) since we are interested in rejecting the null hypothesis and detecting the local clusters with excess risk. If we are interested in estimating a clustering mechanism, we should use some modeling approach rather than spatial scan statistics. Discussion ========== In this paper, we proposed a flexibly shaped spatial scan statistic to detect arbitrarily shaped clusters by amalgamating administrative units. The flexible spatial scan statistic is, via Monte Carlo simulation, shown to have reasonably high powers compared with the circular spatial scan statistic when examined by a newly introduced bivariate power distribution P(l, s). The simulation reveals that the circular spatial scan statistics shows a high level of accuracy in detecting circular clusters exactly and reasonably good power for including some hot-spot regions into the most likely cluster. The flexible spatial scan statistic exhibits no such high power regarding exact identification of clusters but the support of the power distribution is shown to be concentrated in a relatively narrow range of length lon the line s= s\, indicating that an observed significant most likely cluster contains the true cluster with quite high probability. The circular spatial scan statistic, on the other hand, is shown to have zero powers for detecting exactly noncircular clusters that cannot be captured by any circular window. The circular spatial scan statistic is also shown to have a tendency to detect a larger cluster than the true cluster assumed in the simulation even for the case when the true cluster is circular. Furthermore, by introducing the two kinds of cost due to incomplete detection of the true cluster, we could summarize these characteristics in terms of minimizing expected total cost. One of the reviewers suggested a similar cost comparison using the number of people that are incorrectly classified rather than the number of regions since the cost of misclassifying a large region is at least for disease surveillance purposes higher than that of misclassifying a region with smaller population. We think that would be an interesting additional simulation study worth conducting. However, since it can be expected that the result of such a cost comparison strongly depends on the spatial configuration of regions with different population size in the neighborhood of and within the true cluster and thus it requires careful design for creating suitable cluster models from which we can intuitively infer the result to a certain extent, we would like to leave such a simulation study in our future work. The surprising result that Duczmal and Assunção\'s scan statistic detected quite large and unlikely peculiar shaped clusters that had the largest likelihood ratio among the three scan statistics might cast a doubt on the validity of the model selection based upon maximizing the likelihood ratio (5). Such a doubt can also be seen in some simulation results of the circular spatial scan statistic that had non-negligible probabilities of detecting much longer clusters than the true cluster. The flexible spatial scan statistic, on the other hand, is shown not to detect such an unexpected long cluster probably because it has the restriction that our windows are constructed only from members of the (K- 1)-nearest neighbours to the starting region. Nevertheless, these undesirable properties produced by maximum likelihood ratio might suggest the use of a different criterion for model selection. For example, we might consider a penalized likelihood where we consider a penalty for the complexity of the cluster shape, which is also worth future research. All the computations and simulations have been conducted on a PC with Windows XP. For users who are interested in applying the flexible spatial scan statistic, we can provide the software FleXScan \[[@B20]\]. Conclusion ========== The circular spatial scan statistics shows a high level of accuracy in detecting circular clusters exactly and reasonably good power for including some hot-spot regions into the most likely cluster. The proposed flexible spatial scan statistic is shown to have good usual powers plus the ability to detect the noncircular hot-spot clusters more accurately than the circular spatial scan statistic. However, the proposed spatial scan statistic work well for small to moderate cluster size, say up to 30. For larger cluster sizes, the method is not practically feasible and a more efficient algorithm is needed. Appendix: algorithm to find the set Z~2~defined in equation (3) ================================================================= There are probably several procedures to find the set Z~2~that is defined as the set of arbitrarily shaped windows Zwithin a pre-specified maximum length K. The algorithm that we used is described as follows: Step 1. First we set an m× mmatrix A= (a~ij~) such as ![](1476-072X-4-11-i5.gif) and set Z~2~= φand i~0~= 0 Step 2. Let i~0~← i~0~+ 1 and i~0~(= 1, 2,\..., m) be the starting region. Then we create the set ![](1476-072X-4-11-i6.gif) consisting of (K- 1)-nearest neighbours to the starting region i~0~and i~0~itself, i.e., ![](1476-072X-4-11-i7.gif) = {i~0~, i~1~, i~2~,\..., i~K-\ 1~}, where i~k~is the k-th nearest to i~0~. Step 3. We consider all the set Z⊂ ![](1476-072X-4-11-i7.gif), which includes the starting region i~0~. For any given such set Z, repeat the following steps 4--7. Step 4. We divide the set Zinto two disjoint sets: Z~0~= {i~0~} and Z~1~which contains the other regions of Z. Step 5. We make two new sets ![](1476-072X-4-11-i8.gif) and ![](1476-072X-4-11-i9.gif). ![](1476-072X-4-11-i8.gif) consists of the regions of Z~1~that are connected to some regions of Z~0~. On the other hand, ![](1476-072X-4-11-i9.gif) consists of the regions of Z~1~that are not connected to any regions of Z~0~. Then we replace Z~0~and Z~1~by ![](1476-072X-4-11-i8.gif) and ![](1476-072X-4-11-i9.gif), respectively. Step 6. We repeat the step 5 recursively until either Z~0~or Z~1~becomes null first. Step 7. We make a decision as follows. Zis said to be \"connected\" when Z~1~becomes null first and \"disconnected\" when Z~0~becomes null first. If we can find Z\"connected\", Zis added to the set Z~2~. If we find Z\"disconnected\", Zis discarded. Step 8. Repeat the steps 2--7 until we finally get the set Z~2~consisting of arbitrarily shaped windows Zwhose maximum length is K. Now we shall give an example using regions in the Tokyo Metropolitan area shown in Figure [1](#F1){ref-type="fig"}. Let the starting region i~0~= 14. Then, the regions in the set of (K- 1)-nearest neighbours to the region 14 are listed as follows in the ascending order of distance to the region 14, i.e., W~14~= {14, 15, 20, 4, 16, 13, 19, 12, 5,\...}. Suppose that we take a subset Z= {14, 15, 20, 26}. In the first step, we have Z~0~= {14}, Z~1~= {15, 20, 26}. Since a~14,15~= a~14,20~= 1 and a~14,26~= 0, we then have Z~0~= {15, 20}, Z~1~= {26}. Further, because a~15,26~= a~20,26~= 1, these sets are replaced by Z~0~= {26}, Z~1~= φ. So, we can find that the set Z= {14, 15, 20, 26} is \"connected\" and can be a member of Z~2~. If we take a subset Z= {14, 15, 20, 5}, we can find Zis \"disconnected\" because a~14,5~= a~15,5~= a~20,5~= 0, Z~0~= φand Z*~1~= {5} at the final stage. Authors\' contributions ======================= TT proposed the flexibly shaped spatial scan statistic and the bivariate power distribution. KT considered the algorithm given in the appendix, programmed the C++ code and carried out the power simulations. TT wrote the first draft of the manuscript. Both authors interpreted the results and wrote the final version of the paper. Acknowledgements ================ This research was supported in part by Grant-in-Aid for Scientific Research (Grant No. 16300091) from the Ministry of Education, Culture, Sports, Science and Technology of Japan.	PubMed Central	2024-06-05T03:55:58.463681	2005-5-18	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173134/", "journal": "Int J Health Geogr. 2005 May 18; 4:11", "authors": [ { "first": "Toshiro", "last": "Tango" }, { "first": "Kunihiko", "last": "Takahashi" } ] }
PMC1173135	Introduction ============ One of the earliest observations in the study of circadian rhythms was that continuous light (LL) lengthens circadian period in most nocturnal animal species \[[@B1]\]. \"Aschoff\'s Rule\" posits that there is a positive log-linear relationship between the LL intensity and period \[[@B2]-[@B5]\]. In all these studies LL was white light, in one study full-spectrum light \[[@B4]\]. However, we found that mice had slightly longer circadian periods when the monitoring device was a passive infrared (ir) proximity sensor compared to a system using ir beams that crossed the cage. The only obvious difference between the systems was that the proximity sensor had a small, red light-emitting diode (LED) that came on immediately after discerned motion and stayed on for several seconds after motion was not discernable. The first question to be raised is whether a dim red LED can affect the circadian system of mice. The circadian rhythm of locomotor activity in rats is entrained by red light \[[@B6]\]. However, several studies which examined the spectral sensitivity of the photoreceptors in mice suggest little or no response to red light. The peak sensitivity of the photoreceptors that mediate phase shifts in pigmented inbred mouse strains is between 500 nm \[[@B7]\] and 511 nm \[[@B8]\] (blue-green light). The sensitivity of the photoreceptors drop sharply and is vanishingly small at wavelengths above 600 nm (orange light) \[[@B7]-[@B9]\]. In mice lacking rods and cones, the peak sensitivity for phase-shifting is 481 nm, and sensitivity drops to zero at less than 600 nm \[[@B9]\]. In pigmented mice, electroretinographic responses to a flickering monochromatic light and the behavioral responses to a forced-choice discrimination task peak at 510 nm \[[@B10]\]. The light sensitivity in both tests drops sharply as the wavelength approaches 600 nm. Melanopsin, in combination with the classical rod and cone photoreceptors, accounts for the transduction of photic information to the circadian system. We are unaware of studies of the Aschoff effect in rodless and coneless mice, but melanopsin knockout mice have an attenuated Aschoff effect compared to wild-type mice \[[@B11],[@B12]\]. The second question is whether light presented only in response to activity can lengthen period in mice. Pittendrigh and Daan suggested that light pulses during the photosensitive portion of an animal\'s circadian cycle mimic the effect of LL on period \[[@B13]\]. Ferraro and McCormack (1986) confirmed this in rats using feedback lighting (LDfb) \[[@B14]\]. In their LDfb apparatus, the lighting in each rodent\'s cage was controlled by each rodent\'s own locomotor activity. When wheel revolutions reached a certain rate, the cage lighting came on. The lights went out 2 minutes after wheel-running tapered off below the target rate. They compared LL to LDfb at 0.1, 1 and 100 lux of light. They found that circadian period under LDfb obeyed Aschoff\'s rule, and feedback lighting increased circadian period by the same amount as an equivalent irradiance of continuous light. In earlier studies we showed that C57BL/6 and DBA/2 mice differed in their Aschoff affect comparing constant dark to constant full-spectrum LL. At 10 lux, the C57BL/6 mice had an increase in period of 1.20 hours but the period of the DBA/2 mice increased only 0.20 hours \[[@B4]\]. Consequently we predicted that the C57BL/6 mice would have a greater increase in period under the LD feedback regimen than DBA/2 mice. This study tests the hypotheses that a dim red LED provided as feedback to activity elicits an increase in circadian period of locomotor activity and that C57BL/6 and DBA/2 mice have a differential response to the red light stimulus. Methods ======= General housing and care ------------------------ Mice were housed singly in optically clear polycarbonate cages (L × H × W: 11 × 8 × 7 in) with approximately 250 ml of Sani-chip^®^(Harlan Teklad) bedding. They were acclimated under alternating 200 lux light and dark of 12 hours each (LD 12:12) for at least two weeks prior to the start of the study. Food (Teklad 7001 Mouse & Rat Diet 4%) and water were continuously available throughout the study. All animals were maintained in facilities fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care. All research protocols and animal care were approved by the Institutional Animal Care and Use Committee in accordance with the guidelines of the Guide for the Care and Use of Laboratory Animals(Institute of Laboratory Animal Resources, Commission on Life Sciences, National Research Council, 1996). Experimental housing and care ----------------------------- For measurement of circadian period, all test mice were kept in a sound attenuating, ventilated room at a constant temperature (23°C) and under continuous darkness (DD). Sound attenuating, opaque dividers were placed between the test cages. Caretakers wore a Pelican Versabrite headlamp fitted with a red safelight beam diffuser. The diffuser/filter transmitted light greater than 600 nm only. Care in the darkroom consisted of ten min per day and each mouse was inspected for less than a minute. Daily visits occurred at random times between 8 am and 5 pm. Locomotor activity assessment ----------------------------- Daily locomotor activity of the mice was monitored with passive infrared detectors (Ademco, Syosset, NY) mounted over each cage. The passive infrared (ir) proximity sensor works by emitting pulses of ir light, and then measuring the distance to objects from the flight time of the reflected signal. Whenever the distance changes, the detector opens or closes a switch. All detectors were tested to ensure response uniformity. Each detector had a red LED that switched on when motion was detected, then switched off 3 to 5 seconds after movement was no longer sensed. The LED could be disabled with a switch mounted on the motion sensor circuit board. Spectral analysis ----------------- The irradiance spectra of the LEDs were determined using an S.I. Photonics fiber optic spectrophotometer (Tucson, AZ). The distance between the fiber optic input probe and the LED was 18 cm (the depth of the mouse cage). A random sample of eight of the 16 motion sensors used in the study was assessed. A hand-held light meter (UDT Instruments, Baltimore, MD) was used to measure illuminance produced by the LED at about 5 cm from the bottom of the cage and 13 cm from the LED. The test was repeated with the LEDs covered with black electrician\'s tape. Assessment of circadian period of locomotor activity ---------------------------------------------------- Activity events were grouped into 5-minute bins by the Stanford Chronobiology Systems\' (Stanford, CA) data processing and automatic storage system integrated into a Dell computer. Clocklab, the biological rhythm analysis software (Actimetrics, Evansville, IL), was used to calculate the period of locomotor activity of each mouse using linear regression through activity onsets of the last 8 to 10 days of each treatment. Experiment 1: Hypothesis -- LEDfb changes circadian period compared to DD ------------------------------------------------------------------------- ### Mouse husbandry C57BL/6 mice were bred in our facility from mice purchased from Jackson Laboratory (Bar Harbor, ME). Two male and six female C57BL/6 mice between the ages of 185 to 295 d were studied. ### Experimental protocol Four mice were put under motion sensors with the LED enabled (LEDfb), and four were put under sensors with the LED disabled (DD). The locomotor activity of the mice was monitored for two weeks (Stage 1). Then treatment was switched, and activity of the mice was monitored for another two weeks (Stage 2). The circadian period for each stage was calculated from the last 8 to 10 days under a given condition. ### Statistical analysis A factorial ANOVA (SAS ver. 9.1) tested for effect of sex, age, lighting condition (LEDfb compared to DD) and sequence of light treatment (LEDfb first compared to DD first). A post-hoc Tukey\'s Studentized Range test compared periods under different lighting protocols. Experiment 2: Hypothesis -- The effect of LEDfb on circadian period can be eliminated by blocking the light source ------------------------------------------------------------------------------------------------------------------ ### Mouse husbandry Twelve male C57BL/6 mice aged 30 d were purchased from Jackson Laboratory and housed singly. ### Experimental protocol Following acclimation to our facility in LD 12:12 for two weeks, they were moved into DD. All mice were put under motion sensors with the LED enabled (LEDfb), but black electrician\'s tape covered the LEDs of six motion sensors (taped LED) for two weeks. For Stage 2, all the LEDs were turned off for two weeks of DD. For Stage 3, the treatment condition of Stage 1 was switched, i.e. all LEDs were enabled but the LEDs previously covered by tape were uncovered and the previously uncovered LEDs were covered. Mouse activity was monitored continuously throughout the experiment. ### Statistical analysis A one-way ANOVA tested for effect of lighting condition (DD compared to LEDfb and taped LED), with a post-hoc Tukey\'s Studentized Range test. Experiment 3: Hypothesis -- C57BL/6 and DBA/2 mice have different response to LEDfb ----------------------------------------------------------------------------------- ### Mouse husbandry Eight male C57BL/6 mice and eight male DBA/2 mice were purchased from Jackson Laboratory, age 4 weeks. ### Experimental protocol Following acclimation to our facility in LD 12:12 for two weeks, mice were moved into DD. Four mice of each strain (DBA/2 and C57BL/6) were put under motion sensors with the LED enabled (LEDfb), and four of each strain were put under sensors with the light disabled (DD). At the end of Stage 1, they were moved to LD 12:12 for one week. When mice returned to the assessment room under DD, the treatment condition was switched. Activities of all mice were monitored for another two weeks. ### Statistical analysis A two-factor ANOVA tested for effect of strain and lighting condition (LEDfb compared to DD) with a post-hoc Tukey\'s Studentized Range test. The change in period from LEDfb compared to DD (Δτ) was calculated for each mouse and compared by a t-test. Results ======= Light measurements ------------------ The irradiance spectrum of the LED for a sample of 8 proximity sensors was a narrow band centered on 652 nm. Figure [1](#F1){ref-type="fig"} shows a representative spectrum. The illuminance of the LED was one lux. The illuminance of the LED covered with black electrician\'s tape was zero. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The irradiance spectrum of a red LED integrated into the passive-infrared motion sensor circuitry. ::: ![](1740-3391-3-8-1) ::: Experiment 1: LEDfb changes circadian period compared to DD ----------------------------------------------------------- Representative actograms of C57BL/6 mice under DD and dim red LEDfb are shown in Figure [2](#F2){ref-type="fig"}. The mean period in DD was 24.05 ± 0.04 h, and under red light LEDfb it was 24.21 ± 0.04 h. A factorial ANOVA testing for effect of sex, age, lighting condition (LEDfb compared to DD) and sequence of light treatment showed no effect of sex, age or sequence but an effect of lighting condition on period \[F~1,5~= 7.72, p = 0.039\]. A post hoc Tukey\'s test showed longer period with LEDfb compared to DD (p = 0.0095). Figure [3](#F3){ref-type="fig"} shows the effects of LEDfb on the circadian period of locomotor activity. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Double-plotted actograms of C57BL/6 mice under DD and dim red LEDfb. Lighting conditions are shown to the right of each actogram. ::: ![](1740-3391-3-8-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The circadian period of C57BL/6 mice is longer under dim red LEDfb than DD conditions (p = 0.0095). Lines show the mean period for each group. ::: ![](1740-3391-3-8-3) ::: Experiment 2: The effect of LEDfb on circadian period can be eliminated by blocking the light source ---------------------------------------------------------------------------------------------------- Representative actograms of C57BL/6 mice under DD, dim red LEDfb, and LEDs coved with black tape are shown in Figure [4](#F4){ref-type="fig"}. The mean period of 12 mice in DD was 23.96 ± 0.03 h; under taped LEDs, it was 23.93 ± 0.03 h; and, under LEDfb, it was 24.07 ± 0.03 h. There was a significant effect of lighting condition by one-way ANOVA \[F~2,33~= 7.02, p = 0.0029\]. A post-hoc Tukey\'s test showed a longer period under the uncovered LED than under the tape-covered LED (p \< 0.01) or DD (p \< 0.025), as summarized in Figure [5](#F5){ref-type="fig"}. Periods did not differ between DD and tape-covered LED. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Double-plotted actograms of C57BL/6 mice under DD, dim red LEDfb, and LEDs covered with black tape. Lighting conditions are shown to the right of each actogram. Arrows show onset of new lighting conditions. ::: ![](1740-3391-3-8-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Circadian period of C57BL/6 mice is longer under dim red LEDfb than when the light source is covered by black tape (p \< 0.01) or DD (p \< 0.025). Lines show the mean period for each group. ::: ![](1740-3391-3-8-5) ::: Experiment 3: C57BL/6 and DBA/2 mice do not have different responses to LEDfb ----------------------------------------------------------------------------- Representative actograms of DBA/2 and C57BL/6 mice under DD and dim red LEDfb are shown in Figure [6](#F6){ref-type="fig"}. For C57BL/6 mice, the mean period under DD was 23.85 ± 0.07 h; the mean period under LEDfb was 24.00 ± 0.07 h. For DBA/2 mice, the mean period under DD was 23.46 ± 0.14 h; the mean period under LEDfb was 23.78 ± 0.08 h. A two-factor ANOVA testing for effect of strain and lighting condition (LEDfb compared to DD) showed a significant effect of both strain \[F~1,14~= 7.73, p = 0.0147\] and lighting condition \[F~1,14~= 10.99, p = 0.0051\] but no interaction. A post-hoc Tukey\'s test showed longer period with LEDfb compared to DD (p \< 0.025). The C57BL/6 mice had different periods from the DBA/2 mice by post-hoc Tukey\'s test (p \< 0.01). Figure [7](#F7){ref-type="fig"} shows the effects of LEDfb on the circadian period of locomotor activity in the two strains of mice. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Double plotted actograms of DBA/2 and C57BL/6 mice under DD (top) and dim red LEDfb (bottom). The actograms at top and bottom are from one DBA/2 mouse (left) and one C57BL/6 mouse (right). Lighting conditions were separated by two weeks of LD 12:12. ::: ![](1740-3391-3-8-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### The circadian period of both DBA/2 and C57BL/6 mice under DD (filled symbols) and dim red LEDfb (open symbols). Lines show the mean period for each group. Overall, mice had longer period under LEDfb than DD (p \< 0.025), and C57BL/6 mice had longer periods than DBA/2 mice (p \< 0.01). ::: ![](1740-3391-3-8-7) ::: The mean increase in period with LEDfb (period under LEDfb minus period under DD, Δτ) for C57BL/6 mice was 0.15 ± 0.05 h. For DBA/2 mice, Δτ was 0.32 ± 0.13 h. The increase in period with LEDfb did not differ between strains by t-test (p = 0.26). In summary, circadian period was significantly longer under LEDfb (a small, red LED whose intensity was about 1 lux which came on only when a mouse was active) than that under DD in both C57BL/6 and DBA/2 strains of mice. The LEDs gave off red light in a narrow band centered on 652 nm. Covering the LED with black tape blocked the effect of the dim red light. Furthermore, there was no difference in this effect between the two strains. Discussion ========== This study suggests that the circadian system in mice is responsive to long wavelength red light. The result is surprising because recent studies suggest that melanopsin, in combination with the classical rod and cone photoreceptors, account for the transduction of photic information to the circadian system. There are no studies of spectral sensitivity of the Aschoff effect in mice. However, the spectral sensitivity of photoreceptors mediating circadian phase-shifts in mice is vanishingly small at wavelengths above 600 nm \[[@B7]-[@B9]\]. In most organisms, circadian period under LL is a function of both intrinsic period and photic inputs. The Aschoff effect is understood to result from the cumulative phase-shifting effect of LL on the pacemaker \[[@B5],[@B13],[@B15]\]. Thus an effect of red light on circadian period is unexpected. One possible explanation is that there is another photopigment present in mammals that is sensitive to far-red light and affects period rather than phase. It cannot be excluded that period and phase are affected by different light receptors or light receptive pathways. It seems more likely that low sensitivity to red light via the known circadian photopigments has a cumulative period-lengthening effect on the pacemaker. The timing of light exposure could have amplified this effect. Under LEDfb mice received light between circadian time (CT 12) and CT 24, during their active phase. The period-response curves (τRC) of mice have period-shortening between CT 4 and CT 16 and period-lengthening between CT 16 and CT 4 \[[@B16]\]. LEDfb should cause substantial period-lengthening, with minimal period-shortening. This study has several limitations. Only the Aschoff effect was investigated, and this was not under the usual protocol of constant light. Nevertheless, the increase in circadian period of mice under LEDfb was consistent with a prior activity feedback study in rats, where wheel-running triggered full-spectrum illumination resulting in lengthened circadian period \[[@B14]\]. Covering the LED with black tape blocked the effect of the dim red light. We conclude that ultrasound from the LEDs or the electronics associated with their illumination did not produce the effect. Mice emit ultrasonic cries and this is important in maternal behavior \[[@B17]-[@B19]\]. However, it is unlikely that covering the lights with black tape would block ultrasound. Although this remains a possibility, we are aware of no instances in the literature where ultrasound causes either a phase response or a change in circadian period. The C57BL/6 and DBA/2 mice did not differ in the amount period increased with 1 lux red light LEDfb. This is in contrast to an earlier study where C57BL/6 mice had a greater increase in period with 10 lux full-spectrum LL vs DD than DBA/2 mice \[[@B4]\]. One possible explanation is that DBA/2 mice are more sensitive to period-lengthening effects of red light than of full-spectrum light. Another possibility is that the shape of the τRC differs between the two strains, so that for C57BL/6 mice more of the period-lengthening portion of the τRC lies outside of the time when they received light during LEDfb than for the DBA/2 mice. In this case, constant LL would have more period-lengthening effect than LEDfb. The results of this study suggest that investigators cannot use continuous dim red light to simulate DD, and must be judicious in using red \"safe\" lights for animal care in DD. Further studies are needed to determine whether constant red light of this output spectra and intensity lengthens period more than LEDfb, and whether it can phase-shift the circadian rhythms in mice. Conclusion ========== Mice under a dim, long-wavelength red light that came on intermittently when the animals were active had a circadian period that was long compared to their free-running period under DD. Covering the light with black tape blocked the response. Furthermore, under the conditions described, the magnitude of the mean circadian period increase in DBA/2 and C57BL/6 strains of mice was indistinguishable. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= JRH conceived of the study, participated in its design and statistical analysis and drafted the manuscript. ARH carried out experiment 1, designed and carried out experiment 3, and wrote the Methods section. AMH participated in the statistical analysis and designed and set up experiment 2. ARM carried out experiment 2, participated in the statistical analysis and assisted in drafting the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by a Merit Grant to Dr. Mayeda from the Department of Veterans Affairs.	PubMed Central	2024-06-05T03:55:58.472002	2005-5-31	{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173135/", "journal": "J Circadian Rhythms. 2005 May 31; 3:8", "authors": [ { "first": "John R", "last": "Hofstetter" }, { "first": "Amelia R", "last": "Hofstetter" }, { "first": "Amanda M", "last": "Hughes" }, { "first": "Aimee R", "last": "Mayeda" } ] }

PMC1166566

Background ========== Multidrug resistance (MDR) phenotype of malignant cells is the major problem in the chemotherapy of neoplasia. P-glycoprotein (Pgp) activity is recognised to be one of the major mechanisms responsible for MDR. Pgp transports many structurally diverse compounds across the cell membrane and confers the MDR phenotype in tumor cells \[[@B1]\]. A number of signaling pathways participate in the regulation of *MDR1*gene expression and the activity of its product, Pgp \[[@B2]\]. Some of these signaling pathways could participate in coordinated regulation of *MDR1/*Pgp activity, cell proliferation and cell differentiation. It was shown that retinoic acid (RA) can modulate *MDR1*gene expression \[[@B3]-[@B5]\]. Retinoids are known to be involved into the regulation of the cell growth, differentiation and apoptosis. In the last decade retinoids became implicated into the treatment of leukemia and some solid tumors \[[@B6]\]. This approach changed the focus of the haematological diseases treatment from the cytotoxicity of the anti-cancer drugs to the reversal of arrested maturation of leukemic cells. Retinoids act via two families of receptors (RARs -- RARα, RARβ, RARγ) and RXRs (RXRα, RXRβ, RXRγ). There is the evidence that RARα is the crucial receptor mediating the biological effects during retinoid signaling in some cells \[[@B7]\]. Cell differentiation caused by the stable overexpression of receptor RARα was shown to result in constitutive over expression of *MDR1*gene in some cultured cells of solid tumors \[[@B4]\]. However the interconnections between RA/RARα regulated differentiation of leukemic cells and regulation of *MDR1*/Pgp activity remains unclear. In some leukemic cells RA did not influence *MDR1*and/or Pgp activity, while in the others it either augmented or reduced *MDR1*/Pgp expression \[[@B5],[@B8]\]. The aim of this study is to investigate if effects of *all-trans*-retinoic acid (ATRA) on *MDR1*/Pgp activity in leukemic cells are connected with *RAR*α expression and with the leukemic cell differentiation. We isolated sublines of cultured leukemic cells characterized by the stable *RAR*α overexpression and investigated the constitutive and ATRA induced *MDR1*/Pgp activity in these cells. Our data show that various *RAR*α transformed leukemic cell lines acquired more differentiated phenotype. Constitutive level of *MDR1*gene expression increased in one of *RAR*α overexpressing cell sublines. *RAR*α overexpression did not influence Pgp functional activity while Pgp activity induced by ATRA was elevated in all infectants studied. This shows that the main effect of *RAR*α in the cells studied is its influence on the induced functional activity of Pgp. Methods ======= Cell lines and culture ---------------------- Lines of cultured leukemic cells used in the study: H9 cells (acute human T-cell leukemia) \[[@B9]\], KG-1 cell line (cells of acute myelogenous leukemia) \[[@B10]\], K562 cell line (cells obtained from the patient in blast crisis of chronic myeloid leukemia) \[[@B11]\], NB4 (acute promyelocytic leukemia) \[[@B12]\]. Cells were grown in RPMI-1640 medium supplemented with 10% fetal calf serum (Gibco, USA), 2 mM L-glutamine, 50 μg/ml gentamycin at 37°C in a fully humidified atmosphere of 95% air and 5% CO~2~. All the derived cell lines described in this paper were obtained by retroviral infection and selection with the appropriate antibiotic. ATRA (*all-trans*-retinoic acid, Sigma, USA) was added to the culture medium at seeding or 24 hours after seeding (see Legends to Figures). Expression vector and retrovilal infection ------------------------------------------ The PA317/LRARSN retroviral vector-producing cell line was used. All the procedure was described earlier \[[@B4]\]. In brief, the vector used contains a cDNA fragment harbouring the complete coding sequence of the *RAR*α gene driven by the Moloney murine leukemia virus long-terminal repeat as well as the SV40 early promoter-driving neomycin phosphotransferase gene (*neo*) as a selectable marker \[[@B13]\]. The cells (4 × 10^5^per 25-cm^2^flask) were seeded 24 h before infection. Conditioned medium from a retrovirus-producing cell line was filtered through a 0.45-μm membrane (Millipore, USA), diluted 1:1 with medium, containing 1% serum and 8 μg /ml Polybrene and added to the cells for 24 h at 37°C, 5% CO~2~. Further selection were carried out by culturing cells in medium supplemented with 400 μg/ml G418 (Gibco, USA) for at least 21 days. The medium was changed twice a week. The pool of G418-resistant cells was resuspended in culture medium and progressively expanded. Assay of cell growth, apoptosis and differentiation --------------------------------------------------- Cells were seeded into 24-well plates (1 × 10^4^cell per well) and the cell number was counted at days 1, 3, 5 and 8 after seeding. The apoptosis in the populations of the parental and *RARα*infected cell lines was performed using the standard procedure \[[@B14]\]. Cells were collected 24 h after seeding, washed with PBS, and fixed in 70% ethanol overnight at 4°C. Fixed cells were suspended in citric buffer and stained with propidium iodide (5 mcg/ml) in PBS for 1 hour at 4°C. DNA content was subsequently measured by FACScan (Becton Dickinson, USA). The immunophenotype of the cells was evaluated as previously described \[[@B15]\]. Surface expression of the following antigens was determined: CD3, CD5, CD7, CD8, CD11b, CD13, CD15, CD33, CD34, HAE3 and HAE9. In brief, cells were incubated with phycoerythrin-labelled monoclonal mouse antibodies for 20 min at 4°C (Becton, Dickinson), washed with RPMI 1640 medium and analyzed with a flow cytometer (Becton Dickinson). Analysis of rhodamine 123 (Rh123) efflux by the cells ----------------------------------------------------- The technique used in the study was described in \[[@B16]\]. Cells were loaded with 5 μg/ml Rh123 (Sigma) for 10 min at 37°C, washed twice with cold PBS, pH 7.2, and incubated for 30 min in dye-free medium at 37°C. After the completion of incubation, cell were washed twice with cold PBS. Cell fluorescence was measured on a flow cytometer FACScan (Becton Dickinson, USA). Each measurement counted 5000 events. Non-viable cells were gated out of the analysis on the basis of side scatter. RNA isolation and reverse transcriptase polymerase chain reaction (RT-PCR) analysis of RARα and MDR1 genes expression --------------------------------------------------------------------------------------------------------------------- The cells were dissolved in TRI reagent (Sigma, USA). Total RNA was isolated as described in the manufacturer\'s manual. For analysis, aliquots of isolated RNA were denatured with formamide and subjected to electrophoresis in 1.8% agarose gels. The samples with clearly visualized 18S and 28S RNA bands were used for further procedures. First-strand cDNA was synthesized using reverse transcriptase M-MuLV (MBI Fermentas, Russia) with 4 μg RNA as a template, 2.5 ng random hexamers, 0.25 mM of each deoxynucleotide triphosphate (SibEnzyme, Russia), dithiothreitol, 4 Units of RNAase inhibitor (MBI Fermentas, Russia) and 100 Units of M-MuLV RT. The reaction was performed at 42°C for 50 min, and 1/60 volume of reaction mixture was used for amplification. PCR was done in a total volume of 25 μl using the thermocycler \"Tercyc\" (DNA-technology, Russia). The PCR mixture consisted of (NH~4~)~2~SO~4~-containing PCR buffer (\"MBI Fermentas\"), 0.160 mM dNTPs mix (\"MBI Fermentas\"), 2 mM MgCl2, 20 pmoles of each specific primer and 0.8 Unit of Taq-polymerase (\"MBI Fermentas\"). PCR was done as follows: 94°C for 2 seconds, Tm (different for each gene) for 10 seconds, 72°C for 5 seconds. Semi-quantitative PCR analysis of *RARa*and *MDR1*genes expression were performed using oligomers amplifying a 333 bp and 167 bp products, respectively. Specific gene primers used for RT-PCR are given in Table [1](#T1){ref-type="table"}. The amounts of template cDNAs were normalized by PCR amplification of β2-microglobulin cDNA (internal control). The optimal numbers of PCR cycles were 24 for the b2-microglobulin, 26 for RARα-specific product, 33 for *MDR1*(for all cells lines except KG1 and KG1/RAR, for these cells the numbers of PCR cycles *MDR1*-specific product were 26). These numbers of cycles yielded clearly detectable PCR products within an exponential range. PCR products were amplified in separate tubes, resolved by electrophoresis in 2% agarose gel, stained with ethidium bromide and visualized in UV light. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Specific gene primers used for RT-PCR ::: Gene Product size Primer Sequence -------- -------------- -------- ------------------------------- *RARα* 333 bp F 5\'-GTCTTTGCCTTCGCCAACCAG-3\' R 5\'-GCCCTCTGAGTTCTCCAACA-3\' *MDR1* 127 bp F 5\'-CCCATCATTGCAATAGCAGG-3\' R 5\'-GTTCAAACTTCTGCTCCTGA-3\' *β2m* 114 bp F 5\'-ACCCCCACTGAAAAAGATGA-3\' R 5\'-ATCTTCAAACCTCCATGATG-3\' ::: Results ======= Influence of RARα gene overexpression on cell differentiation, proliferation and spontaneous apoptosis ------------------------------------------------------------------------------------------------------ *RARα*gene was introduced into the cultured leukemic cells of diverse types of differentiation as described in Methods. The sublines of H9, KG-1, K562 and NB4 cells characterized by the capability to grow in the medium supplemented with G418 were isolated (H9/RAR, KG-1/RAR, K562/RAR and NB4/RAR). Semi-quantitative RT-PCR revealed more pronounced expression of *RARα*mRNA in all transfected cell lines in comparison with the wild type cells (Fig. [1](#F1){ref-type="fig"}). ATRA (5 μM applied for 48 h) increased *RARα*mRNA in some *RARα*transformed cells (H9/RAR, KG-1/RAR, K562/RAR) to a greater extent than in parental cells (Fig. [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Expression of *RARα*mRNA in parental and *RARα*transfected cell lines. RT PCRM**. k- -- water. ATRA (5 μM) was added to cell cultures 24 h after seeding for 48 h. Then the cells were collected and processed as specified in \"Methods\" (RNA isolation and reverse transcriptase polymerase chain reaction (RT-PCR) analysis of *RARα*gene expression). This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-1) ::: The investigation of the differentiation status of these cells shows that all *RARα*transfected sublines differ from the parental cell populations (Fig. [2](#F2){ref-type="fig"}). *RARα*transfected H9 culture contains more cell variants expressing CD5 and CD8 antigens than parental cell line (Fig. [2A](#F2){ref-type="fig"}). Thus the number of cells with antigens of later lymphoid differentiation markers increased in *RARα*overexpressing H9 cells. There is phenotypic evidence of granulocytic differentiation in KG-1/RAR cell subline as indicated by a reduction in CD13 expression and the increase in the expression of CD11b antigen in comparison with parental cells (Fig. [2B](#F2){ref-type="fig"}). In KG-1/RAR cell population the portion of CD34 cells decreased and the portion of CD33 cells increased (Fig. [2B](#F2){ref-type="fig"}). This also testifies to increased differentiation of these *RARα*overexpressing cells. In K562/RAR population the number of the cells of erythroid differentiation (expressing HAE9 and HAE3 antigens) is larger than in K562 population (Fig. [2C](#F2){ref-type="fig"}). Hemoglobin synthesis is increased in K562/RAR culture more than 5-fold in comparison with parental cells (not shown). In NB4/RAR population the number of the cells of myelogenous differentiation (expressing CD11b and CD15 antigens) is larger than in NB4 population (Fig. [2D](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Comparison of antigen expression by the parental and *RARα*infected cell lines**. Cells were incubated for 30 min at 4°C in the presence of an appropriate monoclonal antibody. After three washes with PBS, cells were incubated for 30 min at 4°C with goat antimouse IgG labeled with phycoerythrin and then analyzed in flow cytometer (Becton Dickinson). ::: ![](1475-2867-5-15-2) ::: The percentage of cells undergoing spontaneous apoptosis increased 2-3-fold in all *RARα*transfected cell populations (Fig. [3](#F3){ref-type="fig"}). This could be connected with more differentiated phenotype of *RARα*transformed cells. It seems that in the population of H9/RAR cells the increased number of apoptotic cells could be at least in part connected with increased expression of CD95 (Fas/APO1): in this *RARα*transformed subline CD95 increased almost 10-fold in comparison with parental cell population (from 2,6% in H9 to 21,4% in H9/RAR culture). However in KG-1 and K562 cell populations the number of CD95 expressing cell did not increase after *RARα*transformation. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Spontaneous apoptosis in the populations of the parental and *RARα*infected cell lines**. Propidium iodide flow cytometry detection of dead cells was performed using the standard procedure. Cells were collected 24 h after seeding, washed with PBS, and fixed in 70% ethanol overnight at 4°C. Fixed cells were suspended in citric buffer and stained with 5 mcg/ml propidium iodide in PBS for 1 hour at 4°C. DNA content was subsequently measured by FACScan (Becton Dickinson, USA). All cells with sub-G~0~DNA content were regarded as dead cells. This figure is representative of 3 separate experiments. ::: ![](1475-2867-5-15-3) ::: As Fig. [4](#F4){ref-type="fig"} shows, *RARα*transfected KG-1, K562 and NB4 cells proliferated more slowly than parental cells. However H9/RAR cells did not demonstrate slower proliferation rate. Thus, more differentiated status of the *RARα*transformed cell populations was not necessary connected with the decrease in the proliferation rates. All *RARα*transformed cells seem to be more sensitive than wild type cells to inhibitory action of ATRA on cell proliferation (Fig. [5](#F5){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Proliferation rates of parental and *RARα*infected cells**. Cells were seeded into 24-well plates (1 × 10^4^cell per well) and the cell number was counted at days 1,3,5,8 after seeding. This figure is representative of 3 separate experiments. ::: ![](1475-2867-5-15-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Influence of retinoic acid (ATRA, 5 μM) on the proliferation of parental and *RARα*transfected cells**. Cells were seeded into 24-well plates (1 × 10^4^cell per well), ATRA was added at seeding and the cell number was counted at days 1, 3, 5 and 8 after seeding. This figure is representative of 3 separate experiments. ::: ![](1475-2867-5-15-5) ::: Influence of RARα overexpression on MDR1 gene activity ------------------------------------------------------ We studied intrinsic and ATRA induced expression of *MDR1*gene in all cell lines by semi-quantitative RT-PCR technique. The basal levels of *MDR1*mRNA varied in different wild type cells: in H9 and NB4 cells constitutive *MDR1*gene expression was not revealed, in K562 wild type cells some *MDR1*mRNA was found, in KG-1 cells the quantity of *MDR1*mRNA was large (Fig. [6](#F6){ref-type="fig"}). It is noteworthy that the optimal number of PCR cycles were 33 for *MDR1*-specific product in all cells while for studies of KG1 and KG1/RAR cells we used 26 PCR cycles. In *RARα*transfected H9 cells the constitutive expression of *MDR1*gene slightly increased, while in KG-1/RAR and NB4/RAR cells the constitutive *MDR1*mRNA quantity was not elevated in comparison with the wild type cells, it seems to be even slightly decreased in K562/RAR (Fig. [6](#F6){ref-type="fig"}). Thus the alterations of the basal level of *MDR1*expression in *RARα*transformed cells seem to vary in different cell lines. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Intrinsic and retinoic acid induced expression of *MDR1*gene in parental and *RARα*transfected cells**. k- -- water. ATRA (5 μM) was added to cell cultures 24 h after seeding for 48 h. Then the cells were collected and processed as specified in \"Methods\" (RNA isolation and reverse transcriptase polymerase chain reaction (RT-PCR) analysis of *MDR1*gene expression). The optimal numbers of PCR cycles were 33 for *MDR1*-specific product for all cells lines except KG1 and KG1/RAR, for these cells the numbers of PCR cycles for *MDR1*-specific product were 26. This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-6) ::: ATRA (5 μM applied for 48 h) increased *MDR1*gene expression in all examined cell lines either in parental or *RARα*transfected cells (Fig. [6](#F6){ref-type="fig"}). In H9/RAR cells effect of ATRA on *MDR1*expression was significantly greater in comparison with parental cells. In other ATRA treated *RARα*transformed cell sublines *MDR1*expression was undistinguishable from ATRA treated parental cells (Fig [6](#F6){ref-type="fig"}). Influence of RARα gene transformation on Rh123efflux by the cells ----------------------------------------------------------------- The retention of Rh123 by the cells is considered as a test for Pgp functional activity \[[@B16],[@B17]\]. Rh123 efflux from the cells was increased in K562/RAR cells in comparison with the parental cell population (Fig. [7B](#F7){ref-type="fig"}, Table [2](#T2){ref-type="table"}). In H9 and KG-1 *RARα*transformed cells there were alterations in the Rh123 retention (Fig. [7A](#F7){ref-type="fig"} and [7C](#F7){ref-type="fig"}): in the populations of H9/RAR and KG-1/RAR cultures the fraction of more dull cells decreased in comparison with parental cultures (mean fluorescence intensity of the cell sublines studied are given in the table [2](#T2){ref-type="table"}). This shows that Pgp activity was not elevated in these *RARα*transformed cell populations and suggests that there is some decrease in Pgp functional activity. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **Evaluation of Rh123 efflux from the parental and cells *RARα*transfected cells**. Cells were loaded with 5 μg/ml Rh123 for 10 min at 37°C, than washed twice with cold PBS, and incubated for 30 min in dye-free medium at 37°C. After the completion of incubation, cell were washed twice with cold PBS. Cell fluorescence was measured on a flow cytometer FACScan (Becton Dikinson, USA). Each measurement counted 5000 events. Non-viable cells were gated out of the analysis on the basis of side scatter. This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-7) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Influence of *RARα*transformation on the intrinsic and induced expression of *MDR1*gene ::: *MDR1*expression Rh123 efflux (mean fluorescence intensity) ------------ ------------------ -------------------------------------------- ---------- ---------- **H9** \- \+ **450** **457** **H9/RAR** \+ ++ **506** **416** K562 \+ ++ **446** **370** K562/RAR \+ ++ **403** **333** KG-1 ++ +++ **65,7** **38,3** KG-1/RAR ++ +++ **90,7** **20,2** NB4 \- ++ **n.d.** **n.d.** NB4/RAR \- ++ **n.d.** **n.d.** ::: There was increase in the portion of Rh123 dull cells after ATRA treatment both in K562 and K562/RAR cell populations (mean fluorescence intensity of both populations decreased approximately on 17%) (Fig. [8C, D](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). ATRA induced Rh123 efflux from H9/RAR cells, while in H9 parental population this drug had no effect (Fig. [8A, B](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). In KG-1/RAR cells ATRA induced very prominent increase in the number of Rh123 dull cells (more than 70% decrease of mean fluorescence intensity), while in the parental cell population ATRA decreased mean fluorescence intensity to a lesser extent (Fig. [8E, F](#F8){ref-type="fig"}, Table [2](#T2){ref-type="table"}). ::: {#F8 .fig} Figure 8 ::: {.caption} ###### **Influence of retinoic acid (ATRA) on Rh123 efflux from the parental and cells *RARα*transfected cells**. ATRA (5 μM) was added to cell cultures 24 h after seeding for 48 h. Than cells were loaded with 5 μg/ml Rh123 for 10 min at 37°C, than washed twice with cold PBS, and incubated for 30 min in dye-free medium at 37°C. After the completion of incubation, cell were washed twice with cold PBS. Cell fluorescence was measured on a flow cytometer FACScan (Becton Dikinson, USA). Each measurement counted 5000 events. Non-viable cells were gated out of the analysis on the basis of side scatter. This figure is representative of 2 separate experiments. ::: ![](1475-2867-5-15-8) ::: Discussion ========== The treatment of leukemia with retinoids is aimed on the induction of leukemic cells differentiation. The question is: are there interconnections between RA/RARα regulated differentiation of leukemic cells and *MDR1*/Pgp activity? In this study we have isolated more differentiated variants of the cultured leukemic cells by the introduction into the cells of *RARα*gene encoding one of RA receptors. All *RARα*transformed leukemic cell populations were characterized by the higher *RARα*gene expression in comparison with the parental cells. All *RARα*transformed leukemic cell populations became more differentiated. This was demonstrated by the studies of the differentiation markers, by the increase in the number of cells dying by spontaneous apoptosis and by the decrease of the proliferation rates of most *RARα*transfected cell sublines. Thus, *RARα*overexpression could result in the increase of the differentiation of various leukemic cell populations. We compared *MDR1*gene expression and Pgp functional activity tested by Rh123 retention in parental and *RARα*transformed cells. The results are summarized in the Table [2](#T2){ref-type="table"}. Increased constitutive (uninduced) expression of *MDR1*gene was found in one of four cell lines after *RARα*transformation (H9/RAR, Table [2](#T2){ref-type="table"}, Fig. [6](#F6){ref-type="fig"}). In the previous experiments with melanoma and hepatoblastoma human cells we have shown that constitutive expression of *MDR1*gene was increased after *RARα*transfection in both *RARα*transformed cell sublines \[[@B4]\]. Thus interconnections between regulation of the basal *MDR1*and *RARα*activities could exist both in the cells of solid tumors and in the leukemic cells. Our data suggest, that in the cell populations of solid tumors *RARα*overexpression could be accompanied by constitutive *MDR1*over-expression more often than in the cells of hematopoietic malignancies. Our study did not reveal the occurrence of the functional Pgp in leukemic cells studied after *RARα*transformation. In H9/RAR cells elevation of the constitutive *MDR1*expression did not lead to the increase in Rh123 efflux (Fig. [7A](#F7){ref-type="fig"}, Table [2](#T2){ref-type="table"}). Some studies also have described discrepancies between Pgp (protein) or *MDR1*mRNA expression and Pgp function in leukemic cells \[[@B18],[@B19]\]. These discrepancies could occur for a variety of reasons. Anyway, our data show that increase in the differentiation of leukemic cell populations induced by *RARα*overexpression did not result in the elevation of constitutive Pgp functional activity. In our previous study we found that *RARα*overexpression did not change Pgp functional activity in two *RARα*transformed sublines of human cells (melanoma and hepatoblastoma) but did change it in the rat cells \[[@B4]\]. It seems that exogeneous *RARα*in the cells of human malignancies does not influence basal Pgp functional activity. In KG-1/RAR characterized by the increased differentiaion (Table [1](#T1){ref-type="table"}) we had not found increase in the constitutive *MDR1*expression and Pgp functional activity decreased (Fig. [7C](#F7){ref-type="fig"}, Table [2](#T2){ref-type="table"}). It is known that blood stem cells and early progenitors expressing CD34 antigen also express high levels of functionally active Pgp \[[@B20]\]. Maturation of these cells is accompanied by the decrease in Pgp expression and even more rapid decrease in Pgp functional activity \[[@B21]\]. It may be suggested that alterations of Pgp function in KG-1/RAR are connected with the differentiation of these cells. The situation with Pgp functional activity induced by ATRA in the cells studied differs from the situation with constitutive activity of this protein. In all three *RARα*transfected cells ATRA had induced Pgp fuctional activity (Fig. [8](#F8){ref-type="fig"}. Table [2](#T2){ref-type="table"}). Moreover, in two *RARα*transformed sublines (H9/RAR and KG-1/RAR) ATRA activated Pgp, while in the parental cells it had either no effect (H9) or activated Pgp to a lesser extent (KG-1) (Table [2](#T2){ref-type="table"}). These data suggest that *RARα*participate in the control of induced, but not in constitutive Pgp functional activity in leukemic cells. The regulation of *MDR1*gene transcription and Pgp functional activities are the complex processes \[[@B1],[@B2]\]. The studies of these processes are underway. Our data show that *RARα*gene overexpression could influence the induced Pgp functional activity in leukemic cells, i.e. could participate in the occurrence of multidrug resistance in the populations of these malignant cells. It seems that this influence could depend on the cell context. Acknowledgements ================ This work was supported by grants 04-04-48613a and 02-04-48200 from the Russian Foundation for Basic Research.

PubMed Central

2024-06-05T03:55:58.152910

2005-5-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166566/", "journal": "Cancer Cell Int. 2005 May 24; 5:15", "authors": [ { "first": "Tatjana P", "last": "Stromskaya" }, { "first": "Ekaterina Y", "last": "Rybalkina" }, { "first": "Tatjana N", "last": "Zabotina" }, { "first": "Alexander A", "last": "Shishkin" }, { "first": "Alla A", "last": "Stavrovskaya" } ] }

PMC1166567

Introduction ============ The prevalence of type 2 diabetes in adults was estimated at 2.8% worldwide in 2000, and predicted to increase to 4.4% by 2030 \[[@B1]\]. Patients with type 2 diabetes have a risk of cardiovascular disease approximately two- to four-times greater than that in the non-diabetic population \[[@B2]\]. Furthermore, their prognosis is worse; in a Swedish study the 5-year mortality rate after myocardial infarction was 55% for patients with diabetes compared with 30% in patients without diabetes (p \< 0.001), and the re-infarction rates were 42% and 25%, respectively (p \< 0.001) \[[@B3]\]. More recent data reflecting the outcome of new evidence-based interventions in acute myocardial infarction demonstrate that the difference between diabetic and non-diabetic subjects is still present, showing a 1-year mortality in males of 22.3% versus 13.0% in males and 26.1% versus 14.4% in females \[[@B4]\]. In the US National Health and Nutrition Examination Survey (NHANES I), the age-adjusted mortality rate in diabetic patients over 9 years of follow-up was double that in non-diabetic patients, and cardiovascular disease accounted for 75% of the excess mortality in men and 57% in women \[[@B5]\]. The elevated cardiovascular risk in patients with type 2 diabetes is primarily attributed to the clustering of atherogenic risk factors, including dyslipidaemia, hypertension, abdominal obesity, left ventricular hypertrophy, and impaired fibrinolysis \[[@B6]\]. For patients with diabetes, European Diabetes Policy Group guidelines published in 1999 and European guidelines for coronary heart disease prevention published in 1998, both recommend that low-density lipoprotein cholesterol (LDL-C) levels should be \<3.0 mmol/L \[[@B7],[@B8]\]. More recent (2003) European guidelines on cardiovascular disease prevention also recognise type 2 diabetes as a risk factor, and recommend more stringent LDL-C reductions to\<2.5 mmol/L \[[@B9]\]. In addition, this goal is recommended by both the American Diabetes Association and the US National Cholesterol Education Program Adult Treatment Panel III \[[@B10],[@B11]\]. Statins are recognized as first-line therapy for cholesterol lowering \[[@B7],[@B11]\], and have been proven to reduce cardiovascular morbidity and mortality in large outcome trials in various populations \[[@B12]-[@B17]\]. The benefits of statin therapy extend to patients with diabetes, as shown by subgroup analyses of patients with diabetes in several of the major statin outcome studies, including the Cholesterol And Recurrent Events (CARE) study, the Scandinavian Simvastatin Survival Study (4S), the Long-Term Intervention with Pravastatin in Ischemic Disease (LIPID) study and the Heart Protection Study \[[@B18]-[@B21]\]. The Collaborative Atorvastatin Diabetes Study (CARDS) recently investigated the effects of lipid lowering with statin therapy specifically in patients with type 2 diabetes \[[@B22]\]. The primary endpoint, time to the first occurrence of acute coronary events, coronary revascularisation or stroke, was significantly reduced by 37% in patients treated with atorvastatin 10 mg compared with placebo (p = 0.001). In addition, LDL-C levels were significantly reduced by 40% in the atorvastatin 10 mg group compared with the placebo group (p \< 0.001) \[[@B22]\]. Despite the proven benefits of statin therapy, studies suggest that many patients with diabetes fail to achieve lipid goals in clinical practice \[[@B23],[@B24]\]. Statins differ in their lipid-modifying efficacy and their ability to enable patients to achieve lipid goals \[[@B25],[@B26]\]. Trials in patients with hypercholesterolaemia have shown that rosuvastatin is more effective than atorvastatin at reducing LDL-C and achieving US and/or European LDL-C goals over treatment periods ranging from 6 to 52 weeks \[[@B25]-[@B30]\]. The URANUS (Use of Rosuvastatin versus Atorvastatin iN type 2 diabetes mellitUS) study is a direct comparison of the effects of rosuvastatin with atorvastatin on LDL-C, other plasma lipids and LDL-C goal achievement in patients with diabetes. This study was designed to reflect the optimal treatment of diabetic dyslipidaemia in the clinical setting, in that statin dose was titrated upwards from the recommended start dose to that required to enable patients to achieve LDL-C goal. Methods ======= Patients -------- Patients (male or female) aged 18 years or more were eligible for the study if they had a history of type 2 diabetes for at least 3 months; were being treated with diet, oral antidiabetic medication, insulin or a combination of these treatments; and had fasting LDL-C of ≥ 3.3 mmol/L and triglycerides (TG) of\<6.0 mmol/L at enrolment. Exclusion criteria included: type 1 diabetes; uncontrolled type 2 diabetes; uncontrolled hypothyroidism or hypertension; nephrotic syndrome or severe renal failure; active liver disease or hepatic dysfunction; active arterial disease (e.g., unstable angina, myocardial infarction, transient ischaemic attack, cerebrovascular accident, coronary artery bypass grafting or percutaneous transluminal coronary angioplasty within 3 months before beginning the study); serum creatinine kinase (CK) levels \>3 × the upper limit of normal (ULN); body mass index \>35 kg/m^2^; and known hypersensitivity to statins. All patients gave written informed consent, and the study was conducted in accordance with the Declaration of Helsinki. Study design ------------ The trial was a randomised, double-blind, parallel-group study (4522SE/0001) conducted in 51 centres in Sweden. The study design is summarised in figure [1](#F1){ref-type="fig"}. Patients meeting the inclusion and exclusion criteria at enrolment entered a 6-week dietary run-in period and all lipid-lowering therapy was withdrawn at least 14 days before the end of this period. Patients with fasting LDL-C ≥ 3.3 mmol/L were then randomised to a starting dose of either rosuvastatin 10 mg or atorvastatin 10 mg for 4 weeks. This was followed by a 12-week period of dose titration, making a total of 16 weeks of treatment. Patients who had not reached the 1998 European guideline goal of LDL-C\<3.0 mmol/L \[[@B7]\] after 4 weeks were titrated up by doubling the statin dose (rosuvastatin 20 mg or atorvastatin 20 mg). Further dose titrations (to rosuvastatin 40 mg, or atorvastatin 40 mg or 80 mg) were performed at 8 weeks and 12 weeks for patients who were still not at their LDL-C goal. Patients whose LDL-C level was below the goal at 4 weeks continued on the initial dose of study medication; if their LDL-C level exceeded the goal at subsequent visits, study medication was up-titrated. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Study design**. ^a^In patients who had not reached the European goal of LDL-C\<3.0 mmol/L after 4 weeks, the statin dose was doubled at each visit, up to a maximum of RSV 40 mg and ATV 80 mg **RSV**: Rosuvastatin, **ATV**: Atorvastatin, **LDL-C**: Low-density lipoprotein cholesterol ::: ![](1475-2840-4-7-1) ::: Concomitant treatment with erythromycin, azole antimycotic agents, vitamin K antagonists, immunosuppressive agents, glitazones or systemic steroids was not permitted during the study. If insulin treatment became necessary, or if the patient took lipid-lowering medication (other than study medication), the patient was discontinued from the trial. Assessments ----------- ### Efficacy The primary endpoint was the percentage change in LDL-C from baseline (randomisation) to 16 weeks. Secondary endpoints included: percentage change in LDL-C from baseline to 4 weeks; percentage of patients achieving the 1998 European LDL-C goal at 4 and 16 weeks; percentage change in total cholesterol (TC), TG, high-density lipoprotein cholesterol (HDL-C), the LDL-C/HDL-C ratio, the non-HDL-C/HDL-C ratio, the TC/HDL-C ratio, apolipoprotein (apo) B, apo A-I and the apo B/apo A-I ratio from baseline to 4 and 16 weeks; and the number of titration steps at 16 weeks. A tertiary endpoint was the difference in overnight urinary albumin excretion (UAE) from baseline to 16 weeks. All patients were instructed to fast for 8 hours prior to giving blood samples. LDL-C levels were measured using a direct method with enzymatic colorimetry (Genzyme Diagnostics, Genzyme Corporation, Cambridge, MA, USA). All analyses were conducted at a central laboratory. ### Safety Adverse events spontaneously reported by the patients, elicited in response to an open question or revealed by observation, were recorded at each visit. Laboratory safety variables included: blood haemoglobin, platelet count, leucocyte count, serum aspartate aminotransferase (ASAT), serum alanine aminotransferase (ALAT), serum alkaline phosphatase, serum bilirubin, CK, serum creatinine and glycated haemoglobin (HbA~1c~). All analyses were performed at a central laboratory. Statistical methods ------------------- In order to have a 90% chance of detecting a difference between the two treatment arms of 6% in the primary endpoint (percentage change in LDL-C from baseline to 16 weeks), 212 patients per arm were required to complete the study. The primary efficacy endpoint was determined using analysis of covariance with change in LDL-C as response variable, treatment and centre as factors, and baseline LDL-C as covariate. Percentage change in other lipid variables from baseline to 4 weeks and 16 weeks, and percentage change in UAE, were analysed in the same way as the primary endpoint. The proportion of patients reaching LDL-C goal was analysed using a Mantel-Haenszel test stratified by centre. All tests were two-sided with a significance level of 5%. In addition, 95% confidence intervals were calculated for the treatment differences in all efficacy variables except the percentage of patients reaching LDL-C goal. All efficacy variables were analysed in the intention-to-treat population (observed data). Analysis of the primary endpoint was also carried out using the last observation carried forward approach. All enrolled patients were evaluated for safety. Results ======= Demographics ------------ A total of 469 patients were randomised, and efficacy data were obtained from 465 patients, 232 in the rosuvastatin group and 233 in the atorvastatin group (figure [2](#F2){ref-type="fig"}). The two groups were well matched at baseline, and demographic details are shown in table [1](#T1){ref-type="table"}. Previous statin treatment was received by 31 patients (13%) in the rosuvastatin group and 39 patients (17%) in the atorvastatin group. Eleven patients in the rosuvastatin group and 12 in the atorvastatin group discontinued during the randomised treatment period (figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Study populations ::: ![](1475-2840-4-7-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient demographics of ITT population ::: **Rosuvastatin (n = 232)** **Atorvastatin (n = 233)** ---------------------------------- ---------------------------- ---------------------------- Gender, male/female (%) 128/104 (55.2/44.8) 136/97 (58.4/41.6) Race, white (%) 229 (98.7) 229 (98.3) Mean age, years (SD) 63.5 (8.8) 65.0 (8.6) Mean weight, kg (SD) 84.8 (14.3) 82.5 (13.5) Mean BMI, kg/m^2^(SD) 29.0 (3.6) 28.4 (3.6) Mean baseline LDL-C, mmol/L (SD) 4.6 (0.85) 4.6 (0.82)^a^ Mean baseline HDL-C, mmol/L (SD) 1.2 (0.27) 1.2 (0.27)^a^ Mean baseline TG, mmol/L (SD) 2.0 (1.0) 2.0 (0.93)^a^ **ITT:**Intention to treat, **SD:**Standard deviation, **BMI:**Body mass index, **LDL-C:**Low-density lipoprotein cholesterol, **HDL-C:**High-density lipoprotein cholesterol, **TG:**Triglycerides ^a^n = 232 ::: Efficacy -------- At the end of the titration-to-goal period, rosuvastatin was significantly more effective than atorvastatin on the primary efficacy measure, reducing LDL-C by 52% compared with 46% in the atorvastatin group (p \< 0.0001) (table [2](#T2){ref-type="table"}). In line with its greater efficacy for LDL-C reduction, significantly more rosuvastatin-treated patients reached the 1998 European LDL-C goal after 16 weeks than atorvastatin-treated patients (94% vs 88%, p \< 0.05). Furthermore, more patients achieved the goal on the starting dose of rosuvastatin than atorvastatin (75% vs 54%) (figure [3](#F3){ref-type="fig"}). The greater ability of rosuvastatin to lower LDL-C was also reflected by the number of dose titrations required by each treatment group; a total of 75 titration steps were required by rosuvastatin-treated patients compared with 155 titrations in the atorvastatin group. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Percentage change from baseline in lipid variables at 16 weeks (ITT population). Doses were titrated from week 4 to week 16 in patients who had not reached the 1998 European LDL-C goal (\< 3.0 mmol/L) ::: **Variable** **Least-squares mean percentage change from baseline to 16 weeks** **Difference (95% CI)** **p-value** ----------------------- -------------------------------------------------------------------- ------------------------- ------------------- ----------- LDL-C -52.3 -45.5 -6.7 (-8.8, -4.7) \< 0.0001 TC -35.4 -31.3 -4.1 (-5.8, -2.4) \< 0.0001 HDL-C 5.3 4.0 1.3 (-1.3, 3.8) NS TG -21.2 -21.1 -0.1 (-5.6, 5.3) NS Non-HDL-C -45.0 -39.6 -5.5 (-7.4, -3.5) \< 0.0001 LDL-C/HDL-C ratio -54.1 -47.0 -7.1 (-9.3, -4.9) \< 0.0001 Non-HDL-C/HDL-C ratio -47.1 -40.9 -6.2 (-8.6, -3.9) \< 0.0001 TC/HDL-C ratio -38.0 -33.1 -5.0 (-6.9, -3.0) \< 0.0001 Apo B -45.2 -40.1 -5.1 (-7.2, -3.1) \< 0.0001 Apo A-I 2.6 -0.2 2.8 (1.0, 4.6) 0.0024 Apo B/apo A-I ratio -46.3 -39.6 -6.7 (-8.9, -4.6) \< 0.0001 **ITT:**Intention to treat, **CI:**Confidence interval, **LDL-C:**Low-density lipoprotein cholesterol, **TC:**Total cholesterol, **HDL-C:**High-density lipoprotein cholesterol, **TG:**Triglycerides, **Apo:**Apolipoprotein, **NS:**Not statistically significant ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Cumulative percentage of patients to 1998 European LDL-C goal of\<3.0 mmol/L \[7\] by dose at 16 weeks**. \*p \< 0.05 rosuvastatin 10--40 mg vs atorvastatin 10--80 mg **RSV**: Rosuvastatin, **ATV**: Atorvastatin, **LDL-C**: Low-density lipoprotein cholesterol ::: ![](1475-2840-4-7-3) ::: During the 4-week fixed-dose period, significantly more patients on rosuvastatin 10 mg had reached the 1998 European LDL-C goal compared with patients on atorvastatin 10 mg (figure [4](#F4){ref-type="fig"}). When data from the fixed-dose period were re-analysed to the more stringent 2003 European LDL-C goal of \<2.5 mmol/L, 65% of patients receiving rosuvastatin 10 mg achieved goal compared with 33% of patients receiving atorvastatin 10 mg (p \< 0.001; figure [4](#F4){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Percentage of patients to 1998 and 2003 European LDL-C goals \[7,9\] at 4 weeks**. \*p \< 0.001 vs atorvastatin **LDL-C**: Low-density lipoprotein cholesterol ::: ![](1475-2840-4-7-4) ::: Rosuvastatin also reduced TC, non-HDL-C, LDL-C/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio significantly (p \< 0.0001) more than atorvastatin after 4 weeks of treatment (table [3](#T3){ref-type="table"}). Both treatments increased HDL-C and decreased TG from baseline to 4 weeks, but there were no statistically significant differences between the groups (table [3](#T3){ref-type="table"}). In addition, rosuvastatin significantly reduced levels of apo B and the apo B/apo A-I ratio, and increased apo A-I levels compared with atorvastatin (p ≤ 0.05) (table [3](#T3){ref-type="table"}). Similar significant effects on TC, non-HDL-C, apolipoproteins and lipid ratios were observed at 16 weeks (table [2](#T2){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Percentage change from baseline in lipid variables at 4 weeks (ITT population) ::: **Variable** **Least-squares mean percentage change from baseline to 4 weeks** **Difference (95% CI)** **p-value** ----------------------- ------------------------------------------------------------------- ------------------------- -------------------- ----------- LDL-C -47.6 -38.5^a^ -9.1 (-11.4, -6.7) \< 0.0001 TC -33.6 -27.9 -5.7 (-7.4, -4.0) \< 0.0001 HDL-C 4.4 2.6 1.8 (-0.5, 4.0) NS TG -19.2 -15.5 -3.7 (-9.5, 2.2) NS Non-HDL-C -42.6 -35.0 -7.6 (-9.6, -5.7) \< 0.0001 LDL-C/HDL-C ratio -49.3 -39.5 -9.8 (-12.2, -7.4) \< 0.0001 Non-HDL-C/HDL-C ratio -44.4 -35.9 -8.5 (-10.8, -6.2) \< 0.0001 TC/HDL-C ratio -35.8 -29.0 -6.7 (-8.7, -4.8) \< 0.0001 Apo B -42.9 -35.3 -7.6 (-9.7, -5.6) \< 0.0001 Apo A-I 2.6 0.8 1.8 (0, 3.5) 0.05 Apo B/apo A-I ratio -43.9 -35.4 -8.5 (-10.6, -6.4) \< 0.0001 **ITT:**Intention to treat, **CI:**Confidence interval, **LDL-C:**Low-density lipoprotein cholesterol, **TC:**Total cholesterol, **HDL-C:**High-density lipoprotein cholesterol, **TG:**Triglycerides, **Apo:**Apolipoprotein, **NS:**Not statistically significant ^a^n = 232 ::: There was no statistically significant difference in UAE rate from baseline to study end, or between the treatment groups, including those patients with baseline microalbuminuria (UAE \>20 μg/min). Safety ------ Both treatments were well tolerated, with overall incidences of adverse events being similar between the treatment groups (51% with rosuvastatin, 53% with atorvastatin). A total of 10 patients experienced serious adverse events (two in the rosuvastatin group, eight in the atorvastatin group), none of which were considered by the investigator to be related to study treatment. Ten patients discontinued because of adverse events, three in the rosuvastatin group and seven in the atorvastatin group. There were no cases of myopathy. Myalgia was reported by 3.4% of the patients in the study; none of the cases were associated with a clinically important elevation in CK (\>5 × ULN). Indeed, there were no clinically important elevations in CK in either group throughout the study and changes in CK were not related to dose of study medication or duration of treatment. The most frequent adverse events overall were nasopharyngitis, myalgia, and inadequately controlled diabetes mellitus (table [4](#T4){ref-type="table"}). There were no clinically relevant changes in ALAT or ASAT (\>3 × ULN). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Adverse events occurring in ≥ 3% of patients in any treatment group ::: **Number (%) of patients with adverse event** ------------------------------------------- ----------------------------------------------- ---------------------------- **Adverse event** **Rosuvastatin (n = 233)** **Atorvastatin (n = 236)** Nasopharyngitis 23 (9.9) 19 (8.1) Myalgia 13 (5.6) 7 (3.0) Inadequately controlled diabetes mellitus 14 (6.0) 11 (4.6) Constipation 9 (3.9) 6 (2.5) Headache 6 (2.6) 7 (3.0) Urinary tract infection 9 (3.9) 2 (0.9) Arthralgia 1 (0.4) 9 (3.8) ::: Discussion ========== Rosuvastatin provided significantly greater LDL-C reductions than atorvastatin, both at the initial dose of 10 mg and also when titrated over the dose range of 10--40 mg for rosuvastatin and 10--80 mg for atorvastatin. In addition, a significantly higher percentage of patients treated with rosuvastatin achieved the 1998 European LDL-C goal (\<3.0 mmol/L), both with the starting dose of 10 mg and after the period of dose titration. These results are consistent with the findings of studies in patients with hypercholesterolaemia that compared rosuvastatin with atorvastatin over 6 weeks \[[@B25]\], 8 weeks \[[@B31]\], 12 weeks \[[@B27],[@B28]\], and 52-week dose titration \[[@B28]\]. Patients with type 2 diabetes commonly have a highly atherogenic lipid profile including elevated LDL-C, increased TG and low HDL-C, which is associated with a high risk of developing cardiovascular disease \[[@B32],[@B33]\]. Statins are recognized as first-line therapy for cholesterol lowering, and their benefits have been shown to extend to patients with diabetes \[[@B18]-[@B22]\]. The present study was designed to reflect the treatment of diabetic dyslipidaemia in the clinical setting, in that statin treatment of patients was titrated upwards from the recommended starting dose to that required to achieve the 1998 European target of LDL-C \<3.0 mmol/L. The population of the present study was compared with patients with type 2 diabetes in the Swedish National Diabetes Registry and was found to be consistent in terms of baseline characteristics such as age, HbA~1c~, body mass index, percentage of smokers, blood pressure, and antidiabetic medication \[[@B34]\]. New European guidelines published in 2003 recommend a more stringent target (LDL-C \<2.5 mmol/L) \[[@B9]\] than that used when the present study was planned. Further analysis of the 4-week (fixed-dose) LDL-C data indicated that rosuvastatin 10 mg treated significantly more patients to the new 2003 European goal of \<2.5 mmol/L than atorvastatin 10 mg. As expected, the absolute percentages of patients achieving the more stringent 2003 goal were lower than the absolute percentages achieving the 1998 goal at 4 weeks, but the greater efficacy of rosuvastatin 10 mg compared with atorvastatin 10 mg remained the same. As more clinical trial evidence becomes available regarding the positive effects of intensive lipid lowering among patients with diabetes, it is likely that even more stringent LDL-C goals will be recommended. Indeed, National Cholesterol Education Program Adult Treatment Panel III recommendations were recently reviewed and a target of LDL-C \<70 mg/dL (1.8 mmol/L) was suggested as a therapeutic option for individuals considered to be at very high risk including those with both type 2 diabetes and established cardiovascular disease \[[@B35]\]. The availability of an agent that enables a large number of patients to achieve LDL-C goal at the starting dose is important given that many patients receiving lipid-lowering therapy fail to existing attain lipid targets due to a lack of dose titration and the use of less effective agents \[[@B36],[@B37]\]. A recent observational study, designed to reflect dyslipidaemia treatment in the clinical setting, evaluated the number of hyperlipidaemic patients with coronary heart disease or diabetes who achieved LDL-C goal with their initial statin dose and whether patients were dose titrated \[[@B38]\]. Less than half (48%) achieved LDL-C \<2.6 mmol/L with their initial dose and, of those who did not achieve goal, only 45% had their dose titrated. Dose titration increases costs and the need for follow-up, which, while necessary, can be time-consuming and inconvenient. The ability of rosuvastatin to enable greater proportions of patients to achieve LDL-C goal, with reduced requirement for dose titration is highly advantageous. The benefits of reaching treatment goals have been demonstrated in the Steno-2 study \[[@B39]\], in which patients with type 2 diabetes were randomised to receive conventional treatment or intensive multifactorial intervention to strict treatment goals (including TC\<4.5 mmol/L). LDL-C levels were reduced by 47% in those receiving intensive therapy, and the risk of both cardiovascular and microvascular events was reduced by approximately 50% compared with conventional treatment \[[@B39]\]. Rosuvastatin was also more effective than atorvastatin in reducing a range of other lipid variables, including TC, non-HDL-C, LDL-C/HDL-C ratio, non-HDL-C/HDL-C ratio, and TC/HDL-C ratio. Reductions in TC and/or TC/HDL-C are particularly relevant given that the Systemic Coronary Risk Evaluation (SCORE) system advocated in the new European guidelines uses these variables to estimate total risk \[[@B9]\]. In the present study, both treatments produced similar increases in HDL-C, which were lower than those observed previously. In the Measuring Effective Reductions in Cholesterol Using Rosuvastatin therapY (MERCURY I) study involving 3,161 patients with hypercholesterolaemia, 8 weeks\' treatment with rosuvastatin 10 mg increased HDL-C by 9.2% and this was significantly greater than atorvastatin 10 mg (6.8%) and atorvastatin 20 mg (5.7%) (p \< 0.01) \[[@B31]\]. In the current study, average baseline HDL-C values were higher than would be expected in the diabetic population, which could partly explain the relatively small increases in HDL-C compared with other studies \[[@B31],[@B40]\]. In the present study, the main apolipoprotein in HDL-C, apo A-I, was significantly increased by rosuvastatin compared with atorvastatin (p \< 0.05). In addition, significant reductions were also observed in apo B and apo B/apo A-I (p \< 0.0001). Changes in apolipoprotein levels may have important implications in the reduction of cardiovascular risk, since results from the Apolipoprotein-related Mortality Risk (AMORIS) study indicate that apo B, apo A-I and apo B/apo A-I are powerful predictors of cardiac events \[[@B41]\]. Taken together with the other changes to lipid variables, the findings of the present study indicate that a less atherogenic lipid profile was achieved with rosuvastatin. Treatment of diabetic dyslipidaemia may also reduce the incidence of microvascular disease including nephropathy \[[@B33]\]. Statins have been shown to have beneficial effects in diabetic nephropathy by reducing the rate of UAE \[[@B42],[@B43]\]. In the present study, statin treatment did not significantly alter the rate of UAE; however, this may reflect the fact that the treatment period was relatively short. Previously, a reduced UAE rate with statin therapy has been observed after at least 6 months\' treatment \[[@B42],[@B43]\]. Both treatments were similarly well tolerated, with no unexpected safety concerns, and tolerability was similar to that previously observed in non-diabetic patient populations \[[@B40],[@B44]\]. In conclusion, rosuvastatin was significantly more effective at reducing LDL-C and achieving European LDL-C goals both during the fixed-dose period and following dose titration than atorvastatin in patients with type 2 diabetes. Competing interests =================== This study was supported by AstraZeneca, Södertälje, Sweden, which provided study expenses and covered the article processing charge. Annica Siewert-Delle is an employee of AstraZeneca. Authors\' contributions ======================= Christian Berne and Annica Siewert-Delle participated in the design and coordination of the study and prepared the manuscript. Both authors read and approved the final manuscript. Acknowledgements ================ We gratefully acknowledge the investigators, their co-investigators and study co-ordinators, and the patients who participated in this trial. In addition to the authors of this publication, the following investigators participated in this trial: R Ekesbo, Dalby; L Fröberg, Höganäs; N Henningsen, Malmö; G Ilestam, Malmö; A Fahlbom, Malmö; J Nielsen, Skivarp; S Nordström, Knislinge; N Nörgaard, Höganäs; G Vatnaland, Ängelholm; A-C Knutsson, Ängelholm; J Wiuff, Oskarshamn; C Sjödin, Växjö; T Svensson, Växjö; G Tygesen, Lagan; O Benéus, Partille; L Benéus, Partille; L Lingetun, Falkenberg; I Wallin, Mölndal; J Alvång, Trollhättan; E Angesjö, Borås; P Hellke, Göteborg; L Nord, Stenungsund; C Andersson, Stenungsund; U Thorslund, Göteborg; M Öhberg, Skene; P E:son Jennersjö, Linköping; O Borgholst, Kungsör; B Cöster, Kristinehamn; G Holmberg, Karlstad; W Meyer, Köping; B Finger, Köping; R Baylis, Köping; E Sundequist-Stockhaus, Karlstad; P Sundin, Örebro; K Vetterskog, Västerås; J-E Andersson, Tyresö; G Widerström, Tyresö; A Bröijersen, Stockholm; B Thorsson, Stockholm; B Eriksson, Gustavsberg; H Noppa, Spånga; A Häggmark, Skärholmen; L Held, Skärholmen; H Salminen, Bagarmossen AB; P Nordström, Bagarmossen AB; R Zlatewa-Cuenca, Stockholm; L Hjelmaeus, Stockholm; E Hammarström, Stockholm; K Brismar, Stockholm; I Bäckström, Märsta; S Hellerstedt, Kungsbacka; L Håkansson, Norrtälje; A Lindh, Åkersberga; R-M Brinkeborn, Uppsala; T Lundmark, Kilafors; U Sundström, Kilafors; R Malmström, Sandviken; E Tönnesen, Uppsala; E Edén, Uppsala; K Åresund, Gävle; O Berglund, Umeå; K Henriksson, Krokom; L Lönneborg, Sundsbruk; M Mullaart, Sundsbruk; P Malm, Östersund; G Strömberg, Trehörningsjö; T Lindén, Västra Frölunda.

PubMed Central

2024-06-05T03:55:58.155173

2005-6-3

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166567/", "journal": "Cardiovasc Diabetol. 2005 Jun 3; 4:7", "authors": [ { "first": "Christian", "last": "Berne" }, { "first": "Annica", "last": "Siewert-Delle" } ] }

PMC1166568

The unbearable lightness of being \[a cardiac imaging specialist\] ================================================================== A Renaissance of cardiac imaging occurred in the 1980s \[[@B1]\]. New technologies allowed the non-invasive description of cardiac function, perfusion, and metabolism in a polychrome, three-dimensional, overwhelming fashion. Almost unlimited resources were devoted to patient care in the economic framework of the affluent society. At the beginning of the 1990s, The Renaissance made its transition into the splendid decadence of the Baroque. The increasing technological burden in clinical cardiology paradoxically did not bring a parallel increase in the quality of care but rather an increase in cost. The economic climate had changed; the illusion of unlimited economic resources had come to an end \[[@B2]\]. Keeping in mind that each test represents a cost, often a risk, and always a diagnostic hypothesis, we can agree that every unnecessary and unjustifiable test is one test too many. Small individual costs, risks, and wastes multiplied by billions of examinations per year represent an important population \[[@B3]\], society \[[@B4]\] and environmental \[[@B5]\] burden. Unfortunately, the appropriateness of cardiac imaging is usually extra-ordinarily low and there is little awareness among patients and physicians of the elementary physical basis, differential costs, radiological doses, and long term risks of different imaging modalities \[[@B6]\]. It is also well known that -- in the words of Bernard Lown -- \"*technology in medicine is frequently untested scientifically, often applied without data relating to cost benefit, and driven by market forces rather than by patient needs*.\" Bernard Lown, 2004 \[[@B7]\]. The cost of cardiac imaging =========================== *\"Ten years ago, medical imaging wasn\'t even in the radar screen for most health insurers. In 2004, it\' s one of the highest cost items in a health plan\'s medical budget, and also one of the fastest growing\"*. (Atlantic info service newsletter, 2004) \[[@B8]\]. As an example, in U.S. during the year 2002, 7.8 million cardiac perfusion scans were performed, with a growth of 40% in the last 3 years \[[@B9]\]. Still in U.S., about 10 million CT scans were done in 1993, and about 60 millions in 2001 \[[@B10]\]. Booz Allen Hamilton projects that spending on diagnostic imaging could grow 28 % by 2005, with utilization growing by 9% per year. No doubt that any assessment of costs of medical imaging should also include the often unquestionable downstream benefits, leading to a reduction of overall costs due to more disabling disease prevented. For instance, ultrasound screening for abdominal aneurysms can reduce the risk of death by more than 50 % in men aged 65--74. When projected to 10 years, that screening costs 13.000 dollars per life year gained -- a highly favourable result in terms of cost-effectiveness \[[@B11]\]. However, it is beyond question that the explosion in imaging costs is also driven by a number of factors which not always improve the cost-effectiveness. High patient demand for the newest diagnostic tests, physicians eager to use the most effective technologies, and possibly even some providers who may boost utilization to help pay off investments in high-tech equipment. All too often physicians may not have adequate controls on the number of high cost imaging tests they are ordering. For a resting cardiac imaging test, being the average cost (not charges) of an echocardiogram equal to 1 (as a cost comparator), the cost of a CT is 3.1x, of a SPECT 3.27x, of a Cardiovascular Magnetic Resonance imaging 5.51x, of a PET 14.03x, and of a right and left heart catheterization 19.96x (Fig [1](#F1){ref-type="fig"}) \[[@B12]\]. For stress cardiac imaging, compared with the treadmill exercise test equal to 1 \[as a cost comparator\], the cost of stress echocardiography is 2.1x and of a stress SPECT scintigraphy is 5.7x \[[@B13]\]. Obviously, the evaluation of costs of medical imaging should also imply the entire clinical context (pre-test likelihood of disease, basic risk profile, target of stress testing, socio-economic characteristics of the health care system, etc.) to be taken into account. Some measures of precision and/or dispersion of each modality as compared to echocardiography may provide a more comprehensive information in the specific health care milieu where these tests are actually performed. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The average costs of CMR and other common cardiac imaging procedures when compared 2 D echocardiography (modified from Pennell, ref. 12). ::: ![](1476-7120-3-13-1) ::: The individual and social risks of imaging ========================================== Current protection standard and practices are based on the premise that any ionising radiation dose, no matter how small, can result in detrimental health effects \[[@B14]\]. These include long-term development of cancer and genetic damage \[[@B15]\]. For the purposes of radiation protection, the dose-response curve for radiation-induced cancer is assumed to be linear at low doses, with no minimum threshold \[[@B16]\]: (Fig. [2](#F2){ref-type="fig"}). The dose of 50 chest X rays (for example, a lung scintigraphy) corresponds to an extra-risk of cancer of about 1 in 20,000 exposed patients. The dose of 500 chest x rays (such as technetium sestamibi scan) corresponds to an extra-risk of about 1 in 2000 exposed patients. The dose of 1 000 chest x rays (associated with a Thallium scan) corresponds to an extra risk of cancer of about 1 in 1 000 exposed patients \[[@B17]\]. The radiation dose and risk associated with some common imaging examinations are expressed in Table [1](#T1){ref-type="table"} as equivalent dose of natural yearly background radiation, extra-risk of fatal cancer in the lifetime and lost life expectancy per exam \[[@B18]\]. Presented data refer to the best available estimates from the radiological Commission of Radiation Protection and conform to their suggested standards for communicating risk to patients. These estimations are a benchmark for the physicians and are incorporated in the European Commission guidelines for medical imaging. These small individual risks multiplied by billion examinations become significant population risk. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Presentation of cancer risk and radiation dose (in multiples of dose from a simple chest x rays) for some common radiological and nuclear medicine examinations (Modified from Picano E, ref. 37) ::: ![](1476-7120-3-13-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Radiation doses and estimated cancer risk from common radiological examinations and isotope scans ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Type of test** **Effective radiation dose (mSv)** **Equivalent period of natural background radiation** **Lifetime additional risk of cancer/examination** **Lost life expectancy** **Equivalent n. of chest x-rays** --------------------- ------------------------------------ ------------------------------------------------------- ---------------------------------------------------- -------------------------- ----------------------------------- Chest radiograph 0.01 A few days Negligible risk 2 minutes 1 Skull radiograph 0.1 A few weeks Minimal risk (1 in 100,000 to 1 in1,000,000) 20 minutes 5 Lung isotope scan 1 A few months to a year Very low risk (1 in 10,000 to 1 in 100,000) 3 hrs 50 Cardiac gated study 10 A few years\ Low risk\ 2 days 500 (4 years) (1 in 2,000) Thallium scan 20 (8 years) (1 in 1,000) 4 days 1000 -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: Use of radiation for medical examinations and test is the largest manmade source of radiation exposure. The medical sources of radiation were about one fifth of the natural radiation in 1987 \[[@B18]\], close to one-half in 1993 \[[@B19]\], and almost 100% of natural radiation in 1997 \[[@B20]\], and the use of procedures with a high load of radiation continues to grow steadily \[[@B21]\]. This impressive amount of ionizing friendly fire translates into a significant population risk. Lifetime risk of developing cancer attributable to diagnostic X-rays is 0.6--3.2% in developed countries \[[@B22]\]. The numbers are striking, but underestimate the biological population burden of medical radiation for three reasons. First, they refer to the radiological volume of 10 years ago -- substantially lower than the current radiological volume. Second, they do not consider the practice of nuclear medicine, which adds a further 10 % to the global radiation burden. Third, in addition to the risk of cancer, one should consider the burden of teratogenesis. The risk of radiation -- induced damage passed onto the offspring is estimated to represent a fifth of the risk of fatal cancer \[[@B23]\]. Unawareness in imaging ====================== Several recent studies clearly prove that not only general practitioners but also cardiologists, orthopaedics and even radiologists and nuclear physicians usually ignore the dose and the risk of what they do: and the more they do, the more they tend to ignore \[[@B24]-[@B28]\]. Radiologists working in an academic US environment frequently underestimate of 100 to 500 times the dose of a common CT and 97 % of UK doctors underestimate of sixteen times the dose of a common CT chest scan. In 1 case out of 10, doctors believe that magnetic resonance employs ionising radiations, and 1 case out of 20 that ultrasound employs ionising radiation \[[@B25],[@B27]\]. The majority of doctors, and even of radiologists, is not aware of the oncogenic risk of common, high dose radiological examinations \[[@B26],[@B28]\]. The vast majority of cardiologists underestimates of 200 up to 1000 times the dose of a stress cardiac perfusion scintigraphy \[[@B28]\]: Figure [3](#F3){ref-type="fig"}. This unawareness generates inappropriateness, which is an endemic and pervasive disease in the world of cardiac imaging. We need more prudence, wisdom and responsibility in indicating and performing (cardiac) imaging tests. Prudence and responsibility should obviously be especially high for more expensive tests, greater for those exposing the patient to significant ionizing radiation \[[@B29]\], and greatest in special subset particularly vulnerable to the damaging effects of ionizing radiation, such as women in reproductive age \[[@B30]\] and children \[[@B31]-[@B33]\]. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### What cardiologists know about dose of a test they prescribe and/ or perform daily. 2 out of 3 physicians underestimate of 100 to 500 times the dose of a cardiac scintigraphy. The dose is equivalent to 500 chest x rays with technetium sestamibi, and to more than 1000 chest x rays with thallium scan (modified from Correia et al, ref 28) ::: ![](1476-7120-3-13-3) ::: From benefit to risk-benefit ============================ According to the International Commission of Radiological Protection \"Medical exposure is the only category in which large reductions in average dose are possible, and it is therefore highly desirable to reduce applications of medical radiation which are of no benefit to the patients and to minimise useless radiation in the course of medical examinations\" \[[@B34]\] In other words it is desirable to adopt a radiation sparing strategy not only for the physician and the patients but also for the society and the environment. Nuclear medicine and X-ray procedures are intended, as stated by the United Nations Annex on Medical Exposures, \"to provide doctors with diagnostic information and in principle conducted with the lowest practicable levels of patient dose to meet clinical objectives\" \[[@B5]\]. The clinical counterpart of this concept is that it is not enough that a test is marginally \"better\" than the other to justify its use: the extra-value should be proportional to the extra-cost and to the extra-risk. Both the physician and the patient should be well aware of the different individual risks and social costs posed by different diagnostic options. As recently stated by guidelines on fluoroscopically guided invasive cardiovascular procedures, \" the core principle governing the use of ionizing radiation is ALARA (as low as reasonably achievable). The ALARA principle recognizes that there is no magnitude of radiation exposure that is known to be completely safe. This principle confers a responsibility on all physicians to minimize the radiation injury hazard to their patients, to their professional staff, and to themselves\" \[[@B35]\]. If the information is comparable, every effort should be done to orient the patient towards non-ionizing testing. If it is true that \"local expertise and availability should guide the selection of imaging techniques\" \[[@B36]\], it is also true that the doses and risks associated with the different diagnostic options should be clearly spelled out to allow the patient and the prescriber to make an informed decision. This policy is encouraged by common sense \[[@B37]\], deontologic code \[[@B38]\], European Commission imaging guidelines \[[@B14]\] and the European law \[[@B39]\]. It propels to use a green technique whenever the information supplied is grossly comparable to the red one \[[@B40]\]. The basic concept underlying this cardiac imaging paradigm shift is obvious but is presently neglected. Medical images of heartbreaking beauty when considering the benefit, can be ambiguous when considering the cost-benefit relationship, and unacceptable when considering the risk-benefit. In this transition, we as physicians and imaging specialists, are the critical link. *\"Health professionals involved in the processes of diagnosis and treatment are the critical link. Training them properly and ensuring intensive information exchange among them are, therefore, probably the most cost-effective ways of achieving patient safety*\" \[[@B4]\]. (International Action Plan for the radiological Protection of patients).

PubMed Central

2024-06-05T03:55:58.157715

2005-5-25

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166568/", "journal": "Cardiovasc Ultrasound. 2005 May 25; 3:13", "authors": [ { "first": "Eugenio", "last": "Picano" } ] }

PMC1166569

Exercise training is an effective intervention that can partially restore the impaired aerobic capacity and musculoskeletal fitness (i.e. muscle strength) found in cardiac transplant recipients \[[@B1],[@B2]\]. However, previous reports have focused exclusively on the effects of exercise training in men. Therefore, the effect of exercise training on these outcomes in female cardiac transplant recipients is not known \[[@B2]-[@B7]\]. Importantly, a majority of female cardiac transplant recipients do not engage in regular physical activity leading to increased levels of fatigue, poor functional status and reduced exercise capacity \[[@B8]-[@B10]\]. Based on this rationale, the aim of this study is to examine the effect that exercise training has on improving aerobic endurance (i.e. distance walked in six-minutes) and lower extremity muscle strength in female cardiac transplant recipients. We hypothesized that exercise training would be a feasible and effective intervention to improve aerobic endurance and lower extremity strength in female cardiac transplant recipients. Methods ======= Subjects and procedures ----------------------- The participants for this study consisted of 20 (51 ± 11 years) clinically stable female cardiac transplant recipients who participated in the University of Alberta Post-Transplant Exercise Rehabilitation program between 1997 and 2003. All assessments and exercise training were performed in the Physical Therapy Department at the University of Alberta Hospital. Ethics approval for this study was obtained from the Biomedical Ethics Board at our University. Outcome Assessments ------------------- The six-minute walk test was performed in accordance with the American Thoracic Society guidelines \[[@B11]\]. In addition, the six-minute walk scores were compared with age-matched norms for healthy females published by Gibbons and associates \[[@B12]\]. Leg-press maximal strength testing was performed on a commercially available leg-press machine with the greatest weight lifted while adhering to strict technique being used as the maximal score. All assessments were repeated following the 12-week training program. Exercise Training Intervention ------------------------------ Exercise training consisted of supervised aerobic (cycling and/or treadmill exercise at an intensity between 12 to 14 on the BORG perceived exertion scale for 30 to 40 minutes/day including warm-up and cool-down, 5 days/week) and lower extremity strength training. Data analysis ------------- Statistical analysis was performed with a one-way analysis of variance. The alpha level was set \"a priori\" at p \< 0.05. Data are presented as mean ± SD. Results ======= Baseline testing was performed 37 ± 27 days after cardiac transplantation. Aerobic Endurance ----------------- At baseline, our participants\' aerobic endurance was 50% lower than age-matched predicted values (Figure [1](#F1){ref-type="fig"}). Twelve weeks of training resulted in a significant increase in aerobic endurance, however, it remained 22% lower than age-predicted values (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Effect of exercise training on the distance walked in six-minutes. \*, p \< 0.05 vs. pre-training; ^†^, p \< 0.05 vs. post-training. ::: ![](1468-6708-6-10-1) ::: Leg Press --------- Leg-press maximal strength increased by 64% after three months of training (pre: 48 ± 16 kg vs. post: 78 ± 27 kg, p \< 0.05). Discussion ========== This is the first study to examine the effect that combined aerobic and strength training have on improving aerobic endurance and musculoskeletal fitness in female cardiac transplant recipients. The main finding of this study is that combined aerobic and strength training is a feasible and effective intervention to partially restore female cardiac transplant recipients\' aerobic endurance and leg-press strength. Cole et al. \[[@B9]\] recently found that female cardiac transplant recipients\' aerobic capacity was 60% lower than age-predicted values. Consistent with this finding, our transplant recipients\' baseline aerobic endurance was 50% lower than age predicted values. Moreover, our participants pre-training leg-press strength was 36% lower than that found in age-matched male cardiac transplant recipients tested in our laboratory \[[@B13]\]. The mechanisms responsible for the impaired cardiovascular and musculoskeletal fitness are likely secondary to abnormalities in cardiac and skeletal muscle function associated with pre-transplant heart failure, post-transplant deconditioning, cardiac denervation or immunosuppressuion therapy \[[@B14]\]. Several research groups \[[@B2],[@B3],[@B13]\] have demonstrated that exercise training initiated in the early post-operative period is associated with an increase aerobic endurance \[[@B13]\], muscle mass \[[@B2]\], muscle strength \[[@B2],[@B13]\] and bone density \[[@B3]\] in male cardiac transplant recipients. This study extends previous investigations by demonstrating that 12 weeks of combined aerobic and strength training are associated with a significant and marked improvement in aerobic endurance and muscle strength in female cardiac transplant recipients. The mechanisms responsible for the improvement in aerobic endurance was not examined in this study, however, they may be due to favorable improvements in mitochondrial oxidative properties \[[@B15],[@B16]\] that increase arteriovenous oxygen difference during exertion as aerobic training does not alter exercise cardiac output in this population \[[@B4],[@B17]\]. The training mediated increase in leg-press strength that we found is likely secondary to the increase in muscle mass that occurs with aerobic \[[@B4]\] or combined aerobic and strength training \[[@B2]\]. The consequence of our training mediated improvement in cardiorespiratory and musculoskeletal fitness is that it may result in a favorable improvement in mortality. Specifically, Kavanagh et al. \[[@B18]\] reported that cardiac transplant recipients with the greatest training-mediated improvement in aerobic capacity and lean body mass had a lower mortality rate 12 years after cessation of the training program. A limitation of our investigation is that we did not have a non-exercise control group. However, our cardiac transplant recipients are required to participate in a supervised 12-week exercise program beginning as an inpatient and completed as an outpatient. Despite this limitation, the improvement in aerobic endurance and leg-press strength associated with our training program is similar to that found in male cardiac transplant recipients who participated in our outpatient exercise rehabilitation program \[[@B13]\]. Summary ======= A majority of female cardiac transplant recipients adhere to a sedentary lifestyle and as result their aerobic capacity is 60% lower than age-predicted values \[[@B9]\]. The effect that exercise training has on improving female cardiac transplant recipients\' aerobic endurance and lower extremity strength is not known. The primary finding of this investigation is that 12 weeks of aerobic and strength training is an effective intervention that can improve aerobic endurance and musculoskeletal fitness in recent female cardiac transplant recipients. Moreover, the improvement in aerobic endurance and leg-press strength is similar to that found in male cardiac transplant recipients after combined aerobic and strength training. In summary, female cardiac transplant recipients should be encouraged to perform aerobic and strength training to increase their aerobic endurance and musculoskeletal fitness. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MH. Conceived the study, performed data analysis and manuscript preparation. KR. Assisted with data collection and manuscript preparation. LF. Performed the exercise rehabilitation training and assisted with manuscript preparation. DK. Assisted with manuscript preparation. DW. Assisted with manuscript preparation. LJ. Assisted with manuscript preparation. WT. Assisted with manuscript preparation. All authors read and approved the final manuscript.

PubMed Central

2024-06-05T03:55:58.159362

2005-5-26

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166569/", "journal": "Curr Control Trials Cardiovasc Med. 2005 May 26; 6(1):10", "authors": [ { "first": "Mark", "last": "Haykowsky" }, { "first": "Kenneth", "last": "Riess" }, { "first": "Linda", "last": "Figgures" }, { "first": "Daniel", "last": "Kim" }, { "first": "Darren", "last": "Warburton" }, { "first": "Lee", "last": "Jones" }, { "first": "Wayne", "last": "Tymchak" } ] }

PMC1166570

Background ========== Human skeletal muscles contain approximately 35% of total human body magnesium, which is an essential cofactor in a number of cell reactions. Magnesium ions influence the equilibria of many reactions involved in cellular bioenergetics by interacting with phosphorylated molecules and interfere with the kinetics of ion transport across plasma membranes \[[@B1]\]. There is considerable evidence that Mg^2+^is actively transported and regulated, although the mechanisms are still largely unknown \[[@B2]\]. In skeletal muscle variations of cytosolic pH, phosphocreatine (PCr) and inorganic phosphate (Pi) concentrations influence the complex multi-equilibrium system of the molecular species binding magnesium ions. As a consequence \[Mg^2+^\] changes considerably in different metabolic conditions such as rest, exercise and recovery, showing an increase matched by a decrease of intracellular pH during exercise and recovery \[[@B3]\]. We assessed the cytosolic pH and the \[Mg^2+^\] by ^31^P MRS at rest, during exercise and post-exercise recovery in the calf muscle of two patients with McArdle\'s disease with muscle glycogen phosphorylase deficiency (McArdle), and two brothers affected by Tarui\'s disease with muscle phosphofructokinase deficiency (PFK). These two type of glycogenosis, being characterized by almost absent activity of enzymes involved in glycogenolysis (McArdle) and glycolysis (PFK) pathways, show in general limited/absent production of intracellular lactic acid, depending on the degree of enzyme deficit \[[@B4],[@B5]\]. As consequence, patients with McArdle\'s and Tarui\'s disease, typically show a decrease or a lack of intracellular acidification during muscle exercise when studied by ^31^P MRS \[[@B6],[@B7]\]. We used these diseases as natural experimental models to study the pattern of free Mg^2+^during exercise and recovery in the absence of intracellular acidification to understand to what extent homeostasis of intracellular free Mg^2+^is linked to pH. Methods ======= Patients -------- We studied 4 patients: two unrelated males both aged 42, with myo-phosphorylase deficiency (named MCArdle I and II respectively) and two brothers aged 18 and 10 years with phosphofructokinase deficiency (named PFK I and II respectively), as detected by histochemical/biochemical analysis of muscle. Ten healthy volunteers (10 males age: 33 ± 15) were recruited as control subjects. Written informed consent was obtained from all subjects. Protocol -------- MR spectra were acquired on a General Electric 1.5 T Signa System whole-body scanner. Radiofrequency pulses at 25.866 MHz with a pulse width of 400 μs and a transmitter power of 0.5 kW were transmitted by a surface coil (20.5 cm diameter; General Electrics) and the resonance signals were collected by a 7.5 cm receiving coil. A data table of 1024 complex points was collected for each FID. The band width was 2 kHz. The delay between transmission and reception was 0.5 ms and the dwell time was 250 μs. The stimulation-response sequence was repeated every 5000 ms (TR = 5000 ms). Magnetic field homogeneity was optimized by shimming the ^1^H water spectrum (FWMH 0.25--0.35 ppm) The spectroscopic measurements were performed according to the quantification and quality assessment protocols defined by the EEC Concerted Research Project on \"Tissue Characterisation by MRS and MRI\", COMAC-BME II.1.3 \[[@B8]\]. Subjects lay supine with a 20.5/7.5 cm diameter transmitter/receiver surface coil centred on the maximal circumference of the right calf muscle. Muscle aerobic incremental exercise consisted of different levels of 1 minute each (12-FIDs) of plantar flexion against a pedal using a pneumatic ergometer \[[@B9]\]. All patients were asked to perform an exercise to reach a PCr depletion of about 50% at the end of exercise. Sixty-four FIDs at rest, and 12 FIDs for each level of work were averaged. During recovery 4-FIDs data blocks (20 s) were recorded for 60 s, while longer time blocks were collected thereafter. The area of each metabolite signal was fitted to a Lorentzian line shape using a time-domain fitting program AMARES/JMRUI\[[@B10]\], the PCr and Pi concentration were calculated by assuming a normal ATP concentration of 8 mM \[[@B11]\]. The cytosolic pH and \[Mg^2+^\] are calculated from the chemical shift of Pi and β-ATP respectively, both measured from the resonance of PCr, using an equation which takes into account the mutual influence between pH and \[Mg^2+^\] \[[@B3]\]. The simultaneous calculation of \[Mg^2+^\] and pH was performed by the specific software package MagicMC, that we developed and made available on the internet \[[@B12]\]. Results ======= ^31^P MRS spectra of human skeletal muscle typically show the peak signal of: phosphomonoesters (PME) which represents the phosphorylated monosaccharide intermediates of glycogenolysis, inorganic phosphate (Pi), phosphocreatine (PCr) and the three phosphate groups α, β, γ of ATP. Figure [1](#F1){ref-type="fig"} shows ^31^P MRS spectra acquired at the end of exercise in the calf muscle of McArdle and PFK patients compared to that of a control subject reaching a similar level of PCr depletion. End exercise spectra of both PFK patients show a marked increase in the PME peak. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **^31^P MRS spectra of calf muscle during exercise.**End-exercise ^31^P MRS spectra of calf muscle acquired in patients and in a control subject reaching a PCr depletion of about 50%. PFK patients showed a marked phosphomonoester (PME) accumulation, although to a different extent. ::: ![](1476-5918-4-7-1) ::: Table [1](#T1){ref-type="table"} reports the rest and end-exercise values of cytosolic \[Mg^2+^\] and pH in patients\' calf muscle, compared to the mean values obtained in a control group of ten subjects with a comparable end-exercise PCr depletion. At rest all patients showed a cytosolic pH not different from control values. Resting cytosolic \[Mg^2+^\] in PFK II patient was 0.45 mM, higher than the control value (0.32 + 0.04 mM) \[[@B3]\], while in PFK I and both McArdle\'s patients was normal. Both McArdle and PFK I patients reached a similar PCr depletion just above 50%, while PFK II patient stopped at a lower degree of PCr depletion. All patients displayed a lack of intracellular acidosis during exercise, showing pH values higher than controls at the end of exercise. Cytosolic free \[Mg^2+^\] at the end of exercise was lower in both PFK patients compared to control values. The variation of free \[Mg^2+^\] from rest to end-exercise (Δ\[Mg^2+^\]) was negative in both PFK patients and in McArdle II patient. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### \[Mg^2+^\] and pH values at rest and end-exercise ::: Rest End-Exercise --------------------------- --------------- --------------- ----------- --------------- --------------- ----------------- McArdle I 0.34 6.98 53.1% 0.39 7.06\* \+ 0.05 McArdle II 0.36 6.95 53.8% 0.32 7.07\* \- 0.04\* PFK I 0.38 6.95 54.1% 0.23\* 7.01\* \- 0.15\* PFK II 0.45\* 6.95 35.3% 0.23\* 7.01\* \- 0.22\* Controls Mean (*n = 10*) 0.31 6.96 47.7% 0.42 6.8 +0.11 S.D. 0.04 0.02 13.8% 0.06 0.09 0.07 control range \[min:max\] \[0.27:0.40\] \[6.94:7.00\] \[22:73\] \[0.35:0.54\] \[6.90:6.67\] \[+0.03:+0.24\] Rest and end-exercise values of cytosolic \[Mg^2+^\] and pH assessed by ^31^P MRS in patients calf muscle, compared to the mean values obtained in a control group of ten healthy subjects with a comparable end-exercise PCr depletion; Δ\[Mg^2+^\]: difference between end-exercise \[Mg^2+^\] and rest \[Mg^2+^\]; \%PCr: percentage of PCr depletion \* denotes values out of the range: control mean ± 2 SD. ::: Figure [2](#F2){ref-type="fig"} reports the patterns of cytosolic free \[Mg^2+^\] and pH obtained in patients during exercise (panel A and B) and recovery (panel C and D) compared with typical patterns from a healthy volunteer with comparable PCr depletion. During exercise and recovery the McArdle patients did not show any significant change (McArdle II) or small change (McArdle I) in free \[Mg^2+^\], while both PFK patients showed decreased free \[Mg^2+^\] during exercise. On the other hand, during recovery the pattern of free \[Mg^2+^\] was different in the two PFK patients, with PFK II showing a moderate increase during early recovery. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **\[Mg^2+^\] and pH patterns during exercise and recovery.**Patterns of cytosolic free magnesium concentration and pH at rest, during exercise and recovery in patients compared with typical patterns from a healthy volunteer with comparable PCr depletion. (A): pattern of \[Mg^2+^\] during exercise; (C): pattern of \[Mg^2+^\] during recovery; (B): pH pattern during exercise; (D): pH pattern during recovery. ::: ![](1476-5918-4-7-2) ::: Figure [3](#F3){ref-type="fig"} report the PME pattern of the two PFK patients during exercise and recovery. PFK I patient shows a slower rate of both PME accumulation and recovery compared to PFK II patient. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **PME patterns of PFK patients during exercise and recovery.**PFK I patient shows both a slower rate of PME accumulation during exercise and a slower recovery of PME after exercise compared to PFK II patient. PME signal comes from the phosphorylated monosaccharide intermediates of glycogenolysis. ::: ![](1476-5918-4-7-3) ::: Discussion ========== In the skeletal muscle variations of cytosolic pH, phosphocreatine and inorganic phosphate concentrations influence the complex multi-equilibrium system of the molecular species which bind magnesium ions. As a consequence free cytosolic \[Mg^2+^\] can change considerably in different metabolic conditions such as rest, exercise and recovery. It has been shown by ^31^P MRS that the increase of cytosolic free \[Mg^2+^\] occurring in skeletal muscle of healthy subjects during exercise and initial recovery is matched by a decrease in cytosolic pH, and the changes in cytosolic free \[Mg^2+^\] were mainly the result of the predominant effect of \[H^+^\]. \[[@B3]\]. This result was attributed to mechanisms of binding competition existing between Mg^2+^and H^+^towards the molecules negatively charged present in the cell cytosol \[[@B3]\]. However, the causal relationship between pH and \[Mg^2+^\] has not been proved yet, as it could be argued that muscular exercise *per se*elicits an increase in cytosolic free \[Mg^2+^\]. Therefore, we used patients with McArdle\'s and Tarui\'s disease as experimental models to study the pattern of \[Mg^2+^\] during exercise and recovery in the absence of intracellular acidification to understand to what extent homeostasis of intracellular free Mg^2+^is linked to pH. Due to the rare nature of these disorders we were able to enrol just two patients for both diseases and therefore we had to deal with a small sample size. The results show that the increase in cytosolic \[Mg^2+^\] occurring in skeletal muscle during exercise is actually the consequence of the increase of H^+^concentration and not of other mechanisms related to muscle contraction. In addition, we found that both PFK patients showed a reduction of \[Mg^2+^\] during exercise concomitant with the PME increase. The decrease of \[Mg^2+^\] also persisted during recovery in PFK I patient who displayed a slower PME recovery. The PME peak in the ^31^P MRS spectra corresponds to the phosphorylated monosaccharide intermediates of glycogenolysis. Therefore, due to the deficit of the phosphofructokinase activity in Tarui\'s disease, the PME accumulation shown by these patients is likely due to the increase of fructose- 6-phosphate, which represents an additional binding site for cytosolic Mg^2+^. As a consequence, we interpret the decrease of \[Mg^2+^\] concomitant with the PME increase as due to the binding of Mg^2+^to fructose-6-phosphate. A previous study (6) reported that the abnormal PME accumulation of PFK patients during exercise was accompanied by a subnormal Pi accumulation. This finding was interpreted as a result of the incorporation of free Pi into phosphorylated glycolytic intermediates. However, both our patients did not show any Pi trap into PME, since we found that the sum of PCr and Pi was constant for the whole exercise duration, while the total phosphates signal increased proportionally to PME increase. Therefore, the \[Mg^2+^\] decrease found in PFK patients cannot be ascribed to a diminished Pi build-up. Moreover, PFK II patient displayed a larger decrease of \[Mg^2+^\] from rest to end-exercise compared to PFK I patient. Accidentally, PFK II patient also had a smaller PCr breakdown, nevertheless, this cannot be the cause of the larger decrease of \[Mg^2+^\], as a lower \[PCr\] corresponds to lower calculated \[Mg^2+^\] (3). Conclusion ========== Our results show that: i\) free \[Mg^2+^\] is strongly linked to pH in skeletal muscle homeostasis as previously suggested by a study in healthy volunteers \[[@B3]\], and by computer simulation on a chemical model mimicking muscle cell cytosol \[[@B13]\]; ii\) the decrease of free \[Mg^2+^\] during exercise in both PFK patients suggests that \[Mg^2+^\] is influenced by the accumulation of fructose- 6-phosphate, an additional binding site for cytosolic Mg^2+^, as shown by the accumulation of the phosphomonoesters peak in the ^31^P MRS spectra of these patients. iii\) ^31^P MRS is a suitable tool for the *in vivo*assessment of free cytosolic \[Mg^2+^\] in human skeletal muscle during rest, exercise and recovery; Authors\' contributions ======================= EM participated in the study design, performed the post-processing and statistical analysis, RL and CT participated in the study design and coordinated the data collection, AM participated in the study design, BB participated in the coordination of the study, SI participated in the study design, coordinated the study, and drafted the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by Ricerca Fondamentale Orientata (ex quota 60%) and by Programmi di Ricerca Scientifica di Rilevante Interesse Nazionale -- Cofin (ex quota 40%).

PubMed Central

2024-06-05T03:55:58.160085

2005-6-6

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166570/", "journal": "Dyn Med. 2005 Jun 6; 4:7", "authors": [ { "first": "Emil", "last": "Malucelli" }, { "first": "Raffaele", "last": "Lodi" }, { "first": "Andrea", "last": "Martinuzzi" }, { "first": "Caterina", "last": "Tonon" }, { "first": "Bruno", "last": "Barbiroli" }, { "first": "Stefano", "last": "Iotti" } ] }

PMC1166571

Innate immune system ==================== The first line of immune defence mainly operates by detection of a broad range of molecular patterns foreign to mammalian tissues, called pathogen-associated molecular patterns that induce the activation of the innate immunity and inflammatory response \[[@B1]\]. The constitutive expression of a limited set of pattern recognition receptors by many cell types of the innate immunity does not require clonal expansion of specific cell populations. These germ cell-encoded proteins recognize microbial pathogens or ligands from damaged tissues based on shared molecular structures and induce host responses that localize the spread of infection and enhance systemic resistance to infection. Therefore, the expression of a limited number of highly active genes during the activation of innate immunity is able to induce rapid (minutes to hours) efficient defensive immune responses \[[@B2]\]. Several cell types contribute to innate immunity and the mononuclear phagocyte lineage plays a pivotal role in innate immunity. Monocytes, macrophages and their tissue-differentiated derivatives, such as microglia in the nervous system, express pattern recognition receptors, namely various scavenger and Toll-like receptors \[[@B3]\]. These receptors induce transmembrane signals that activate NF-kB and mitogen dependent protein kinase pathways \[[@B4]\]. Toll-like receptor activation also induces the expression of a wide number of genes encoding proteins, such as cytokines, with regulatory functions upon leukocyte activation and tissue inflammation \[[@B5]\]. Therefore, the capacity of each individual organism to regulate the activation of innate immunity and local inflammatory responses is crucial for initiating defensive action against pathogens, limiting tissue damage and enhancing fast recovery and tissue healing. The inflammatory response ========================= In response to cell injury elicited by trauma or infection the inflammatory response sets in, constituting a complex network of molecular and cellular interactions directed to facilitate a return to physiological homeostasis and tissue repair. The response is composed of both local events and a systemic activation mediated by cytokines. If tissue health is not restored or in response to stable low grade irritation, inflammation becomes a chronic condition that continuously damages the surrounding tissues. In fact, during chronic inflammatory immune responses, tissue injury and healing proceed simultaneously. The collateral damage caused by this type of inflammation usually accumulates slowly, sometimes asymptomatically for years but can eventually lead to severe tissue deterioration \[[@B6]\]. Cytokines are largely secreted molecules that act on the surrounding microenvironment by providing cell to cell signalling. Cytokines are components of a large, complex signalling network. The effects of cytokines on target cells may be inhibited or enhanced by other cytokines, hormones, and cytokine-receptor antagonists and circulating receptors. Tumor necrosis factor α (TNF-α), Interleukin-1 (IL-1) and IL-6 are the classical pro-inflammatory cytokines. Their ability to activate both local and systemic effects is well established. Locally, they contribute to the activation of the inflammatory cells and together with chemokines, which induce the expression of adhesion molecules, cause their local recruitment. When the causes of the inflammatory reaction are of a high intensity, the production of cytokines is increased and they are released in the circulation provoking the \"acute phase response\". On the other hand, \"inhibitory\" cytokines such as IL-10 damp down the activation of some effector functions of T lymphocytes and mononuclear phagocytes, by inhibiting the release of pro-inflammatory cytokines and therefore turning off the inflammatory processes \[[@B7],[@B8]\]. The acute phase response e.g. leukocytosis, fever, somnelence, anorexia and acute phase proteins synthesis, such as C reactive protein (CRP) in the liver, is a highly conserved inflammatory response which is rapidly activated by infections or trauma via pattern recognition molecules. Acute phase protein concentration rapidly increases after infection, and their production is controlled primarily by IL-6- and IL-1-type cytokines. The acute phase proteins provide enhanced protection against microorganisms and modify inflammatory responses by affecting cell trafficking and mediator release. The more traditional view of acute phase proteins has extended beyond that of opsonization of microorganisms. Some acute phase proteins have anti-inflammatory effects while others have important effects on leukocyte activation and trafficking \[[@B9]\]. CRP, named for its capacity to precipitate the somatic C-polysaccharide of Streptococcus pneumoniae, was the first acute-phase protein to be described, and is an exquisitely sensitive systemic marker of inflammation and tissue damage. It is a member of the pentraxin family of plasma proteins, which are part of the lectin fold superfamily of calcium-dependent ligand-binding and lectin (carbohydrate-binding) proteins. In healthy blood donors, the median concentration of CRP is 0.8 mg/l, but following an acute-phase stimulus, values may increase 10 000-fold. In an apparently healthy population the median baseline value is slightly higher and tends to increase with age with females showing slightly higher circulating concentrations. In most, but not all diseases, the circulating value of CRP reflects on-going inflammation much more accurately than do other biochemical parameters of inflammation, such as plasma viscosity or the erythrocyte sedimentation rate \[[@B10]\]. Innate immune system during ageing ================================== Age related changes in clonotypic immune system are well documented. For instance, involution of the thymus gland is an early feature of the immune reconfiguration and several age- related alterations of T or B cell compartments have been described \[[@B11]-[@B13]\]. Studies on age-associated changes of the innate immunity are not as advanced as those of the clonotypic immune system. However, investigations from aged mice showed a functional decline of monocytes and macrophages, a low expression level of Toll-like receptors from activated splenic and peritoneal macrophages and an altered secretion of several chemokines and cytokines \[[@B14],[@B15]\]. Aged macrophages also contribute to an impaired the proliferative response of activated peripheral T lymphocytes \[[@B16]\]. Aged phagocytes, such as macrophages and neutrophils, showed an impaired respiratory burst and reactive nitrogen intermediate production with a decreased ability to destroy pathogens. Moreover, aged dendritic cells were less efficient in activating both T and B cell populations and aged NK cells showed a decreased ability in killing tumour cells \[[@B17],[@B18]\]. However, not all immune activities are decreased during ageing. In fact, the \'in vitro\' production of pro-inflammatory cytokines, such as IL-1, IL-6 and TNF-α, by mitogen activated peripheral blood monuclear cells from elderly persons was higher than that from young donors \[[@B19]\]. Cytokine production may also be up-regulated \'in vivo\' in old subjects resulting in an abnormal elevation of pro-inflammatory cytokines during inflammatory responses \[[@B20]\]. An age- related increase of IL-6 levels has also been reported in plasma, serum and supernatants from healthy elderly and centenarians \[[@B19],[@B21],[@B22]\]. It is interesting to note that \'in vitro\' addition of IL-1 and TNF-α to fibroblasts induced an accelerated senescent phenotype which was rescued by anti-oxidant addition \[[@B23]\]. Inflammatory responses during ageing ==================================== A dramatic increase in mean life span and life expectancy, coupled with a significant reduction in early mortality, has lead to a large increase in the number of elderly people in modern societies. This demographic phenomenon has been paralleled by an epidemic of chronic disease usually associated with advancing age \[[@B24]\]. Most age-related diseases have complex aetiology and pathogenic mechanisms. The clinical diagnosis and therapy of these diseases requires a multidisciplinary medical approach with progressively increased costs. A body of experimental and clinical evidence suggests that the immune system is implicated, with a variable degree of importance, in almost all age-related or associated diseases. Both innate and the clonotypic immune system are usually involved in the pathogenesis of these chronic diseases \[[@B11],[@B25]\]. However, inflammatory responses appear to be the prevalent triggering mechanism driving tissue damage associated with different age-related diseases and the term \"Inflammaging\" has been coined to explain the underlining inflammatory changes common to most age-associated diseases \[[@B26]\]. This new term indicates that ageing is accompanied by an age-dependent up-regulation of the inflammatory response, due to the chronic antigenic stress which bombards the innate immune system thorough out life and potentially triggers the onset of inflammatory disease. Chronic inflammation is considered to be involved in the pathogenesis of all age-related diseases: Alzheimer\'s disease, atherosclerosis, diabetes, sarcopenia and cancer all have important inflammatory components. Inflammaging, i.e. the up-regulation of a variety of anti-stress responses at the cellular and molecular level, is the consequence of the body\'s ability to counteract and modulate the effects of a variety of stressors, which cause the accumulation of molecular and cellular scars \[[@B26],[@B27]\]. However, as recently discussed \[[@B28]\], a wide range of different aetiological factors is likely to contribute to increased low-grade inflammatory activity in elderly populations including a decreased production of sex steroids, smoking, subclinical disorders such as arteriosclerosis, asymptomatic bacteruria and a higher relative/absolute amount of fat tissue, which has been suggested to produce pro-inflammatory cytokines \[[@B29]\] (see below). Furthermore, increased levels of circulating inflammatory mediators may result from a constant, low-grade activation of cytokine-producing cells or a dysregulated cytokine response following stimulation \[[@B28]\], which does not readily become damped down. On the other hand, a recent hypothesis suggests that the reduction in lifetime exposure to infectious diseases and other sources of inflammation -- the cohort mechanism -- may also contribute to the historical decline in old-age mortality \[[@B30]\]. Recent studies have linked an individual\'s exposure to past infection to levels of chronic inflammation and to increased risk of heart attack, stroke, and cancer. For example, the risk of heart attack is correlated with serum levels of inflammatory proteins such as CRP \[[@B31]\]. Within individuals, CRP levels are also correlated with the number of seropositivities to common pathogens, suggestive of infection history \[[@B32]\]. Low-grade increases in levels of circulating TNF-α, IL-6, soluble IL-2 receptor (sIL-2R), and CRP and low levels of albumin and cholesterol, which also act as inflammatory markers, are strong predictors of all-cause mortality risk in several longitudinal studies of elderly cohorts. The effects of inflammatory mediators upon survival are independent of pre-existing morbidity and other traditional risk factors for death (smoking, blood pressure, physical exercise, total cholesterol, co-morbidity, body mass index, and intake of anti-inflammatory drugs) suggesting that cytokines may induce exaggerated pathological processes. Furthermore these molecules appear to be sensitive markers of pre-clinical disorders in elderly populations \[[@B28],[@B33]-[@B39]\]. In the following sections we will focus on the role of innate immunity and inflammation in the most common diseases of elderly in order to direct the reader\'s interest to these emerging topics in modern gerontology and geriatrics investigations. Frailty and sarcopenia ====================== Frailty has been defined as an age-related decline in lean body mass, decreased muscle strength, endurance, balance and walking performance, low activity and weight loss accompanied by a high risk of disability, falls, hospitalisation and mortality \[[@B28],[@B40]\]. It has been suggested that this syndrome reflects a metabolic imbalance caused by overproduction of catabolic cytokines and by the diminished availability or action of anabolic hormones, resulting from ageing itself and the presence of associated chronic conditions \[[@B28],[@B41]\]. Sarcopenia is obviously a central part of the frailty syndrome \[[@B28]\]. Sarcopenia is the loss of muscle mass and strength that occurs with normal ageing. In elderly, there is a decline in a variety of neural, hormonal, and environmental trophic signals to muscle. However, the most important endogenous cause may be the irreversible age-related loss of α-motor units in the central nervous system \[[@B42]-[@B44]\]. The most important environmental cause of sarcopenia is lack of physical activity which is an age-related phenomenon which shows universal decline with age in industrialized societies; obesity and type 2 diabetes being concomitant epidemics (see below). An emerging issue in the past few years for sarcopenia research has been whether there is also an increased catabolic signal, driven by systemic inflammation \[[@B44]\]. Systemic low-grade inflammation has been associated with decreased muscle mass as well as the development of functional disability in elderly populations \[[@B33],[@B44]-[@B47]\]. TNF-α has special effects that may contribute directly to sarcopenia \[[@B44]\], including increased basal energy expenditure, anorexia, loss of muscle and bone mass \'in vivo\' and association with wasting/cachexia in chronic inflammatory disorders. Consistent with these findings, muscle protein synthesis was inversely related to local levels of TNF-α protein in skeletal muscles in frail very old humans \[[@B48]\]. In a recent study of nursing home residents aged 85--96 years, systemic low-grade activation of the TNF-α system at baseline was inversely correlated to muscle strength after resistance training for 12 weeks, demonstrating that TNF-α could also be a limiting factor for training-induced improvement in muscle strength in very old people \[[@B49]\]. The role of IL-6 in sarcopenia is not clear. Epidemiological studies have reported that IL-6 is strongly associated with functional disability and loss of muscle mass but experimental studies have not been able to link IL-6 to sarcopenia \[[@B28],[@B44]\]. Human investigations have shown that inflammation impairs muscle strength in elderly people \[[@B50]\] while Insulin Growth Factor 1 and IL-6 plasma levels were synergistically related to disability and mortality in older women \[[@B51]\]. A complicating factor in untangling the cause-effect relationships underlying sarcopenia is the ability of IL-6, TNF-α, physical inactivity, abdominal obesity, and other factors to cause insulin resistance in the elderly \[[@B52]-[@B54]\]. Loss of control of the anabolic action of insulin on muscle would be another factor favouring sarcopenia, resulting in the elevation of one or more catabolic signals. Thus, the weight gain that occurs in most adults during middle age may actually predispose to sarcopenia as they age \[[@B44]\]. Obesity, the metabolic syndrome, type II diabetes ================================================= Western societies face a health care crisis of epidemic proportion, since the number of people with the metabolic syndrome is rising exponentially. This condition constitutes a major challenge for public health professionals in the field of preventive medicine and estimates suggest that more than 40 million US adults will be affected by the syndrome \[[@B55]\]. There is not known survival advantage of morbid obesity; on the other hand increased body fat is linked to high mortality \[[@B56]\]. In the past, humans have been plagued by famine when survival advantage would have been conferred by genes favouring an available energy source: the so called \'thrifty gene\' hypothesis. These genes led to genetic selection toward insulin resistance in the peripheral tissue in order to preserve glucose supply of the brain during starvation \[[@B57]\]. Human obesity shows a clear genetic component, which is usually polygenic and polymorphic variances in a number of \'thrifty\' genes could contribute to different susceptibilities to obesity and diabetes \[[@B58],[@B59]\]. Metabolic syndrome may be detected by 5 clinical diagnostic criteria 1) abdominal adiposity, 2) hypertriglyceridemia, 3) low high density lipoprotein, 4) hypertension, 5) fasting hyperglycaemia \[[@B55]\]. An involvement of inflammation with pathogenic mechanisms influencing the development of the metabolic syndrome has been suggested \[[@B55],[@B60],[@B61]\] and elevated CRP, IL-6 and TNF-α, associated with visceral adiposity, have been reported in this syndrome \[[@B61]\]. Adipocytes constitutively express the pro-inflammatory TNF-α \[[@B54]\] which decreases after weight loss \[[@B62]\]. Further work in this area has confirmed increased plasma concentrations of CRP, IL-6, and plasminogen activator inhibitor-1 (PAI-1) \[[@B63]-[@B65]\]. Therefore, a complex interplay may exist between inflammatory responses and general metabolism in atherosclerosis, cardiovascular disease, metabolic syndrome. It is becoming clear that the increased adipose tissue is not a simple reservoir for excess nutrients, but rather an active and dynamic organ capable of expressing several cytokines and other fat-derived peptides (FDP). Some FDP may have a role in the development of the metabolic syndrome but there is no evidence that these FDP are directly causing inflammation. It has been suggested that high levels of inflammatory factors are markers for obesity/abdominal obesity seen with aging, but some of these may not necessarily have a causative role in the development of inflammation \[[@B29]\]. Alternatively, a positive correlation was found between the body mass index and the percentage of resident macrophages, suggesting that fat tissue growth is associated with a recruitment of blood monocytes, responsible for cytokine production \[[@B66]\]. Insulin resistance is due to the reduced ability of peripheral tissues to properly respond to the activation induced by insulin. It is a key feature in the pathogenesis of type II diabetes and this condition may precede by 10--20 years the onset of hyperglycaemia and the clinical manifestation of the disease. Recent data suggests that a defect in insulin activation of glucose transportation in muscle cells could be induced by serine kinase cascade activation, which is down stream mediators of tissue inflammation factors \[[@B67]\]. Data from the Insulin Resistance Atherosclerosis Study showed that insulin resistance, as assessed by frequently sampled glucose tolerance tests, correlated with high blood levels of CRP, fibrinogen and PAI-1 and levels of these inflammatory factors were predictors of type II diabetes development \[[@B68]\]. Increased blood concentrations of TNF-α and IL-6 were associated with obesity and type II diabetes \[[@B69]\]. Finally, results of the population study from the European Prospective Investigation into Cancer and Nutrition Potsdam indicated a significant interaction between plasma IL-1β, IL-6 and type II diabetes development. In fact subjects with detectable IL-1β levels and increased levels of IL-6 showed an independently elevated risk of developing the disease \[[@B70]\]. Inflammation may predispose to a pre-diabetic state by increasing insulin resistance, since pre-diabetic subjects showed increased plasma levels of inflammatory proteins without primary defects of beta cell functions \[[@B71]\]. Sub-clinical inflammation was found significantly related to insulin resistance in a high risk group for diabetes, i.e. in subjects with positive family history of diabetes, obesity and hyper or dyslipoproteinemia \[[@B72]\]. Recent results from the INCHIANTI population study showed that subjects in the upper tertile of insulin resistance had increased serum levels of TNF-α, IL-1R antagonist and IL-6 and low levels of sIL-6R \[[@B73]\]. TNF-α causes an inhibition of auto-phosphorylation of tyrosine residues of the insulin receptor (IR) and an induction of serine phosphorylation of insulin receptor substrate-1, which in turn causes serine phosphorylation of the IR in adipocytes and inhibits tyrosine phosphorylation \[[@B74]\]. More recently, IL-6 has been shown to inhibit insulin signal transduction in hepatocytes \[[@B53]\]. Therefore, cytokines show relevant metabolic effects. Novel data have now appeared showing that the concomitant presence of the promoter polymorphisms of TNF-α and IL-6, linked to high production of these cytokines increases the risk of conversion to type 2 diabetes in obese subjects with impaired glucose tolerance response \[[@B75]\]. As discussed by Dandona et al. \[[@B69]\], two mechanisms might be involved in the pathogenesis of inflammation. Glucose and macronutrient intake causes oxidative stress and inflammatory changes. Chronic overnutrition (obesity) might thus be a pro-inflammatory state with oxidative stress. The increased concentrations of TNF-α and IL-6, associated with obesity and type 2 diabetes, might interfere with insulin action by suppressing insulin signal transduction, which in turn might promote inflammation. In fact, insulin reduces ROS generation by mononuclear cells, suppresses NADPH oxidase expression and intranuclear NF-kB binding, induces IkB expression and suppresses some inflammatory molecules \[[@B69],[@B76]\]. The well-known beneficial effects of caloric restriction on longevity in animal models by inducing reduced visceral fat mass might also induce a reduced secretion of multiple metabolically active factors, which are potentially responsible for the development of insulin resistance. This decrease in fat mass and its beneficial effects observed in ageing animal models might apply also to human ageing and its related pathology \[[@B77]\]. Cancer diseases =============== The majority of cancer occurs in subjects over the age of 65 years. Cancer rates increase sharply with age in both sexes: the incidence of cancer is 12--36 times higher in individuals aged 65 years or older compared with individuals aged 25--44 years, and 2--3 times more common than in persons aged 45--64 years. It is worth noting that 70% of deaths attributable to all cancer occur in men and in women aged 65 years or older whereas 35% cancer deaths in men and 46% of cancer deaths in women occur in those aged 75 years or older. The relationship between ageing and cancer is similar for most cancers, and it is well described by the multistage model. So, ageing might be considered not as a determinant of cancer per se, but as a surrogate marker of the duration of exposure to relevant carcinogenic factors \[[@B78]-[@B80]\]. Chronic inflammation induced by either biological, chemical, mechanical or physical injuries has been associated with increased incidence of cancer in different human tissues \[[@B81]\]. For instance, inflammatory bowel disease, ulcerative colitis and Crohn\'s disease are clinical conditions predisposing to cancer development of the large bowel or terminal ileum \[[@B81],[@B82]\]. Helicobacter pylori microorganisms are associated with atrophic gastritis, mucosal dysplasia, gastric adenocarcinoma and an unusual form of gastric lymphoma \[[@B81],[@B82]\]. Schistosome and trematode infections have been associated with cancers of the bladder and the biliary tracts \[[@B81]\]. The recognition of a role for inflammation in the natural history of a tumor has been known stretching from the mid-19th century \[[@B83]\]. The inflammatory microenvironment of tumors is characterised by the presence of host leucocytes both in the supporting stroma and in the tumor area \[[@B84]\]. The question whether the inflammatory infiltrate helps or hinders tumors is still open. In fact, inflammatory cells and cytokines found in tumors can contribute to tumor growth, progression, and immunosuppression as well mounting an effective host anti-tumor response. However, in most cases inflammation plays a role in the development of solid cancer. That is clearly demonstrated by a prospective, nested case-control study of a cohort of 22,887 adults followed for 11 years. A total of 172 colorectal cancer cases were identified. Up to 2 controls (n = 342) were selected for each case and matched by age, sex, race, and date of blood draw. Plasma CRP concentrations were higher among all colorectal cases combined compared with controls and the risk of colon cancer was higher in persons in the highest vs. lowest quartile of CRP \[[@B85]\]. Furthermore, in another recent study, a total of 174 patients considered to have undergone curative resection for cancer were studied. The results show that raised circulating concentrations of CRP, whether measured before or after operation, predict overall and cancer-specific survival in patients undergoing potentially curative surgery for colorectal cancer suggesting that the presence of a systemic inflammatory response predicts a poor outcome \[[@B86]\]. This data supports the hypothesis that inflammation is a risk factor for the development of some kind of solid tumors such as colon cancer. Cancer susceptibility and severity may also be associated with functional polymorphisms of cytokine genes involved in regulation of inflammation. In particular, as discussed in a recent review \[[@B79]\], a considerable body of data indicates that particular cytokine polymorphisms, especially those involving IL-6 and IL-10 genes, may influence susceptibility to, and in some cases prognosis in neoplastic diseases. It is intriguing that these two cytokines are involved in longevity \[[@B87]-[@B89]\]. Discrepant results might depend on confounding factors that affect case-control studies \[[@B90]\] or as discussed by Caruso et al. \[[@B79]\] might also depend on pleiotropic action of cytokines. It is paradigmatic in this respect that the favourable effect of high producer IL-10 genotype on hepatocarcinoma induced by HBV \[[@B91]\], may be due to the relative lack of control of HBV infection by immunosuppressive effects of high IL-10 levels. On the other hand, several and diverse mechanisms may link inflammation to cancer. 1\) Free radical production, derived by oxygen or NO metabolic pathways; the respiratory burst of leukocytes, and the arachidonic acid cascade activation are strongly implicated in DNA modification and protein damage which in turn promotes carcinogenesis \[[@B92]\]. 2\) High constitutive hyper-expression of NF-kB, an ubiquitous transcription factor with regulatory effect upon different inflammatory, apoptotic and oncogenic genes \[[@B82]\]. 3\) Alterations of the p53 gene induced by NO derived radicals promoting clonal expansion of aberrant or mutated cells \[[@B92]\]. 4\) Induction of angiogenesis by inflammatory factors favouring cancer progression \[[@B92]\]. 5\) Increased release of key pro-inflammatory factors and some cytokines, such as IL-1β, TNF-α and interferon (IFN)-γ, implicated in both the regulation of inflammation and the development of cancer \[[@B92],[@B93]\]. Atherosclerosis and cardiovascular diseases =========================================== Atherosclerosis and its complications are a major problem contributing to large sections of morbidity and mortality in old people. Cardiovascular disease is the leading worldwide cause of morbidity and death in Western societies \[[@B94]\]. However, our understanding of pathogenic mechanisms underling atherosclerosis and its complications is still incomplete, since more than half of patients with atherosclerosis do not show classical risk factors, such as hypercholesterolemia, hypertension, history of smoking, diabetes, obesity and sedentary life style \[[@B94],[@B95]\]. On the other hand, atherosclerosis, formerly considered an inocuous lipid storage disease, actually involves an ongoing inflammatory response (Figure [1](#F1){ref-type="fig"}). Recent advances in basic science have established a fundamental role for innate immunity in mediating all stages of this disease from initiation through progression and, ultimately, the thrombotic complications of atherosclerosis \[[@B96]-[@B98]\]. Clinical studies have shown that the emerging biology of inflammation in atherosclerosis applies directly to human patients. Elevation in markers of inflammation predicts outcomes of patients with acute coronary syndromes, independently of myocardial damage. In addition, low-grade chronic inflammation, as indicated by levels of the inflammatory marker CRP and cytokines, prospectively defines risk of atherosclerotic complications, thus adding to prognostic information provided by traditional risk factors. In fact, levels of CRP or IL-6 have been suggested as significant predictive risk factors for future development of cardiovascular events \[[@B31],[@B95],[@B99]-[@B103]\]. Increased levels of serum IL-1β have also been associated with high risk of congestive heart failure and angina pectoris \[[@B104]\]. Altered level of IL-1β is also suggested to be implicated in chronic inflammation underlining high blood pressure \[[@B105]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic representation of inflammatory mechanisms involved in pathogenesis of atherosclerosis and plaque formation. Monocytes and macrophages are the protagonists of atherosclerotic processes. ::: ![](1742-4933-2-8-1) ::: An initiating event is the accumulation of lipids in the vessel wall, which subsequently will become modified and triggers an inflammatory process. Low-density lipoproteins (LDL) are taken up by macrophages through scavenger receptors, leading to foam cells; the lipid deposits become oxidized forming pro-inflammatory lipid peroxides. Monocytes are attracted from the blood and differentiate into macrophages that take up the modified LDL and form lipid-laden foam cells, which is the first hallmark of atherosclerotic plaque development. Later on, inflammatory mediators increase, other immune cells are attracted, and smooth muscle cells are activated and become involved. More advanced stages of plaque development are characterized by increased deposition of extra cellular lipid cores, fibrous material, and often necrosis. Subsequently, these macrophages are further activated, leading to the production of a wide range of cytokines and growth factors. Myocardial infarction may occur as a result of erosion or uneven thinning and rupture of the fibrous cap, often at the shoulders of the lesion where macrophages enter, accumulate and are activated and where apoptosis may occur \[[@B7],[@B96]-[@B98]\]. Because genetic traits contribute significantly to the risk of coronary heart disease \[[@B106]\], a number of studies have now addressed the hypothesis that allelic variations in genes of innate immunity may increase the risk of this disease \[[@B107],[@B108]\]. Differences in the genetic regulation of inflammatory processes might partially explain why some people, but not others, develop the disease and why some develop a greater inflammatory response than others. Accordingly, common gene polymorphisms controlling high production of inflammatory molecules have been associated with atherosclerosis and a good control of inflammation might play a protective role against atherosclerosis \[[@B107]-[@B110]\]. Recently, a combination of alleles in different inflammatory genes has been associated with increased risk of developing acute myocardial infarction in elderly men \[[@B111]\]. In addition to innate immunity, clonotypic immunity plays also a role in atherosclerosis. Recent findings support the hypothesis that a crucial component of atherosclerosis is represented by T cell-mediated immune responses that are inappropriate in terms of time of onset, intensity, and target \[[@B112]\]. On the other hand, the most direct evidence for the critical role for T cells, IFN-γ, and IFN-γ-driven molecules in atherosclerosis is provided by mice with combined deficiencies of apolipoprotein E (apoE) and the IFN-γ receptor, in which the development of atheroma is significantly reduced in comparison to mice with only apoE deficiency, whereas exogenous IFN-γ enhances atherosclerosis \[[@B113],[@B114]\]. In contrast, the role of B cells has remained unclear. B-cells are not always present and may be protective. In fact, recent studies suggest that this cell type may inhibitthe development of vascular pathology in models of atherosclerosisand restenosis \[[@B115]\]. Finally, Wick et al. \[[@B116]\] have developed an *autoimmune-inflammatory*concept of atherogenesis based on experimental and clinical models. They argue that atherosclerosis may be the price humans pay for pre-existing immunity to microbial or autologous heat shock protein (HSP) 60. In summary, their autoimmune hypothesis of atherogenesis postulates that HSP60 expression is induced in arterial endothelial cells as a response to the action of stress factors, notably the classical atherosclerotic risk factors. Autologous HSP60 epitopes exposed on the surface of stressed endothelial cells are recognized by either pre-existing humoral and cellular anti-microbial HSP60 immune reactions or by invoking *bona fide*autoimmunity based on exposure of altered autologous HSP60 epitopes. The authors accept the importance of well-established atherosclerotic risk factors during atherogenesis, but assign a new role for them in the earliest stages of the disease, viz. acting as stressors \[[@B116]\]. Brain degenerative diseases: Alzheimer\'s disease ================================================= Alzheimer\'s disease (AD), a heterogeneous and progressive neurodegenerative disease which in Western societies mainly accounts for clinical dementia, is expected in the USA to rise from 4.6 today to 16 millions cases in 2050 \[[@B117]\]. Neuro-pathological hallmarks of AD are neuronal and synapsis loss, extracellular amyloid deposits (neuritic plaques) and intracellular deposition of degenerate filaments (neurofibrillary tangles) \[[@B118]\]. Major clinical manifestations of the disease are memory loss and cognitive impairment \[[@B119]\]. Inflammation clearly occurs in pathologically vulnerable regions of the AD brain, and it does so with the full complexity of local peripheral inflammatory responses. In the periphery, degenerating tissue and the deposition of highly insoluble abnormal materials are classical stimulants of inflammation. Likewise, in the AD brain damaged neurons and neuritis, highly insoluble Aβ42 peptide deposits and neurofibrillary tangles provide obvious stimuli for inflammation. Senile plaques in AD brains are associated with reactive astrocytes and activated microglial cells; cytokines and acute phase proteins are also overexpressed in microglia and astrocytes surrounding neuropathological lesions in AD brains. Inflammatory factors, such as cytokines, chemokines, complement components and acute phase proteins co-localize as secondary components in neuritic or senile plaques, or are over-produced in AD brains. Finally, activated microglia surrounds senile plaques and areas of neurodegeneration \[[@B120],[@B121]\]. There is accumulating evidence that Aβ peptide may promote or exacerbate inflammation by inducing glial cells to release immune mediators. Moreover, microglial and astroglial cells surrounding mature plaques in AD brains have been found to express activation markers. Enriched populations of human microglial cells isolated from mixed cell cultures prepared from embryonic human telencephalon tissues are able to express constitutively mRNA transcripts for cytokines and chemokines and treatment with pro-inflammatory stimuli as lipopolysaccharide or Aβ peptide leads to increased expression of mRNA levels of these inflammatory molecules \[[@B122]\]. The role of inflammation is further emphasized by a number of epidemiological studies demonstrating that the long-term use of nonsteroidal anti-inflammatory drugs may protect against AD. There are now a several published observational studies demonstrating that people who are known to be taking anti-inflammatory drugs considerably reduce their odds of developing AD and population studies have confirmed this negative association \[[@B123]\]. However, alternative hypothesises have been proposed. In particular, this effect has been hypothesised as relating to the ability of these drugs to inhibit angiogenesis. In fact, the brain endothelium secretes the precursor substrate for the β-amyloid plaque and a neurotoxic peptide that selectively kills cortical neurons. So, antiangiogenic drugs targeting the abnormal brain endothelial cell might be able to prevent and treat this disease \[[@B124]\] The long-term prospective association between dementia and the well known inflammation marker CRP was evaluated in a cohort of Japanese American men. These subjects were seen in the second examination of the Honolulu Heart Program (1968--1970) and subsequently were re-examined 25 years later for dementia in the Honolulu-Asia Aging Study (1991--1996). In a random subsample of 1,050 Honolulu-Asia Aging Study cases and noncases, high-sensitivity CRP concentrations were measured from serum taken at the second examination; dementia was assessed in a clinical examination that included neuroimaging and neuropsychological testing and was evaluated using international criteria. Compared with men in the lowest quartile (\<0.34 mg/L) of high-sensitivity CRP, men in the upper three quartiles had a 3-fold significantly increased risk for all dementias, mainly Alzheimer\'s disease and vascular dementia. These data support the view that inflammatory markers may reflect not only peripheral disease, but also cerebral disease mechanisms related to dementia, and that these processes are measurable long before clinical symptoms appear \[[@B125]\]. On the other hand, several other investigations have shown increased blood levels of some cytokines, such as IL-1β and IL-6, and acute phase proteins α-1-antichymotrypsin, (ACT) in patients with clinical AD \[[@B126]-[@B129]\]. Therefore, altered immune responses in the brain and the peripheral blood appeared to be associated with the disease. Finally, plasma levels of ACT also correlated with the degree of cognitive impairment in AD patients from a case-control study \[[@B126]\] suggesting that peripheral markers of inflammation or impaired immune responses could be used for monitoring the progression of the disease. Elevated levels of IL-6 in both brain homogenates and peripheral blood from AD patients have also been reported \[[@B130]\]. These findings suggest that an important, but still largely unknown, interplay between brain and peripheral immune responses may exist in the disease. In conclusion, the brain lesions associated with AD, which are referred to as neurofibrillary tangles and senile plaques, are characterized by the presence of a broad spectrum of inflammatory mediators, produced by resident brain cells, including neurons. Although secondary to the fundamental pathology caused by the presence of tangles and plaques, there is strong evidence that inflammation exacerbates the neuronal loss. Accordingly, several reports have appeared indicating that the risk of AD is substantially influenced by several polymorphisms in the promoter region, and other untranslated regions, of genes encoding inflammatory mediators. Alleles that favour increased expression of the inflammatory mediators or alleles that favour decreased expression of anti-inflammatory mediators are more frequent in patients with AD than in controls. The polymorphisms are fairly common in the general population, so there is a strong likelihood that any given individual will inherit one or more of the high-risk alleles \[[@B121],[@B127],[@B129]-[@B135]\] (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Alzheimer\'s disease: amyloid deposition is one the main pathogenetic mechanism. Accumulation of Aβ peptide may be caused by 1) gene mutations (PS1, PS2 and APP human mutations in familial Alzheimer\'s disease) 2) genotype (and/or phenotype) favoring unbalanced inflammatory responses (pro-inflammatory genotype/anti-inflammatory genotype). ::: ![](1742-4933-2-8-2) ::: Conclusion ========== In this presentation we have presented evidence linking innate immunity with the pathogenesis of age-related chronic degenerative diseases. Inflammation may be a pivotal and common background shared by the majority of these pathological conditions. Although evidence already presented has focused on innate immunity, clonotypic immunity has support as an emerging mechanism related to the pathogenesis of these diseases \[[@B116]\]. The evidence that a decline in mortality exists when comparing Swedish cohorts born in the middle of the 18^th^century to those in the 20^th^suggests important links between mortality in old age and in early life. Cohort mortality during childhood has been linked to cohort mortality in old age, implying that early exposure to infection is important in determining and imprinting the cohort morbidity phenotype \[[@B30]\]. Thus, chronic inflammatory mechanisms carry the imprint of early-life infections into later life morbidity and mortality. Evidence presented suggests that serological indicators of infection and inflammation in present day populations are related to vascular disease and other morbidities of ageing. Such inflammatory responses can be induced by invading pathogens, as well as by trauma or internal tissue injury. Thus, adaptive responses to short term infections or injury can become maladaptive as life progresses -a double-edged sword that evolutionary biologists refer to as antagonistic pleiotropy \[[@B136]\]. The process of life for the individual is the struggle to preserve its biological and immunological integrity. However, the preservation of the integrity of the organism comes with the price of responsiveness to systemic inflammation \[[@B137]\] which must be finely tuned otherwise dysregulation becomes a damaging accompanynt. With ageing, the reason why the innate immune system becomes over-activated is not clear, but increased exposure to infectious agents or cumulative damage to tissues could spark the change. Inflammation is not \'per se\' a negative phenomenon: it is the response of the immune system to pathogenic viruses or bacteria. Thorough out evolution, man has been set to live about 40 or 50 years but in today\'s world the immune system is active for several decades compared with the past centuries. A long period of activitation may lead to chronic inflammation which inexorably damages several/all organs and is the phenotype linked to both ageing and chronic disease. In cardiovascular disease, conventional risk factors remain important, but differential baselines in inflammatory status may explain why cholesterol levels seem not always directly associated with cardiovascular disease \[[@B22],[@B109]\]. Low-grade inflammation is also associated with parameters such as obesity, smoking, and physical inactivity, so inflammatory mediators constitute a link between life style factors, infections and physiological changes in the process of ageing on the one hand, and risk factors for age-related diseases on the other \[[@B28]\]. This immune response also depends on the genetic background of individuals. In fact, emerging evidence suggests that polymorphic alleles of inflammatory cytokines, involved in high cytokine production, are related to \'unsuccessful\' ageing as noted previously with respect to atherosclerosis and AD. On the other hand, controlling inflammatory status may enhance individual chance of achieving \'successful\' ageing. So, major findings reporting a relationship between cytokine polymorphisms and longevity suggest that those individuals who are genetically predisposed to produce low levels of inflammatory cytokines or high levels of anti-inflammatory cytokines may have an increased capacity to reach the extreme limit of human life-span \[[@B87]-[@B89],[@B109],[@B138],[@B139]\]. In other words, age-related diseases are \"the price we pay\" for an active immune system that defends us in youth but may harm us later on \[[@B139],[@B140]\]. Such data support the notion that antagonistic pleiotropy \[[@B136]\] plays a relevant role in diseases and longevity. Pro-inflammatory genotypes may therefore be both friends and foes. They are an important and necessary part of the normal host responses to pathogens, but the overproduction of inflammatory molecules may aggravate immune-inflammatory-related disease and contribute to earlier death. An immune system evolved to control pathogens is likely to favour a highly charged pro-inflammatory response programmed to resist fatal infections with avoidance of early mortality. However, persons with low responder genotypes may risk earlier mortality in response to serious infection but if they survive may respond less aggressively to age-related disease development. Such conditions might result in an increased chance of longevity in an environment with reduced pathogenic antigen load and/or adequate medical treatment \[[@B139],[@B141],[@B142]\]. Further studies on this field will open the way for new diagnostic approaches for early diagnosis of relevant pre-clinical states of age-related diseases. It may be possible, before clinical manifestations appear, that anti-inflammatory or other treatments might play a decisive role in preventing or significantly retarding the manifestation of the disease. Other studies focused on clarifying the specific contribution of each immune factor to a given disease will also contribute to the discovery of new drugs and/or innovative intervention protocols specific for these diseases related to ageing. Acknowledgements ================ Original work of Authors was supported by grants from the Italian Ministry of Education, University and Research, excofin 40% and 60%, to FL, GC, GCR, DL, CF and CC and FIRB to CF and CC, CURA financial support to FL is also acknowledged. Funds from EU 5FP T-CIA project to CF and Ministry of Health to GCR and CF are also acknowledged. The collaboration between thee \"Gruppo di Studio sull\'immunosenescenza\" coordinated by Prof. C. Caruso and INRCA was enhanced by a cooperation contract (Longevity and elderly disability biological markers).

PubMed Central

2024-06-05T03:55:58.161543

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166571/", "journal": "Immun Ageing. 2005 May 18; 2:8", "authors": [ { "first": "Federico", "last": "Licastro" }, { "first": "Giuseppina", "last": "Candore" }, { "first": "Domenico", "last": "Lio" }, { "first": "Elisa", "last": "Porcellini" }, { "first": "Giuseppina", "last": "Colonna-Romano" }, { "first": "Claudio", "last": "Franceschi" }, { "first": "Calogero", "last": "Caruso" } ] }

PMC1166572

Introduction ============ In nursing rabbits, sexual receptivity and fertility achieved after artificial insemination is depressed during the period of lactation, presumably through a hormonal antagonism between prolactin and gonadotropin release \[[@B1]-[@B3]\]. Several studies have demonstrated that separation of the doe from its litter for short periods of time, prior to artificial insemination, is very effective in stimulating ovarian activity in the mother \[[@B4]-[@B6]\], with endocrine changes that may explain the activation of ovarian function \[[@B7]-[@B9]\] Thus, separation of the does from their litters could be an effective procedure to augment breeding efficiency under farm conditions \[[@B5]-[@B7]\]. It must be noted that the published studies on the endocrine changes taking place in the doe after litter separation have been performed at single time points in the daily cycle, usually in the morning, which is an important drawback in view of the circadian nature of the secretion of the pituitary hormones involved \[[@B10],[@B11]\]. Indeed a number of circadian functions have been examined in rabbits \[[@B12]-[@B17]\], but there is no information on 24-h pattern of hormone release. This prompted us to undertake the present study whose aim was to examine the effect of litter separation for 48 h on 24-h changes in plasma prolactin (PRL), follicle stimulating hormone (FSH) and luteinizing hormone (LH) levels of the doe. Specifically, we sought to answer two questions: (i) did suppression of a major neuroendocrine and circadian stimulus like the stimulation of nipples by the lactating pup affect the 24-h changes in gonadotropin and prolactin release?; (ii) could the changes in circulating hormone levels be related to augmentation of breeding efficiency found after litter separation from the doe? Materials and methods ===================== Animals ------- The study was performed in 84 multiparous, lactating Californian x New Zealand White crossbreed female doe rabbits. Animals were housed in the research facilities of the Animal Production Department, Universidad Politécnica de Madrid, under controlled light-dark cycles (LD 16:8, light on at 8:00 h), housed in individual metal cages, fed at libitum using a commercial pellet diet (Lab Rabbit Chow, Purina Mills, Torrejón de Ardoz, Madrid, Spain) and having access to tap water ad libitum. The study was performed according to the CEE Council Directives (86/609, 1986) for the care of experimental animals. Groups of 6--7 female lactating rabbits were maintained with their litters or separated from them for 48 h, starting at different times (i.e., at 09:00, 13:00, 17:00, 21:00, 01:00 or 05:00 h). Ninety five percent of the does suckled the pups between 03:30 and 04:30 h during the dark period, as has been previously described \[[@B16]\]. On day 11 post-partum, the does were killed by decapitation at 6 different time points throughout a 24-hour cycle starting at 0900 h. Blood was collected from the cervical wound and the plasma was separated to measure prolactin and gonadotropin concentration. Hormone assays -------------- Plasma prolactin, FSH and LH levels were measured in duplicate samples by specific RIA methods \[[@B18]\] using AFP-991086, AFP-472176 and AFP-3120489 antibodies for prolactin, FSH and LH respectively, supplied by the National Institute of Health (NIH, Bethesda, MD, USA) and Dr. A. F. Parlow (Harbour-UCLA Medical Center, CA, USA). Hormones were labeled with ^125^I by the Chloramine T-method \[[@B19]\]. The antibody titers used were 1:62,500 for prolactin, 1:45,000 for FSH and 1:250,000 for LH assays, respectively. The volume of plasma used was 10 μl (prolactin assay), 75 μl (FSH assay) and 100 μl (LH assay). *Staphylococcus aureus*(prepared by the Department of Plant Physiology, U.A.M., Madrid, Spain) was used to precipitate the bound fraction \[[@B18]\]. The assays were previously validated in our laboratory \[[@B18]\]. All samples were measured in the same assay run to avoid inter-assay variations. The limits of detection for prolactin, FSH and LH were 0.125, 0.48 and 0.05 ng/mL respectively. The intra-assay coefficient of variation, calculated from a pool of plasma measured ten times in the same assay, was \< 5%. Statistical analysis -------------------- After determining that the homogeneity-of-variance assumption was tenable and that the distribution appeared unimodal and nonskewed, the statistical analysis of results was performed by a two-way factorial analysis of variance (ANOVA). Generally, the factorial ANOVA included assessment of the group effect (i.e. the occurrence of differences in mean values between control and separated groups), of time of day effects (the occurrence of daily changes) and of the interaction between the two factors (separation and time, from which inference about differences in timing and amplitude could be obtained). A one-way ANOVA followed by Student-Newman-Keuls\' test was then employed to show which time points were significantly different within each experimental group to define the existence of peaks. A Student\'s t test was performed to assess differences between the experimental groups at particular time intervals. P values lower than 0.05 were considered evidence for statistical significance. Results ======= Figure [1](#F1){ref-type="fig"} depicts the plasma prolactin levels through a 24-h cycle in control does and in does separated from their litters for 48 h. Analyzed as a main factor in a factorial ANOVA, significant time of day changes occurred (F = 35.4, p \< 0.00001) with absence of any significant effect of litter separation. Both in control and separated does, plasma level of prolactin changed throughout the day showing two maxima, at 0500 -- 0900 h and at 1700 -- 2100 h, respectively (p \< 0.001, Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **24-h changes in plasma prolactin levels in female lactating rabbits**. Groups of 6--7 animals maintained with their litters or separated from them for 48 h were killed by decapitation on day 11 post-partum, at 6 different time points throughout a 24-h cycle, starting at 0900 h. The dark bar indicates scotophase duration. Results are the means ± SEM. Letters indicate the existence of significant differences between time points in each group after a one-way ANOVA followed by a Student-Newman-Keuls\' test, as follows: ^a^p \< 0.05 vs. 09:00 and 17:00 h, ^b^p \< 0.01 vs. 09:00 h, p \< 0.05 vs 05:00 h. For further statistical analysis, see text. ::: ![](1740-3391-3-9-1) ::: Figure [2](#F2){ref-type="fig"} shows the 24-h changes in plasma FSH concentration in control does and in does separated from their litters for 48 h. Analyzed as a main factor in a factorial ANOVA, litter separation augmented FSH levels by 37 % (F = 104.6, p \< 0.00001). A significant effect of time of day and a significant interaction \"time of day x litter separation\" were found (F = 41.1 and 23.3, p \< 0.00001, respectively), i.e., a single maximum in the first half of the light period was seen in controls whereas two maxima, at 0500 -- 0900 h and at 1700 -- 2100 h, respectively, were found after litter separation. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **24-h changes in plasma FSH levels in female lactating rabbits**. Groups of 6--7 animals maintained with their litters or separated from thrm for 48 h were killed by decapitation on day 11 post-partum, at 6 different time points throughout a 24-h cycle, starting at 0900 h. The dark bar indicates scotophase duration. Results are the means ± SEM. Asterisks indicate significance differences with control at that particular time interval (Student\'s t test, \* p \< 0.05, \*\* p \< 0.01). Letters indicate the existence of significant differences between time points in each group after a one-way ANOVA followed by a Student-Newman-Keuls\' test, as follows: ^a^p \< 0.01 vs. all remaining groups, ^b^p \< 0.05 vs. 17:00 h, ^c^p \< 0.01, vs 09:00, 13:00, 21:00 and 05:00 h, p \< 0.05 vs. 17:00 h. For further statistical analysis, see text. ::: ![](1740-3391-3-9-2) ::: Figure [3](#F3){ref-type="fig"} displays the 24-h changes in plasma LH concentrations. A significant effect of litter separation and time of day was apparent (F = 16.1, p \< 0.0001 and F = 2.51, p \< 0.03, factorial ANOVA). Litter separation brought about a small albeit significant 16% increase in mean circulating LH values. As shown by the significant interaction \"time of day x litter separation\" found (F = 29.4, p\< 0.00001), litter separation disrupted the plasma LH rhythm by phase-shifting its maximum by 12 h, from 1300 h in controls to 0100 h in separated mothers (Figure [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **24-h changes in plasma LH levels in female lactating rabbits**. Groups of 6--7 animals maintained with their litters or separated from them for 48 h were killed by decapitation on day 11 post-partum, at 6 different time points throughout a 24-h cycle, starting at 0900 h. The dark bar indicates scotophase duration. Results are the means ± SEM. Asterisks indicate significance differences with control at that particular time interval (Student\'s t test, \* p \< 0.05, \*\* p \< 0.01). Letters indicate the existence of significant differences between time points in each group after a one-way ANOVA followed by a Student-Newman-Keuls\' test, as follows: ^a^p \< 0.05 vs 21:00, 01:00 and 05:00 h, ^b^p \< 0.05, vs 05:00 h, ^c^p \< 0.01 vs. 09:00, 21:00, 01:00 and 05:00 h, ^d^p \< 0.05 vs. 21:00, 01:00 and 05:00 h. For further statistical analysis, see text. ::: ![](1740-3391-3-9-3) ::: Discussion ========== The questions posed in the Introduction may now be answered. First, our results indicate that the 24 h patterns of plasma FSH and LH, but not of prolactin, changed significantly in nursing rabbits after litter separation for 48 h. Second, litter separation disrupted the 24-h rhythmicity of plasma FSH and LH concentration and caused a moderate increase in their concentration (when assessed as the mean 24-h values). Previous studies in the rabbit showed that litter separation for short periods of time augmented sexual receptivity and fertility of the doe \[[@B4]-[@B9]\]. Thus, the changes in gonadotropins reported herein could be a reflection of the same mechanisms involved in the massive release of gonadotropin releasing hormone (GnRH), LH and FSH triggered by mating or artificial insemination in litter-separated mothers. The rabbit exhibits an unusual form of maternal care, with a single and very short visit (3--5 min) every day to nurse \[[@B20]\]. This daily nursing visit of the doe is extremely regular, with some individuals showing a day-to-day variability of only a few minutes. Estrogen, androgen, progesterone and prolactin promote the onset of this behavior in does \[[@B21]\] while its maintenance relies on stimuli from the litter (i.e., maternal responsiveness is altered or abolished by prevention of mother/young contact at parturition or during early lactation). From a number of studies on the distribution of estrogen, androgen and prolactin receptors, quantification of expression of immediate-early genes, and lesions of structures of the olfactory system, it was concluded that rabbits rely on the same hormonal and extrahormonal factors that stimulate maternal behavior in other mammals except for the very peculiar circadian nursing pattern that is unique to rabbits \[[@B22]-[@B24]\]. Since the early observations by McNeilly and Friesen \[[@B25]\] it is known that postpartum blood levels of prolactin are similar in lactating and postpartum nonlactating females. Such an observation was confirmed in the present study in which plasma prolactin levels, measured at six time intervals in a 24-h cycle (the closest to nursing time was at 05:00 h), were essentially similar in control and litter-separated does. In lactating females, suckling evoked an immediate increase (3- to 5-fold) in circulating prolactin levels, an effect mimicked by the tactile stimulation of the teats \[[@B25]\]. Likewise, in hares, prolactin levels increased significantly during lactation only after suckling stimuli \[[@B26]\]. It must be noted that, in contrast to rabbits, plasma prolactin levels are significantly changed by nursing in most species, the suckling stimulus being an effective masking signal for the 24-h release of prolactin \[[@B27],[@B28]\]. This does not occur in the doe, the circadian changes of plasma prolactin levels remaining essentially unchanged after litter separation (presumably because of the very short nursing period). This suggests that the circadian secretion of prolactin and the prolactin response to physical stimulation of the nipples are independent phenomena that occur throughout the nursing period. Previous reports using single sampling procedures \[[@B8]\] indicated that litter separation decreased thr doe\'s prolactin levels and did not affect FSH. Discrepancies are possibly dependent on the sampling frequency and time of day examined. Collectively, the results underline the importance of performing 24-h studies to have a more precise picture of the hormonal changes. Litter separation disrupted the 24-h rhythmicity of both FSH and LH significantly. McNeilly \[[@B29]\] suggested that a reduction of plasma LH levels found during the light period could be coupled to an increase of pulsatile pattern of hypothalamic GnRH release. In the present study, the does exhibited, after litter separation, an inverse 24-h pattern of LH release with the lowest values during the light phase of daily photoperiod. Presumably, the disrupted 24-h rhythmicity of LH (and FSH) are linked to the greater mating or artificial insemination-induced release of LH and FSH found in does separated from their litters. In summary, the present study demonstrates the existence of 24-h variations in circulating prolactin, LH and FSH levels in nursing does. Litter separation for short periods of time, an effective procedure to stimulate ovarian activity prior to artificial insemination, markedly influences the secretory patterns of FSH and LH, a finding that can be related to the higher reflex ovulatory response to mating or artificial insemination observed in does separated from their pups. The specific value of the present study in terms of augmenting breeding efficiency should be further explored.

PubMed Central

2024-06-05T03:55:58.166008

2005-6-2

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166572/", "journal": "J Circadian Rhythms. 2005 Jun 2; 3:9", "authors": [ { "first": "Pilar", "last": "Cano" }, { "first": "Vanessa", "last": "Jiménez" }, { "first": "Maria P", "last": "Álvarez" }, { "first": "Mario", "last": "Alvariño" }, { "first": "Daniel P", "last": "Cardinali" }, { "first": "Ana I", "last": "Esquifino" } ] }

PMC1166573

Background ========== The residents of the Arab countries in the Persian Gulf region have become more sedentary and have dramatically changed their diet over the last two decades. They consume more fat, meat, sugar, rice and wheat flour than before \[[@B1]-[@B4]\]. This has resulted in a rise in obesity, diabetes, and cardiovascular disease prevalence. There is a need to study the relation between diet and chronic diseases in this population, but there is no customized instrument to do so. The Food Frequency Questionnaire (FFQ) is the most commonly used method to assess diet in relation to chronic disease \[[@B5]\]. Based on a complete search of the literature (using Medline medical subject heading and text words) and personal communication with nutritionists in United Arab Emirates and Kuwait Universities, we decided to develop a SFFQ and accompanying food composition database for UAE and Kuwait. Subjects and methods ==================== Population ---------- Our sample consisted of 326 apparently healthy persons of Arab origin, between the ages of 18 and 65 years, living in UAE and Kuwait in 2003--2004. Of these 126 provided information using the 24-hour dietary recall, and 200 participated in the testing of the long-FFQ in the SFFQ development. Development of the semi-quantitative food frequency questionnaire ----------------------------------------------------------------- To develop the SFFQ we went through the following steps: construction of a food list, definition of portion sizes, and assignment of frequency of consumption, pilot test of long-FFQ and assembling the selected food list into SFFQ. 24-hour Dietary Recall (DR) --------------------------- Qualified nutritionists interviewed the participants. The nutritionists asked the participants what they ate in the previous 24-hours in direct chronological order from the first foods in the morning to the last foods before breakfast on the day of the interview. To standardize data collection, we prepared a manual of procedures for the interviewers. The interviewers used a food atlas modified from \"Food portion sizes: A photographic Atlas\" \[[@B6]\] containing coloured pictures of 8 different portion sizes of foods commonly eaten in UAE and Kuwait. We constructed a long food list from information we obtained from the 24-hour dietary recalls, and supplemented it with foods derived from popular cookbooks, and suggestions of experienced local dieticians. We obtained the range of portion sizes from the 24-hour dietary recall and cookbooks. For items such as eggs we considered one of those items as a portion. For fruits such as banana or orange, a medium size was considered to be a \"unit\" of the fruit. For foods with non-unitary measures, such as grapes, we considered 1/2 cup to be one unit. For other foods one unit of a utensil that was mostly reported by respondents was assumed as one portion size. To get an estimate of the usual portion we used the mode of the portion size distribution for each food reported. We formatted the SFFQ based on the pattern of the Harvard FFQ. We organized the main foods by the traditional food groups into seven food categories: 1. Milk, milk products and fats (including milk with different amounts of fat, Labnah, etc.), 2. vegetables (fresh or cooked), 3. fruit, 4. meat, eggs and meat products (including meat organs), 5. cereal and cereal product, 6. beverages 7. sweet and baked goods (Faaloodah, baklava, etc.) and nuts like almond, and hamoose. To distinguish the fibre content of the diet and the quality of carbohydrates, we differentiated between the types of consumed cereal and cereal products (white or wholemeal bread, whole grain). In the last section of the FFQ we asked about type of cooking oils and fats and also, minerals including calcium, iron, zinc and multivitamins. We used 9 categories to assess frequency of intake varying from \"never or less than once a month to 6 or more times per day. For each food item, participants indicated their average frequency of consumption over the past year of a specified serving size by checking 1 of the 9 frequency categories. For foods that contain an extremely high amount of a particular nutrient but are used infrequently, such as liver, we re-categorized the options for frequency of intake. For instance, we used eliminated options of higher intake (once per day) but distinguished between never and less than once a month at the lower end. We will compute the daily intake based on the midpoint of the reported frequency category for each food item; for example we will take a response of \"2--4/week\" to be 3/7 or 0.43 times/day. Although most fruits and vegetables are available year round in UAE and Kuwait, their intakes may vary depending upon cost and cultural preferences of foods by season. For this reason we designed specific questions for fruit and vegetable consumption by season. We determined the length of the season from local experts who used their experience. We pilot tested the long-FFQ among 200 participants from the same populations (but not those who participated in the 24 hr DR). The objectives were to determine the completeness of the food list and to shorten the food list by deleting foods that were not commonly consumed. Based on the analysis of the pilot we completed the food list and created SFFQ. ### Food Composition Database We will use the food composition database to convert intakes of foods into nutrients. We constructed one nutrient database for UAE and Kuwait, as most foods are similar in both countries. We extracted the nutrient contents of the food items from Table SR17 of the USDA food database, which is available online, as the starting point to establish the database <http://www.nal.usda.gov/fnic/foodcomp/Data/SR17/sr17.html>. From SR17 we chose those varieties of food items which are not very specific to a region and are more representative, for example; for apple we chose \"apples, raw, with skin (NDB No:09003)\" or for orange, we chose \"oranges, raw -- all commercial varieties (NDB No: 09200)\". We used Statistical analyses SPSS 10.1 for windows (SPSS Inc, Chicago IL) for all analyses. Results ======= There were 326 participants with different occupations in Kuwait and UAE in our study. The populations were similar in UAE and Kuwait with respect to age, sex distribution, and BMI, but Kuwaiti women had higher education levels than those in UAE (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). No significant differences were noted between female and male respondents with respect to BMI (P = 0.4, 0.5 for UAE and Kuwait, respectively). However, in UAE, women had a higher mean BMI than men while in Kuwait men were heavier than women. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of UAE participants ::: **Men (N = 35)** **Women (N = 91)** **Overall (N = 126)** --------------------- ------------------ -------------------- ----------------------- Age 43.0 ± 11 34.3 ± 10 37.0 ± 11 BMI (kg/m^2^) 27.4 ± 4 28.7 ± 7 28.3 ± 7 **Education** None 22.7% 10% 13% Primary school 9.0% 8% 9% Secondary school 23% 30% 31% Trade School 32% 3% 2% University 23% 49% 45% **Income (Dirham)** \<5000 12.5% 9% 11% 5001--10000 12.5% 63% 60% 10001--15000 19% 24% 22% \>15000 12.5% 4% 6% ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Demographic characteristics of Kuwaiti participants ::: **Men (N = 56)** **Women (N = 145)** **Overall (N = 201)** ------------------ ------------------ --------------------- ----------------------- Age 45.6 ± 14.0 37.3 ± 12 39.6 ± 13.3 BMI (kg/m^2^) 26.6 ± 7 25.9 ± 5 26.0 ± 6 **Education** None 0% 0% 0% Primary school 0% 5% 4% Secondary school 32% 18% 21% Trade School 5% 4% 4% University 64% 72% 70% **Income** \<500 68% 31% 61% 501--1000 32% 54% 36% \>1000 0% 15% 3% ::: Daily food intake ----------------- The estimated daily intakes of seven major food groups among Kuwaiti and UAE men and women are shown in Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}. On average the participants in UAE reported eating 3.4 servings/d of fruits and 3.1 servings/d of vegetables versus 2.8 servings/d of fruits and 3.2 servings/d of vegetables in Kuwait. Cereals are an important staple in the diet of both countries and the participants reported eating cereals 4.8 times/d in UAE and 5.3 times/d in Kuwait. All participants reported consuming cereals at least once per day. Meat was consumed nearly two times/day in both countries and among the meat group, poultry was consumed more often than red meat or fish. The mean intake of dairy products was 2.2/day in UAE and 3.4/day in Kuwait. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Average daily intake of main foods estimated by long-FFQ reported by UAE participants (Men = 24, Women = 76) ::: **Sex** **Foods** **Min** **Max** **Mean** **Std. Deviation** --------- ----------------------------- --------- --------- ---------- -------------------- Women Fruits 0.2 8.5 3.0 1.7 Vegetables 0.4 7.9 3.0 1.5 Dairy products 0.0 5.2 2.1 1.0 Meat 0.6 4.2 1.7 0.7 Cereals and cereal products 0.7 10.7 4.7 2.2 Beverages 0.0 9.5 3.4 2.1 Baked goods 0.0 3.4 0.5 0.5 Men Fruits 1.8 8.8 4.8 2.0 Vegetables 0.3 6.6 3.7 1.6 Dairy products 0.4 7.0 2.5 1.3 Meat 0.6 4.4 2.3 0.9 Cereals and cereal products 1.5 6.6 5.0 1.2 Beverages 1.3 8.5 5.0 1.9 Baked goods 0.0 2.1 0.8 0.6 Overall Fruits 0.2 8.8 3.4 1.9 Vegetables 0.2 7.9 3.1 1.6 Dairy products 0.2 7.1 2.2 1.1 Meat 0.6 4.4 1.9 0.8 Cereals and cereal products 0.7 10.8 4.8 2.0 Beverages 0.0 9.5 3.8 2.2 Baked goods 0.0 3.4 0.6 0.5 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Average daily intake of main foods estimated by long-FFQ reported by Kuwaiti participants (Men = 22, Women = 78) ::: **Sex** **Foods** **Min** **Max** **Mean** **Std. Deviation** --------- ----------------------------- --------- --------- ---------- -------------------- Women Fruits 0.4 8.1 2.8 1.7 Vegetables 0.0 10.5 3.3 2.2 Dairy products 0.0 9.7 3.3 2.3 Meat 0.2 9.9 1.8 1.3 Cereals and cereal products 0.7 20.1 4.9 3.2 Beverages 0.0 11.1 2.9 2.3 Baked goods 0.0 5.0 1.0 1.0 Men Fruits 0.5 7.3 3.1 1.8 Vegetables 0.3 6.5 3.0 1.9 Dairy products 0.4 6.6 3.5 1.7 Meat 0.4 6.6 2.5 1.6 Cereals and cereal products 2.3 12.8 6.5 2.4 Beverages 0.0 9.3 4.0 2.7 Baked goods 0.0 5.7 1.2 1.5 Overall Fruits 0.4 8 2.8 1.7 Vegetables 0.00 10.5 3.2 2.1 Dairy products 0.00 9.7 3.4 2.3 Meat 0.2 9.9 1.9 1.4 Cereals and cereal products 0.7 20.1 5.3 3.1 Beverages 0.0 5.7 1.0 1.1 Baked goods 0.0 11.0 3.1 2.4 ::: Frequency of consumed food -------------------------- Tables 6 and 7 show the frequency of consumption of some foods in UAE and Kuwait. About 32% of respondents in UAE and 26% of Kuwaiti respondents reported that they had at least one glass of milk daily on average in the past year. Sixty-eight percent of UAE participants and 48% of Kuwaiti participants reported consuming rice at least once per day. 67% and 51% of people (UAE and Kuwaiti respectively) ate an egg at least 2 times/week. Overall in UAE, 86% of participants did not eat chicken with skin while in Kuwait 42% of participants did not eat chicken with skin. Among fruits, apples, oranges and bananas were consumed very frequently. Nutrient database ----------------- To convert local mixed dishes to nutrients, we created a new nutrient database appropriate for local foods (Table [5](#T5){ref-type="table"}). Therefore, a local nutritionist collected 2 recipes from each low, middle and high-income family. We supplemented these recipes from the \"Food composition: Kuwaiti composite dishes\" \[[@B7]\] and other popular cookbooks. The average amount of ingredients from those recipes was used to create a base recipe for the nutrient database. We matched ingredients of the recipes with the appropriate food items in the USDA database to obtain nutrient content, taking cooking method into account. The SFFQ for UAE is in Appendix I and the SFFQ for Kuwait is in Appendix II. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Nutrient composition per serving (100 g) of some commonly eaten foods in UAE and Kuwait ::: **Vitamins (mg)** ------------------------------ ----- ------ ------ ------ ------- ------- ------ ------------------- ------ ------- ------- ------ ------ ------ ----- **[Mixed dish]{.underline}** Kofta 173 3.1 11.1 15.0 0.6 63.1 14.1 0.09 0.26 1.7 12.6 0.1 23.7 128 1.6 Qouzi 206 15.0 11.2 10.5 0.6 5.1 36.7 0.1 0.1 0.9 2.3 0.3 18.0 90.3 1.4 Marga Laham 95 7.1 4.0 7.5 0.8 19.0 4.6 0.07 0.2 0.7 7.0 0.04 7.1 65.4 0.8 Jereesh 226 29.7 7.0 12.9 5.2 13.9 22.7 0.2 0.2 0.7 3.5 0.4 21.1 252 2.1 **[Sweets]{.underline}** Balalett 210 35 4.7 6.6 0.1 Trace 10 0.01 0.06 Trace Trace 0.5 28.5 105 1.5 **Elba** 264 30.6 12.6 11.3 Trace 206 25 0.1 0.1 1.3 1.1 0.6 522 356 2.1 ::: Discussion ========== In this paper, we have described the development of a semi-quantitative FFQ and food composition database for the Arab population in UAE and Kuwait. The goals of diet assessment in epidemiologic studies are to obtain a measure of usual rather than current diet, and rank people by intake, in contrast to clinical settings where current absolute intake is more important. The FFQ has been developed with these purposes in mind and has become the standard method to collect dietary data in studies of chronic disease all over the world. We opted to use a semi-quantitative FFQ, which estimated food intake in categories rather than the exact frequency, because it has been shown that there is minimal loss of information in estimating nutrient intakes using food intake categories \[[@B8]\]. We also asked the participants about intakes of pre-specified portion sizes rather than asking them to estimate their regular portion size. Correlations for nutrient intake calculated using the FFQ with and without taking portion sizes into account were over 0.9 \[[@B5]\]. The advantage of using categories to estimate food intake and pre-specified portion sizes is that the SFFQ becomes easier to administer, and likely, more reliable. We did not attempt to make a comprehensive list of foods to include in the SFFQ. Rather, we kept items in the SFFQ if they were nutrient rich, consumed frequently and discriminated intakes between individuals. The other criterion we considered together with the nutrient content (including caloric value) was the presence of other substances of interest, for instance, caffeine. Most FFQs have between 100--150 items \[[@B9]\] and our SFFQ has 153 (UAE) and 152 (Kuwait) items. Increasing the number of items in the FFQ has been shown to increase over-reporting \[[@B10]\]. To estimate nutrient intake from SFFQ, there is a need for a food composition table listing the average nutrient content of foods contained in the SFFQ. To obtain nutrient intake we multiplied the average nutrient content of a specified portion of food listed in the food composition table by the average frequency of intake reported in the SFFQ. The food composition table can be a substantial source of variation in the estimation of nutrients using the SFFQ. As no nutritional database has ever been gathered in UAE or Kuwait, we used the US Department of Agriculture nutrient database as our standard to estimate nutrient content. The advantages of this approach are: First, the USDA food composition database is probably the most comprehensive in the world. For example, there are 26 categories of spinach including different types of spinach, raw spinach, and spinach cooked in a variety of ways \[[@B11]\], allowing us to choose the most appropriate one. Second, the nutrient estimation assays have been done in a standardized manner. Third, it has the largest number of nutrients reported. Fourth, the USDA food composition database is continually updated. Last, UAE and Kuwait import foods from all around the world and a mixture of food items from different regions are available in any market. For mixed dishes that were not listed in the USDA database, we calculated nutrient intake by analyzing recipes. Moreover, there are nearly 150 food composition tables in use around the world and their values are primarily based on USDA \[[@B12]-[@B14]\], and even European countries include nutrient information from USDA in their food composition tables \[[@B15],[@B16]\]. Finally, similar approaches have been taken by other investigators in Israel, \[[@B17]\] and Costa Rica \[[@B18]\]. A limitation of this study is that the age groups represented in the UAE and Kuwait sample are mostly \<50 years for both males and females. Thus, the overall impression in dietary habits is biased towards this younger group. For example, the consumption of rice as well as dates might be underestimated. The way to make it more accurate is of course to repeat it (validate). Another limitation of this study is that most participants from both countries were women and some foods, which men may eat, may be underestimated. However, a nutritionist with experience in those countries reviewed the food lists to ensure their completeness. Conclusion ========== The validated questionnaire and food composition database will not only be useful tools for our own study, but they will also be assets that other researchers in the region can use or adapt to suit their needs. We have enclosed two SFFQs in this article so other researchers in the field of public health can use this comprehensive SFFQ. We are currently evaluating SFFQs and the validated SFFQ will be available online for all public health researchers in the region. List of Abbreviations ===================== SFFQ Semi-Quantitative Food Frequency Questionnaire 24 hr DR 24 hour Dietary Recall UAE United Arab Emirates PURE Prospective Urban and Rural Epidemiologic SPSS Statistical Packages for Social Sciences Authors\' contributions ======================= MD Participated in design of study, coordinated and performed statistical analysis, drafted the manuscript AM Participated in design of study, performed statistical analysis, helped to draft the manuscript NH Coordinated study in Kuwait, helped to draft the manuscript AY Facilitated data collection in UAE, helped to draft the manuscript FN Coordinated study in UAE SY Participated in design of study, helped to draft the manuscript All authors read and approved the final manuscript. Acknowledgements ================ The authors acknowledge the contribution of Mrs. Sumathy Rangarajan who supported the fieldwork during the data collection phase and are grateful to the nutritionists who collected the dietary data in Kuwait and UAE and Mrs. Pam Mackie for secretarial support. We are grateful to the participants of the study for their cooperation. We also wish to thank Dr. Yusuf (Director of Population Health Research Institute) for all his support and guidance. We also acknowledge the \"**Sheikh Hamdan bin Rashid Al Makhtoum Award for Medical Sciences\"**for their **Grant no.**MGR-16/2001--2002 for the UAE portion of the study.

PubMed Central

2024-06-05T03:55:58.167288

2005-5-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166573/", "journal": "Nutr J. 2005 May 27; 4:18", "authors": [ { "first": "Mahshid", "last": "Dehghan" }, { "first": "Nawal", "last": "Al Hamad" }, { "first": "AfzalHussein", "last": "Yusufali" }, { "first": "Fathimunissa", "last": "Nusrath" }, { "first": "Salim", "last": "Yusuf" }, { "first": "Anwar T", "last": "Merchant" } ] }

PMC1166574

Background ========== Apoptosis is a type of programmed cell death and is a natural phenomenon occurring when the cells are subjected to stress such as DNA damage, death signals or lack of growth factor. Apoptotic stimuli allow an intracellular cascade of signals such as the caspases, a family of cysteine proteases implicated in the cleavage of a number of important proteins which results in cell disassembly and cell death, phagocytosis and removal of cell debris by immune cells. Apoptosis plays an important role during embryo implantation in rodents where morphological characteristics of apoptosis are observed in endometrial epithelial cells at the embryo implantation site \[[@B1]-[@B3]\]. Moreover, this phenomenon also occurs during late pregnancy, especially during regression of the decidua basalis (DB) in the rat endometrium \[[@B4],[@B5]\]. Two decidual zones are formed during pregnancy: the primary decidual zone on the antimesometrial side of the uterus and the secondary decidual zone (or antimesometrial decidua) which is formed following expansion of the primary decidual zone \[[@B5],[@B6]\]. The secondary decidual zone eventually transforms stromal cells in the mesometrial region to form the DB that regresses following day 14 of pregnancy \[[@B7]\]. Whether the phenomenon of growing size of embryo is a cause or correlation to the increase of apoptosis remains to be elucidated. The first members of the transforming growth factor-beta (TGF-β) superfamily were identified on the basis of their ability to induce a transformed phenotype of certain cells in culture \[[@B8]\]. They are now known as multifunctional polypeptides involved in the regulation of cell proliferation and differentiation, immunoregulation, angiogenesis and the regulation of extracellular matrix \[[@B9],[@B10]\]. They act via cellular signalling through Smads and Phosphorylated-Smads (P-Smads), the active form of Smads. Those proteins are translocated to the nucleus and activate transcription factors which in turn activate caspases and other regulation proteins \[[@B11],[@B12]\]. Another characteristic of TGF-β is its capacity to induce apoptosis in several cell types \[[@B7],[@B13]\]; in fact, TGF-β was shown to have a pro-apoptotic function mediated by caspases \[[@B14]-[@B16]\]. Genes encoding the three isoforms are localized on different chromosomes and the isoforms molecular weights are slightly different: 15, 12.5 and 12 KDa for TGF-β1, β2 and β3 respectively; they share 80 % sequence identity and are produced in latent forms which are activated into a 112-amino acid mature peptide \[[@B17]\]. Multiplicity of TGF-β isoforms and sequence conservation within each form through evolution suggests important specific roles. Moreover, it has been demonstrated that TGF-β1, -β2 and -β3 are differently expressed in the mouse uterus \[[@B18]\] and porcine conceptus-maternal interface \[[@B19],[@B20]\]. The uterus is a hormone-dependent organ and it is subject to an abundant amount of cellular proliferation and cell death. Studies have shown that apoptosis is increased in the rat endometrium during implantation and regression of the decidual basalis in the rat \[[@B1],[@B2],[@B21]\]. The mRNA for TGF-β1 has been shown to be present within the uterus during rat pregnancy and was localized to the luminal and glandular epithelial cells during early and late pregnancy \[[@B22]\]. TGF-β1 and -β2 mRNAs were also found in the mouse uterus during pregnancy \[[@B23]-[@B25]\]. Expression of TGF-β2 and TGF-β3 mRNAs were also shown to be expressed in the mouse periimplantation uterus \[[@B18]\]. A study showed that TGF-β1 and -β2 treatments on cultured endometrial rat stromal cells induced apoptosis \[[@B7]\]. These studies suggest that TGF-β isoforms might be involved specifically in the control of apoptosis in the uterus during pregnancy. However, the mechanisms involved in the control of apoptosis in the endometrium during pregnancy are poorly documented. Studies suggesting the importance of TGF-β isoforms at specific time of pregnancy were done at the mRNA level and it is important to determine their presence at the protein level. Thus, the aim of this study was to determine the expression and the developmental expression of TGF-β1, β2 and β3 proteins in the rat uterus throughout pregnancy and to further determine *in vitro*, the effect of TGF-β in determining decidual cell fate. We found that the three isoforms of TGF-β were expressed and regulated differently in epithelial and stromal endometrial cells during rat pregnancy and showed using primary decidual cell cultures the involvement of TGF-β in the regulation of programmed cell death. Moreover, the present study shows that TGF-β signals through Smad2, which coincide with XIAP and Akt down regulation and induction of apoptosis. Methods ======= Reagents -------- TGF-β1 (sc-146, lot \# F262, 200 μg/ml), TGF-β2 (sc-90, lot \# B202, 200 μg/ml) and TGF-β3 (sc-82, lot \# A222, 200 μg/ml) polyclonal antibodies were purchased from Santa Cruz Biotechnology, Inc (Santa Cruz, CA, USA). CDC47/MCM7 antibody was obtained from Medicorp (Montréal, QC, Canada). Phospho-Akt (Ser 473), Akt, XIAP, Cleaved caspase-3, and Phospho-Smad2 (Ser 465 / 467) antibodies were obtained from Cell Signaling Technology (Beverly, MA, USA). The Keratin 8/18 antibody used to determine cell culture purity was donated by Dr Monique Cadrin (Univ. of Québec at Trois-Rivières, QC, Canada). Anti-Smad2/3 antibody was purchased from Calbiochem (San Diego, CA, USA). Vectastain ABC Kit for rabbit IgG was purchased from Vector Laboratories Inc. (Burlingame, CA, USA). *In Situ*Cell Death detection kit (TUNEL), POD and DAB substrate was obtained from Roche (Laval, QC, Canada). TGF-β1 recombinant protein was purchased from Biosource (Cat \# PHG9104, lot \# 16865-01S, 5 μg, diluted at 50 μg/ml, QC, Canada). Animals ------- Sprague-Dawley female rats, 200--225 g, were obtained from Charles River Laboratories Canada. Animals were maintained on standard chow and water, which were available *ad libitum*, in animal facilities illuminated between 6:00 h and 20:00 h. All procedures were performed in accordance with guidelines of the Canadian Council on Animal Care for the handling and training of laboratory animals and the Good Health and Animal Care Committee of the Université du Québec à Trois-Rivières. Male and female mice were mated overnight and confirmation of pregnancy was determined by vaginal smears and/or the presence of a vaginal plug (day 1). Rats were killed on day 2, 4, 5, 6, 8, 10, 12, 14, 16, 18 and 20 of pregnancy at 10:00 h in the morning and at 18:00 h for days 5.5 and 6.5. Six to 8 different rats were used for each time of pregnancy. Uteri were collected and fixed for immunohistochemical staining (IHC) and apoptotic cell death detection by \[TdT\]-mediated deoxyuridinetriphosphate nick end-labeling (TUNEL) or endometrial protein extracts collected for Western blot analysis. Rat pretreatments and decidual endometrial stromal cell culture --------------------------------------------------------------- A total of 10 rats were ovariectomized and then allowed to recover from surgery for a minimum of 10 days. They were pre-treated with physiological doses of estradiol (1,3,5(10)-Estratriene-3,17β-diol, Sigma-aldrich) and progesterone (Laboratoire Mat, PQ) to induce decidualization as described previously \[[@B26]\]: 1) 0.2 ug estradiol injection per day for three days (in the morning, day -2,-1 and 0); 2) On the third day (day 0 of pseudopregnancy), another injection in the afternoon of estradiol (0.2 μg) and progesterone (1 mg) was performed; 3) No treatment for 2 days (day 1 and 2 of pseudopregnancy); 4) Injections of estradiol (0.1 μg) and progesterone (4 mg) for three days (day 3, 4 and 5 of pseudopregnancy); 5) Another injection of estradiol (0.1 μg) in the afternoon on day 7 (day 4 of pseudopregnancy); 6) Rats were killed on day 8 (day 5 of pseudopregnancy). All endometrial stromal cells collected for cultures were recovered from rats treated with the protocol described above. Uteri were removed and horns taken and immerged in HBSS solution containing HEPES (20 mM), penicillin (100 units/ml), streptomycin (100 μg/ml) and fungizone (1,25 μl/ml) (Invitrogen, ON, Canada). Further manipulations were performed in a sterile environment. The uterine horns were transferred into a sterile petri containing HBSS, slit longitudinally and immersed in trypsin type I solution (0.3%) (Roche Diagnostics, QC, Canada) in HBSS and agitated for 60 minutes at room temperature. Uterine horns were then vortexed at maximum for 5 sec and supernatant containing epithelial cells was discarded. Uterine horns were washed three times with 2.5 ml of HBSS and immersed in a HBSS solution containing trypsin type I (0.03%), DNAse I (0.016%) and collagenase type II (0.064%) for 15 minutes at 37°C in a water bath. Uterine horns were then vortexed at maximum for 5 sec. The supernatant containing stromal cells was transferred into a sterile falcon tube containing 150 μl of FBS D.C (Dextran-Charcoal extracted). Uterine horns were washed two times with 2.5 ml of HBSS and the supernatant was mixed with stromal cells. Uterine horns were discarded and stromal cells were centrifuged at 1000 g for 5 minutes. Cells were washed two times with HBSS and centrifuged. The supernatant was discarded and cells diluted with DMEM-F12 (Ph 7.1) (Invitrogen, ON, Canada) containing 2.438 g/L NaHCO~3~, 10% FBS D.C. and gentamycine 50 μg/ml. Cells were incubated at 37°C in an atmosphere of 5% CO~2~. Cells were plated in 6-well plates (Corning plates) at a density of 50% (4 × 10^5^cells per well). The medium was changed two hours after the first incubation in order to eliminate epithelial cell contamination from stromal cell cultures. The purity of stromal cells was more than 97%: cell culture contamination with epithelial cells was evaluated by cellular morphology and immunofluorescence using a Keratin 8/18 antibody. Three to 5 days after plating (more than 90% of confluency reached), cells were treated for 24 hours in the presence or absence of increasing doses of TGF-β recombinant protein. Total proteins from treated cell cultures were extracted using TRIZOL (Invitrogen, ON, Canada). For Western blot analyses, 15 μg of total protein was used for each analysis. Immunohistochemical staining ---------------------------- The uterus was fixed in 4% paraformaldehyde solution and embedded in paraffin. Tissue sections 7 μm thick were mounted on polylysine-coated slides, deparaffinized, rehydrated, and then heated in 10 mM citrate buffer (pH 6) containing triton X-100 (Sigma-Aldrich) 0.1% (v/v). After two washes with PBS, slides were then incubated with 0.3 % hydrogen peroxide in methanol for 30 min to quench endogenous peroxidase activity. After washing with PBS, tissues were incubated with blocking serum (Vectastain ABC Kit) at room temperature for 1 h. Then, a primary antibody diluted in blocking serum (TGF-β1, β2 or β3; 1:50 dilution or CDC47/MCM7; 1:100 dilution) was added to the slides and incubated at 4°C overnight in a humidified chamber. After washing 5 min. in PBS, tissue sections were incubated for 30 min. with 3 μg/ml biotinylated antibody (anti-rabbit or anti-mouse). Subsequently, slides were washed with PBS and incubated with avidin-biotin complex reagent containing horseradish peroxidase for 30 min. Slides were washed with PBS for 5 min and color development was achieved using DAB substrate. The tissue sections were counterstained with haematoxylin. Negative controls were performed using the same protocol but substituting the primary antibody with normal rabbit IgG (Vector Laboratories Inc., Burlingame, CA, USA). Immunofluorescence ------------------ Tissues were prepared as described in the immunohistochemical section. Cleaved caspase-3 antibody was diluted 1:100 in blocking serum and slides were incubated at 4°C overnight. After washing twice for 5 min. in PBS, tissue sections were incubated for 30 min. at room temperature with 2 mg/ml Alexa Fluor 488 donkey anti-rabbit (1:50). Subsequently, slides were washed with PBS and mounted. Negative controls were performed using the same protocol but substituting the primary antibody with normal rabbit IgG. Sections were examined using an OlympusBX60 microscope equipped with a Coolsnap-pro CF digital camera (Carsen Group, ON, Canada). TdT-mediated deoxyuridinetriphosphate nick end-labeling (TUNEL) --------------------------------------------------------------- Tissue sections were deparaffinized, rehydrated and rinsed with PBS. They were incubated with proteinase K (20 μg/ml) for 30 min. at room temperature. Slides were washed twice with PBS, the endogenous peroxidase was quenched with 0.3 % hydrogen peroxide in methanol for 30 min. The slides were rinsed and incubated with 10 mM citrate solution for two min on ice. Then, tissue sections were rinsed with PBS and incubated with TdT labelling reaction (*In Situ*Cell Death Detection, POD) for 30 min at 37°C in humidified environment. Slides were washed three times in PBS and tissue sections were blocked with 3% BSA for 20 min. at room temperature. Converter-POD solution was added to the slides and incubated for 30 min. at 37°C in humidified environment. Slides were washed for 5 min. in PBS, colour development was achieved using DAB substrate and counterstained with haematoxylin. Negative controls were performed using the same protocol without TdT enzyme. Protein extraction and Western analysis --------------------------------------- Protein homogenates from pregnant endometrium were isolated according to a protocol previously described \[[@B27]\]. Briefly, uteri from Day 2 to Day 20 pregnant rats were rapidly excisedand placed in ice-cold saline until dissected. Uteri were carefully laid on a glass plate and placed on the stage of a dissecting microscope. In early pregnancy (Day 2 to 5.5), total endometrium was scraped using a microscope glass and collected. Uteri from Day 6 to 10 the placenta and decidua were at an early stage of differentiation and could not be reliably separated. For this reason, DB dissectedfrom animals between these days of pregnancy contain some chorioallantoic cells, but antimesometrial decidua, choriovitelline tissues, fetus, and myometrium were removed. Even though we carefully dissected DB from these tissues, it is a possibility that a contamination with some antimesometrial decidual cells that regress to form the deciduas caspularis (DC) would occur. This is an important fact that we need to take into consideration. In uteri collected from Day 12 to 20 pregnant rats, DB were isolated by gently separating the placenta and myometrial regions with 23-gauge needles. Additionally, the DB began to regress on Day 14 and became too thin to reliablydissect after Day 17. The protocol for DB isolation was described previously by Ogle and George \[[@B28]\]. Endometrial cells from pregnant animals were homogenized using a pipette in RIPA lysis buffer (PBS 1× pH 7.4; 1% Nonidet P-40; 0.5% Sodium deoxycholate; 0.1% SDS; Protease Inhibitor Cocktail Tablets (Roche Diagnostics Canada, PQ)). Homogenates were centrifuged (12,000 × g for 20 min at 4°C) to remove insoluble material. The supernatant was recovered and stored at -20°C pending analysis. Protein content was determined with the Bio-Rad DC Protein Assay. Protein extracts (50 μg) were heated at 94°C for 3 min, resolved by 10% SDS-PAGE and electrotransferred to nitrocellulose membranes using a semidry transfer (Bio-Rad, Mississauga, ON). The membranes were then blocked 2 h at room temperature with PBS containing 5 % milk powder, then incubated with anti TGF-β 1-2-3 1:1000 ; P-Smad2 (Ser 465 / 467) 1:1000 and Smad 2/3 1:1000 and subsequently with horseradish peroxidase-conjugated anti-rabbit or anti-mouse secondary antibody (1:3000; room temperature for 45 min). All membranes were stripped with Restore western blot stripping buffer (Pierce, \# 21059, lot \# FH71541), reprobed with an antibody specific to β-actin which was used as an internal standard. Peroxidase activity was visualized with the Super signal^®^West Femto maximum sensitivity substrate (Pierce, Arlington Heights, IL, USA), according to the manufacturer\'s instructions. Signal was visualized using the Biochemi Imaging System (UVP, CA, USA). Densitometrical analyses were performed (protein of interest and β-actin) using the GelDoc 2000 and the Quantity One software (Bio-Rad, Mississauga, ON, Canada). Results are expressed as a ratio (protein of interest/β-actin) to correct for loading for each endometrial sample. Hoechst and trypan blue exclusion staining ------------------------------------------ Following TGF-β treatment, both floating and attached cells were resuspended in PBS containing Hoechst 33258 for 24 hours at 4°C or resuspended in trypan blue solution (0,4%) for 5 minutes. Hoechst nuclear staining was viewed and photographed using a Olympus BX60 fluorescence microscope and a Coolsnap-Pro CF digital Camera (Carsen Group, ON, Canada). Cells with typical apoptotic nuclear morphology (nuclear shrinkage, condensation and fragmentation) were identified and counted using randomly selected fields on numbered photographic slides, of which the \"counter\" was not aware of the treatment, so as to avoid experimental bias. A minimum of 200 cells per treatment group were counted in each experiment. For trypan blue exclusion test, blue cells were counted under a regular microscope and were counted as non-living cells. Statistical analysis -------------------- Western analyses of pregnant animals were repeated six to eight times (6 to 8 different endometrial extract per day of pregnancy from 6 to 8 different rats). Endometrial extracts from each rat were assessed individually. Western analyses of cultured decidual cells were repeated 5 times for each TGF-β dose (for each culture experiment, decidual cells were recovered from a pool of ten different ovariectomized/treated rats). Results subjected to statistical analyses were expressed as mean ± SEM. Data were subjected to one-way ANOVA (PRISM software version 4.0; GraphPad, San Diego, CA). Differences between experimental groups were determined by the Tukey\'s test. Results ======= Apoptosis expression during pregnancy ------------------------------------- In order to confirm the presence of apoptosis in DB, the presence of the activated form of caspase-3 was measured by Western analysis (Fig. [1A](#F1){ref-type="fig"}) and immunofluorescence (Fig. [1B](#F1){ref-type="fig"}) using a day 14 pregnant uterine section. TUNEL measurement was also performed using a day 14 pregnant uterine section (Fig. [1C](#F1){ref-type="fig"}). As demonstrated by Western analysis, Fig. [1A](#F1){ref-type="fig"} clearly demonstrates that apoptosis was present in the endometrium at day 14: the 17 KDa cleaved caspase-3 fragment was significantly and gradually increased from day 8 and was maximal at day 14. Cleaved caspase-3 fragment was observed in the cytoplasm of apoptotic cells as demonstrated by immunofluorescence (Fig. [1B](#F1){ref-type="fig"}) and TUNEL positive cells were also found in the in the endometrium at day 14 (Fig. [1C](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Detection of apoptosis in pregnant endometrial tissues as demonstrated by Western and TUNEL analyses. A) Apoptosis as determined by Western analysis of cleaved caspase-3 (one blot presented out of 6). β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis of the 17 KDa cleaved fragment. B) Cleaved caspase-3 immunofluorescence at day 14 of pregnancy. Magnification: control: 10× ; inset and Day 14: 40×. C) TUNEL analysis at day 14 of pregnancy. B) and C) One representative section is presented out of 6 experiments. C-: negative control. Magnification: control: 10× ; inset and Day 14: 40×. Arrows indicate positive staining. ::: ![](1477-7827-3-20-1) ::: Expression of TGF-β1, β2 and β3 in the rat uterus during pregnancy ------------------------------------------------------------------ To document the presence and expression of TGF-β proteins in the uterus throughout pregnancy, IHC and Western analyses were performed on uterine sections and lysate of pregnant rats respectively. It was important to document the presence of TGF-β proteins since the information found in the literature shows principally mRNA expression of different TGF-β isoforms in specific periods of pregnancy rather than in the whole gestation period. Western blot analyses shown in figure [2](#F2){ref-type="fig"} demonstrate that TGF-β1 and β2 are both expressed in a similar pattern. Their expression is increased following implantation (days 5.5 to 6.5) and is maximal during regression of the DB (day 14; p \< 0.01). However, the localization of the expression of those two isoforms during pregnancy is slightly different (Fig. [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"} respectively): immunohistochemicals analysis confirms that during early pregnancy the signal is found in both epithelial and stromal cells, but during late pregnancy TGF-β1 is expressed mainly in stromal cells while TGF-β2 is located in epithelial cells. It is interesting to see that both TGF-β1 and β2 isoforms are clearly present in the epithelial cells and in stroma at the time of implantation (day 5.5) all around the uterine lumen surrounding the implanting conceptus. On the opposite, TGF-β3 has not been found during early pregnancy (Fig. [2](#F2){ref-type="fig"} and [5](#F5){ref-type="fig"}). However, TGF-β3 was increased and present at the time of DB regression (days 12 to 16; p \< 0.001) suggesting that its action might be limited to decidual regression during pregnancy. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### TGF-β1, -β2 and -β3 expressions in rat endometrium during rat pregnancy. Total endometrial proteins were collected at different days of pregnancy. β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis. Data represent the mean ± SEM of six independent experiments (six different rats). C+: positive control (day 14 endometrial protein extract). ::: ![](1477-7827-3-20-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Immunohistochemistry of TGF-β1 in rat endometrium during pregnancy. IHC shown are from one representative experiment and were repeated 6 times using 6 different uterine sections from 6 different rats per day of pregnancy. Representative days of pregnancy are presented (A: day 4; B: day 5.5; C: day 6.5; D: day 10; E: day 14; F: negative control in which primary antibody was absent). le: luminal epithelium; em: embryo; s: stroma. Magnification: 40×. ::: ![](1477-7827-3-20-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Immunohistochemistry of TGF-β2 in rat endometrium during pregnancy. IHC shown are from one representative experiment and were repeated 6 times using 6 different uterine sections from 6 different rats per day of pregnancy. Representative days of pregnancy are presented (A: day 4; B: day 5.5; C: day 6.5; D: day 10; E: day 14; F: negative control in which primary antibody was absent). le: luminal epithelium; em: embryo; s: stroma. Magnification: 40×. ::: ![](1477-7827-3-20-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Immunohistochemistry of TGF-β3 in rat endometrium during pregnancy. IHC shown are from one representative experiment and were repeated 6 times using 6 different uterine sections from 6 different rats per of day pregnancy. Representative days of pregnancy are presented (A: day 4; B: day 5.5; C: day 6.5; D: day 10; E: day 14; F: negative control in which primary antibody was absent). le: luminal epithelium; em: embryo; s: stroma. Magnification: 40×. ::: ![](1477-7827-3-20-5) ::: Expression of Smad2 and Phospho-Smad2 during pregnancy ------------------------------------------------------ TGF-β signal transduction is mediated intracellularly by Smad proteins, including the R-Smads (receptor regulated Smads including Smad2), the I-Smads (inhibitor Smads), and the Co-Smad (common Smad). The triggering event in Smad activation is the type I receptor-dependent sequential phosphorylation of the two C-terminal serine residues in Smads \[[@B29]\]. Therefore, phosphorylation of the C-terminus of receptor-activated Smads, particularly Smad2, is crucial for initiation of the TGF-β signaling. Western blot analyses of Smad2 and phospho-Smad2 on endometrial cell lysate from pregnant rats were carried out to confirm that TGF-β isoforms present in the endometrium during pregnancy have an activity on those cells. The results demonstrate that Smad2 expression is regulated throughout pregnancy (Figure [6](#F6){ref-type="fig"}). Its expression is high from day 5 to 10 and is dramatically reduced from day 12 to the end of pregnancy (p \< 0.05). However the levels of phospho-Smad2, the activated form, are high on days 12 and 14 and are gradually reduced to the end of pregnancy. The presence of high levels of phospho-Smad2 correlates with the high expression of the three TGF-β isoforms and the presence of apoptosis, with an exception on day 2 of pregnancy (Figure [6](#F6){ref-type="fig"}). Although high levels of phospho-Smad2 are observed, levels of TGF-β1 and -2 isoforms are higher when compared to days 4 and 5 and caspase-3 cleaved fragment is also present. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Smad2 and phospho-Smad2 (P-Smad2) expressions in rat endometrium during pregnancy. Total endometrial proteins were collected at different days of pregnancy. β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis. Data represent the mean ± SEM of six independent experiments (six different rats). ::: ![](1477-7827-3-20-6) ::: TGF-β action *in vitro*on decidual endometrial stromal cell fate ---------------------------------------------------------------- As demonstrated previously by Moulton in 1994 \[[@B7]\], TGF-β1 induces DNA cleavage in rat decidual cells. To confirm that TGF-β1 is responsible for the induction of apoptosis, three different techniques were used to measure apoptosis (Figure [7](#F7){ref-type="fig"}). Hoechst nuclear staining (Figure [7A](#F7){ref-type="fig"}) and TUNEL analysis (Figure [7C](#F7){ref-type="fig"}) clearly demonstrated that TGF-β1 induced apoptosis in a dose-dependent manner (p \< 0.0001). Apoptosis was increased to 20% at 1 ng/ml of TGF-β1 and up to 30% at 10 ng/ml. Trypan Blue exclusion staining assay was used to test for viability and cell death; even though this test is not an apoptotic specific test, it shows a direct effect of TGF-β1 on cell survival and viability. A dose-dependent increase of cell death was observed in response to TGF-β1 treatment (Figure [7B](#F7){ref-type="fig"}). ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Effect of TGF-β1 (ng/ml) on cell survival in cultured rat endometrial cells as demonstrated by Hoechst staining, TUNEL and trypan blue exclusion analyses. A) Apoptosis as determined by Hoechst nuclear staining. Data represent the mean ± SEM of six independent experiments. All doses are significantly different from control (p \< 0.001). B) Cell viability as determined by trypan blue exclusion assay. Data represent the mean ± SEM of six independent experiments. 1, 5 and 10 ng/ml TGF-β1 doses are significantly different from control (p \< 0.0001). C) Apoptosis as determined by TUNEL assay. Representative fields are presented out of 6 experiments. Arrows indicate positive staining. ::: ![](1477-7827-3-20-7) ::: Effect of TFG-β on Akt, P-Akt, CDC-47 and XIAP expression in vitro on decidual endometrial stromal cells -------------------------------------------------------------------------------------------------------- To further determine how TGF-β might act at the intra-cellular level to induce apoptosis in decidual cells, experiments were carried out to determine the possible interaction of TGF-β and the PI3K/Akt survival pathway. Recent studies have suggested and demonstrated that TGF-β directly acts with Smad3 to regulate the sensitivity to TGF-β induced apoptosis \[[@B30],[@B31]\] and another study showed that TGF-β exerts a largely inhibitory effect on basal meningioma proliferation possibly through Smad 2/3 \[[@B32]\]. Since Smad3 is directly involved in TGF-β signaling, a similar mechanism might be involved in the control of decidual cell fate. Another study has shown that Akt activity might in turn be affected by the presence or absence of inhibitor of apoptosis proteins such as XIAP \[[@B33],[@B34]\]. Figure [8](#F8){ref-type="fig"} shows that Phospho-Smad2, the activated form, is significantly increased in response to TGF-β1. However, the concentration-response relationship in term of Smad phosphorylation appears to be biphasic, with 1 ng/ml stimulating, while 10 ng/ml having a significant effect compared to control (p \< 0.05) but lower than the 1 ng/ml dose. There was no significant difference observed in term of Smad2 protein expression in response to TGF-β1. As demonstrated in Figure [8](#F8){ref-type="fig"}, the proliferation marker CDC-47 was significantly decreased in response to TGF-β1 treatment. Figure [9](#F9){ref-type="fig"} demonstrates that Phospho-Akt, the active form of Akt, was highly expressed in control cells, indicating that this pathway is active and important in cell proliferation and cell survival. However, in the presence of TGF-β1, Phospho-Akt was significantly decreased suggesting an interaction of TGF-β and the PI3K/Akt survival pathway. Total Akt protein expression was not influenced by TGF-β1 treatment. Moreover, XIAP protein expression, a well known inhibitor of apoptosis protein, was significantly reduced in response to increasing doses of TGF-β. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Expression of Smad2, P-Smad2 and CDC47 in cultured rat endometrial cells *in vitro*as demonstrated by Western blot analyses in response to TGF-β1 (ng/ml). β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis and are the mean ± SEM of four independent experiments. \*Significantly different from control (p \< 0.05). ::: ![](1477-7827-3-20-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Expression of Akt, P-Akt and XIAP in cultured rat endometrial cells *in vitro*as demonstrated by Western blot analyses in response to TGF-β1(νγ/μλ). β-actin blots shown were used as controls to correct for loading in each lane. Blots shown are from one representative experiment. Graphics represent Western blot densitometrical analysis and are the mean ± SEM of four independent experiments. \*Significantly different from control (p \< 0.05). ::: ![](1477-7827-3-20-9) ::: Discussion ========== Transforming growth factor-β isoforms have been known to be expressed and regulated differently in several types of tissues such as ovine uterus \[[@B35]\], human colon carcinoma \[[@B36]\] and at the porcine conceptus-maternal interface \[[@B19]\]. They are also recognized as pro-apoptotic factors in many cell types including fetal rat hepatocytes \[[@B14]\] and human leiomyoma smooth muscle cells \[[@B37]\]. However, little is known about the role of TGF-β isoforms in the rat uterus during pregnancy. The aim of the present study was to investigate the pro-apoptotic functions of TGF-β isoforms in rat uterus during pregnancy and also to determine more precisely the different patterns of expression of those isoforms in pregnant endometrium. The results presented in this study demonstrate that, as observed in other tissues, TGF-β isoforms (β1, β2 and β3) are differently regulated in rat uterus throughout pregnancy. It is already known that apoptosis is induced during embryo implantation and decidualization in rodents \[[@B1],[@B2],[@B7]\]. Our results confirm that apoptosis is induced in the pregnant rat uterus especially during regression of the decidua basalis. Caspases are well known executioners of apoptosis \[[@B15],[@B38]\]. The highest concentration of cleaved caspase-3 protein, the activated form of caspase-3, was found on day 14 of pregnancy, at the time of DB regression. The presence of active caspase-3 shows that this pathway might be important to induce cleavage of critical survival proteins and to further stimulate apoptosis during the regression of DB. However, caspase-3 activation was weakly observed at the time of implantation and this might be explained by the fact that, although uterine epithelium undergoes degeneration in the presence of embryo \[[@B4]-[@B6]\], only some epithelial cells undergo apoptosis during this process at the implantation site. Western blot and IHC analyses might not be sensitive enough techniques to detect the presence of small levels of cleaved caspase-3 in epithelial cells at implantation. It is also a possibility that degeneration of the epithelium may also involve non-apoptotic pathways. The logical pursuit of this study was to determine the expression of TGF-β isoforms during those critical stages of pregnancy. The present results are consistent with previous work published in regards to TGF-β isoforms in other systems and in mouse uterus \[[@B18],[@B22],[@B39]\], where expression and regulation is different for each isoform. TGF-β1 was already known to induce apoptosis in human \[[@B40]\] and rat \[[@B7]\] endometrial stromal cells. In the present study, TGF-β1 and TGF-β2 were found throughout pregnancy and were particularly strong during apoptotic phases such as regression of the DB. On the other hand, using both IHC and Western analyses, TGF-β3 protein was undetectable in the early stages of gestation, suggesting that TGF-β3 may not be required in the regulation of cell death during embryo implantation and early pregnancy. This result supports a study performed by Das et al. \[[@B18]\] showing that TGF-β3 mRNA was absent during early pregnancy. Nevertheless, the three mammalian forms of TGF-β are strongly expressed during regression of the DB suggesting they could all play an important role during the regulation of programmed cell death to induced regression of DB. Those three isoforms are also expressed differently in distinct cells of the same tissues; similarities have been observed in human normal and malignant prostate epithelial cells \[[@B39]\] as TGF-β2 and TGF-β3 are more expressed in epithelial cells than in stromal cells during late pregnancy. The strong expression of TGF-β1 during rat early pregnancy (day 5.5 to 6.5) may be explained by the fact that this isoform could be necessary to initiate embryo implantation processes during the periimplantation period, a similar situation that is observed during trophoblast invasion at the time of human embryo implantation \[[@B41]\]. The next logical step of the present study was to determine if the TGF-β isoforms present during pregnancy were active. A lot of information can be found in many physiological systems regarding TGF-β and its cellular receptors \[[@B8],[@B10]\] and Smad proteins responsible for its intracellular signal transduction \[[@B42]-[@B44]\]. Recent studies have suggested and demonstrated that TGF-β directly acts with Smad3 to regulate the sensitivity to TGF-β induced apoptosis \[[@B30],[@B31]\] and another study showed that TGF-β exerts a largely inhibitory effect on basal meningioma proliferation possibly through Smad 2/3 \[[@B32]\]. Since Smad3 is directly involved in TGF-β signaling, a similar mechanism might be involved in the control of decidual cell fate. Since it is well known that TGF-β signals through Smads proteins, Western blot and IHC analyses were performed on pregnant endometrial cell lysates to measure Smad2 and phospho-Smad2 (the activated form) to test this hypothesis. The results showed that Smad2 protein was increased and stronger at the time of embryo implantation and that Smad2 phosphorylation was gradually increased during regression of the DB, which correlates with the presence of TGF-β isoforms. These results suggest that TGF-β isoforms present during these two critical stages of pregnancy might act through Smads proteins to induce apoptosis or to induce other genes known to be regulated by TGF-βs. Smad2 protein is relatively low during regression of the DB compared to early stages of pregnancy. This result is supported by a recent study showing that increased Smad2 expression is probably caused by the invasion of the trophoblast resulting in the formation of the first decidual zone \[[@B45]\]. However, Smad2 phosphorylation is not as much increased during early pregnancy as compared to late pregnancy. It is again a possibility that, since only epithelial cells undergo apoptosis at the time of embryo implantation, the possible increase of Smad2 phosphorylation induced by TGF-β might only be observed in a small number of cells which was undetectable with the techniques used. To better understand the effect of TGF-β at the cellular level, decidual cell cultures were used to further investigate the interaction of TGF-β and the PI 3-K/Akt survival pathway. Although a recent study showed the possibility that cells obtained following artificial decidualization might be different from those obtained from pregnant animals \[[@B46]\], this model is an excellent alternative to obtain sufficient material to test the role of TGF-β and to correlates the data with the physiologic situation. It has been demonstrated recently that Akt phosphorylation is directly induced by 17β-estradiol in the ovariectomized rat uterus indicating that sex steroids have an important influence on endometrial cell fate \[[@B47]\]. Very recent studies have shown that TGF-β might directly block Akt activity through Smad activation \[[@B30]\]. Thus, stromal decidual cells were treated with different doses of TGF-β1 to determine if Smad activation might in turn block Akt survival pathway to induce cell death. TGF-β1 induced Smad2 phosphorylation in cultured endometrial stromal cell *in vitro*and triggered apoptosis in a dose-dependent manner, which was accompanied by a reduction of cell proliferation, confirming TGF-β as an apoptotic factor in rat decidual endometrial stromal cells. In response to TGF-β1, Akt phosphorylation was significantly decreased indicating that Akt activity inhibition might be an important mechanism for TGF-β-induced apoptosis in this model. Other studies have shown that Akt activity might in turn be affected by the presence or absence of inhibitor of apoptosis proteins such as XIAP \[[@B33],[@B34]\] and cIAP-1 \[[@B48]\]. The results support the hypothesis that TGF-β action via Smads not only acts at the transcriptional level to induce production of apoptotic factors in decidual cells but also at the protein level to block activation of survival factors. Akt phosphorylation was recently shown to be regulated by XIAP, a well known inhibitor of apoptosis protein, in human ovarian epithelial surface cells and in rat granulosa cells \[[@B33],[@B34]\]. The present results show a possible interaction (direct or indirect) between TGF-β1 and XIAP protein expression. TGF-β1 reduced XIAP expression *in vitro*in a dose dependent manner. Recent studies have demonstrated that XIAP can act as a cofactor in the regulation of gene expression induced by TGF-β and is independent of Smad4 \[[@B14],[@B49]\]. Because TGF-β1 reduced Akt phosphorylation and that XIAP was shown to induce Akt phosphorylation \[[@B34]\], it is possible that TGF-β action in this case might be through Smad2 activation and XIAP gene expression which in turn would act on Akt phosphorylation. Further experiments will be necessary to have a better understanding of the interactions between TGF-β and Akt survival pathway. In particular, how specifically Smads act on XIAP expression and how XIAP acts on the PI 3-K/Akt survival pathway activity. Other factors such as Smac/Diablo, a XIAP intracellular inhibitor, which was demonstrated as being regulated by 17β-estradiol in the rat during estrous cycle \[[@B50]\], might also be a putative candidate for regulation of TGF-β activity in decidual cells. Conclusion ========== In conclusion, this study demonstrates that the three isoforms of TGF-β are differently localized and regulated in endometrial cells of pregnant rats, particularly at the time of implantation and regression of the DB. The present study also showed that TGF-β plays an important role in the control of cell survival and cell death and that it may interacts with the PI 3-K/Akt survival pathway through Smad activation to allow apoptosis induction. Further experiments will be necessary to understand more precisely the effect of other Smads and/or co-Smads proteins during TGF-β-induced apoptosis. Other investigations will also be required to better understand the specific roles of TGF-β2 and TGF-β3 at the intracellular and molecular level *in vitro*to determine how and if these isoforms control cell survival through Smad signal transducers. Authors\' contributions ======================= CS drafted the paper. CS, PLC, GFF, VL and MCD performed the experiments. EA conceived the study, participated in its design and coordination, and wrote final version of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work has been supported by a grant from NSERC (238501-01). Eric Asselin is a chercheur-boursier from the Fond de la Recherche en Santé du Québec (FRSQ) and New Investigator of the Canadian Institute of Health Research of Canada (CIHR). Marie-Claude Déry is recipient of a FRSQ and NSERC studentships. We are grateful to Mrs Rollande Caron for the contribution of her precious time and expertise to our projects. We thank also Mrs Sophie Parent and Mrs Daphne Efford for reviewing the manuscript.

PubMed Central

2024-06-05T03:55:58.170260

2005-5-31

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166574/", "journal": "Reprod Biol Endocrinol. 2005 May 31; 3:20", "authors": [ { "first": "Carl", "last": "Shooner" }, { "first": "Pierre-Luc", "last": "Caron" }, { "first": "Guylaine", "last": "Fréchette-Frigon" }, { "first": "Valérie", "last": "Leblanc" }, { "first": "Marie-Claude", "last": "Déry" }, { "first": "Eric", "last": "Asselin" } ] }

PMC1166575

Background ========== The vertical transmission of human immunodeficiency virus type 1 (HIV-1) accounts for more than 90% of all HIV-1 infections in children. HIV-1 infected pregnant women can transmit the virus to their infants during all stages of their pregnancy, including prepartum (trans-placental passage), intrapartum (exposure of infants\' skin and mucous membranes to contaminated maternal blood and vaginal secretions) and post-partum (via breast milk) at an estimated rate of 30% \[[@B1]-[@B4]\]. However, the rate of vertical transmission can be reduced by antiretroviral therapy during pregnancy. The risk of vertical transmission increases with several parameters, including advanced maternal disease status, low maternal CD4 cell count, high maternal viral load, recent infection of the mother, prolonged exposure of infant to ruptured membranes during parturition, and higher viral heterogeneity in the mother \[[@B5]-[@B8]\]. Viral heterogeneity is one of the classical means by which HIV-1 evades the host immune system. The heterogeneity of HIV-1 is attributed to the error-prone reverse transcriptase (RT) enzyme, which is responsible for converting the single stranded viral genomic RNA to double-stranded DNA that integrates into the host chromosome. As reverse transcription is the first step of the viral replication cycle \[[@B9]\], errors made at this stage ensures propagation of the erroneously copied genome to form the quasi-species of HIV-1 found in the infected individuals. These quasi-species infect other uninfected target cells and the cycle of error-prone reverse transcription continues. We have previously demonstrated that HIV-1 sequences from transmitting mothers (mothers who transmitted HIV-1 to their infants) were more heterogeneous compared with HIV-1 sequences from non-transmitting mothers (mothers who failed to transmit HIV-1 to their infants) \[[@B10]\]. This finding further suggests that the reverse transcription step that is responsible for generation of viral heterogeneity, may also play an important role in vertical transmission. The RT gene is unique in that it is also exposed to the same mutating effects of the RT enzyme as other part of the HIV-1 genome. Therefore, we sought to examine HIV-1 RT sequences from five infected mother-infant pairs following perinatal transmission. The HIV-1 RT shows significant sequence and structural similarity to other viral reverse transcriptases as well as viral and bacterial RNA polymerases \[[@B11]-[@B13]\]. HIV-1 RT is a heterodimeric protein comprising of two subunits, 66 kDa and 51 kDa. It is encoded as a Gag-Pol precursor, Pr160^gag-pol^, which is cleaved by viral protease to yield the Gag protein and the viral polymerase which codes for RT \[[@B9],[@B14]\]. The larger subunit (p66) of the heterodimer acts as an RNA-dependant DNA polymerase, a DNA-dependant DNA polymerase and has RNase H activity associated with the C-terminus \[[@B15],[@B16]\], whereas the p51 subunit lacks the C-terminus RNase H activity, is folded differently from the p66 subunit and is thus inactive \[[@B17]-[@B20]\]. The p66 is folded to form a structure similar to a right hand with palm, finger and thumb subdomains \[[@B21]-[@B23]\] that are connected to the RNase H by the \"connexion\" subdomain \[[@B22],[@B24],[@B25]\]. Each domain has several secondary structural elements which are critical for primer binding, template binding \[[@B14],[@B22],[@B23],[@B26],[@B27]\] and nucleotide recruitment \[[@B28]\]. More specifically, the aspartate residues at position 110, 185 and 186 are believed to be the active sites of the polymerase and are located in the palm subdomain at the bottom of the DNA binding cleft \[[@B14],[@B16],[@B20],[@B28],[@B29]\]. Mutations in this subdomain and the active site abolish the enzymatic activity of HIV-1 RT \[[@B2],[@B19],[@B22],[@B30]-[@B32]\] and alter viral replication, which may also affect HIV-1 mother-to-infant transmission. In this study, we characterized the HIV-1 RT quasi-species from five mother-infant pairs following vertical transmission, including a mother with infected twin infants. We show that the open reading frame of the RT gene was highly conserved in the sequences from five mother-infant pairs. In addition, there was a low degree of heterogeneity and high conservation of functional domains essential for RT activity. These findings may be helpful in the understanding of the molecular mechanisms of HIV-1 vertical transmission. Results ======= Patient population and sample collection ---------------------------------------- Blood samples were collected from five HIV-1-infected mother-infant pairs following perinatal transmission, including samples from a set of twins (IH1 and IH2) in the case of mother H. The demographic, clinical and laboratory findings on these mother-infant pairs are summarized in Table [1](#T1){ref-type="table"}. The Human Subjects Committee of the University of Arizona, and the Institutional Review Board of the Children\'s Hospital Medical Centre, Cincinnati Ohio, approved this study. Written informed consent was obtained for participation in the study from mothers of infected mother-infant pairs. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic, Clinical, and Laboratory Parameters of HIV-1 Infected Mother-Infant Pairs ::: Patient Age Sex CD4+ cells/mm3 Length of infection ^a^ Antiviral drug Clinical Evaluation ^b^ --------- --------- ----- ---------------- ------------------------- ---------------- ---------------------------------------------- MB 28 yr 509 11 mo None Asymptomatic IB 4.75 mo M 1942 4.75 mo None Asymptomatic, P1A MC 23 yr 818 1 yr6 mo None Asymptomatic IC 14 mo F 772 14 mo ZDV Symptomatic AIDS;P2A,D1,3,F MD 31 yr 480 2 yr6 mo None Asymptomatic ID 28 mo M 46 28 mo ddC^c^ Symptomatic AIDS, P2AB,F; failed ZDV therapy MF 23 yr 692 2 yr10 mo None Asymptomatic IF 1 wk M 2953 1 wk ZDV Asymptomatic,P1A MH 33 yr 538 5 mo None Asymptomatic IHT1 7 mo F 3157 7 mo ACTG152 Hepatosplenomeglay lymphadenopathy IHT2 7 mo F 2176 7 mo ACTG152 Hepatosplenomegaly lymphadenopathy M: mother; I: infant. ^a^Length of infection: The closest time of infection that we could document was the first positive HIV-1 serology date or the first visit of the patient to the AIDS treatment Center, where all the HIV-1 positive patients were referred to as soon as an HIV-1 test was positive. Therefore, these dates may not reflect the exact dates of infection. ^b^Evaluation for infants is based on CDC criteria, ^c^ddC, Zalcitibine ::: Phylogenetic analysis of RT sequences of mother-infant isolates --------------------------------------------------------------- We first performed multiple independent polymerase chain reaction (PCR) amplifications from peripheral mononuclear cells (PBMC) DNA of five mother-infant pairs and obtained 10 to 14 clones from each patient followed by nucleotide sequencing of these clones. We then performed the phylogenetic analysis by constructing a neighbor-joining tree of the 132 RT sequences from these mother-infant pairs, including the set of twins from mother H and the reference strain NL4-3, as shown in Figure [1](#F1){ref-type="fig"}. A model of evolution was optimized for the entire nucleotide sequence data set using the approach outlined by Huelsenbeck and Crandall \[[@B33]\]. The model of choice was incorporated into PAUP \[[@B34]\] to estimate a neighbor-joining tree and the tree was bootstrapped 1000 times to ensure fidelity. The phylogenetic tree demonstrated that the RT sequences from five mother-infant pairs were well discriminated in separate clusters and that the mother and infant sequences were generally separated in distinct subclusters. However, there was some intermingling between mother and infant sequences in pair C. Furthermore, the formation of separate subclusters of RT sequences from twins of mother H suggests that the there was probably compartmentalization of HIV-1 in the two fetuses causing independent evolution. We also compared our mother-infant pairs\' RT sequences with the RT sequences of several clades present in the HIV databases and found that our RT sequences grouped with clade or subtype B sequences (not shown). The data on phylogenetic analysis indicate that the epidemiologically linked mother-infant sequences are closer to each other than epidemiologically unlinked sequences and that there was no PCR cross contamination. It is important to note that the mother-infant pairs grouped in the same subtree, even when some of the infants\' ages were more than 2 to 3 years, suggesting that the epidemiological relationships are maintained in mother-infant pairs no matter how long the infection in the infants has progressed. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Phylogenetic analysis of HIV-1 RT of 132 RT sequences from five mother-infant pairs, including B, C, D, F and H. The neighbor-joining tree is based on the distance calculated between the nucleotide sequences from the five mother-infant pairs. Each terminal node represents one RT gene sequence. The numbers on the branch points indicate the percent occurrence of branches over 1,000 bootstrap resamplings of the data set. The sequences from each mother formed distinct clusters and are well discriminated and in confined subtrees, indicating that the variants from the same mother-infant pair are closer to each other than to other sequences and that there was no PCR cross-contamination. These data were strongly supported by the high bootstrap values indicated on the branch points. ::: ![](1742-4690-2-36-1) ::: Coding potential of RT gene sequences ------------------------------------- The multiple sequence alignments of the deduced amino acid sequences of HIV-1 RT genes from five mother-infant pairs, B, C, D, F, mother H and her twin infants IH1 and IH2 are shown in Figures [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"}, [6](#F6){ref-type="fig"}, and [7](#F7){ref-type="fig"}, respectively. These sequences were aligned with consensus subtype B RT sequence (CON B). We found that 115 of the 132 sequences analyzed contained a complete RT open reading frame (ORF), with an 87.2% frequency of intact RT open reading frames thus indicating that the coding potential of the RT ORF was maintained in most of the sequences in 1680 bp sequenced. Moreover, the infected mothers\' sequences showed a frequency of 85.5% of intact RT ORF while infants demonstrated a frequency of 88.5%. Several clones in mother-infant pair B and mother H were found to be defective due to a single nucleotide substitution, insertion or deletion resulting either in frame-shift or stop codons. The RT sequences also displayed patient and pair specific amino acid sequence patterns. Several amino acid motifs changes were observed in majority of the mother-infant pairs\' sequences, including a glutamic acid (E) or proline (P) at position 122, an arginine (R) at 277, and a threonine (T) or serine (S) at 376 and 400. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother-infant pair B involved in vertical transmission. In the alignment, the top sequence is the consensus RT sequence of subtype or clade B (CON B) to which mother-infant pair-B RT sequences are aligned. In mother-infant pair B sequences, each line refers to a clone identified by a clone number with M referring to mothers and I referring to infants. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT activity are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother-infant pair C in reference to consensus subtype B (CON B) RT sequence. In the alignment, the top sequence is CON B RT sequence and the bottom sequences are mother-infant pair C sequences (M refers to mother sequences and I to sequences). The number of clones sequenced is represented with clone numbers. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Spanning arrowheads indicated above the alignment shows relevant amino acid motifs and domains essential for RT function. ::: ![](1742-4690-2-36-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother-infant pair D. The patient sequences are aligned in reference to consensus RT sequence of HIV-1 subtype or clade B (CON B) at the top. In the mother-infant pair sequences, each line refers to a clone identified by a clone number with M referring to mother and I to infants. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT activity are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase gene from mother-infant pair F. In the alignment, the top sequence (CON B) is the consensus subtype B RT sequence and the bottom sequences are from mother-infant pair F sequences (M stands for mother sequences and I for infant sequences and the number of clones for mother and infant are indicated by clone number). The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT functions are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase (RT) gene from mother H, who had given birth to infected twins, H1 and H2 (alignment shown in Figure 7). In the mother H sequences, each line refers to a clone identified by a clone number with M referring to mother. The mother sequences are aligned in reference to consensus RT sequence of HIV-1 subtype or clade B (CON B) shown at the top. The structural elements of RT are indicated above the alignment. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Spanning arrowheads indicated above the alignment shows relevant amino acid motifs and domains required for RT activity. ::: ![](1742-4690-2-36-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Multiple sequence alignment of deduced amino acids of HIV-1 reverse transcriptase gene (RT) from infected twin infants, H1 and H2 of mother H (alignment shown in Figure 6). In the alignment, the top sequence is the consensus subtype B RT sequence (CON B) and the bottom sequences are of infants H1 and H2 represented by I and clone numbers. Dots represent amino acid agreement with CON-B and substitutions are shown by single letter codes for the changed amino acid. Stop codons are shown as x and dashes represent gaps or truncated protein. Relevant amino acid motifs and domains essential for RT activity are shown by spanning arrowheads indicated above the alignment. ::: ![](1742-4690-2-36-7) ::: Variability of RT gene sequences in mother-infant isolates ---------------------------------------------------------- The degree of genetic variability of RT sequences, measured as nucleotide and amino acid distances based on pairwise comparison (as described in Methods), was determined for the five mother-infant pairs\' sequences, and is shown in Table [2](#T2){ref-type="table"}. The nucleotide sequences of RT within mothers (mothers B, C, D, F and H) differed by 0.80, 1.76, 1.37, 1.21 and 2.90% (median values), respectively, ranging from 0 to 3.46%. The variability in the infant sets (infants B, C, D, F, H1 and H2) was similar to the mother sequences and differed by 0.80, 1.49, 1.37, 1.31, 0.64 and 1.24% (median values), respectively, ranging from 0 to 2.21%. Interestingly, the variability between epidemiologically linked mother and infant sets (pairs B, C, D, F and H) was also on the same order of 1.05, 1.7. 1.74, 1.22 and 1.45 (median values) respectively, ranging from 0 to 4.48%. Moreover, the amino acid sequence variability of RT within mothers (mothers B, C, D, F and H) differed by 1.26, 2.81, 1.98, 1.26 and 2.27% (median values), respectively, ranging from 0 to 5.51%. The variability within infants (infants B, C, D, F, H1 and H2) differed by 1.44, 2.35, 1.80, 1.62, 1.44 and 1.62% (median values), ranging from 0 to 4.57%, and between mother-infant pairs (pairs B, C, D, F and H) by 1.44, 2.90, 2.53, 1.44 and 2.17% (median values), ranging from 0 to 6.47%, respectively. We also determined sequence variability between epidemiologically unlinked individuals and found that the nucleotide distances ranged from 0 to 9.1% (median 5.4%) and amino acid from 0 to 12.4% (median 6.34%). The variability in general was lower between epidemiologically linked mother-infant pairs\' sequences than epidemiologically unlinked individuals, suggesting that epidemiologically linked mother-infant pair sequences are closer to each other. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distances in the RT sequences within mother sets, within infant sets, and betweenmother-infant pairs ::: **Nucleotide distances** -------------------------- -------------------- -------------------------------- ---------- ----------- --------- ---------- ---------- ----------- --------- ---------- ---------- **Within mothers** **Within infants** **Between mother and infants** Pair Min Med Max Pair Min Med Max Pair Min Med Max MB 0.0 0.80 2.10 IB 0.0 0.80 1.30 B 0.0 1.05 2.05 MC 0.0 1.76 3.46 IC 0.0 1.49 2.17 C 0.0 1.70 3.26 MD 0.0 1.37 2.21 ID 0.0 1.37 2.21 D 0.0 1.74 4.48 MF 0.0 1.21 1.54 IF 0.0 1.31 2.93 F 0.0 1.22 2.08 MH 0.0 2.90 2.60 IH1 0.0 0.64 1.34 H 0.0 1.45 3.30 IH2 0.0 1.24 1.75 **Total** **0.0** **1.34** **3.46** **Total** **0.0** **1.48** **2.21** **Total** **0.0** **1.32** **4.48** **Amino acid distances** **Within mothers** **Within infants** **Between mother and infants** Pair Min Med Max Pair Min Med Max Pair Min Med Max MB 0.0 1.26 4.61 IB 0.0 1.44 2.72 B 0.0 1.44 4.57 MC 0.0 2.81 5.51 IC 0.0 2.35 4.01 C 0.0 2.90 5.51 MD 0.0 1.98 3.83 ID 0.0 1.80 4.57 D 0.0 2.53 6.47 MF 0.0 1.26 2.35 IF 0.0 1.62 3.09 F 0.0 1.44 3.09 MH 0.0 2.27 3.09 IH1 0.0 1.44 2.17 H 0.0 2.17 6.27 IH2 0.0 1.62 2.72 **Total** **0.0** **1.52** **5.51** **Total** **0.0** **1.42** **4.57** **Total** **0.0** **2.90** **6.47** M: mother; I: infant. Min: Minimum; Med: Median; Max: Maximum. Totals were calculated for all pairs together ::: We also investigated if the low variability of RT sequences seen in our mother-infant pair isolates is due to errors made by LA Taq polymerase used in our study. We did not find any errors made by the LA Taq polymerase when we used a known sequence of HIV-1 NL 4--3 for PCR amplification and DNA sequencing of the RT gene. Dynamics of HIV-1 RT gene evolution in mother-infant isolates ------------------------------------------------------------- The maximum likelihood estimates and chi square tests performed by Modeltest 3.06 \[[@B35]\] suggested different models of evolution for each patient sample. The estimates of genetic diversity of RT sequences from the five mother-infant pairs were determined by using the Watterson model, assuming segregating sites and the Coalesce method assuming a constant population size. The estimates of genetic diversity shown as theta values (estimated as nucleotide substitutions per site per generation) are shown in Table [3](#T3){ref-type="table"}. The levels of genetic diversity among infected mothers and infants, as estimated by Watterson method, ranged from 0.012 to 0.025 and 0.009 to 0.021, respectively. Similar results were obtained when the mother-infant pair populations were analyzed by the Coelesce method, with the values ranging from 0.020 to 0.058 in mothers and from 0.016 to 0.060 in infants. These data suggest that the mother and infant populations evolved very slowly and at similar rates. The differences observed in the estimates of genetic diversity between and mothers and infants sequences are not statistically significant. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Estimates of genetic diversity of HIV-1 RT within mother sets and infant sets ::: **MOTHERS** **INFANTS** ------------- ------------- ----------- ----------- ----------- -------- ----------- ----------- **N** **θ~w~** **θ~c~** **θ~w~** **θ~c~** Mother B 12 0.015 0.038 Infant B 12 0.014 0.033 Mother C 12 0.025 0.058 Infant C 13 0.021 0.060 Mother D 11 0.017 0.042 Infant D 10 0.019 0.040 Mother F 14 0.012 0.029 Infant F 12 0.018 0.053 Mother H 14 0.020 0.020 Infant H1 11 0.009 0.016 Infant H2 11 0.015 0.044 **Totals** **63** **0.018** **0.037** **69** **0.016** **0.041** N -- number of RT clones sequenced. θ~w~-- genetic diversity as calculated by the Watterson method; θ~c~-- genetic diversity as calculated by the Coelesce method. Totals were indicated as an average of all values. ::: Rates of accumulation of nonsynonymous and synonymous substitutions ------------------------------------------------------------------- Selection pressure on the RT gene was estimated as a ratio of accumulation of non-synonymous to non-synonymous substitutions using the Nielsen and Yang model \[[@B36]\] as implemented in codeML \[[@B37]\]. Although there are several models to predict the rate of positive selection, most of these models assume that all sites in a sequence are under the same selection pressure with the same underlying dN/dS ratio \[[@B38]\]. As substitutions of critical regions of a protein can lead to deleterious mutations, it is unrealistic to make assumptions about equal degree of selection throughout the protein. In cases where positive selection is operating on proteins, it has been shown that only a limited number of amino acids may be responsible for adaptive evolution. In such a case, methods that estimate dN/dS ratios over an entire sequence may fail to detect positive selection even when it exists \[[@B39]\]. The codeML method uses the codon as a unit of evolution as opposed to a nucleotide, and thus allows us to estimate the percentage of positions that are being positively selected instead of averaging the rates of positive selection over the entire gene \[[@B39]\]. This method also provides the percentage of mutations that are conserved, neutral or positively selected based on dN/dS values of 0, 1 or \> 1, respectively. The dN/dS values as well as the proportions of each site category estimated using the Nielsen and Yang model are shown in Table [4](#T4){ref-type="table"}. As described in the methods, a dN/dS value of greater than 1 suggests positive selection. The percentage of the substitutions being positively selected is shown in column p3. Except for viral populations in infants C and F, all isolated populations were associated with dN/dS ratio \>1, indicating positive selection. In case of infants C and F, there was no positive selection on the mutations and most of the substitutions were neutral. All mothers generally displayed a higher proportion of positively selected p3 sites as compared to the infants. Although the dN/dS values for infant H1 and H2 seem higher than mother H, closer observation shows that the percentage of sites undergoing positive selection is higher in the mother than in the twin infants. Table [4](#T4){ref-type="table"} shows that in mothers, over half the sites (66.6%) belong to the conserved p1 category, whereas the frequency of neutral and positively selected sites was equally distributed. This is in contrast to the viral population from the infants where the conserved site category (p1) had a frequency of only 36.5% and close to half the sites (55.7%) belongs to the neutral p2 category. Statistical analysis revealed that only the proportion of the neutral p2 category was significantly different between mothers\' and infants\' sequence viral populations (p \< 0.05). This is signified by the case that all the sites in Infant F belonged to the p2 category. Higher proportion of p2 sites in infants have also been shown in the nef gene product in these same mother infant pairs \[[@B40]\]. The variable (positively selected) sites (p3) in the mothers\' sequences were associated with dN/dS ratios that ranged from 2.34 to 8.9, with viral sequence populations from three mothers (MD, MF, MH) that displayed a dN/dS ratio of below three. This is in contrast to the infants\' viral populations that were either associated with a dN/dS of below 1, indicating no directional selection (IC and IF), a dN/dS ratio between 3 and 4 (IB and ID) or a very high dN/dS ratio as found in the sequences isolated from the twins H1 and H2. This analysis showed that the RT gene in both the mothers and infants is under positive selection pressure. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### dN/dS values in HIV-1 RT sequences within mother sets and within infant sets. ::: **MOTHER** **INFANT** ------------ ------------ ---------- ---------- ---------- ----------- ----------- -------- ---------- ---------- --------- ----------- **N** **P1** **P2** **P3** **dN/dS** **N** **P1** **P2** **P3** **dN/dS** Mother B 12 53 18.8 27 8.9 Infant B 12 41 42 16 3.31 Mother C 12 55.5 43 1.3 6.09 Infant C 13 0 81.2 18.8 0.01 Mother D 11 70.6 5.7 23.6 2.52 Infant D 10 74.8 19.2 5.9 4.44 Mother F 14 81.7 7.8 10.4 2.67 Infant F 12 0 100 0 0.001 Mother H 14 72 0 27 2.34 Infant H1 11 47 50 2.8 14.04 Infant H2 11 56 42 0.6 16.58 **Totals** **66.5** **15.1** **18.4** **4.50** **69** **36.5** **55.7** **7.8** **6.39** N -- number of RT clones sequenced.; P1 = proportion of conserved codons as a percent; P2 = proportion of neutral codons as a percent; P3 = proportion of positively selected codons as a percent. dN/dS = ratio of synonymous to non-synonymous at P3 sites. Totals were calculates as an average of all values. ::: Analysis of functional domains of RT in mother-infant pairs ----------------------------------------------------------- HIV-1 RT is a heterodimeric protein comprising of two subunits, p66 and p51. The larger subunit of the heterodimer acts as an RNA-dependant DNA polymerase, a DNA-dependant DNA polymerase and an RNase H that is associated with the C-terminus \[[@B15],[@B16]\]. The p66 is folded to form a structure similar to the right hand with palm, finger and thumb subdomains \[[@B21],[@B23],[@B32]\] that are connected to the RNase H by the \"connexion\" subdomain \[[@B22],[@B24],[@B25]\]. Each domain has several secondary structural elements, which are critical for primer binding, template binding \[[@B14],[@B22],[@B23],[@B26],[@B27],[@B41]\] and nucleotide recruitment \[[@B28]\]. The active sites of the polymerase comprise of aspartic acid (D) residues at positions 110, 185 and 186, which are located in the palm subdomain at the bottom of the DNA binding cleft \[[@B22],[@B23]\]. Mutations of these aspartic acid residues abrogates the polymerase activity of RT \[[@B22],[@B23],[@B29],[@B32]\]. These aspartate residues of the RT active site were conserved within the five mother-infant pairs RT sequences. Furthermore, the D185 and D186 that form a part of an essential highly conserved YMDD \[[@B32],[@B42],[@B43]\] motif involved in binding to the 3\'OH of the primer strand \[[@B14],[@B26]\], were highly conserved in our mother-infant pairs\' RT sequences (Figures [2](#F2){ref-type="fig"} to [7](#F7){ref-type="fig"}). The amino acids at positions 73--90 that constitute the template grip required for positioning and binding the RT template near the active site of the RT \[[@B23]\], were also conserved in most of our RT sequences. The primer grip responsible for primer binding extends from amino acids 227 to 235 \[[@B22],[@B23]\] and these amino acids were also conserved in the mother-infant RT sequences. The K263, K353 and R358 that form salt bridges with the phosphate groups \[[@B14],[@B21],[@B22],[@B30],[@B44]\] of the template and primer were found to be conserved in most of the RT sequences analyzed. The thumb subdomain of RT is comprised of two anti-parallel α helices, αH and αI, which bind to the opposite strand of dsDNA. The αH also directly inserts into the minor groove of the DNA \[[@B14],[@B22],[@B41]\]. Both these helices were generally conserved in our mother-infant RT sequences. The connexion subdomain that links the RT to the RNase H and forms the floor of the template binding cleft \[[@B22],[@B24],[@B25],[@B42]\], showed some substitutions, including V293I, A376S and A400T in our mother-infant RT sequences. Mutations at positions H361 and Y501 reduces RNase H activity \[[@B24]\]. Examination of the five mother-infant pairs\' sequences revealed that these two positions were intact in all RT sequences (Figures [2](#F2){ref-type="fig"} to [7](#F7){ref-type="fig"}). Furthermore, the RNase H active sites contain four acidic amino acid residues, D443, E478, D498 and D549 \[[@B22],[@B24],[@B25],[@B41],[@B42]\], which were highly conserved in our mother-infant pairs sequences. In addition, several substitutions were seen in regions of RT that are not known to have critical function. The relevance of these changes is not known. Mutations associated with anti-retroviral drug resistance --------------------------------------------------------- Several naturally occurring mutations in the *pol*gene in treatment-naïve patients have been reported \[[@B45],[@B46]\], although most of these mutations are not seen in our RT gene sequences. In addition, these mutations found in treatment-naïve patients were usually seen in non-subtype B infections and our patient population was from subtype B infected individuals. These changes were usually in amino acids where the mutations did not actually confer nucleoside reverse transcriptase inhibitor (NRTI) drug resistance but were accessory mutations \[[@B46]-[@B48]\]. Several amino acid changes in RT seen in patients undergoing NRTI therapy are selected primarily with zidovudine (ZDV) treatment. These mutations referred to as thymidine analog mutations (TAMs) include M41L, D67N, K70R, L210N, T215Y/F and K219Q \[[@B47],[@B49]\]. Since most of our infected mothers were treatment naïve but infants were actively on ZDV therapy or on other drugs (Table [1](#T1){ref-type="table"}), we examined the RT sequences for ZDV resistant mutations (Figure [2](#F2){ref-type="fig"}). Several TAMs associated with drug resistance were observed in our infants C and D who were either on prolonged or failed ZDV therapy. These mutations included M41L in three clones from infant C and two clones in infant D, D67N and K70R in five clones from infant C, L210W in one clone from infant D and T215F in seven clones from infant D and K219Q in four clones from infant C and D. In addition, one clone from infant C had all the above mutations, indicating significant resistance to ZDV \[[@B46],[@B50]\]. Although Mother C was not on any antiretroviral therapy two clones had TAMs at M41L and K219Q positions, suggesting that these mutations were naturally occurring. It is interesting to note that the infant of this mother yielded several clones with these two mutations. An R211K mutation known as an accessory mutation associated with NRTI resistance \[[@B46]\] was also observed in all mother-infant pair H clones. Immunologically relevant mutations in the CTL epitopes of RT ------------------------------------------------------------ The cytotoxic T lymphocyte (CTL) responses have been shown to exert significant immune pressure during HIV-1 infection. Strong CTL responses are maintained in long-term nonprogressors and these responses correlate with decrease in viral load \[[@B51]-[@B55]\]. It has been shown that transmitting mothers have larger numbers of CTL escape variants as compared to non-transmitting mothers \[[@B56]\], emphasizing that CTL escape variants may become a part of circulating virus that influences vertical transmission \[[@B56],[@B57]\]. Several regions in the RT gene have been shown to elicit strong CTL responses during HIV-1 infection. The CTL eptitope, TVLDVGDAY, between amino acid positions 107--115 <http://www.hiv.lanl.gov/content/immunology/ctl_search>, is highly conserved among known HIV-1 isolates \[[@B57]\]. This epitope contains the amino acid D110 which is part of the RT active site. This epitope was highly conserved in most of the mother-infant RT clones sequenced (Fig. [2](#F2){ref-type="fig"}). Another motif, TAFTIPSI, between amino acid positions 128--135 is an HLA-B51 restricted epitope <http://www.hiv.lanl.gov/content/immunology/ctl_search>. This epitope is present in the palm region consisting of positions A129 and I135 as anchor residues \[[@B57]\]. This motif was mostly conserved in the RT sequences of the five mother-infant pairs analyzed. In addition, I135T mutation decreases CTL response but increasing concentration of mutant peptide re-establishes appropriate responses \[[@B57]\]. The I135T mutation was seen in several of our mother-infant pair\'s D sequences. The next motif AIFQSSMTK from amino acid positions 158--166, comprising of I159, F160, K166 anchor residues and recognized by several HLA types, is conserved among known HIV-1 isolates and believed to be associated with vertical transmission \[[@B56],[@B57]\]. Our mother-infant pairs\' RT sequences showed conservation in this motif. Another CTL epitope YPGIKVRQL from positions 271--279 has been reported to be conserved in transmitting mothers and infants with several natural occurring variants \[[@B56]\], was also found to be conserved in our mother-infant pairs\' RT sequences. In addition, a P272H mutation that causes significant loss of CTL response for this epitope \[[@B56]\] was not seen in any of the RT clones analyzed. Discussion ========== In this study, we show for the first time that reverse transcriptase open reading frames from five mother-infant pairs following perinatal transmission were maintained with a frequency of 87.2%. The functional domains required for reverse transcriptase activity in HIV-1 replication were highly conserved in most of the mother-infants sequences. We also demonstrate a low degree of sequence variability and estimates of genetic diversity for reverse transcriptase genes after mother-to-infant transmission. However, epidemiologically unlinked individual\'s sequences were more heterogeneous than epidemiologically linked mother-infant pair\'s sequences. Several motifs in reverse transcriptase responsible for primer and template binding and positioning and motifs involved in nucleotide recruitment were conserved in all mother-infant pairs\' sequences. The data we show here are comparable to those of our previously analyzed conserved genes, including *gag*P17MA, *vif*, *vpr*, *tat*and *nef*\[[@B58]-[@B62]\]. Our findings suggest that an intact and functional reverse transcriptase open reading frame is essential for HIV-1 replication in mothers and their infants and low degree of viral heterogeneity is maintained following vertical transmission. The RT open reading frame was maintained in 115 of the 132 sequences (1680 base pairs sequenced), whereas 17 sequences contained stop codons (Figure [2](#F2){ref-type="fig"}). The frequency of conservation in five mother-infant pairs was found to be 87.2%. The comparison of the RT sequences with those of other conserved genes from HIV-1 infected mother-infant pairs showed comparable frequency of conversation, including *gag*p17 (86.2%), *vif*(89.8%), *vpr*(92.1%), *tat*(90.9%), *nef*(86.2%) and *vpu*(90.12%). There was no significant correlation between the conservation of RT open reading frame and disease progression in mothers and infants \[[@B63]-[@B65]\]. Several amino acid motifs were found to be a signature characteristic of each mother-infant pair, even in older infants where infection has progressed for more than 3 years. Phylogenetic analysis of the RT sequences revealed that the five mother-infant pairs were well discriminated, separated and confined within subtrees (Fig. [1](#F1){ref-type="fig"}), indicating that the epidemiologically linked mother-infant pairs were closer to each other and that there was no PCR product cross-contamination \[[@B66],[@B67]\]. In addition, most of the mother and infant sequences of the same pair formed separate subclusters, with little intermingling between sequences of mother and infant in some pairs. In some mother-infant pairs, minor variants of the mothers seem to be predominating in the infants, which was also seen in our previous V3 region analysis \[[@B68]\]. We also observed intermingling of sequences in mother-H and her infected twins, indicating that different mother\'s variants were transmitted to the twins. With respect to viral heterogeneity, there was a low degree of genetic variability in the RT sequences from mother-infant pairs estimated by several methods. Similar levels of genetic diversity were seen in other conserved genes of the same mother-infant pairs, including *gag*, *vif*, *vpr*and *tat*\[[@B59]-[@B61],[@B69]\]. The low degree of genetic variability was observed in RT sequences of mothers and maintained in the infants following transmission, suggesting the essential nature of this gene in viral pathogenesis. It is important to note that the mother-infant pairs retained the same epidemiological relationship, even when some of the infant\'s age was more than 2 to 3 years. We believe this is an important finding that the epidemiological relationships as well as certain signature sequence motifs are maintained in mother-infant pairs or transmitter-recipient partners no matter how long the infection has progressed. This information may be critical in terms of vaccine development. Examining the motifs of the deduced amino acid sequences of the RT gene from five mother-infant pairs, we found that the essential motifs required for RT activity were mostly conserved in our mother-infant pairs\' sequences (Figure [2](#F2){ref-type="fig"}). The sites essential for primer binding, template binding, positioning of template and primer, which are located in α-Helix H and α-Helix I \[[@B22],[@B23]\], were are all conserved in RT sequences (Figure [2](#F2){ref-type="fig"}). Specifically, the amino acids involved in recruitment of nucleotides during reverse transcription \[[@B28]\] were mostly conserved. The active sites of the polymerase are located in the palm subdomain at the bottom of the DNA binding cleft comprising of aspartic acid (D) residues at positions 110, 185 and 186 were conserved within the five mother-infant pairs\' RT sequences. Furthermore, the D185 and D186 also form a part of an essential YMDD motif, which is highly conserved in known HIV-1 isolates \[[@B14],[@B22],[@B23],[@B26],[@B32],[@B43]\], was also conserved in our mother-infant pairs\' RT sequences analyzed. Some of the amino acids of the connexion subdomain that are critical for RNase H activity and replication \[[@B9],[@B24],[@B25]\] are conserved in our RT sequences with several substitutions of compatible nature, including V293I, K358R, A376S, and A390T. These substitutions were located in the regions of the connexion that forms the base of the binding cleft. It is possible that such mutations in the binding cleft may change the size of the cleft and affect fidelity of the reverse transcriptase without affecting the active site. Further assessment also shows that our RT sequences harbor mutations in the connexion and RNase H subdomains that are not at the critical sites required for RT activity. The implications of these mutations can be studied by performing the biological characterization of these RT clones in the context of HIV-1 replication. It would be interesting to determine whether the degree of genetic variability and conservation of RT functional domains in non-transmitting mothers and compare their sequences with the data presented here. Nonetheless, the data described here suggest that functional domains of the RT enzyme, including reverse transcriptase, DNA polymerase and RNase H, were highly conserved in our five mother-infant pair sequences. In terms of CTL epitopes in the RT gene, Wilson et al., have shown that the transmitting mothers have larger numbers of CTL escape variants as compared to non-transmitting mothers but the transmitted viruses carrying epitopes are not escape variants \[[@B56]\]. It is possible that the CTL responses studied are tissue specific and a representation of peripheral blood, and the virus and the CTL variants in the placenta, birth canal, and breast milk are different \[[@B70]\]. In addition, there is evidence suggesting that Nef and Pol specific CTLs found in breast milk showed no detectable responses in peripheral blood. Although several previously defined CTL motifs in the RT gene \[[@B56],[@B57]\] were conserved in our RT sequences, other mutations that either abrogated or improved the CTL responses \[[@B56],[@B57]\] were not seen in our sequences. The possibilities exist that the mutants observed in the CTL epitopes in our study may contribute to differential responses in a tissue specific manner and thus influence vertical transmission. While antiretroviral treatment during pregnancy has reduced the risk of vertical transmission in the United States, HIV-1 infection in children, as a result of perinatal transmission, is still increasing rapidly in developing countries. There is a global need of better preventive strategies of HIV-1 vertical transmission. If we characterize the properties of the transmitted viruses, we can then develop interventions against the properties of the transmitted viruses. We have already shown that the minor genotypes with R5 phenotypes are transmitted from mothers to infants and are initially maintained in the infants with the same properties \[[@B71]\]. Additional data on the properties of HIV-1 from mothers and infants following perinatal transmission presented in this study may aid in a better understanding of the molecular mechanisms of vertical transmission and development of effective strategies for prevention and control of HIV-1 infection in children. Conclusion ========== We have demonstrated that an intact and functional RT gene was maintained in infected mother-infant pairs following perinatal transmission. In addition, there was a lower degree of viral heterogeneity and estimates of genetic diversity in epidemiologically linked mother-infant pairs compared with epidemiologically unlinked individuals. Several amino acid motifs were found as a signature sequences in each mother-infant pair. We also found that the functional motifs of RT responsible for reverse transcription, DNA polymerization and RNase H were highly conserved in mother-infant RT sequences. These findings support the notion that RT is essential for HIV-1 replication in mothers and their infected infants. Methods ======= PCR amplification, cloning and nucleotide sequencing ---------------------------------------------------- Peripheral blood mononuclear cells (PBMCs) were isolated by a single step Ficoll-Hypaque procedure (Pharmacia-LKB) from whole blood samples of HIV-1-infected mother-infant pairs. DNA was isolated as described previously \[[@B68]\]. The HIV-1 RT gene was amplified by a two-step PCR method, first using outer primers RT1 (5 GTACAGTATTAGTAGGACCTACACCTGTC, 2470 to 2498, sense) and RT2 (5\'AAAATCACTAGCCATTGCTCTCCAATTAC, 4307 to 4279, antisense) and then with nested primers RT3 (5\'TGGAAGAAATCTGTTGACTCAGATTGG, 2507 to 2533, sense) and RT4, (5\'TTCTCATGTTCTTGGGCCTTATCT, 4270 to 4244, antisense). Equal amounts of PBMC DNA (approximately 25 to 50 copies from each patient) as determined by end-point dilution was subjected to multiple (5 to 8) independent PCRs to obtain clones that were sequenced and analyzed. PCRs were performed according the modified procedure of Ahmad et al., \[[@B68]\] in a 25 μl reaction mixture containing 2.5 μl of 10X PCR buffer (100 mM Tris-HCL, pH 8.3, 100 mM KCl, 0.02% Tween 20), 2.5 mM MgCl~2~, 400 μM each of dATP, dCTP, dGTP and dTTP, 0.2 to 1.0 μM of each of outer primers, and 2.5 U of TaKaRa LA *Taq*polymerase (TaKaRa Biomedicals, Shiga, Japan). The reactions were carried out at 94°C for 30s, 45°C for 45s and 72°C for 3 min for 35 cycles, with the last cycle allowing for seven minutes of additional polymerization. After the first round of PCR, 4μl of the first-PCR product was used for nested PCR, using inner primers and same reagents at 94°C for 30s, 52°C for 45s and 72°C for 3 min for 35 cycles. We used negative control with each PCR amplification and a known HIV-1 DNA, pNL4-3, to assess errors generated by the LA *Taq*polymerase. To avoid contamination, all samples, reagents and PCR products were stored separately and dispensed in a separate room free of all DNA used in the lab. The PCR products were then visualized on a 1% agarose gel, excised ad extracted by using a QIAquick Gel Extraction kit (Qiagen Inc.). These DNAs were cloned into the TA cloning system (pCR 2.1-TOPO vector, Invitrogen Inc.) and transformed into chemically competent TOP10 cells (Invitrogen Inc.). The white colonies were screened for correct size inserts and 10 to 14 clones from each patient obtained from multiple independent PCRs were initially manually sequenced and then sequenced using University of Arizona Biotechnology Center automated system. Sequence analysis ----------------- The nucleotide sequences of HIV-1 RT gene (approximately 1680 bp) from five mother-infant pairs were analyzed with the Wisconsin package 10.1 version of the Genetics Computer group (GCG) and were translated to corresponding deduced amino acid sequences (560 amino acids). A multiple sequence alignment was performed for the nucleotide and amino acid sequences with a reference HIV-1 consensus clade or subtype B RT sequences with a gap-opening penalty of 10 and a gap extension penalty of 5 using Clustal X. The transitions were not weighted and the amino acids were scored using a BLOSUM matrix. A model of evolution was optimized for the entire nucleotide sequence data set using the approach outlined by Huelsenbeck and Crandall \[[@B33]\]. Likelihood scores for different models of evolution were calculated using PAUP \[[@B34]\] and a chi square test was performed by Modeltest 3.06 \[[@B34],[@B35],[@B40],[@B72]\]. Using the Model test and Akaike Information Criterion \[[@B72]\], all the null hypotheses were rejected except a GTR+G model. The five rate categories were as follows: R (A-C) = 2.962, R (A-G) = 10.5176, R (A-T) 1.3663, R (C-G) = 0.6563, R (C-T) 12.5484, R (G-T) = 1. A gamma distribution with the shape parameter (α) of the distribution estimated from the data matrix via maximum likelihood was used to account for the rate of heterogeneity. This shape parameter α was = 0.7775. The model of choice was incorporated into PAUP \[[@B34]\] to estimate a neighbor-joining tree and the tree was bootstrapped 1000 times to ensure fidelity. Models to represent patterns of evolution of variants of each patient population were identified and were used to estimate corrected pairwise nucleotide distances using PAUP \[[@B34]\]. Amino acid distances were also estimated using the Jukes-Cantor model with the Wisconsin package 10.1 of GCG. The minimum, median and maximum nucleotide and amino acid distances for each patient and linked patient pairs were calculated from these data (Table [2](#T2){ref-type="table"}). To analyze the evolutionary processes acting on the RT gene, we estimated the ratio of non-synonymous (dN) to synonymous (dS) substitutions by a maximum likelihood model using codeML, a part of the PAML \[[@B37]\] package. The Nielsen and Yang \[[@B36]\] model considers the codon instead of the nucleotide as the unit of evolution and incorporates three distinct categories of sites. Every mutation is three times more likely to cause a nonsynonymous than a synononymous substitution and codeML accounts for this bias. The first category p1 represents the sites that are conserved and invariable where dN/dS = 0. The second category p2 represents neutral sites where dN/dS = 1 and represents sites at which the dN and the dS are fixed at the same rate. The third category p3 represents sites that are under positive selection where the dN have a higher rate of fixation than dS proportionally and dN/dS \>1. The dynamics of HIV-1 evolution was assessed using techniques of population genetics. In population genetics, genetic diversity is defined as θ = 2N~*ei*~μ, where N~*ei*~is the inbreeding effective population size and μ is the per nucleotide mutation rate per generation. The Watterson model based on segregating sites and the Kuhner model assuming constant population size were used to estimate differences in genetic diversity, using the program Coalesce, <http://inbio.byu.edu/faculty/kac/crandall_lab> which is part of the Lamarc software package. The tree files and the data matrixes from PAUP were used to estimate θ values as a measure of genetic diversity. ### Nucleotide sequence accession numbers The sequences have been submitted to GenBank with accession numbers [AY560388](AY560388) to [AY560528](AY560528). Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= VS carried out the PCR, cloning, and sequencing. VS and TH performed the sequence analysis by computer programs. VS and NA participated in the experimental design, data interpretation and writing of the manuscript. All the authors read and approved the final manuscript. Acknowledgements ================ This work was supported by grants to NA from the National Institute of Allergy and Infectious Disease (AI 40378, AI 40378-06) and the Arizona Disease Control Research Commission (ADCRC-7002, 8001). We thank Raymond C. Baker, Children\'s Hospital Medical Center, Cincinnati, Ohio and Ziad M. Shehab Department of Pediatrics, University of Arizona College of Medicine for providing HIV-1-infected mother-infant pairs blood samples. We thank members of Ahmad Lab, including Tiffany Davis and Kamlesh Patel for their help in cloning of the RT genes and Rajesh Ramakrishnan, Roshni Mehta and Brian Wellensiek for critically reading this manuscript and providing helpful suggestions.

PubMed Central

2024-06-05T03:55:58.173935

2005-5-26

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166575/", "journal": "Retrovirology. 2005 May 26; 2:36", "authors": [ { "first": "Vasudha", "last": "Sundaravaradan" }, { "first": "Tobias", "last": "Hahn" }, { "first": "Nafees", "last": "Ahmad" } ] }

PMC1166576

Findings ======== Two independent screens identified TRIM5 as a potent retrovirus restriction element that targets select viruses after entry into primate cells \[[@B1],[@B2]\]. The biochemical basis for specificity of restriction is only evident in cells of the owl monkey where HIV-1 CA is recognized by the C-terminal cyclophilin domain that is unique to the TRIM5 orthologue found in this genus \[[@B2]-[@B4]\]. In all other primates, including humans and macaques, potent CA-specific restriction is conferred by the TRIM5α isoform \[[@B1],[@B5]-[@B9]\], which possesses a C-terminal SPRY domain \[[@B10]\]. The mechanism by which TRIM5α selects retroviruses bearing particular CAs for restriction is unknown, though the TRIM5α SPRY domain is required for restriction and variation in SPRY amino acid residues determines the CA-specificity of given TRIM5α orthologues \[[@B9],[@B11]-[@B13]\]. Conventional biochemical and two-hybid experiments failed to detect an interaction between TRIM5α and CA (SS and JL, unpublished data). The observation that non-infectious virus-like particles saturate TRIM5α-mediated restriction \[[@B14]\], but only if the particles bear a mature core from a restriction-sensitive virus \[[@B15],[@B16]\] suggests that the TRIM5α SPRY domain recognizes a complex structure unique to the core of susceptible virions. Consistent with this model, expression within target cells of *gag*, *gag-pol*, or *gag*fragments encoding CA, CA-NC, or ubiquitin-CA-NC fusions, failed to block restriction activity (David Sayah and JL, unpublished data). Retrovirion cores can be liberated from the viral membrane envelope by detergent \[[@B17]\]. HIV-1 virion cores were prepared with several different detergents and mixed with recombinant TRIM5 orthologues. After TRIM5 enrichment by affinity chromatography, CA associated with owl monkey TRIMCypA, as reported with other methods \[[@B3],[@B4]\], but not with the equally potent HIV-1 restriction factor rhesus macaque TRIM5α (SS and JL, unpublished data). We then selected murine leukemia virus (MLV) for study because, relative to HIV-1, MLV CA remains tightly associated with viral reverse transcription (RT) and preintegration complexes \[[@B18],[@B19]\]. MLV strains bearing an arginine at CA residue 110 (so-called N-MLV) are highly susceptible to restriction by human TRIM5α whereas MLV virions bearing glutamate in this position (B-MLV) are completely resistant to restriction \[[@B5]-[@B8]\]. VSV G-pseudotyped N- and B-tropic MLV virions were produced as previously described \[[@B20]\] and, after normalization on non-restrictive *Mus dunni*cells, N-MLV was roughly 100-fold less infectious than B-MLV on HeLa cells (Figure [1A](#F1){ref-type="fig"}). Full-length human TRIM5α was then produced as a GST-fusion protein in 293T cells and mixed with purified N-MLV virions. CAp30, the major MLV core protein constituent, associated with TRIM5α (Figure [1B](#F1){ref-type="fig"}). CAp30 from B-MLV virions did not associate with TRIM5α (Figure [1B](#F1){ref-type="fig"}) demonstrating that TRIM5α binding was specific for restriction-sensitive CA. CAp30 did not associate with TRIM5 lacking the SPRY domain (Figure [1B](#F1){ref-type="fig"}), indicating that the SPRY-domain is required for CA-recognition. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Human TRIM5α binds CA from restricted MLV virions. (A) HeLa cells were infected with VSV G-pseudotyped, N- and B-tropic MLV-GFP vectors after normalization for RT activity and infectivity on non-restrictive *Mus dunni*tail fibroblasts. The percentage of infected (GFP-positive) cells was determined by flow cytometry. (B) 293T cells were transfected with plasmids encoding glutathione S-transferase (GST) fusions with full-length TRIM5α or with TRIM5 lacking the SPRY domain. Cells were lysed (50 mM Tris pH 8.0, 150 mM NaCl, 1% NP-40, 0.1% SDS) and mixed for 2 hrs at 4°C with virions (N-MLV or B-MLV) that had been concentrated by acceleration through 25% sucrose. GST fusions and associated proteins were enriched on glutathione-sepharose beads and immunoblotted with goat anti-MLV CA antibody (CA pull-out), or anti-GST antibody (bottom panel). Unbound CA remaining in the binding reaction was probed with anti-MLV CA antibody (CA input). TRIM5 protein domains fused to GST are indicated schematically on the bottom left: RF, ring finger; BB, B box; CC, coiled-coil. ::: ![](1742-4690-2-40-1) ::: Retroviral restriction specificity thus seems to be determined by TRIM5α binding to CA in a process that requires the SPRY domain. The fact that TRIM5α recognized retroviral CA presented by detergent-stripped virion cores, but not free CA protein, suggests that the SPRY domain recognizes a complex surface of multimerized CA. Once cores of restriction-sensitive viruses are singled out by the SPRY domain, TRIM5α blocks retroviral RT \[[@B1]\] by a mechanism that awaits elucidation. Our findings bring us one step closer to understanding how the potent antiviral activity of TRIM5α might be harnessed to block HIV-1 infection in people. List of abbreviations ===================== HIV-1, human immunodeficiency virus; MLV, murine leukemia virus; TRIM, tripartite motif protein; RT, reverse transcriptase; CA, retroviral capsid protein; GST, glutathione S-transferase; RF, ring finger domain; BB, B box domain; CC, coiled-coil domain. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= SS and JL conceived the experiments and wrote the manuscript. SS performed the laboratory work. Both authors read and approved the final manuscript. Acknowledgements ================ This work was supported by NIH grant AI 36199 to J.L.

PubMed Central

2024-06-05T03:55:58.179481

2005-6-20

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166576/", "journal": "Retrovirology. 2005 Jun 20; 2:40", "authors": [ { "first": "Sarah", "last": "Sebastian" }, { "first": "Jeremy", "last": "Luban" } ] }

PMC1166577

Introduction ============ Several chronic lung diseases are associated with an exaggerated accumulation and activity of neutrophils locally in the human bronchoalveolar space \[[@B1]\]. These lung diseases include acute, severe asthma, chronic bronchitis and chronic obstructive pulmonary disease (COPD) \[[@B1]\]. Neutrophils may play a pathogenic role in these diseases through the release of free oxygen radicals and by producing enzymes that increase the local proteolytic burden. Thus, via these compounds, neutrophils may contribute to pathogenic hallmarks such as non-specific bronchial hyper-responsiveness and hypersecretion, as well as to epithelial damage and tissue remodelling \[[@B1]\]. There is evidence that, among the proteolytic enzymes in the lungs, matrix metalloproteinase (MMP)-9 plays an important pathogenic role and this is most likely due to its capacity to cleave structural proteins such as collagens and elastin \[[@B2]\]. In support of this, local MMP-9 is increased in asthma and in COPD \[[@B3]-[@B8]\]. It is believed that neutrophils constitute an important source of MMP-9 in the lungs, even though macrophages, bronchial epithelial cells and fibroblasts may also produce MMP-9 under certain conditions \[[@B2]\]. Recent studies on mice *in vivo*and on human cells *in vitro*are compatible with activated T-lymphocytes contributing to the local accumulation of neutrophils in the bronchoalveolar space through the release of the cytokine interleukin (IL)-17A. However it remains unknown whether lymphocytes producing IL-17A actually reside within the bronchoalveolar space of humans \[[@B1]\]. In the bronchoalveolar space of mice and rats, local administration of recombinant IL-17A protein increases the number of neutrophils substantially \[[@B1],[@B9]\]. Systemic pre-treatment with a neutralizing anti-IL-17A antibody attenuates endotoxin-induced neutrophil accumulation in the same compartment of mice \[[@B10]\]. It is likely that IL-17A causes this neutrophil accumulation in part indirectly through the induced production of neutrophil-mobilizing cytokines, such as granulocyte-macrophage colony-stimulating factor (GMC-SF), IL-6 and IL-8, in local cells, including bronchial epithelial cells, fibroblasts and smooth muscle cells \[[@B1],[@B10]\]. In addition, there are now studies on mice and rats indicating that IL-17A may control the local proteolytic burden in the bronchoalveolar space *in vivo*through its impact on neutrophil accumulation \[[@B11],[@B12]\]. We recently showed that organic dust in a swine confinement, a strong stimulus of neutrophilic inflammation, causes a substantial increase in the concentration of free, soluble IL-17A protein in the human bronchoalveolar space \[[@B13]\]. Interestingly, this increase in local IL-17A protein is associated not only with an accumulation of neutrophils but also with an accumulation of lymphocytes. The current study was undertaken to determine whether these lymphocytes can produce IL-17A in humans. The study was also undertaken to address whether production of IL-17A is associated with an increased local proteolytic burden from neutrophils. Exposure to organic dust in a swine confinement was utilised as a pro-inflammatory stimulus and intracellular expression of MMP-9 was targeted as a sign of proteolytic burden in the bronchoalveolar space of human healthy volunteers. Methods ======= Study design ------------ The study protocol was approved by the Ethics Committee at Karolinska Institute (Diary N. KI 95:347). Six healthy, non-smoking volunteers gave their written consent to participate in the study after receiving both written and oral information. All the participants were non-atopic, non-asthmatic as determined by history and questionnaire. They were exposed to organic dust while weighing pigs in a swine confinement (ie. a barn) during three hours, as previously described \[[@B14]-[@B16]\]. The procedure incorporates exposure to aerosolised organic dust, with a concentration of approximately 24 mg dust and 1.2 μg endotoxin per m^3^, as assessed in a recent study \[[@B13]\]. Normal lung function and normal airway responsiveness were ascertained in all subjects prior to exposure. The subjects underwent bronchoscopy with bronchoalveolar lavage (BAL-see below) 2 weeks before and 24 hours after the exposure. Lung function ------------- Spirometry was performed in accordance with the criteria of the American Thoracic Society using a low-resistance rolling-seal spirometer (OHIO model 840; Airco, Madison, WI, USA) \[[@B17]\]. Swedish reference values were used \[[@B18],[@B19]\]. Bronchial responsiveness ------------------------ The methacholine provocation test was conducted prior to exposure to ascertain a normal airway function, as previously described \[[@B20]\]. Briefly, after the inhalation of the diluent (vehicle) alone, doubling concentrations of methacholine were given starting at 0,5 and ending at 32 mg/ml. The test was stopped either at a FEV~1~decrease of 20%, compared to the value registered after the inhalation of the diluent, or after inhalation of the highest methacholine provocation (32 mg/ml). The cumulative dose causing a 20% decrease of FEV~1~(PD~20~FEV~1~) was calculated. Bronchoalveolar lavage ---------------------- Bronchoscopy and bronchoalveolar lavage were performed according to standard procedures previously described \[[@B13]\]. In short, after pre-medication with morphine-scopolamine, a flexible fibreoptic bronchoscope (Olympus Type 4B2; Olympus Optical Co. Ltd., Shinjukuku, Tokyo, Japan) was inserted through the nose under local anaesthesia with lidocaine (Xylocain R, Astra, Södertälje, Sweden). The bronchoscope was wedged in a middle lobe bronchus and a total of 250 ml of sterile saline at 37°C was instilled in 5 aliquots of 50 ml. After each instillation, the BAL suspension was gently aspirated and collected in a siliconized plastic bottle kept on ice. After sterile filtration (70 μm filter), the BAL suspension was centrifuged (Hereus Omnifuge 300 g, 10 min, +4°C). The cell-free supernatant was kept frozen (-80°C) until analysis. The cell pellet was resuspended (phosphate-buffered saline \[PBS\]: 5 ml). Viability of cells was assessed using trypan blue exclusion and the total cell number (ie. concentration) was determined for each sample in a Bürker chamber. BAL cytospin samples were prepared and stained (May-Grünwald-Giemsa) and differential cell counts were performed using light microscopy (Zeiss Axiopaln 2, Carl Zeiss, Jena, Germany). Messenger RNA for IL-17A ------------------------ Remaining cells from the resuspended pellet were centrifuged (see above) and the supernatant was removed. The cell pellet was re-suspended (PBS: 50 μl plus RNAlater from Ambion Inc, Austin, USA: 250 μl per 3 × 10^6^cells), transferred into microtubes, (3 × 10^6^cells per tube) and incubated (24 hrs, +4°C). The RNeasy minikit was used according to the manufacturer\'s protocol with the following modification: The eluted RNA was treated with Dnase I (Rnase-Free Dnase set: 30 min, 37°C). After this, the steps from adjusting binding conditions to eluting RNA were repeated. The eluted total RNA was measured for quantity and purity using spectrophotometry (SpectraMaxPlus^®^Molecular Devices Corporation Sunnyvale, CA, USA; absorbance at 260 and 260/280 nm respectively) and then stored frozen (-83°C) until purification of Total RNA. The RT-PCR ELISA technique was employed to quantify relative changes in the IL-17A mRNA gene transcript. A one-step RT-PCR was utilized with a Gene Amp PCR system 2400 (Perkin Elmer, Wellesley, MA, USA) for amplification. Each RT-PCR of 50 μl was conducted using 100 ng of total cellular RNA and 30 pmol of each primer, 10 U RNase inhibitor (recombinant RNasin, Promega Corporation, Madison, WI, USA), 200 μM PCR DIG Labelling mix (20 μM dATP, dGTP, dCTP each plus 19 μM dTTP plus 1 μM DIG-dUTP), 5 mM DDT, 10 μl RT-PCR reaction buffer (1.5 mM Mg+), 1 μl Titan enzyme mix (AMV reverse transcriptase+Taq DNA polymerase+Pwo DNA polymerase) or 0,5 μl Expand high fidelity PCR system (Taq DNA polymerase+Pwo DNA polymerase) for DNA controls (all reagents from Roche Diagnostics). Reverse transcription was performed (30 min, 50°C). The IL-17A and the house-keeping gene HPRT were subsequently annealed (55°C). IL-17A and HPRT were then amplified (35 and 30 cycles, respectively) and the DIG labelled PCR product was stored frozen (-20°C) before the detection step. RT-PCRs were performed in duplicate. Gene sequences were accessed from NCBI Database and accession number used was for human (h) HPRT V00530 and for hIL-17 U32659. Scandinavian Gene Synthesis AB (Köping, Sweden) provided the oligonucleotides. The gene primer sequences used for RT-PCR and ELISA detection: hHPRT Sense (5\') 5\'CGT CGT GAT TAG TGA TGA TGA AC3\' Antisense (3\') 5\'GCA AAG TCT GCA TTG TTT TGC CA3\' Internal probe 5\'GAG GCC ATC ACA TTG TAG CCC TCT GTG3\' hIL-17 Sense (5\') 5\' GTG AAG GCA GGA ATC ACA ATC 3\' Antisense (5\') 5\' ACC AGG ATC TCT TGC TGG AT 3\' Internal probe 5\' CAG AGT TCA TGT GGT AGT CCA CGT TCC CA 3\' The DIG labelled PCR products were denaturated and hybridized with the biotinylated internal probe specifically designed to hybridize with each gene PCR product, and immobilized on streptavidin coated microtiter plates (3 hrs, 42°C), utilising the PCR ELISA DIG-detection kit (Roche). After washing, the bound PCR-products were incubated (30 min, 37°C) with an anti-DIG-horseradish peroxidase conjugated antibody followed by reaction (20 min) with the substrate 2,2\'-azino-di(3-ethyl benzthiazoline sulfonate) (ABTS). The absorbance was then measured using spectrophotometry (see above, at 405/492 nm) in an ELISA plate reader (Labsystems Multiscan Multisoft, Vanda, Finland). The expression of transcripts for IL-17 mRNA was normalized to the expression of HPRT transcripts and shown as percent of the house-keeping gene (% HPRT). In parallel with tested samples, control samples for PCR, probe specificity, DNA contamination, hybridisation and sample dilution were run. Immunocytochemistry (ICC) ------------------------- ### General procedure BAL cells were fixed with formaldehyde (2%, 30 min, on ice) and washed twice in buffer prior to making cytospin preparations. Air-dried samples were stored frozen (-80°C) until further use. After thawing, samples were treated with donkey serum (10%) to avoid unspecific binding. Endogenous biotin was blocked using the Biotin Blocking System (DAKO corporation, Glostrup, Denmark). ### IL-17A immunoreactivity (IR) The intracellular expression of IL-17A protein was assessed utilising cytospin slides incubated with a polyclonal goat anti-human IL-17A antibody (R&D: 1 hr). As secondary antibody, a biotinylated F(ab\')~2~fragment donkey anti-goat IgG (Jackson ImmunoResearch laboratories Inc) was used, followed by alkaline phosphatase-conjugated streptavidin (DAKO). All the solutions above were supplemented with saponin (Sigma: 0.1 %), to permeabilise cell membrane. Bound antibodies were visualised with Vector Red Alkaline Substrate Kit (Vector Laboratories, Inc. Burlingame, CA, USA). Mayer\'s Hematoxylin (Sigma) was used for counterstaining. ### MMP-9 immunoreactivity The intracellular expression of MMP-9 protein was assessed utilising cytospin slides incubated with a polyclonal goat anti-human MMP-9 (R&D: 1 hr). The secondary antibody and the detection system were the same as for IL-17A (see above). Human neutrophilic lipocalin ---------------------------- The concentration of the neutrophil-specific activity marker human neutrophilic lipocalin (HNL) was measured in cell-free BAL fluid using a solid phase, double-ligand radioimmunoassay as described elsewhere \[[@B21]\]. Data analyses ------------- Data are presented as median (range) values unless otherwise stated. The analysis of differences between measurements prior to and after the exposure, were conducted utilizing the sign test, assuming binominal distribution. To avoid falsely positive conclusions, the significance level at each test was set to 0.05/3 (ie. 0.0167), since three key variables (IL-17A mRNA, IL-17 IR and MMP9 IR) were compared in the current material. *n*refers to number of human subjects. Results ======= Clinical characteristics prior to exposure ------------------------------------------ Four males and 2 females with a median age of 22 (19--24) were included in the study. There were no apparent gender-related differences in pulmonary function of the subjects. FEV~1~was 96 (92 -- 105) % of predicted value and vital capacity 98 (90--102) % predicted value. All subjects had normal bronchial responsiveness to metacholine (PD~20~FEV~1~for metacholine: 5.3 \[1.7--15\] mg). Cells in BAL fluid ------------------ The BAL recovery volume was similar before (195 \[170--234\] ml) and after (196 \[190--212\] ml) the exposure. The exposure caused more than a two-fold increase in the total BAL cell number, more than a five-fold increase in the number of lymphocytes and more than a ten-fold increase in the number of neutrophils.(Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Concentration of BAL cells. BAL cells before and after exposure to organic dust in a swine confinement. Data are presented as 1 × 10^6^/L median (range). ::: **Cell type** **Before** **After** --------------- ------------------- --------------------- Total cells 71,3 (57,9--80,9) 165,1 (98,2--365,6) Lymphocytes 4,4 (1,2--10.0) 23,1 (9,8--61,6) Neutrophils 2,3 (0,8--3,6) 23,1 (6,6--167,3) ::: Messenger RNA for IL-17A ------------------------ Before the exposure, 4 out of 6 persons displayed no detectable levels of mRNA for IL-17A, whereas all six subjects displayed detectable and increased levels after the exposure (fig. [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Messenger RNA for IL-17A in BAL cells**. Messenger RNA for IL-17A in BAL cells (percentage of the house-keeping gene HPRT), measured 2 weeks before and 24 hours after exposure to organic dust in a swine confinement. Data are shown as individual (rhombs) plus median (bold horizontal lines) values. \*: p \< 0,0167; n = 6 ::: ![](1465-9921-6-44-1) ::: IL-17A immunoreactivity ----------------------- IL-17A immunoreactivity was mainly detected in BAL lymphocytes (fig. [2](#F2){ref-type="fig"}). In addition, the percentage of lymphocytes positive for IL-17A immunoreactivity (fig. [3](#F3){ref-type="fig"}) tended to increase in 5 out of 6 subjects after the exposure. In contrast, a much lower fraction of macrophages expressed a weak signal for IL-17A immunoreactivity and there was no pronounced increase in this signal after the exposure (data not shown). BAL neutrophils did not display IL-17A immunoreactivity (data not shown) ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Immunocytochemical detection of IL-17A protein in BAL lymphocytes**. BAL lymphocytes expressing IL-17A IR (brown arrows) after exposure to organic dust in a swine confinement, detected using immunocytochemistry. ::: ![](1465-9921-6-44-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **BAL lymphocytes expressing IL-17A protein**. Percentage of BAL lymphocytes expressing IL-17A IR before and after the exposure to organic dust in a swine confinement. Data are shown as individual (rhombs) plus median (bold horizontal lines) values. NS: p \> 0,0167; n = 6 ::: ![](1465-9921-6-44-3) ::: MMP-9 immunoreactivity ---------------------- Neutrophils were the prevailing cell type in BAL fluid expressing MMP-9 immunoreactivity (fig. [4](#F4){ref-type="fig"}) and this signal was substantially increased after exposure (fig. [5](#F5){ref-type="fig"}). However, the fraction of neutrophils (% all neutrophils) expressing MMP-9 immunoreactivity was not increased after exposure (data not shown). Very few macrophages expressed MMP-9 immunoreactivity and this signal was consistently weak (data not shown). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Immunocytochemical detection of MMP-9 protein in BAL neutrophils**. Abundance of BAL neutrophils positive for MMP-9 IR (purple arrows) after exposure to organic dust in a swine confinement. ::: ![](1465-9921-6-44-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **BAL neutrophils expressing MMP-9 protein**. BAL neutrophils expressing MMP-9 IR (percent of total cells) before and after exposure to organic dust in a swine confinement. Data are shown as individual (rhombs) plus median (bold horizontal lines) values. \*: p \< 0,0167; n = 5. ::: ![](1465-9921-6-44-5) ::: HNL --- The exposure did not cause any substantial increase in the concentration of free, soluble HNL in the BAL fluid, even though there was a weak average trend in this direction (data not shown). Furthermore, the amount of HNL per neutrophil tended to be decreased after exposure (data not shown). Discussion ========== The current study demonstrates that, in previously healthy human volunteers, exposure to organic dust substantially increases the messenger RNA for IL-17A in BAL cells. Our study also demonstrates that lymphocytes constitute the predominant cell type that expresses intracellular IL-17A protein among BAL cells. The demonstrated increase in IL-17A mRNA is associated with an increased absolute number of MMP-9 expressing BAL neutrophils at the same time point. The aerosolised organic dust in a swine confinement, that was utilised as a pro-inflammatory stimulus in our study, is a more complex stimulus than endotoxin, because of its mixed content of grain dust, ammonia, fungi, Gram-positive and Gram-negative bacteria \[[@B15]\]. It has previously been shown that exposure to this type of organic dust causes severe inflammation in the bronchoalveolar space of humans \[[@B13],[@B15],[@B16]\]. Our current study now adds to these previous studies novel evidence of *de novo*synthesis of IL-17A by local cells within the bronchoalveolar space of humans \[[@B1]\]. Furthermore, our study forwards lymphocytes as a local source of IL-17A protein in the human bronchoalveolar space, because the intracellular immunoreactivity for this specific protein was found mainly in certain BAL lymphocytes. Noteworthy, even though this immunoreactivity should be regarded as a qualitative assessment primarily, it was increased eight-fold by average after exposure with only one subject lacking a detectable increase. Clearly, this is compatible with lymphocytes accounting for production of IL-17A protein in the bronchoalveolar space of humans. Indeed, our current results are supported by our previous demonstration of an increase in the final consequence of the increased transcription and intracellular loading of protein, namely in free, soluble IL-17A protein in cell-free BAL fluid \[[@B13]\]. It is also noteworthy that in our current study, certain but not all lymphocytes in the human bronchoalveolar space express IL-17A protein after exposure to organic dust. This observation is compatible with previous studies on activated mouse CD4^+^and CD8^+^lymphocytes from BAL and spleen as well as human CD4^+^and CD8^+^lymphocytes from blood \[[@B1],[@B22]-[@B28]\]. Taken together, these observations warrant new specific cellular targets for additional analysis in future studies on human airways. We assessed intracellular immunoreactivity for MMP-9 as a marker of proteolytic burden, both before and after exposure to organic dust. Even though some of the BAL macrophages displayed a very weak expression of MMP-9 immunoreactivity, as has previously been reported in a different inflammatory setting \[[@B29]\], we found the most pronounced increase in and expression of MMP-9 immunoreactivity consistently among BAL neutrophils. This observation fully supports that neutrophils can constitute an important source of MMP-9 in the human bronchoalveolar space *in vivo*, at least after exposure to organic dust. At the same time point after the exposure, the MMP-9 expression in neutrophils was associated with an increase in mRNA for IL-17A, even though the fraction (percentage) of neutrophils expressing MMP-9 was not increased. This is compatible with IL-17A indirectly contributing to the local proteolytic load, mainly through its increasing effect on the number of neutrophils within the human bronchoalveolar space; a type of mechanism supported by previous studies on the bronchoalveolar space of mice *in vivo*\[[@B9],[@B12]\]. We also measured the concentration of free, soluble HNL protein in BAL fluid, as an assessment of neutrophil-specific activity \[[@B30]\]. However, we detected only a weak average trend towards a modest increase in the HNL concentration after exposure to organic dust and this increase did not even correspond to the substantial increase in neutrophil number; the amount of HNL per neutrophil was actually decreased after the exposure. Furthermore some subjects displayed even a decrease in the total HNL concentration after exposure. Moreover, separate and preliminary assessments of the myeloperoxidase (MPO) concentration in BAL fluid displayed a weak increase not corresponding to the increase in neutrophil number (data not shown). Similar findings on MPO and neutrophils have previously been published \[[@B31]\]. Taken together, all these findings imply that, in terms of MPO and HNL, there is actually no true increase in the average activity of each accumulated neutrophil in the human bronchoalveolar space after exposure to organic dust. Again, from a mechanistical point of view, this finding in the bronchoalveolar space of humans is supported by our recently published findings on the effect of recombinant IL-17A in the bronchoalveolar space of mice *in vivo*\[[@B12]\]. This IL-17A induces an increase of MMP-9 protein and in corresponding gelatinase activity, that is due to an increased number of neutrophils rather than an increased amount of MMP-9 per neutrophil \[[@B12]\]. In conclusion, supported by our recent study showing a corresponding increase in free, soluble IL-17A protein \[[@B13]\], the current study adds novel evidence for organic dust inducing *de novo*synthesis of IL-17A protein in a subset of local lymphocytes within the human bronchoalveolar space. Our current study on humans also adds evidence compatible with induced *de novo*synthesis of IL-17A being associated with an increased proteolytic burden due to a local accumulation of neutrophils rather than an increased activity in each of these inflammatory cells. Interventional studies will be required to determine whether targeting IL-17A is therapeutically beneficial in lung diseases characterised by excessive accumulation of neutrophils. Authors\' contributions ======================= LP, KL, AL -- study design and coordination SI, LP, KL, PV -- exposure in swine farm, laboratory work SI, AL -- data analysis and interpretation of results SI, LP, KL, PV, AL -- preparation and revision of the manuscript Acknowledgements ================ The present study was financially supported by Göteborg University, the Karolinska Institute, the Swedish Heart-Lung Foundation, the Swedish Medical Research Council (K2002-74X-09048-13A), the Vårdal Foundation and, finally, by Glaxo-Wellcome, Sweden. No support was obtained from the tobacco industry. The excellent technical support by Carina Malmhäll, B Sci, and Margareta Sjöstrand, PhD, is gratefully acknowledged. The support with statistical analysis by Associate Professor Kerstin Wiklander, Department of Mathematics, the Chalmer\'s University of Technology in Göteborg is also acknowledged.

PubMed Central

2024-06-05T03:55:58.180495

2005-5-25

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166577/", "journal": "Respir Res. 2005 May 25; 6(1):44", "authors": [ { "first": "Stefan", "last": "Ivanov" }, { "first": "Lena", "last": "Palmberg" }, { "first": "Per", "last": "Venge" }, { "first": "Kjell", "last": "Larsson" }, { "first": "Anders", "last": "Lindén" } ] }

PMC1166578

Introduction ============ Transcription, the first step of information flow from DNA, is regulated by sequence specific DNA-protein interactions. The regulation depends on the presence of *cis*-acting elements. The best examples of *cis*-acting elements are promoters. Other well-known examples in bacteria include the Shine Dalgarno (SD) sequence, sigma 32 binding site, LexA binding site, etc. In bacteria, promoters recognized by sigma factors initiate transcription. The responses of an organism to various stimuli are mediated by changes in gene expression patterns. These changes are initiated by promoter-sigma factor interactions and regulated by other *cis*-acting elements. Thus, families of co-regulated genes are under the control of the same promoter. Though core promoters are small words (6--8 bases), certain changes that are permissible in promoter sequences have little or no effect on their activity. This means that a few closely-related sequences, in the right context, can function as promoters. Identifying promoters is a challenging yet rewarding problem; challenging because promoters can differ subtly in sequence and still retain function, and rewarding because it can shed light on an organism\'s life style. Computational approaches are required since experimental methods for identifying promoters are not applicable on a genome-wide scale. In most instances, computational identification or prediction of promoters involves model-based searches. The model is, by and large, derived from prior data. Techniques using artificial neural networks \[[@B1]\] or genetic programming methodologies \[[@B2]\] are also used, and require prior experimental data. Using prior data for identifying new candidates is also known as dictionary-based searching. Databases of experimentally verified *cis*-acting elements are available for promoter prediction \[[@B3]\] through dictionary-based approaches. These approaches are biased towards the best-characterized promoter in the initial dataset, though non-redundant data sets have been used recently \[[@B4]\]. A paucity of experimental data can compromise the efficiency of these methods. The success of dictionary-based methods is directly dependent on the relatedness of the database to the query. It has also been observed that, while using dictionary-based methods, taking into account the local genomic landscape for generating Markov profiles improved the prediction quality in eukaryotes \[[@B5]\]. Another method that has been applied to both simpler and larger genomes is the comparative genome analysis method. It is observed that functional regions, albeit non-coding, are conserved across species and genera. Analyses of this kind have been used for yeast \[[@B6],[@B7]\], higher eukaryotes \[[@B8],[@B7]\] and bacterial regulons \[[@B9],[@B7]\]. In *Saccharomyces cerevisiae*, the distribution of certain words across the genome is non-random. For example, some words appear to be preferred in regions upstream \[[@B10]\] or downstream \[[@B11]\] of genes. Analyses showed that such words occurring preferentially near the genes represent functional elements. Though non-random usage of k-sized words in bacterial genomes has been documented \[[@B12],[@B13]\] in genomic contigs, studies have not focused on the upstream regions of prokaryotic genes. We have developed a method that uses preferential occurrence of k-sized words within specific (gene-proximal) regions in a given genome to predict *cis-*acting elements. This method does not use a dictionary or database for initiating searches. The method can be applied to any genome of which the gene co-ordinates are known. Its advantage is that there is no extrapolation of data. This allows unique families of *cis*-acting elements for a given genome to be determined. Inter-genome comparison can establish the functionality of conserved words across genera. The results of oligonucleotide profiling as applied to the genome of *E.coli*K12 \[[@B14]\] are presented. Comparative analyses of the resultant oligonucleotide profiles show that a subset of preferred hexanucleotides in *E.coli*-K12 is conserved across two other genomes, those of *Salmonella typhi*and *Yersinia pestis*\[[@B15],[@B16]\]. We suggest a function for the ubiquitous hexanucleotides that are preferentially present in -100 regions and are neither single-base relatives of TATAAT or AGGA nor CTAG-containing, and we propose a novel model for bacterial promoters. Results and Discussion ====================== The results of oligonucleotide profiling, as performed for *E.coli*K12 genome, are discussed. The word size was restricted to six. For higher word sizes the word occurrence frequency was low. Smaller words were not used since the intra-word Markov dependencies, if any, are statistically invalid \[[@B17]\]. Word occurrences were analyzed in four contiguous sequence sets, F4 through F1 (Fig. [1a](#F1){ref-type="fig"}),. A threshold of 200% (two-fold increase in occurrence over the genomic average) was set to identify signals for *cis*-acting elements. The average occurrence of a random hexanucleotide in a sequence set is 4.6% of its genomic total and the standard deviation is 0.573. It can be seen that a two-fold increase (9.2%) is more than six times the standard deviation (σ) above the average. Any hexanucleotide that had at least 9.2% of its overall genomic occurrence within any of the four fragments analyzed was termed \"enriched\" in that respective region. Such enrichment was more pronounced in the gene-proximal regions (-1 to -100 region) than in the distal regions (-300 to -400). In the three random sequence sets (controls), only once did we find enrichment (a CTAG-containing element). Fig. [1a](#F1){ref-type="fig"} schematically illustrates this procedure. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### \(A) A schematic representation of the procedure used for profiling, incorporating the definition of the four fragments F1, F2, F3 and F4 used in this study. (B) Comparison of the occurrence distribution in the random control (series 1), F4 (series 2), F3 (series 3), F2 (series 2) and F1 (series 4). (C) Number of words whose occurrence is greater than μ+Nσ, where N is on the x axis. (D) Distribution of the three classes of oligonucleotides in the four fragments: TATAAT for class 1, AGGAGG for class 2 and AAAAAA for class 3. ::: ![](1742-4682-2-20-1) ::: The preferential occurrences of hexanucleotides within the controls and the fragments under study are contrasted in Table [1](#T1){ref-type="table"}. The distributions of hexanucleotide occurrence in control 1 and fragments (F1-F4) are shown in Fig. [1b](#F1){ref-type="fig"}, while Fig. [1c](#F1){ref-type="fig"} shows the number of hexanucleotides with frequencies N × (σ) more than average. The units on the X-axis are N (N times σ) and 200%. ::: {#T1 .table-wrap} Table 1 Threshold(C)/Region(R) \%\>μ+2 \%\>μ+3 \%\>μ+4 \%\>200 ------------------------ --------- --------- --------- --------- Control1 184 42 18 1 Control2 155 30 9 0 Control3 164 46 16 0 F1 229 66 24 7 F2 341 116 44 0 F3 662 387 236 14 F4 1112 834 634 190 ::: The method retrieved 183 hexanucleotides that were enriched in the -100 region. These included the Pribnow box (TATAAT), SD consensus (AGGA), the LexA binding site (CTATNN), sigma 32 binding site one-base relative (CCCTAT) and CTAG-containing regulatory elements \[Supplementary Information 1\]. The CTAG-containing elements are known to be core repressor binding regions in the Trp, Met and MalPQ operons and the treA gene \[[@B18]-[@B20]\]. They occur at high frequency near the rRNA gene clusters \[[@B12]\]. However, in the rest of the genome, we find their distribution to be roughly uniform (data not shown). Certain trends are apparent in the usage of enriched oligonucleotides by bacterial genomes. The occurrence of some oligonucleotides increases gradually with proximity to genes (class I oligonucleotides), while others (class II oligonucleotides) peak near the genes. A third class comprises non-specifically preferred oligonucleotides (Class III oligonucleotides). Class I Oligonucleotide ----------------------- Bacteria are expected to have limited number of promoter elements and to have them near genes. The Pribnow box in *E.coli*is a representative promoter. The overall frequency of the Pribnow box is lower than the genomic average (1067 occurrences as against the genomic average of \~1400). Here, we analyze: the occurrence of the Pribnow box and its single base substitution relatives, the distribution of the Pribnow box within the -100 region, and the position-dependency of other bases on the Pribnow box in its vicinity. For this analysis, Pribnow box occurrences in the -100 region alone were taken into account for four strains of *E.coli*. ### Occurrence of Pribnow box This analysis shows that the occurrence of the Pribnow box increases gradually as one goes closer to genes. Furthermore, seven of its one-base substitution relatives figure in the enriched list \[Table [2](#T2){ref-type="table"}\]. Most of these one-base relatives show a gradual but definite increase in their occurrence as we move nearer the genes \[Table [2](#T2){ref-type="table"}\]. This gives an idea as to how an element that has a function similar to the Pribnow box would behave in other genomes. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Occurrence of single base relatives of TATAAT in *E.coli*genome. F1:-301 to -400; F2:-201 to -300; F3: -101 to -200; F4: -1 to -100. Those elements that are enriched (\> = 200%) are marked by an asterisk in the last column. ::: Hex Total Occ. F4 F3 F2 F1 Occ. % in F1 ENR -------- ------------ ----- ----- ----- ----- -------------- ----- TCTAAT 595 29 30 47 46 7.731092 TAGAAT 507 25 23 36 71 14.00394 \* TATAGT 681 19 38 30 87 12.77533 \* TATAAG 879 45 32 67 102 11.6041 \* TACAAT 903 42 56 71 109 12.07087 \* TATAAC 1590 65 80 94 121 7.610063 TATGAT 1539 63 111 94 127 8.252112 TATACT 620 39 40 62 131 21.12903 \* CATAAT 2448 90 106 115 132 5.392157 TATAAT 1036 58 68 78 134 12.93436 \* TGTAAT 1870 74 80 129 140 7.486631 TATCAT 2082 101 96 128 164 7.877041 TATATT 1943 87 130 145 168 8.646423 TATTAT 2280 118 124 155 178 7.807018 GATAAT 3735 125 140 172 201 5.381526 TATAAA 2304 98 144 177 236 10.24306 \* TTTAAT 3671 182 205 219 249 6.782893 TAAAAT 2947 142 159 204 268 9.093994 AATAAT 4132 188 201 287 287 6.945789 ::: ### Distribution of Pribnow box Analyses show that the maximal number of strong minimal promoters occur within the -100 region and that the Pribnow box prefers the -30 to -70 position, centering around -40 \[Fig. [2a](#F2){ref-type="fig"}\]. The report by Collado-Vides *et al*. shows that \~80% of the 800 genes analyzed have their promoters in the -100 region. In fact, the highest concentration of promoters that they report is at the -40 region \[[@B21]\], which we corroborate. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Addressed promoter model. (A) Occurrence distribution of TATAAT, AGGAGG and AAAAAA within the -100 region using a 30-base window: -1 to -30, -10 to -40, -20 to -50, \..., -70 to -100. (B) A schematic comparison of the classical and the addressed promoter models. Blue peaks represent the canonical promoter. Red background (where present) represents the address. ::: ![](1742-4682-2-20-2) ::: ### Markov dependency analysis of sequences surrounding Pribnow box Markovian analysis of TATAAT-containing sequences (within the -100 region) was done for *E.coli*. For analysis, such sequences were taken from all four *E.coli*genomes (K12, O157:H7, EDL933 and CFT073) to improve statistical significance (TATAAT occurred only 128 times in the -100 region of the K12 genome). The results showed that TTGACA is preferred between positions -32 and -27. Further, it was seen that, with G at -14, the occurrence of TTGACA decreased, (All corresponding data points are highlighted in the Supplementary Information 2 file.) This has been reported by analysis of experimentally characterized promoters \[[@B22]\]. These correlations validate the results of oligonucleotide profiling with respect to the sigma 70 binding site. Class II Oligonucleotides ------------------------- AGGA- (SD consensus) and CTAG-containing hexanucleotides belong to this class. Unlike the Class I oligonucleotides, Class II oligonucleotides show a steep increase in occurrence in the -100 region. This is expected in the case of the Shine-Dalgarno sequence (AGGA), since it should lie within 30 base pairs upstream of the ORF start site (owing to geometric constraints imposed by the ribosomal complex). Another example of this class is the tetranucleotide CTAG, representing all the hexanucleotides that contain it. CTAG kinks DNA when bound by proteins \[[@B23]\], making it a likely candidate for a regulatory site. CTAG also has low genomic frequency, uniform distribution and a preference for the -100 region. This might imply a global regulatory function. Class III Oligonucleotides -------------------------- Certain oligonucleotides not only have a more than average genomic frequency but are also more common in the -100 region. Many of these are A/T rich oligonucleotides, which are known to bend DNA when present in stretches \[[@B24]\]. The presence of such A/T repeat elements upstream \[[@B25]\] and downstream \[[@B26]\] of the canonical promoter is necessary. They are evidently not stand-alone signals. We propose that they are facilitator elements that are necessary but not sufficient for promoter recognition and function. The set of such oligonucleotides that were readily distinguished as facilitators is given, along with their distribution, in Supplementary Information 3. They occur preferentially up to -100 and beyond. We find this significant since a recent report shows that DNA of size 90 base pairs can bend upon itself in a sequence-dependent manner \[[@B24]\]. Though all 64 A/T containing hexanucleotides were found to occur more frequently than the genomic average, only 18 of them were enriched in the -100 region. Thus, the increased occurrence of Class III hexanucleotides is not an artifact of increased base frequency. It transpires that the genome increases the bending capacity of the -100 region by preferential usage of certain oligonucleotides. The occurrence of hexanucleotides representing each of the three classes is shown in Fig. [1d](#F1){ref-type="fig"}. TATAAT is used to represent class I, AGGA-containing hexanucleotides to represent class II and AAAAAA to represent class III. Protein Binding Capacity of the -100 region: Evidence from NDB -------------------------------------------------------------- We analyzed the occurrence of enriched hexanucleotides in a protein-bound state in the NDB database \[[@B27]\]. Of the \~130 hexanucleotides that are neither TATAAT-related (1 base substitution oligonucleotides) nor AGGA- or CTAG-containing, 112 have at least one occurrence in the database, bound to proteins \[data not shown\]. Most of them occurred more than once in the database in a protein-bound state. These results show the propensity of the genome to increase the protein-interacting capacity of the -100 region and hence increase the activity of this region. Dependency Analysis ------------------- A position-specific probability matrix (PSPM) was created for enriched oligonucleotides that were not TATAAT related or AGGA/CTAG-containing. This matrix was used to determine the tendency of hexanucleotides to assume specific consensus words within the -100 region of the genes. Secondary matrices were derived by anchoring the first base in the PSPM. The consensus words derived from these matrices are given in Supplementary Information 4. For each secondary matrix, two more character states were chosen for anchoring on the basis of their prominence,. The results show a strong preference for tetra-A signals, TATA-containing signals and GGA-containing signals. Inter-genome comparison of hexanucleotide usage profiles -------------------------------------------------------- Conservation of DNA sequence across genomes has been established as a pointer to functionality. This method has been used to identify regulatory regions in *Saccharomyces*\[[@B6]\] by sequence comparison among different species. We see that the logic extends beyond conservation of sequences and patterns to that of oligonucleotide profiles. We have compared the profile of enriched hexanucleotides between *E.coli*, *Salmonella enterica*and *Yersinia pestis*to test its validity. The *E.coli*and *Salmonella*profiles shared 110 enriched oligonucleotides out of 160 in *Salmonella typhi*. *Yersinia pestis*, whose profile had 97 enriched oligonucleotides, shared 66 of them with *E.coli*. Of those that were conserved across genomes, the AGGA-containing and CTAG-containing hexanucleotides, TATAAT, and the LexA binding site were prominent (Supplementary Information 5). While conservation of hexanucleotides usage implies functionality, the converse may not be true and might reflect unique regulatory / facilitator elements for each genome. Role of facilitator elements in promoter identification and the Addressed Promoter Model ---------------------------------------------------------------------------------------- Classical promoters in bacteria are sigma factor binding sites. The sequence that is known to bind to sigma factor with maximal affinity in vitro is taken to be the strongest promoter. DNA footprinting experiments do not allow us to assess the importance of the surrounding sequences. It is clear from the profiles that the strongest promoters have limited occurrence in the genome. Most genes are controlled by sigma 70 in *E.coli*\[[@B28]\], and only \~12% of the overall strong consensus occur in a region where they are maximally effective \[[@B21]\]. The question to be addressed is how a sigma factor (Sigma 70 in this case) can distinguish the promoter from non-specific promoter-like signals (degenerate -10 and -35 like signals in non-functional places in the genome). The sigma factor could not read every one of the possible signal combinations since this would result in enormous loss of time in bacterial genomes. In larger genomes, given the small size and degeneracy of the promoters, it is possible that the sigma factor would recognize a false signal on most occasions. To account for the efficiency of promoter recognition in the organism, we propose the addressed promoter model, where the sigma factor binding element is an information-dense peak (specific information) within a plateau of moderate information density (different but related words). The peak and the plateau together constitute the promoter. The plateau is formed by class III oligonucleotides that have the capacity to bend DNA. The facilitators are an integral part of the promoter. The presence of facilitators, which occur in greater frequencies around the core promoter, will serve as addresses for the core promoter. These addresses act as homing segments that allow the transcription factor to recognize the core promoter and bind to it. This model immediately suggests a way of identifying *cis*-acting regions in eukaryotes, where greater genome sizes and more degeneracy are seen. The extension of this logic would be to view enhancers and other regulatory regions in large eukaryotic genomes as local landscapes rather than as sequence motifs. While the protein binding sites would still be sequence motifs, their occurrence in a particular landscape may prove to be the determining factor for their activity. This accords with the observation of Huang et. al. \[[@B5]\] that local genomic landscape information affects the prediction quality of promoter elements. To illustrate this model, we have analyzed the distribution of one representative element from each of the three classes. The distribution was studied in a 30-base sliding window with a 10-base pitch. The representative elements are TATAAT (Class I), AGGAGG (Class II) and AAAAAA (Class III). The distribution is shown in Fig. [2a](#F2){ref-type="fig"}. It can be seen that AAAAAA forms a plateau around the TATAAT peak. The classical model and the addressed promoter model are contrasted in Fig. [2b](#F2){ref-type="fig"}. Conclusion ========== This method for identifying regulatory regions in DNA is powerful. Its strength is its ability to use the genomic sequence as a control. This obviates the need for data extrapolation from related genomes. The method can identify functional elements that can be experimentally characterized. Application of this method to the *E.coli*K12 genome reveals the presence of at least three classes of *cis*-acting elements. The occurrence, distribution and dependencies of these elements have been analyzed. Most of the profile data correlate with existing experimental evidence. The canonical sigma70 promoter has been analyzed in further detail in four *E.coli*genomes. The information derived from *E.coli*K12 using this method suggests that the functionality of a promoter is determined not only by the sequence of the core promoter element but also by its local milieu. We note that the occurrence of proposed facilitator elements extends just beyond the length known for DNA to bend upon itself (90 bp) and this, together with other reports about AT-rich tracts in the vicinity of the canonical promoter, suggests that the sigma factor recognizes a promoter more efficiently if it is present in the \"address\" region. This immediately explains why the transcription process is efficient in spite of the degeneracy that the promoter exhibits. We see that the occurrence of facilitators is not an artifact of increased base frequencies. The occurrence of many of the enriched hexanucleotides as protein-bound DNA complexes in the NDB database is indicative of their protein-interacting ability. This reflects on the protein binding capacity of the gene proximal regions in *E.coli*K12. The limitation of this method is its inability to pick up rare regulatory elements. In small genomes the method is known to give false positives, and in degraded genomes it picks up false negatives. In such cases, comparative analysis with related genomes will give valuable information. Methods ======= Sequence Extraction ------------------- Published genome sequences from the NCBI database <http://www.ncbi.nlm.nih.gov>(.fna file) were used. The start sites of genes given in the annotation file (.ptt file) were used for extracting upstream sequences of all the genes. Upstream sequences were taken only from their respective strands (+ strand for + genes and vice versa) because of the directionality of promoters. Four such fragments were taken from upstream of each gene, viz. -1 to -100, -101 to -200, -201 to -300 and -301 to -400. The distance between any two genes was not given importance because of the possibility that regulatory and transcriptional start sites may be present in the coding region of the preceding gene. Profiling --------- For every gene in the *E. coli*K12 genome, four contiguous DNA fragments from the corresponding strand were extracted. The length of each fragment was 100 bases. The fragments were named F4 through F1, where F1 is the gene-proximal fragment. There are 4311 genes in *E.coli*. Four sequence sets, one each for F1, F2, F3 and F4, were created for all the genes. Each of these sequence sets covers approximately 4.6% of the genome. Occurrence of all hexanucleotides was counted on both strands of the genome and the four upstream-sequence sets. The Compseq program from the EMBOSS \[[@B29]\] suite was used for this purpose. Any word that was non-functional was expected to be distributed equally across the sequence sets. Thus, for a non-functional word in the upstream context, we expected approximately 4.6% of its genomic occurrence in any of the sequence sets. Since *cis*-acting elements are gene-proximal, we expected their occurrence to be higher in F1 than elsewhere. We set a threshold (T) of 200% in word frequency to identify signals. Given a standard deviation of 0.56, it is apparent that a 200% increase (9.2% of genomic occurrence) is more than 6σ, which is significant. Words whose frequency in a given sequence set was 9.2% or more were termed \"enriched\" in the corresponding fragment. All analyses were carried out using Perl 5.6.1<http://www.perl.com> scripts on a Mandrake Linux 9.1 platform. The complete dataset is available in an in-house MySql <http://www.mysql.org>-based server. Markov Dependency Analysis -------------------------- We analyzed the character-state probabilities of all the words (137 words) for which a function could not be assigned. For this, we created a position-specific probability matrix (PSPM). The PSPM was derived from a position-specific frequency matrix (PSFM), which is defined as follows. For a word size of L, a PSFM is a 4 × L matrix M, where each element M~i,j~\[i ∈ {A,T,G,C} and j ∈ {1,2, \... L}\] is the number of times the character state i occurs at position j. In this case, L = 6. If S is the sum of all occurrences of words, then the PSPM is related to the PSFM as given below: PSPM = (1/S) × PSFM Such a matrix was used to derive consensus words preferred in the -100 region. From the PSPM, four sub-matrices were derived by anchoring the various character states (A, C, G, and T) at the first position. Further dependencies were analyzed by subsequent anchoring of two more positions, based on their prominence in the sub-PSPMs, to their representative character states. Markov Analysis for TATAAT-dependent Signals -------------------------------------------- For each occurrence of the Pribnow box within the -100 region, the preceding 50-base region was extracted. The PSPM was created for the sequence set as described above, where the value of L is 50. Different profiles were created by anchoring the base profile at all positions with all four bases. This was used to analyze the dependency of upstream signals on TATAAT. This analysis was done on a sequence set collated from all the four strains of *E.coli*. Authors\' contributions ======================= KS gave the core idea for oligonucleotide profiling, analysis of occurrence and proposed the model. ASNS worked with KS in profiling and analyzing the statistical significance of results, and KrS worked with KS in analyzing the distribution of words in gene proximal regions. GM was involved in analysis of results and critically analyzing the manuscript. PG is the group leader. Acknowledgements ================ The authors would like to thank Ms. Anishetty for discussions, and to acknowledge the financial support given by Council for Scientific and Industrial Research, Government of India and Department of Biotechnology, Government of India through the BTIS programme. We also extend our thanks to the developers of EMBOSS for making it available free of cost. We wish to acknowledge the contribution of the Free Software Foundation, MySQL, PERL community and others for making valuable software available free.

PubMed Central

2024-06-05T03:55:58.182968

2005-5-31

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166578/", "journal": "Theor Biol Med Model. 2005 May 31; 2:20", "authors": [ { "first": "Karthikeyan", "last": "Sivaraman" }, { "first": "Aswin Sai Narain", "last": "Seshasayee" }, { "first": "Krishnakumar", "last": "Swaminathan" }, { "first": "Geetha", "last": "Muthukumaran" }, { "first": "Gautam", "last": "Pennathur" } ] }

PMC1166579

Background ========== Colorectal cancer remains a leading cause of cancer-related mortality and morbidity in the Western world. In the UK it is the third commonest cancer, after lung and breast, with more than 35,000 new cases diagnosed each year and the second commonest cause of cancer-related mortality with approximately 16,000 deaths per year (CancerStats: Large Bowel -- UK, Cancer Research UK 2003, <http://www.cancerresearchuk.org>\[[@B1]\]). Complete surgical resection is the cornerstone of curative treatment. Other treatment modalities, such as chemotherapy and radiotherapy, administered before or after surgery, are used to counteract disease progression particularly in Stage III and IV disease \[[@B2],[@B3]\]. Despite continuing technical advances in the use of these techniques the improvement in overall five-year survival rate has been limited (5--6%). This fact, along with the high proportion of patients who present with tumours not amenable to surgery, drives the search for additional modalities of anti-cancer treatment. Immunotherapy, the use of the adoptive immune system to target cancer cells, holds much promise in this regard. The concept of immune surveillance, whereby the host immune system detects and removes cancerous cells, arose from the observation that immunodeficiency predisposes to the development of malignancies. Furthermore, it has been proposed that rare reports of solid tumour regression are the result of an anti-tumour immune response. Early work suggested that such anti-tumour responses were impotent against colorectal cancer \[[@B4]-[@B6]\]. However, evidence that colorectal malignancies may generate tumour specific antigens (TSA\'s) has encouraged considerable efforts to develop means of harnessing the host immune system to combat colorectal cancer. The use of immunotherapeutic approaches in colorectal cancer has evolved substantially (not reviewed here) and now focuses on the use of tumour specific antigens. This involves either passive immune therapy with antibodies targeted directly to tumour cells or by active immune therapy via vaccination with tumour cells, tumour cell lysates, peptides, carbohydrates, gene constructs encoding proteins, or anti-idiotypic antibodies that mimic TSA\'s. Despite some successes these techniques are still in their infancy and many hurdles remain, largely because our understanding of the biology of host-tumour interactions in colorectal cancer remains elementary. In the last ten years a distinct subset of colorectal cancer, with characteristics of widespread microsatellite instability, has emerged as a group in which immune responses may be pronounced. Clearly, if this type of colorectal cancer does indeed have particular immunogenic features, its study should provide vital advances in our understanding of colorectal tumour immunology. Here we review the literature in relation to putative immune responses in colorectal cancers with microsatellite instability. Microsatellite instability (MSI) -------------------------------- It is now clear that colorectal cancers may arise by a number of distinct molecular pathways. The \"classical pathway\" characterised by chromosomal instability (CSI) and a series of genetic events that drives the progression from normal colon through adenomatous polyp to invasive cancer accounts for 50--60% of cancers \[[@B7]\]. However, this is but one of many alternate pathways and one of these, involving mismatch repair dysfunction, leads to colorectal cancer characterised by microsatellite instability. High-degree microsatellite instability (MSI-H) is a feature of most colorectal cancers arising as part of the hereditary non-polyposis colorectal cancer (HNPCC) syndrome and also 15--20% of sporadic colorectal cancers \[[@B8]\]. Microsatellites are short segments of repetitive DNA bases found throughout the genome but mainly located in intronic (non-coding) segments. Due to their repetitive nature they are prone to insertion and deletion mutations during DNA replication, which commonly introduces such errors at a rate of 1 in 100,000 base pairs copied. The majority (99%) of these errors are detected and corrected by the inherent proof reading capabilities of DNA polymerase enzymes. A small proportion of mismatches that are missed by this mechanism are normally identified and corrected by a group of proteins collectively termed the mismatch repair (MMR) system \[[@B9]\]. Dysfunction of the MMR system allows the accumulation of replication errors and leads to microsatellite instability \[[@B10]\], defined as \"a change of any length due to either insertion or deletion mutations of repeating units in a microsatellite\" \[[@B11]\]. Mismatch repair dysfunction --------------------------- The molecular mechanisms that underlie MMR dysfunction are different in familial and sporadic colorectal cancers. In HNPCC colorectal cancers a germline mutation within an individual mismatch repair gene is inherited and a subsequent \"second hit\" \[Knudson\'s \'two-hit\' hypothesis \[[@B12]\]\] in the normal allele leads to mismatch repair deficiency. The commonly affected genes are hMLH1, hMSH2 and to a lesser extent PMS2, and their involvement predicts the development of extensive microsatellite instability \[[@B13]\]. By contrast, cancer cells with isolated mutations in hMSH3 and hMSH6 continue to maintain some mismatch repair function \[[@B14]-[@B16]\]. In sporadic MSI colorectal cancers genetic alterations in MMR genes are relatively rare. Instead epigenetic modulation of MMR genes, particularly hMLH1, such as hypermethylation of the promoter region leads to silencing and MMR deficiency \[[@B17],[@B18]\]. Hypermethylation is characteristic of a subset of colorectal cancers designated the CpG island methylator phenotype (CIMP+) that is relatively common in right sided tumours \[[@B19]\]. Sporadic MSI-H colorectal cancer appears to represent a subset of the methylator phenotype in which global hypermethylation silences MMR genes. These cancers have biological properties quite distinct from that of HNPCC tumours, in which MMR dysfunction is the sole abnormality, whereas widespread changes in methylation targets the expression of numerous other genes not involved in the MMR pathway \[[@B20]\]. All colorectal cancers display some degree of microsatellite instability if enough markers are studied \[[@B21]\]. In 1997 an International Collaborative Group laid down guidelines for the identification and assessment of microsatellite instability \[[@B22]\]. They classified tumours into microsatellite stable (MSS) and two groups of microsatellite unstable cancers, discriminated by their degree of instability: high (MSI-H) or low (MSI-L). Two mononucleotide microsatellite markers (BAT-25 and BAT-26) and three dinucleotide markers (D2S123, D5S346 and D17S250) were chosen as a panel to be studied for assignation of MSI status. Tumours displaying instability in two or more markers were designated MSI-H, instability in only one marker was designated MSI-L and if no markers were affected the tumour was designated MSS. There is now a growing consensus that the MSI-H clinico-pathological phenotype is characterised particularly by instability in the mononucleotide markers \[[@B21]\]. Colorectal cancers with MSI-H clearly demonstrate distinct biological behaviour whereas MSI-L cancers are less well characterised and are often grouped with cancers that show no instability (MSS) \[[@B23]\]. The focus of this review is the clinico-pathological phenotype characteristic of sporadic MSI-H colorectal cancer. ### MSI-H colorectal cancer Sporadic MSI cancers do not harbour some of the characteristic features or gene mutations associated with CSI cancers, such as aneuploidy and *p53*mutation. A number of human genes implicated in pathways regulating growth control and apoptosis, for example *IGF-IIR, TGFβ-RII*and *BAX*, contain microsatellite repeats within the coding regions of their transcribed sequences. More recently the *BRAF*gene has been noted to be commonly mutated in sporadic MSI-H, as well as in serrated polyps, suggesting that serrated polyps are likely precursors of these cancers \[[@B20]\]. It has been suggested that instability of such sequences plays a role in the progression of disease, presumably by altering gene activity \[[@B8],[@B24]\]. Altered regulation of all or some of these genes may be involved in cancer development but precise mechanisms are yet to be identified. Colorectal cancers with MSI-H also have distinct clinico-pathological characteristics. Sporadic MSI-H tumours are associated with the right side of the colon and display marked lymphocyte infiltration into tumour epithelium, stroma, peri-tumoral cuffs and Crohn\'s type reactions \[[@B25]\]. Sporadic MSI-H colorectal cancers have also specifically been associated with medullary-type poor differentiation, high mucin content and reduced lymph node involvement \[[@B20],[@B25]\]. ### MSI-H and improved survival Early reports on microsatellite instability noted an association of MSI with improved patient survival \[[@B26],[@B27]\]. Some small studies have failed to reproduce this finding \[[@B28]-[@B30]\] but generally the association of MSI-H with improved survival in sporadic cancers has been confirmed in larger studies \[[@B31]-[@B33]\]. Many further studies have followed that confirm improved prognosis for patients with sporadic MSI-H colorectal cancer, particularly in lymph node positive cancers, and the link to improved survival has now been confirmed in a systematic review \[[@B34]\]. One explanation offered for the improved survival of MSI-H colorectal cancer patients has been that these cancers may be more sensitive to adjuvant chemotherapy. Indeed early retrospective analyses suggested that cancers with high-degree microsatellite instability had improved response to 5-Fluorouracil (5FU)-based adjuvant chemotherapy \[[@B35]-[@B37]\]. However, these data were retrospectively derived from non-randomised groups and the effects of small numbers, younger age of patients receiving chemotherapy, and different responses of right and left-sided tumours might have confounded the findings. In contrast, two studies have reporting retrospective data derived from randomised control trials of chemotherapy conclude that 5-FU regimes confer no benefit in sporadic MSI-H colorectal cancers and may even be detrimental \[[@B38],[@B39]\]. There is clearly a need for robust prospective randomised controlled trials to evaluate the effects of 5-FU, and other agents, on survival with respect to MSI status. An alternative explanation for the improvement in patient survival is one of enhanced host immune response. An immunohistochemical study noted that the improved survival of patients with MSI-H was associated with the higher frequency of activated tumour infiltrating lymphocytes in these cancers \[[@B40]\]. Consequently, it has been suggested that these lymphocytes may actually represent a host immune response that contributes to improved survival and subsequent work has confirmed a possible link \[[@B41]\]. The hypothesis that colorectal cancers with MSI-H are more immunogenic than microsatellite stable cancers is very attractive, as study of these cancers may provide crucial insights into colorectal cancer immunology. The further elucidation of immune responses to colorectal cancer *in vivo*is crucial to the continuing development of immunotherapy. ### MSI-H and tumour immunogenicity Current concepts in solid tumour immunology identify certain features that are fundamental to the stimulation of an anti-tumour response \[Reviewed in \[[@B42]\]\]. Successful immune response requires a cancer to possess antigens that can be recognised as non-self during immune surveillance. In MSI-H colorectal cancer the accumulation of errors in microsatellites, that reside in gene exons, leads to the generation of a large number of abnormal peptides due to frameshift mutations: the so-called \"mutator phenotype\" \[[@B43]\]. Clearly, such an array of abnormal peptides might represent a pool of TSA\'s that render MSI-H tumours inherently more detectable by the host immune system. Studies on abnormal peptides characteristic of MSI-H have demonstrated their potential for immune recognition. A peptide generated by frameshift mutations in TGFβRII, characteristic of the mutator phenotype, has been used to pulse dendritic cells, which in turn induced activated cytotoxic (CD8+) T cells \[[@B44]\]. One cytotoxic T cell clone demonstrated specific anti-tumour activity against a cell line expressing this peptide in *in vitro*conditions. The same group have also identified MHC class-II restricted epitopes by induction of helper (CD4+) T cells from the peripheral blood mononuclear cells of normal volunteers and MSI-H colorectal cancer patients \[[@B45]\]. They also demonstrated that tumour infiltrating lymphocytes derived from an HNPCC MSI-H cancer recognised this frameshift-mutation-derived peptide but unusually the infiltrate in this tumour was predominantly CD4+ T cells. More recently, another group have identified several other frameshift-mutation-derived peptides as potential tumour specific antigens (TSA\'s) by using SEREX (serological analysis of recombinant cDNA expression cloning) \[[@B46]\]. They also detected IgG antibodies specific to these peptides only in the MSI-H cancer patients. In an HNPCC patient they demonstrated an antibody specific to a mutated CDX2 peptide that was found in the tumour tissue of the patient and this antibody was not seen in any of the other patients. Interestingly, the antibody disappeared seven years after curative resection. These observations lend significant support to the notion that MSI-H TSA\'s are effectively recognised by the host immune response and an immune response is induced, but more work is needed to confirm that findings in HNPCC patients are mirrored in sporadic MSI-H cancers. Tumour specific antigens must be appropriately presented to the host immune system if a specific immune response is to occur. This normally requires the presentation of antigen in conjunction with the appropriate Major Histocompatibility Complex (MHC) molecules that interact with the T cell receptor (TCR): Class I MHC for presentation to cytotoxic (CD8+) T cells and Class II MHC for presentation to helper (CD4+) T cells \[[@B47]\]. It has been argued that MSI-H colorectal cancers are incapable of eliciting cytotoxic responses *in vivo*due to the truncation of β~2~-microglobulin that is characteristic of these tumours \[[@B48]\]. The β~2~-microglobulin protein is an essential co-factor for MHC Class I and its defective function would render effective presentation of antigens impossible. However, whilst this defect negates antigen presentation by the tumour itself there is now considerable evidence that the direct presentation of antigen by a tumour may be relatively unimportant. Instead, the concept of \"cross-priming\" in which APC\'s such as dendritic cells, and to a lesser extent macrophages, pick up antigens released by dead tumour cells and subsequently present them to T cells may be more important *in vivo*\[[@B49],[@B50]\]. Thus, β~2~-microglobulin mutations may be of relatively little importance. Instead a pool of abnormal peptides generated by the mutator phenotype may be recognised and harvested as TSA\'s by immature dendritic cells. On acquisition of non-self antigens these dendritic cells mature into professional APC\'s, with a natural propensity to appropriately present antigen and co-stimulate T lymphocytes \[[@B50]\]. Antigen presenting cells may obviate the need for direct presentation by tumour cells and cross-priming may involve the use of HLA Class II machinery. Immunohistochemical studies have demonstrated that HLA Class II expression is increased in MSI-H colorectal cancer compared to MSS cancers, although this was not associated with improved survival \[[@B51]\]. Additionally, molecular analyses using single subtraction hybridisation \[[@B52]\] and oligonucleotide microarrays \[[@B53]\] have also confirmed increased Class II mRNA levels in MSI-H colorectal cancer. These findings support the *in vitro*studies of tumour infiltrating lymphocytes in MSI-H colorectal cancer, mentioned earlier, that demonstrated HLA Class II restricted T helper cell activity and the presence of dendritic cells \[[@B44],[@B45]\]. Either of these two populations may provide an alternative pathway by which cytotoxic T cells may be recruited and stimulated indirectly, but more work is needed to ascertain whether cytotoxic populations themselves are capable of recognising TSA\'s. It has recently become evident that appropriate co-stimulation is pivotal to effective antigen presentation \[[@B54]\]. Co-stimulation refers to the interaction between surface molecules on the antigen-presenting cell (APC), such as CD80 (B7.1) and CD86 (B7.2), and surface molecules on the T cells (CTLA-4, OX40, CD40-L). If these molecules do not bind each other during antigen-presentation then co-stimulation fails and the T cells become anergic, i.e. ineffective and tolerogenic \[[@B54]\]. A microarray analysis comparing the gene expression profiles of MSI-H colorectal cancers to MSS counterparts demonstrated increased signal intensity of CD80 and CD86 in the former group \[[@B53]\] but further studies to validate these findings are still required. After appropriate antigen-presentation and co-stimulation lymphocytes are activated in a tumour-specific response \[[@B49],[@B54]\]. They may then migrate to the tumour, infiltrate it and release mediators that can induce tumour cell death \[[@B47]\]. ### MSI-H colorectal cancer and lymphocyte infiltration For many years the presence of a pronounced infiltrate of lymphocytes in colorectal cancer has been associated with improved prognosis \[[@B55],[@B56]\] but conclusions have been confounded by inconsistencies. Lymphocytes associated with colorectal cancer can be sub-divided into those that invade tumour epithelium (intra-epithelial lymphocytes, IEL\'s), those confined to the stroma or those aggregated around the tumour (peri-tumoral lymphocytes). Previous studies on the significance of tumour infiltrating lymphocytes have been confusing due to variability in the definition of the subset being studied. More recently it has been shown that the presence of IEL\'s is positively associated with an improved survival in colorectal cancer in general \[[@B57]\] and specifically MSI-H colorectal cancer \[[@B41]\]. It has been suggested that they IEL\'s may represent evidence of an immune response to tumour antigens. In a study specific to MSI-H colorectal cancer, Dolcetti and colleagues demonstrated increased immunostaining for IEL\'s when compared to MSS cancers \[[@B58]\]. These lymphocytes were shown to be predominantly CD8+ in keeping with a cytotoxic lymphocytic response and these findings have been confirmed by other groups using immunohistochemistry and RT-PCR \[[@B41],[@B59]\]. However, debate continues as to whether these lymphocytes are of the αβT cell receptor lineage that are involved in tumour infiltration from the circulation or whether they are derived from γδ T cells that normally reside in the gastrointestinal tract \[[@B60]\]. The αβT cell phenotype represents lymphocytes that are targeting the tumour after antigen priming in lymphoid tissue, whereas the γδ T cells might represent a proliferation of lymphocytes resident in tumour epithelium. These characteristics of the lymphocyte infiltrates need further clarification. One immunohistochemical study, involving relatively small numbers and two groups that differed considerably in size (n = 17 v n = 7), has noted the increased presence of CD8+ CD103+ (α^E^β~7~+) lymphocytes in MSI-H colorectal cancer compared to MSS cancers \[[@B61]\]. The presence of CD103+ cells was much higher in tumour tissue than in adjacent colon in both groups and the authors\' postulate that this integrin subunit, which binds to E-Cadherin, has a role in migration of lymphocytes from stroma to tumour epithelium. They suggest that the differences in lymphocyte infiltration between the two groups are likely to be due to increased expression of CD103 by local lymphocytes in MSI-H cancer that promotes infiltration into the tumour. They also postulate that local mediators released by the tumour, such as TGFβ-I rather than a systemic immune response, leads to CD103 upregulation, but they provide no evidence to support this. The expression of CD103 may well facilitate lymphocyte infiltration from stroma to tumour, but whether the difference in IEL numbers between MSI-H and MSS is due to tumour-derived mediators or an immune response is unclear from this study \[[@B61]\]. The increased expression of cytolytic mediators by IEL\'s in MSI-H colorectal cancer suggests that these cytolytic T-cells are indeed activated. Increased expression of perforin, granzyme B and granulysin in MSI-H colorectal cancers \[[@B53],[@B59],[@B62]\] suggests that IEL infiltrates are attempting to induce tumour cell death. This is lent further credence by the finding that the activation marker IL2Rα is also expressed more significantly in these tumours \[[@B59]\]. No single activation marker has been shown to correlate with survival in studies to date but this may well due to the complexities of immune responses. The process of lymphocyte infiltration and anti-tumour activity is also influenced by the differential activity of helper (CD4+) T lymphocyte subsets and local cytokine profiles. Immune responses are promoted by the Th1 subset of CD4+ lymphocytes and their associated cytokines \[Interferon (IFN)-γ, Tumour necrosis factor (TNF)-α, Interleukin (IL)-2 and IL-12). By contrast the Th2 subset of CD4+ lymphocytes and cytokines such as IL-10 promote tolerance to unrecognised antigens and prevent successful antigen-specific responses. A microarray analysis performed in our laboratory has demonstrated the upregulation of several pro-inflammatory cytokines in MSI-H colorectal cancer \[[@B53]\]. As well as the common players this analysis has identified potentially crucial roles for IL-18 and IL-15, both of which are capable of co-ordinating innate and acquired responses \[[@B63],[@B64]\]. Clearly, activity of these interleukins should promote the likelihood of a successful immune response. This microarray analysis also demonstrated that the heat shock protein family is up-regulated in MSI-H colorectal cancer \[[@B53]\]. This is highly significant in the context of previous work by Srivastava and others that has demonstrated the role of heat shock proteins in innate and antigen-specific immune responses \[[@B65],[@B66]\]. Heat shock proteins act as chaperones that are actively involved in regulating the folding of newly synthesised proteins \[[@B67]\]. Indeed their name derives from the observation that these proteins are over-expressed in response to heat shock, and other stresses, that lead to native protein unfolding. The native fold of a protein is encoded by its amino acid sequence and it is therefore attractive to suggest that cancers with high-degree microsatellite instability, that generate large numbers of abnormal proteins with consequent misfolding, might excite a heat shock protein response. Heat shock proteins are used as natural adjuvants in immunotherapy trials and their up-regulation in MSI-H tumours may reflect a similar function in naturally occurring immune responses \[[@B68],[@B69]\]. Our microarray findings have been validated using quantitative real-time RT-PCR across several genes of interest and demonstrate clear differences in the tumour biology of MSI-H and MSS cancers at the mRNA level. Tumour escape and counter-attack -------------------------------- To counteract immune surveillance solid cancers may employ mechanisms to evade the immune system. Several possible methods of immune escape have been identified in colorectal cancer, such as the down-regulation of MHC proteins and co-stimulatory molecules \[[@B70],[@B71]\]. The available evidence suggests that MSI-H colorectal cancers demonstrate higher levels of MHC Class II machinery than MSS counterparts, although the inefficacy of HLA Class I molecules in MSI-H tumours has been proposed as a means of tumour escape \[[@B48]\]. The significance of this in light of the potential for immune cross-priming is controversial, as discussed earlier. Alternative mechanisms of immune escape identified in colorectal cancer include the release of mediators that suppress the function of tumour infiltrating lymphocytes: so-called \"tumour counter-attack\". In the past *in vitro*studies of lymphocytes infiltrating colorectal cancer have failed to demonstrate specific anti-tumour activity even though analysis of their receptor usage strongly suggested antigen-specificity \[[@B72],[@B73]\]. These studies did not discriminate tumours according to microsatellite status but one explanation offered for the findings was that the presence of anti-lymphocyte agents, released by the tumour itself, neutralized the activity of tumour infiltrating lymphocytes. The fatty acid synthase (Fas)/Fas-Ligand system has been implicated as a pathway of colorectal cancer counter-attack \[[@B74]\]. Fas-L released by colorectal cancers may bind to Fas receptors on infiltrating lymphocytes and induce their apoptosis. Infiltrating lymphocytes may release Fas-L themselves but colorectal cancers evade the detrimental effects of this by uncoupling their Fas receptors from the intra-cellular machinery that mediates cell death. An immunohistochemical study examining membrane-bound Fas-L expression found that, contrary to expectation, Fas-L expression is not increased in MSS cancers and thus does not explain the less prominent lymphocyte infiltrates seen in these cancers \[[@B75]\]. The authors conclude that some other, as yet unidentified, mechanism of counter-attack must exist in these cancers to account for the reduced infiltrate in MSS colorectal cancer rather than attributing the difference to variable degrees of immunogenicity. MSI-H colorectal cancer and apoptosis ------------------------------------- Dolcetti and colleagues showed that MSI-H colorectal cancers stained significantly more heavily for markers of apoptotic cell death \[[@B76]\]. Subsequent follow-up analysis of the patients from whom the tumours had been removed showed that increased apoptosis counts correlated well with improved survival \[[@B40]\]. However, Michael-Robinson and co-workers also studied apoptosis in these cancers using TUNEL and though they confirmed the presence of more apoptotic cells, they found no co-localisation between these cells and infiltrating lymphocytes \[[@B77]\]. They concluded therefore that increased T cell infiltrates were not likely to be responsible for more marked tumour cell death. However, it may be argued that the release cytokine mediators of target cell death by infiltrating lymphocytes may render the need for direct co-localisation obsolete. In our opinion as MSI-H colorectal cancer does appear to be infiltrated by an activated cytotoxic T cell population and the tumours show high levels of apoptosis the two are likely to be linked but we acknowledge the need for further clarification. An alternative explanation for the increased apoptosis observed in MSI-H colorectal cancer has been the theory of \"effete malignancy\". This concept proposes that the mutator phenotype affects peptides or pathways involved in cell processes fundamental to life. Thus, the accumulation of mutations adversely influences the tumour cells\' viability and hence, leads to apoptotic cell death \[[@B78]\]. The anti-apoptotic bcl-2 protein has been noted to be under-expressed in MSI-H colorectal cancers \[[@B79]\] but the pro-apoptotic gene BAX also appears to be a common target for mutation in the MSI pathway \[[@B80]\], an event that may promote clonal expansion and tumourigenesis. These observations confirm the complex role of apoptosis in tumour development and progression. The apoptosis observed in these cancers, whether induced by effete malignancy or an immune response, is of significance for another reason. The death of tumour cells releases intra-cellular proteins into the tumour micro-environment. This would be a vital step in the promotion of TSA release, recognition and presentation to the host T lymphocytes. It might also release heat shock proteins, upregulated in MSI-H colorectal canecr, which may promote both innate and antigen-specific arms of the immune response. The importance of dendritic cells in immune responses has already been mentioned \[[@B49]\] and immunohistochemical analysis of colorectal cancer has demonstrated the presence of dendritic cells at tumour margins \[[@B81]\]. The presence of this population has not been evaluated in MSI-H cancers specifically but the presence of dendritic cell infiltrates has been associated with reduced tumour progression in colorectal cancer \[[@B82]\]. Conclusion ========== Clearly, the mutator phenotype may influence the function of several cellular processes and these may include those that are designed to nullify the host response. Equally it has been suggested that it is the effect of microsatellite instability on other cellular processes, unrelated to tumour immunology, which leads to the survival benefit observed in MSI-H colorectal cancer. Effete malignancy may, for instance, limit a tumour\'s ability grow and survive \[[@B78]\]. Alternatively processes critical to tumour metastasis may be adversely affected by the mutator phenotype. The down-regulation of VEGF in MSI-H colorectal cancer is a one example of how these tumours may suffer reduced metastatic potential \[[@B83]\]. Certainly many of the biological characteristics of MSI-H, such as diploidy, wild-type p53 expression, activating β-catenin mutations and the CIMP+ phenotype, have all been associated with improved prognosis \[[@B84]\]. However, the effects of all these factors and that of an immune response may be additive, rather than exclusive and they may all contribute to the improvement in outcome. There is now an accumulation of evidence that promotes the argument that MSI-H colorectal cancer is subject to a marked, possibly antigen-specific, immune response. We acknowledge that this may not be the sole determinant of improved prognosis although it seems likely to be a major influence. We believe that MSI-H may be a natural paradigm of host-tumour interactions in immunogenic colorectal cancers and as such further studies are required to clarify the nature of immune responses in these tumours. The use of heat shock proteins and abnormal peptides, derived from frameshift mutations in coding microsatellites, already hold promise in advancing the field of immunotherapy \[[@B85],[@B86]\]. Ultimately the demonstration of specific anti-tumour activity in cytotoxic lymphocytes from patients with MSI-H colorectal would provide the best evidence in support of such a response. Such studies should enhance our understanding of the immune responses in MSI-H tumours and the tumour immunology of colorectal cancer in general. This may be crucial to the advance of efforts in the field of immunotherapy. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AB searched and evaluated the literature and drafted the manuscript. SAB and SD revised the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This study was funded by the Bowel and Cancer Research (Registered charity no. 328667, 4th Floor Alexandra Wing, The Royal London Hospital, Whitechapel, London E1 1BB, UK) the funding source had no influence over the content or conclusions of this manuscript.

PubMed Central

2024-06-05T03:55:58.185575

2005-5-12

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166579/", "journal": "World J Surg Oncol. 2005 May 12; 3:26", "authors": [ { "first": "Ayan", "last": "Banerjea" }, { "first": "Stephen A", "last": "Bustin" }, { "first": "Sina", "last": "Dorudi" } ] }

PMC1166580

Epidemiology of hepatocellular carcinoma ======================================== Background ---------- Hepatocellular carcinoma (HCC) is one of the most common malignant tumors, representing more than 5% of all cancers. The estimated annual number of cases exceeds 500,000 \[[@B1]\], with a mean annual incidence of around 3--4% \[[@B2]\]. In terms of relative frequencies, HCC ranks as the fifth most common cancer in the world, it is also the fifth among men and eighth among women; it is the second among cancers of the digestive tract after stomach cancer \[[@B3]\]. The incidence of HCC varies considerably with the geographic area because of differences in the major causative factors. The geographic areas at the highest risk are located in Eastern Asia, with age-adjusted incidence rates (AAIR) ranging from 27.6 to 36.6 per 100,000 in men; Middle Africa (AAIR 20.8--31.1/100,000) and some Western African countries (30--48/100,000). The geographic areas at lowest risk are Northern Europe, Australia, New Zealand and the Caucasian populations of North and Latin America (AAIR 1.5--3.0). In Southern Europe, AAIR is around 10 per 100,000 in men \[[@B3]\]. The most powerful risk factor for development of HCC is the existence of liver cirrhosis, regardless of its etiology \[[@B4]\]. Among cirrhotics, viral infection and high alcohol intake are associated with the highest risk \[[@B5]-[@B8]\]. Of the primary hepatitis viruses, only hepatitis B and C viruses cause HCC \[[@B9]\]. Hepatitis A and E viruses do not produce long-term pathological sequelae. Although hepatitis D virus (HDV) always occurs as co-infection with hepatitis B virus and leads to severe acute or chronic hepatic disease, there is controversy whether it increases the carcinogenic potential \[[@B10],[@B11]\]. Risk factors for development of HCC ----------------------------------- ### Hepatitis B Hepatitis B virus (HBV) infection is a major public health problem. It is estimated that two billion people have been infected worldwide and 360 million suffer from chronic HBV infection \[[@B12]\]. Over 520,000 die each year, 50,000 from acute hepatitis B, 470,000 from cirrhosis and liver cancer. In South-East Asia hepatitis B is mostly acquired perinatally from an infected mother. In sub-Saharan Africa, it is mostly acquired in early childhood by horizontal infection, whereas in Northwestern Europe, North America and Australia infection is mainly through sexual contact or needle sharing among injecting drug users, with a peak incidence in the 15--25 age group \[[@B12]\]. Infection acquired perinatally and in early childhood is usually asymptomatic, becoming chronic in 90 and 30% of cases, respectively. In adults, infection resolves in \>95% with loss of serum HBsAg and the appearance of anti-HBs. Chronic infection is characterized by the persistence of HBsAg for more than 6 months. Acute hepatitis B usually results in complete recovery with little if any risk of HCC. In cases with persistent HBV infection, HBV is one of the most important risk factors for HCC. Chronic HBV infection presents as one of three potentially successive phases: *immune tolerant*, *immune active*and *low- or non-replicative*. In the *immune tolerant phase*, serum HBsAg and HBeAg are detectable, serum HBV DNA levels are high, serum aminotransferases are normal or minimally elevated. In the *immune active phase*, serum HBV DNA levels decrease and serum aminotransferase levels increase. Flares of aminotransferases may be observed, in some patients these flares are followed by HBeAg-anti-HBe seroconversion. Following this conversion, in the *low*- *or non-replicative phase*the HBV replication persists but at a very low level suppressed by the host immune response. HBV DNA in serum is undetectable by conventional, non-PCR based techniques. This phase is also called the \'*inactive carrier state*\'. It may lead to resolution of HBV infection where HBsAg becomes undetectable and anti-HBs is detected, anti-HBc staying positive as sign of contact with the virus. Recently it has been reported that HBV DNA can persist in the serum and liver tissue even after negativation of HBsAg \[[@B13]\]. Recent advances in molecular technology have allowed the isolation of HBV variants that either cannot produce HBeAg or produce it less efficiently, based on precore stop codon mutation and mutations in the core promoter region respectively. In patients with HBV variants, progressive liver damage occurs in parallel with relatively high levels of viremia. In perinatally infected people, the immunotolerant phase lasts till the age of 15--35 years, after which hepatitis flares may occur, leading eventually to viral remission. In patients infected during later childhood or adulthood, there is no immunotolerant phase. Most studies on the risk of developing HCC in chronic HBV infection have been performed in the Far East. Here, most patients acquired the HBV infection as newborn infants \[[@B14]\]. It has been noted that the probability of acquiring HCC increases with severity of liver disease. The annual risk of HCC is 0.5% for asymptomatic HBsAg carriers and 0.8% for patients with chronic hepatitis B \[[@B15]\]. Patients with HBV-cirrhosis have a 1000 times higher risk of developing HCC compared to a HBsAg negative control group \[[@B16]\]. The incidence of HCC in compensated cirrhosis due to HBV from Asia was 2.7%. In Japan, the mean interval between the time of initial infection with HBV and the occurrence of HCC is 50 years. As most people here are infected at birth, HBV related liver cirrhosis usually develops in patients in their 40\'s and HCC in their 50\'s \[[@B17]\]. Few adequate studies have been performed in the West to address the issue of the incidence of HCC in persons who are positive for HBsAg. Most of the studies in Western countries have included small numbers of HBsAg positive patients and/or have not specifically analyzed the group of HBsAg carriers. There is also lack of uniformity in the timing of initiation of follow-up monitoring. In a cohort of 350 Western European patients with compensated cirrhosis followed for a mean period of 6 years, the 5-year cumulative incidence of HCC was 6% \[[@B18],[@B19]\]. The incidence was 2.2% in a series of 179 untreated Caucasian patients \[[@B19],[@B20]\]. In a retrospective analysis of cirrhotic European patients with HBV infection, the 5-year incidence of HCC was 9% irrespective of HBeAg or HBV DNA status at the time of diagnosis of cirrhosis \[[@B21]\]. The hepatitis B replication status seems to play an important role in determining the risk of development of HCC \[[@B22]-[@B24]\]. A recent study found that whereas the relative risk of HCC among men with HBsAg alone was 9.6 compared to those without HBsAg, the risk increased to 60.2 when they were positive for both HBsAg and HBeAg \[[@B23]\]. Another analysis showed that the level of HBV DNA is a prognostic marker for HBV-related HCC and that HCC patients with a less favorable course appear to either clear the virus poorly or to have a greater level of virus production \[[@B24]\]. It was recently demonstrated that positivity for anti-HBc alone in absence of HBsAg and anti-HCV is not rare in Japanese patients with HCC, which may indicate that HBV virus might be involved in so-called non-B HCC \[[@B25]\]. The entire nucleotide sequences of HBV genomes have been classified into 8 genotypes (A-H), with predominance of genotypes A and D in Western countries, and B and C in Southeast Asia and the Far East \[[@B26]-[@B29]\]. Several studies from the Far East evaluated the association between distinct genotypes and severity of liver disease. Genotype C was shown to be associated with the development of liver cirrhosis and HCC in Taiwan \[[@B30]\], China \[[@B31]\] and Japan \[[@B32]\], whereas genotype B was shown rarely to be associated with the development of HCC in China and Japan. In contrast, in Taiwan genotype B is the predominant type in patients with HCC who are younger than 35 years \[[@B30]\]. Another study from Taiwan showed that patients with genotype C had a greater tumor recurrence rate after curative resection of HCC compared with those with genotype B \[[@B33]\]. It was also shown that the likelihood of presence of T1762/A1764 mutations in the basal core promoter parallels the progression of liver disease, and that this mutation is found more frequent in HBV genotype C than B patients \[[@B34]\]. PreS deletions were shown to be more frequent in patients with HBV genotype C, and associated with more advanced disease such as liver cirrhosis and hepatocellular carcinoma \[[@B35]\]. ### Hepatitis C Hepatitis C is also a major public health problem. There are more than 170 million people infected worldwide \[[@B36]\]. Approximately 80% of HCV infected patients develop chronic hepatitis C. About 20% of these patients will develop severe chronic hepatitis C and cirrhosis, which becomes detectable in the second and third decade after infection. The natural history of chronic hepatitis C infection is characterized by a predominantly asymptomatic course and a variable clinical outcome. For these reasons it is difficult to define the rate of progression to cirrhosis and HCC. The risk of cirrhosis in chronic hepatitis C is less than 10% in women infected at a young age and \>30% in men infected after the age of 40 over a 20 year period \[[@B37],[@B38]\]. Five prospective studies from Europe and the US have shown that during the first 10--15 years after initial infection, liver cancer is a rare occurrence \[[@B39]-[@B43]\]. In patients with hepatitis C, there is an increased risk of HCC coinciding with the establishment of cirrhosis with yearly incidence between 3--8% \[[@B6],[@B7],[@B44]-[@B47]\]. In Japan, the mean interval between infection and development of HCC is 30 years \[[@B48]\]. A study from the US shows a long time lag (mean 28 years, range 8--42) between transfusion-associated hepatitis and development of HCC \[[@B49]\]. There is conflicting information on the relationship between HCV genotype and progression to HCC in longitudinal studies. It is suggested by some authors that genotype 1b (most prevalent in Europe and Japan) is associated with a higher incidence of HCC than infection with other genotypes \[[@B50],[@B51]\]. In other studies, however, this was not observed \[[@B52],[@B53]\]. ### Coinfection of HBV and HCV Both HBV and HCV are transmitted parenterally and coinfection is not uncommon in intravenous drug users and in countries with a high prevalence of HBV \[[@B54]\]. Coinfection of HBV and HCV seems to result in more severe liver disease than either infection alone \[[@B55]\]. The risk of developing HCC in subjects with both infections has been investigated in a meta-analysis of 32 epidemiological studies between 1993 and 1997 \[[@B56]\]. The odds ratio for development of HCC in HBsAg positive, anti-HCV/HCV RNA negative subjects was 20.4; in HBsAg negative, anti-HCV/HCV RNA positive subjects 23.6; and subjects positive for both markers 135. These data suggest a more than additive but less than multiplicative effect of HBV and HCV coinfection on the relative risk for HCC. The viruses may act through common as well as different pathways in the carcinogenic process. It has been reported that HBV DNA is still present after seroconversion of HBsAg in patients with hepatitis B. Several reports indicate that prior HBV infection, characterized by presence of anti-HBc, affects the development of HCC in patients infected with HCV \[[@B57]-[@B59]\]. Given these data, in patients with chronic HCV infection, serologic markers of past HBV infection should be checked, not just HBsAg. Other authors, however, were not able to document any adverse event of occult HBV infection on the clinicopathologic course of chronic HCV infection \[[@B60]\]. In case of coinfection with HBV (whether active or past), a more aggressive surveillance to detect early HCC could be suggested \[[@B61]\]. However, to date screening and surveillance programs have not demonstrated a significant survival benefit. In view of the role of HBV as cofactor in the development of HCV related cirrhosis and HCC, vaccination of patients with chronic hepatitis C against HBV has been advocated with the presumption of avoiding additional liver injury \[[@B62],[@B63]\]. ### Coinfection of HBV and HDV Verme *et al*\[[@B11]\] suggested that HBsAg positive patients with HDV superinfection develop cirrhosis and HCC at an earlier stage (mean age 48 year) than HBsAg carriers without HDV infection (mean age 62 years). ### Coinfection of HBV and HCV with HIV Coinfection of HBV and HCV with HIV is common because these diseases share the same routes of transmission. Recently a series of HCC in HIV-HCV coinfected patients was published, indicating an unusually rapid development of HCC in these patients \[[@B64]\]. This is not surprising, as chronic hepatitis C is more aggressive in HIV positive subjects, leading to cirrhosis and end-stage liver disease in a shorter period of time \[[@B65]\]. ### Coinfection of HCV and *S. mansoni* An Egyptian study showed that Schistosoma infection increased the risk of HCC, only in the presence of HCV, whereas isolated *S. mansoni*infection does not \[[@B66]\]. ### Role of alcohol consumption in HBV or HCV infection Reports suggest that HBV and ethanol act synergistically to promote HCC \[[@B67],[@B68]\]. Habitual heavy drinking was reported to be a significant risk factor for HCC in patients with HCV-related liver cirrhosis by multiple logistic regression analysis \[[@B57]\]. A recent study showed synergism between alcohol drinking and HBV or HCV infection, with approximately a twofold increase in the odds ratio for each hepatitis virus infection for drinkers\' \> 60 g/d, with a more than additive but less than multiplicative risk \[[@B69]\]. Although two case-control studies did not show a relationship of alcohol consumption with the occurrence of HCC \[[@B70],[@B71]\], another case-control study found a positive interaction between HBsAg positivity and HCV RNA positivity and heavy alcohol intake in the development of HCC \[[@B72]\]. Furthermore, Hassan et al. \[[@B73]\] showed synergistic interaction (more than additive) between heavy alcohol consumption ≥ 80 ml/d and chronic HBV or HCV infection (odds ratio 53.9) and insulin or non-insulin dependent diabetes mellitus (odds ratio 9.9). Incidence of HBV- and HCV-related HCC worldwide ----------------------------------------------- Chronic hepatitis B and C infection are responsible for the great majority of cases of HCC worldwide \[[@B9]\]. They also account for the peculiar geographical distribution of the tumor. The relative frequencies of HBV and/or HCV related HCC in the world is illustrated in Table [1](#T1){ref-type="table"}\[[@B17],[@B72],[@B74]-[@B93]\]. The worldwide incidence of HCC varies and is predominantly related to the regional prevalence of chronic viral hepatitis and its associated chronic liver disease and cirrhosis. Aflatoxin intake has a role in the genesis of HCC only in patients who have pre-existing chronic hepatitis B \[[@B84]\]. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Relative frequencies of HBV and HCV related HCC in the world ::: **Author \[reference\]** **Country** **Era** **Sample size** **HBsAg + (%)** **Anti-HCV + (%)** **HBsAg/anti HCV + (%)** **Other (%)** --------------------------- -------------- ---------- ----------------- ----------------- -------------------- -------------------------- --------------- Chen, 1990 \[74\] Taiwan NR 66 **35 (53.0)** 15 (22.7) 7 (10.6) 9 (13.6) Chuang, 1991 \[75\] Taiwan NR 128 **87 (68.0)** 13 (10.1) 12 (9.4) 16 (12.5) Lee, 1992 \[76\] Taiwan NR 326 **233 (71.5)** 31 (9.5) 10 (3.1) 52 (15.9) Jeng, 1991 \[77\] Taiwan NR 129 **62 (48.1)** 29 (22.5) 19 (14.7) 19 (14.7) Leung, 1992 \[78\] Hong Kong 1986--90 424 **341 (80.3)** 16 (3.8) 15 (4.0) 52 (12.3) Nishioka, 1990 \[79\] Japan NR 180 64 (35.6) **80 (44.4)** 11 (6.1) 25 (13.9) Saito, 1990 \[80\] Japan NR 253 49 (19.4) **136 (53.8)** 2 (0.8) 66 (26.1) Kiyosawa, 1990 \[17\] Japan 1958--89 83 19 (22.9) **51 (61.4)** 10 (12.0) 3 (3.6) Hassan, 2001 \[81\] Egypt 1995--96 33 5 (15.2) **25 (75.8)** NR NR Kew, 1990 \[82\] South Africa NR 380 137 (36.1) 63 (16.6) 47 (12.4) 127 (33.4) Yu, 1990 \[83\] USA 1984--89 58 22 (37.9) **36 (62.1)** NR NR Di Bisceglie, 1991 \[84\] USA 1987--88 99 7 (7) 12 (12) 1 (1) **79 (79)** Hadziyannis, 1995 \[85\] Greece 1991--92 65 **33 (50.8)** 5 (7.6) 3 (4.5) 23 (38.3) Colombo, 1989 \[86\] Italy 1975--88 132 19 (14.4) **64 (48.5)** 22 (16.7) 27 (20.5) Levrero, 1991 \[87\] Italy 1980--88 167 38 (22.8) **82 (49.1)** 15 (9.0) 32 (19.2) Simonetti, 1992 \[88\] Italy 1982--88 212 15 (7.1) **133 (62.7)** 18 (8.5) 46 (21.7) Donato, 1997 \[72\] Italy 1995--96 172 37 (21.5) 65 (37.8) 4 (2.3) 66 (38.4) Stroffolini, 1998 \[89\] Italy 1996--97 1083 125 (11.5) **771 (71.2)** 55 (5.1) 132 (12.2) Bruix, 1989 \[90\] Spain NR 96 4 (4.2) **67 (69.8)** 5 (5.2) 20 (20.8) Nalpas, 1991 \[91\] France 1982--89 55 3 (5.5) **28 (50.9)** 9 (16.3) 15 (27.3) Van Roey, 2000 \[92\] Belgium 90s 154 37 (24.0) **62 (40.0)** NR 55 (36.0) Haydon, 1997 \[93\] UK 1985--94 80 13 (16.3) 22 (27.5) 2 (2.5) **43 (53.8)** NR: not reported; Bold: predominant cause ::: In the Far East and sub-Saharan Africa, where HBV is highly endemic, HBV is the main cause of HCC. In areas with an intermediate rate of liver tumors such as Southern Europe, Egypt and Japan, HCV is the predominant cause of HCC. Here HCC is mostly discovered at an older age in patients with longstanding cirrhosis due to HCV. In regions with a low incidence of HCC such as Northern Europe and the United States, HCC related to HCV or HBV infection are found in a minority of cases and the tumor is often related to other factors such as alcoholic liver disease. In these low endemic areas, HCC is usually discovered at an older age in patients with longstanding cirrhosis due to alcohol abuse \[[@B72]\]. In France, ethanol is still the leading cause of cirrhosis and was responsible for 60% of all HCC causes during the last decade \[[@B8]\]. Time trends in the incidence of HCC ----------------------------------- An important epidemiological fact is the rising incidence of HCC in developed countries during the last two decades \[[@B79],[@B89],[@B95],[@B99]\](Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Time trends on the incidence of HCC in the world ::: **Author \[reference\]** **Country** **Number/100,000 era 1** **Number/100,000 era 2** ------------------------------- ------------- -------------------------- -------------------------- El Serag, 1999 \[95\] USA 1976--80: 1.4 1991--95: 2.4 El Serag, 2000 \[96\] USA 1993--95: 2.3 1996--98: 7.0 Benhamiche, 1998 \[97\] (men) France 1976--79: 7.5 1992--95: 10.2 Stroffolini, 1998 \[89\] Italy 1969: 4.8 1994: 10.9 Law, 2000 \[98\] (men) Australia 1983--85: 2.1 1995--96: 4.0 Nishioka, 1991 \[79\] Japan 1968--77: 9.5 1984--85: 16.0 Yoshizawa, 2002 \[99\] Japan 1980: ca 10 2000: ca 40 ::: In [Japan]{.underline}, the HCC-related mortality rate has sharply increased since 1975 from 10/100,000 to almost 40/100,000 in 2000 \[[@B99]\]. An analysis of the Shinshu University Hospital (Japan) showed a change in etiology of the HCC \[[@B100]\]. Whereas in the 1971--1980 decade, hepatitis B was the predominant cause of HCC, in the 1991--1995 period hepatitis C was largely predominant (Table [3](#T3){ref-type="table"}). However, the total numbers of yearly deaths because of HCC in HBsAg carriers\' stays constant, approximately 10% in the survey conducted in 1995. The rapid increase of mortality due to HCC in Japan is mainly attributable (ca 80%) to persistent infection with HCV \[[@B99]\]. The hepatitis C epidemic in Japan originated due to intravenous drug use by the young generation after World War II during the late 40s and early 50s. It spread in the general population due to remunerated blood donors. Abrogation of paid blood donation in 1968, exclusion of blood units contaminated with HBV in 1973 and HCV in 1989 decreased the risk of posttransfusion hepatitis from \> 50% in the 60s to almost zero at present. The incidence of HCV in Japan is decreasing. As the interval between the time of the initial infection with the hepatitis C virus and the development of HCC is 30 years \[[@B79]\], the growing incidence of HCC in Japan is expected to reach a plateau around the year 2015, and then to decrease \[[@B99]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Changing causes of HCC in Japan, 1971--95 ::: **Author \[reference\]** **Era** **Sample size** **HBsAg + (%)** **Anti-HCV + (%)** **HBsAg/anti HCV + (%)** **Other (%)** -------------------------- ---------- ----------------- ----------------- -------------------- -------------------------- --------------- Kiyosawa, 1992 \[100\] 1971--80 112 **60 (54%)** 38 (34%) 5 (4%) 9 (8%) 1981--90 267 82 (31%) **159 (59%)** 4 (2%) 22 (8%) 1991--95 162 21 (13%) **126 (78%)** 5 (3%) 10 (6%) Bold: predominant cause ::: Also in [Italy]{.underline} the mortality rate of HCC is rising \[[@B89]\] from 4.8/100,000 in 1969 to 10.9/100,000 in 1994, reflecting the large cohort of subjects infected with HCV through iatrogenic route during the 50s and 60s when glass syringes were commonly used for medical treatment. Likewise in [Australia, France and the United States of America (US)]{.underline} the HCC mortality is increasing, most probably because people infected with HCV have grown old and reach the cancer-bearing age \[[@B95]-[@B98]\]. In the [US]{.underline}, an increase of about 80% in the incidence of HCC over the past 20--30 years is described, it is estimated that approximately 15,000 new cases occur each year. Also in [France]{.underline} the incidence of HCC is steadily and markedly increased, the estimated number being about 4,000 per year \[[@B101]\]. Although the prevalence of HCV is declining in developed countries because of the decline in incidence in the 90s, the number of persons infected for ≥ 20 years is expected to increase substantially before peaking in 2015 \[[@B102]\]. Analysis of long-term serial HCV samples from the US and Japan suggest that HCV was introduced into the US population around 100 years ago and widely disseminated in the 1960s. In contrast, HCV was introduced in Japan \> 100 years ago and widely disseminated in the 1930s and 40s. The HCV genotype 1b population in Japan started to decrease around 1995 whereas HCV genotype 1a in the US is still growing exponentially. It is predicted that an increased HCC prevalence will occur in the US over the next two to three decades \[[@B103]\]. The reasons advocated for explaining the increased incidence of HCC are the increased rate of HCV infection and an improvement of the clinical management of cirrhotic patients. Enhancing the survival of patients with advanced cirrhosis leads to an increased incidence of HCC. In fact, a decade ago, most of the deaths in cirrhotic patients were due to digestive hemorrhage or bacterial infections, two conditions that are now efficiently prevented and cured \[[@B104]\]. Therefore, HCC has become the leading cause of death in patients with cirrhosis. Screening tests for HCC in patients with chronic viral hepatitis ---------------------------------------------------------------- Despite knowledge of the risk factors for HCC, screening of HCC is controversial, as there have been no randomized controlled studies demonstrating the efficacy of screening for HCC. As HCC mostly occurs in patients with cirrhosis, or at least advanced fibrosis, most studies have been performed in these patients at risk. The most frequently used tests have been serum alfa-fetoprotein (AFP) and hepatic ultrasound (US). There is one non randomized prospective cohort study suggesting that HCC was detected earlier and was more often resectable in patients who had twice yearly screening with serum AFP and hepatic US than in patients who had usual care \[[@B105]\]. Twenty-four studies, which included patients with chronic hepatitis B or C or both, addressed the sensitivities and specificities of screening tests \[[@B106]\]. Serum AFP for detection of HCC was evaluated in 19 studies. They were relatively consistent in showing that the sensitivity of serum AFP for detecting HCC increases from very low levels to moderately high levels of 60 to 80% as the threshold value decreased from 400 to 10 ng/mL, with corresponding specificity decreasing from 100 to 70--90%. A threshold between 10 and 19 ng/mL seems most appropriate as sensitivity usually is moderately high (45 to 100%), with a specificity of 70 to 90%. It has been shown that AFP is not always specific for HCC and titers can increase with flares of active hepatitis \[[@B107]\]. Seven studies evaluated screening with US, reporting high specificity of 95--100%, but variable sensitivity, varying from 11--99% \[[@B94]\]. A surveillance study combining US and AFP in 1,125 patients with HCV, HBV or both, reported a sensitivity of 100% when using a serum AFP \> 10 ng/mL together with US, compared with a sensitivity of 75% using only AFP \> 10 ng/mL and a sensitivity of 87% when using US alone \[[@B108]\]. Computed tomography and magnetic resonance imaging have a high sensitivity and specificity in detecting HCC, but are too expensive to be used in surveillance \[[@B1]\]. The surveillance intervals studied varied from 3 to 12 months. In a study of patients with hepatitis B, the most rapidly growing tumor increased from 1 to 3 cm in 5 months \[[@B109]\]. The ideal time for re-screening has not been identified. Some investigators suggest a 4--5 month interval, others have suggested that a 6-month interval may be most appropriate \[[@B109],[@B110]\]. It is suggested that in case of concomitant HBV and HCV infection serum AFP levels should be obtained every 3 months, and that persistent AFP levels should prompt an aggressive imaging search for HCC \[[@B61]\]. It can be concluded that screening patients with advanced chronic hepatitis B or C with AFP and US may improve detection of HCC, but further studies are needed whether screening improves clinical outcomes. Pathogenesis of hepatitis B and C-induced hepatocellular carcinoma ================================================================== Introduction ------------ Epidemiologic data indicate that chronic hepatitis B and C are independent risk factors for development of HCC \[[@B7],[@B16]\]. Furthermore, animal models confirm the oncogenic potential of HBV and HCV in the liver: transgenic mice for hepatitis B and C \[[@B110],[@B111]\], and natural models such as the woodchuck infected with the woodchuck hepatitis virus, a hepadnavirus closely related to the HBV \[[@B112]\]. Carcinogenesis is believed to be a multistage process, occurring through a sequence of steps termed *initiation*, *promotion*and *progression*. This process evolves over several or many years. Tumor *initiation*begins in cells through mutations induced by exposure to carcinogens. DNA changes, maintained during successive cell divisions, activation of oncogenes and inactivation of suppressor genes lead to dysregulation of the cell division and to immortalization \[[@B113]\]. Tumor-initiated cells have a decreased responsiveness to both intercellular and intracellular signals that maintain normal cellular architecture and regulate homeostatic growth. Tumor *promotion*results in a further selective clonal expansion of initiated cells. During tumor *progression*, pre-malignant cells continue to develop progressive phenotypic changes and genomic instability (dysplasia), culminating as overt carcinoma \[[@B115]\]. More than 80% of HCC originate in cirrhotic livers. Macronodules (macroregenerative nodules and adenomatous hyperplasia), irregular hepatocyte regeneration, and some hyperplastic foci are considered as precancerous \[[@B116]-[@B119]\]. Large cell dysplasia and small cell dysplasia are considered to be risk factors for development of HCC \[[@B120]-[@B122]\]. HBV and HCV can be implicated in the development of HCC in an indirect way, through induction of inflammation, necrosis and chronic hepatocellular regeneration, or directly by means of viral proteins or, in the case of HBV, by creating insertional mutations by integration in the genome of the hepatocyte. Indirect carcinogenicity of HBV and HCV --------------------------------------- In most patients with chronic hepatitis B and/or C the occurrence of HCC is preceded by a process of longstanding inflammation. It is probable that malignant transformation is related to continuous or recurring cycles of hepatocyte necrosis and regeneration \[[@B123]\]. The resulting accelerated cell turnover rate may act as a tumor promotor by increasing the probability of spontaneous mutations or damage to DNA by exogenous factors. The accelerated rate of cell division leaves less time for altered DNA to be repaired before the cell divides again, resulting in transmission of altered DNA to the daughter cells. In this way a series of mutations may accumulate in individual cells over time. This process can lead to focal uncontrolled liver cell growth and eventual malignant cell transformation \[[@B115],[@B124]\]. Another mechanism of induction of malignant transformation is the generation of mutagenic reactive oxygen species as a result of the inflammatory process, such as nitric oxide (NO), superoxide anion (O~2~^-^), hydroxyl radical (OH•) and hydrogen peroxide (H~2~O~2~) \[[@B124]\]. Evidence for a causal role for chronic necro-inflammation is provided by transgenic mice into which HBV preS/S genes have been introduced. These mice overproduce pre S1 protein that accumulates in the endoplasmatic reticulum of hepatocytes, producing severe and prolonged injury to these cells, initiating a response characterized by inflammation, regenerative hyperplasia and transcriptional deregulation that progresses ultimately to neoplasia \[[@B125]\]. Patterns of gene expression in cirrhosis and hepatocellular carcinoma have recently been shown to be of value in predicting prognosis. Kim *et al*could identify, using the complementary DNA microarray, a 273-gene signature that distinguished high risk types of cirrhosis (hepatitis B, hepatitis C, hereditary hemochromatosis) from low risk types (autoimmune hepatitis, PBC, alcoholic liver diseases) \[[@B126]\]. The same 273-gene signature was present in samples from patients with proven HCC. A subset of 30 genes was most significantly altered in both the high risk types of cirrhosis and the HCC patients. The TACSTD1, a gene associated with HCC development in other studies, is a lead gene in this gene signature. Lee *et al*could identify a limited number of genes that accurately predicted survival in a series of 91 HCC patients \[[@B127]\]. The genes involved are implicated in cell proliferation and apoptosis, but also in ubiquitination and histone modification. Delpuech *et al*identified distinct patterns of gene expression according to the viral aetiology \[[@B128]\]. Finally, Hann *et al*could demonstrate the presence of antibodies to differentially expressed genes in hepatitis B and C, and this appeared to be linked with decreased survival \[[@B129]\]. These discoveries not only increase our insight in hepatocarcinogenesis, but may ultimately lead to the development of clinically valuable preneoplastic and prognostic blood markers. Direct carcinogenicity of HBV and HCV ------------------------------------- ### Hepatitis B A significant proportion of HBV-related HCCs arise in an otherwise normal liver, implicating that the virus can also be directly oncogenic \[[@B124]\]. It has been demonstrated that HBV integrates into the DNA of the host cells. This integration may dysregulate the control mechanisms on the cell cycle by chromosomal abnormalities, production of viral proteins or alteration of human genes and proto-oncogenes. It is, however, controversial whether viral integration plays an important role in the process leading to development of HCC. The hepadnaviral integration process appears to involve recombination mechanisms that do not preserve the viral genome sequence. Thus it is impossible for the viral integrant to function as a template for subsequent virus replication. Several studies suggest that DNA integration sites are at random and that integration occurs at random times during the course of a chronic viral infection \[[@B130],[@B131]\]. HBV integration can be present in chronically infected liver tissue without evidence of HCC \[[@B132]\]. Non-neoplastic hepatocytes may have a similar pattern of rearrangement of viral sequences following integration into human DNA. Chromosomal DNA instability --------------------------- Several studies have shown that HBV DNA integration enhances chromosomal instability. In many hepatic tumors large inverted duplication insertions, translocations and micro- and macrochromosomal deletions have been associated with HBV insertion \[[@B133]-[@B136]\]. These changes can result in loss of important cellular genes, sometimes involving tumor-suppressor genes and other genes involved in the regulation of regeneration and growth processes. ### Trans-activation of cellular genes HBV DNA may induce malignant transformation in another way. Mammalian hepadnaviruses contain a gene (the HBX gene), of which the protein (HBX protein) can *trans*-activate several cellular promotors and upregulate their expression of different cellular and viral genes \[[@B137],[@B138]\]. Integrated HBX, even when truncated, frequently encodes functionally active *trans*-activator proteins \[[@B139]\]. This protein has been shown to transform mouse fetal hepatocytes into a full malignant phenotype \[[@B140]\]. There are studies in transgenic mice with the HBX gene that developed multifocal areas of altered hepatocytes, adenomas and HCCs \[[@B110]\]. In contrast to mammalian hepadnaviruses associated with HCC, avian hepadnaviruses such as the duck hepatitis virus or heron hepatitis virus, lack the HBX gene and are not associated with HCC \[[@B123]\]. A gene that may be affected by the HBX gene is the p53 tumor suppression gene. This gene has been shown to play an important role in hepatocarcinogenesis. It is considered to negatively regulate the cell cycle. The HBX protein has been shown to complex p53 protein and to inhibit its function \[[@B141],[@B142]\]. In a transgenic mouse model it was shown that HCC development correlates with p53 binding to HBX \[[@B143]\]. ### Oncogenes It has been proposed that HBV acts as an insertion mutagen by integrating into the host genome and activating the cellular proto-oncogenes *c-myc*, *ras*and *c-fos*\[[@B144]\]. The preS2/S gene is integrated in most HCCs associated with HBV. When 3\'-truncated it generates a truncated protein that is oncogenic by *trans*-activating proto-oncogenes *c-myc*and *c-fos*\[[@B145]\]. ### Growth factors Growth factors and their receptors function as positive or negative modulators of cell proliferation and differentiation. Insulin-like growth factor-II and transforming growth factor-β expression correlate with HBX protein expression in animal models \[[@B146],[@B147]\], suggesting *trans*-activation of these growth factors facilitating tumor formation. ### Role of PreS mutations PreS deletion mutants accelerate the storage of large envelope proteins in hepatocyte cytoplasm, which could induce cytotoxic effects toward the development of end-stage liver disease \[[@B148]\]. The accumulation of large envelope protein can activate cellular promoters by inducing endoplasmic reticulum stress \[[@B149]\]. Furthermore, pre-S1 sequences can stimulate the transcription of transforming growth factor α (TGFα). Coexpression of TGFα and HBsAg could accelerate hepatocellular carcinogenesis by stimulation of hepatocyte proliferation \[[@B150]\]. ### Allelic loss of chromosome 4q Allelic loss of chromosome 4q is one of the most frequent genetic aberrations found in HCC. It was found to be associated with HBV-related hepatocarcinogenesis, probably by inactivation of a putative tumor suppressor gene included in it \[[@B151]\]. Hepatitis C ----------- In contrast to HBV, HCV is an RNA virus that lacks a reverse-transcriptase enzyme and cannot integrate into the host genome. Thus, insertional mutagenesis can be excluded as a pathogenic mechanism for the development of HCC associated with chronic HCV infection. The molecular pathogenetic mechanisms by which HCV contributes to cell transformation remain unclear. One possibility is that the development of HCC is simply related to chronic necro-inflammatory liver disease. Overall, 97% of patients with HCV markers and HCC have cirrhosis \[[@B152],[@B153]\], and most of the remainder develop HCC in the presence of chronic hepatitis. An alternative mechanism of HCV-induced hepatocarcinogenesis may be that HCV has a direct oncogenic action. Viral replication might cause inappropriate expression of two growth factors that may be implicated in hepatic carcinogenesis: transforming growth factor-α and insulin-like growth factor II \[[@B154],[@B155]\]. The non-structural HCV protein *NS3*has both protease and helicase activity. HCV may therefore induce genomic instability and favor mutations through its helicase activity \[[@B156]\]. The protein also has an activity similar to protein kinase A, and could disturb cellular homeostasis \[[@B157]\]. The *HCV envelope protein E2*and the non-structural protein *NS5A*inhibit RNA-dependent protein kinase, key mediator of the antiviral, antiproliferative and anti-oncogenic effect of interferon \[[@B158]-[@B160]\]. The *HCV core protein*has characteristics that imply that this protein could function as a gene-regulator \[[@B161],[@B162]\]. The presence of the protein in transgenic mice can induce HCC \[[@B111]\]. After mutation, the HCV core protein can also inhibit tumor suppressor genes such as p53, as has been demonstrated in hepatic oncogenesis \[[@B163]-[@B165]\]. It has recently been shown that the HCV core protein induces nuclear factor κB (NF-κB), thereby suppressing TNF-α-induced apoptosis \[[@B166]\]. This anti-apoptosis may be a mechanism by which HCV leads to viral persistence and possibly to hepatocarcinogenesis. Prevention of hepatocellular carcinoma caused by viral hepatitis ---------------------------------------------------------------- ### Primary prevention The most effective tool to prevent HCC is avoidance of the risk factors such as viral infection by HBV or HCV. Any action diminishing the potential transmission of contaminated blood products (uncontrolled blood transfusion, needle sharing, invasive procedures without proper health standards) will decrease the likelihood of viral spread. The major advance has come from the availability of an effective vaccine that protects against HBV. In 1969, Taiwan was an hyperendemic area of HBV infection with a high rate of HBsAg positivity, 19% of the population being infected before the fourth decade of life. In 1976, HBsAg prevalence was \> 80% in HCC in Taiwan \[[@B167]\]. In 1984 a program to control cirrhosis and HCC began. All neonates born to HBsAg positive mothers were given hepatitis B vaccine in order to counter perinatal infection. In 1986 all neonates were included in the program. As a consequence, there was a decrease in HBsAg positivity in six-year-olds from 10.6% in 1983--1984 to 0.8% in 1993--1994. There was a parallel decline in incidence of childhood HCC (6--14 years old), in the cohort born between 1980 and 1984. The incidence of liver cancer in children between 6 and 14 years old decreased to zero for children born in 1986 and 1987 \[[@B168]\]. The decline of HCC in children after universal vaccination can be considered as an early indicator of the effectiveness of vaccination in reducing the rate of HCC. Since the incidence of HCC in Taiwan peaks in the sixth decade of life, it may take 40 years or longer to see an overall decrease in the rate of HCC as a result of the vaccination program. Vaccination against HBV should become a health priority together with the promotion of adequate health standards. Unfortunately, there is no vaccine against HCV. Up to now, the only effective method to prevent its transmission is the avoidance of contamination with infective blood products. Prevention of HCC in patients with previously acquired risk =========================================================== Introduction ------------ Chronic viral carriage is one of the main risk factors for the development of HCC. Effective antiviral treatments have been developed in recent years and this has changed the management of viral infection. Interferon-alfa is still considered the reference therapy for HBeAg positive chronic hepatitis B. However, its efficacy is limited, with seroconversion from anti-HBe negative to anti-HBe positive in up to 40%. Only \<10% of patients become HBsAg negative \[[@B169]\]. Other possible treatments are antiviral drugs such as lamivudine and adefovir dipivoxil \[[@B12]\]. For the treatment of chronic hepatitis C, interferon-alfa monotherapy yielded only limited response. Combination with ribavirin led to a significant increase in sustained viral response to about 40% in treatment-naïve patients \[[@B170],[@B171]\]. Recently, the combination of peginterferon-alfa and ribavirin improved the sustained viral response rate to nearly 60% in treatment-naïve patients \[[@B172],[@B173]\], and is now considered the reference treatment. It is under debate whether interferon-alfa-based treatments are effective in declining the incidence of HCC in chronic hepatitis B and C. Anti-oncogenic effects of interferon-alfa ----------------------------------------- HCC prevention by interferon-alfa might be the result of several direct or indirect mechanisms. Interferon has an antiproliferative and pro-apoptotic effect \[[@B174]\]. Interferon inhibits the expression of the *c-myc*oncogene and induces the expression of anti-proliferative factors and tumor suppressor genes \[[@B175]-[@B177]\]. In experimental animal models, the anti-neoplastic potential of interferon was demonstrated in already established tumors. In a transgenic mouse model it was demonstrated that early and prolonged administration of interferon diminished the severity of preneoplastic lesions and slowed down the development of HCC \[[@B178]\]. Interferon-alfa also could indirectly reduce the oncogenic risk by inhibition of synthesis of viral proteins which potentially dysregulate the cell cycle, and by enhancing the immune system eliminating not only infected hepatocytes but also initiated or fully malignant cells. Furthermore, interferon-alfa has an antifibrotic and anti-angiogenetic effect, which could also have an influence on tumor development \[[@B179]\]. Interferon and antiviral treatment ---------------------------------- ### Noncirrhotics In patients with chronic hepatitis B, clearance of the HBeAg after treatment with interferon-alfa is associated with improved clinical outcome in terms of survival and development of complications of cirrhosis \[[@B180]\]. Another study confirmed these results and showed a reduction of incidence of HCC in the responders \[[@B181]\]. As most of these patients were non-cirrhotics at entry of the study, the prophylactic effect of interferon on development of HCC can be explained by prevention of cirrhosis development. In Chinese patients with chronic hepatitis B infection, however, interferon-alfa was of no long-term benefit in inducing HBeAg conversion, or in the prevention of HCC and other cirrhosis-related complications \[[@B182]\]. ### Cirrhotics Seven studies investigated the possible effect of interferon treatment on development of HCC in patients with already established cirrhosis \[[@B183]-[@B189]\] (Table [4](#T4){ref-type="table"}). A meta-analysis was performed on these studies \[[@B190]\]. Interferon seemingly decreased the rate of HCC in all trials, while a significant difference was observed in 2 studies \[[@B183],[@B186]\]. Virologic response was strongly associated with reduced risk for HCC in the studies of Oon \[[@B183]\] and Mazzella \[[@B184]\], suggesting that arrest of viral replication is a critical factor. Subgroup analysis in relation to ethnic origin of patients (European, Oriental) showed no preventive effect of interferon on the development of HCC in the European patients \[[@B190]\]. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Studies of treatment with interferon-α for prevention of HCC in patients with hepatitis B-related cirrhosis ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Author \[reference\]** **Country** **Type of study** **Interferon regimen (duration in weeks)** **Follow-up (range) in months** **Sample size** **Rate of HCC (n/n)** **Significance** -------------------------- ----------------------------------------- ------------------- -------------------------------------------- --------------------------------- ----------------- ---------------------------------- ------------------ Oon, 1992 \[183\] Singapore NRCT, P 10 MU daily, 10 days/month (12) 12 (12--60) T 600\ T: 0/600 (0%)\ Significant C 180 C: 10/180 (5.6%) Mazzella, 1996 \[184\] Italy NRCT, P 10 MU tiw (26) 49 (12--119) T 34\ T: 2/34 (5.9%)\ Not significant C 28 C: 4/28 (14.3%) Fattovich, 1997 \[185\] Europe NRCT, P ≥ 300 MU (12--52) 84 (80--92) T 40\ T: 3/40 (7.5%)\ Not significant C 50 C: 4/50 (8.0%) Ikeda, 1998 \[186\] Japan NRCT, P 12 MU/wk (26) 84 (6--168) T 94\ T: 10/94 (10.6%)\ Significant C 219 C: 51/219 (23.3%) IHCSG, 1998 \[187\] Argentina, Germany, Italy, Saudi Arabia NRCT, P 9--30 MU/wk for 3--30 months (36--250) T 49\ T: 8/49 (16.3%) C: 18/97 (18.6%) Not significant C 97 Benvegnù, 1998 \[188\] Italy NRCT, P 6--10 MU (20--26) 72 T 10\ T: 0/10 (0%)\ Not significant C 18 C: 4/18 (22.2%) Di Marco, 1999 \[189\] Italy NRCT, P 655 MU 93 (6--180) T 26\ T: 2/26 (7.7%)\ NR C 60 C: 6/60 (10%) ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- NRCT: non-randomized controlled trial P: prospective T: treated C: controls MU: million units NR: not reported ::: It should be noted that the studies are very heterogeneous and that none of them were randomized controlled trials, so that the results should be interpreted with caution. A recent study showed a significant reduction of the risk of HCC in patients with chronic hepatitis B and advanced fibrosis or cirrhosis, treated with lamivudine for a maximum of five years, compared to placebo \[[@B191]\]. Interferon treatment in HCV patients and HCC prevention ------------------------------------------------------- ### Noncirrhotics Three studies assessed whether interferon treatment prevents the development of HCC in noncirrhotic patients with chronic hepatitis C \[[@B192]-[@B194]\] comprising 3,798 noncirrhotic patients treated with interferon-alfa monotherapy. Pooled together, the incidence of HCC was 60/2,532 (2.37%) in sustained virological responders and 76/1,266 (5.29%) in nonresponders. In a study of 291 noncirrhotic patients with chronic hepatitis C who were nonresponders to interferon therapy and followed for 6--117 months after therapy, the incidence of HCC was significantly lower in patients who received \> 500 MU of interferon. Patients with a transient response (i.e. relapse after end of treatment) had a significant lower rate of HCC development (4/166 = 2.4%) than nonresponders (12/125 = 9.6%) \[[@B195]\]. This anti-oncogenic benefit can presumably be explained by an arrest or slowing down of the cirrhogenic process. ### Cirrhotics The findings of 13 studies of interferon treatment and development of HCC in HCV-infected patients with compensated cirrhosis are summarized in Table [5](#T5){ref-type="table"}\[[@B45],[@B184]-[@B186],[@B194],[@B196]-[@B204]\]. Only 3 studies were randomized \[[@B199],[@B201],[@B202],[@B204]\], the remainders were observational cohort studies. Statistical combination of data is not possible because of different definitions of response (biochemical, virological), different dose schedules for interferon and different duration of follow-up. All studies showed a lower risk for development of HCC in the interferon-treated patients, suggesting that interferon may prevent HCC in compensated cirrhosis caused by hepatitis C. The overall result was largely influenced by three Japanese studies \[[@B194],[@B198],[@B201],[@B202]\], which had the highest incidence of HCC in untreated patients (5--6% per year). This may be explained by intensiveness of the screening programs, but also by genetic, environmental or viral factors. Four European studies failed to document a significant reduction in risk of developing HCC \[[@B45],[@B196],[@B199]\]. In the studies of Fattovich *et al*\[[@B196]\] and Bruno *et al*\[[@B45]\], interferon-alfa treatment showed a decrease in incidence of HCC in univariate analysis. However, this was not present in multivariate analysis. In the study of Fattovich \[[@B196]\], a very low natural incidence of HCC was observed, rendering difficult to show a significant decrease. The prospective randomized controlled trial of Valla *et al*\[[@B199]\] also failed to show a significant effect of interferon treatment on the development of HCC. However, the number of patients in this study was limited and the follow-up relatively short. Also a recently published randomized controlled study from Italy comprising 51 interferon-treated and 71 untreated patients with compensated hepatitis C-cirrhosis, failed to demonstrate any reduced risk in development of HCC after a mean follow-up of 96.5 months \[[@B204]\]. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Studies on treatment with interferon-α for prevention of HCC in patients with HCV-related cirrhosis ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Author \[reference\]** **Country** **Type of study** **Interferon regimen (duration in weeks)** **Follow-up (range) in months** **Sample size** **Rate of HCC (n/n)** **Significance** ----------------------------------- ----------------------------------------- ------------------- -------------------------------------------- --------------------------------- ----------------- ----------------------- ------------------ Mazzella, 1996 \[184\] Italy NRCT, P 3 MU tiw (52) 32 (12--71) T 193\ T: 5/193 (2.6%)\ Significant C 91 C: 9/91 (9.9%) Fattovich, 1997 \[196\] Europe NRCT, P ≥ 200 MU 60 (1--153) T 193\ T: 7/193 (3.6%)\ Not significant C 136 C: 16/136 (11.8%) Bruno, 1997 \[45\] Italy NRCT, P 6 MU tiw (26) 68 (60--84) T 82\ T: 6/82 (7.3%)\ Not significant C 81 C: 14/81 (17.3%) Serfaty, 1998 \[197\] France NRCT, P 3 MU tiw (48) 40 (6--72) T 59\ T: 2/59 (3.4%)\ Significant C 44 C: 9/44 (20.1%) IHCSG, 1998 \[187\] Argentina, Germany, Italy, Saudi Arabia NRCT, R 9--30 MU/wk (3--30 months) (36--250) T 232\ T: 2/232 (0.9%)\ Significant C 259 C: 48/259 (18.5%) Imai, 1998 \[198\] Japan NRCT, R 480 MU (26) 48 (3--65) T 32\ T: 8/32 (25%)\ Significant C 20 C: 7/20 (35%) Benvegnù, 1998 \[188\] Italy NRCT, P 3--6 MU tiw (26--52) 72 T 75\ T: 4/75 (5.3%)\ Significant C 77 C: 20/77 (26.0%) Valla, 1999 \[199\] France RCT 3 MU TIW (48) 40 (37--53) T 47\ T: 5/47 (10.6%)\ Not significant C 52 C: 9/52 (17.3%) Yoshida, 1999 \[194\] Japan NRCT, R 480 MU (23) 52 T 230\ T: 33/230 (14.3%)\ NR C 107 C: 29/107 (27.1%) Okanoue, 1999 \[200\] Japan NRCT, R 3--10 MU qd or tiw (16--24) 1--7 years T 40\ T: 7/40 (17.5%)\ NR C 55 C: 22/55 (40.0%) Nishiguchi, 1995/2001 \[201,202\] Japan RCT 6 MU tiw (12--24) 104 (31--110) T 45\ T: 12/45 (26.7%)\ Significant C 45 C: 33/45 (73.3%) Gramenzi, 2001 \[203\] Italy RCT, P 741 MU 72 T 72\ T: 6/72 (8.3%)\ Significant C 72 C: 19/72 (26.4%) Testino, 2002 \[204\] Italy RCT 3 MU tiw (52) 96.5 ± 18 T 51\ T: 15.51 (29.4%)\ Not significant C 71 C: 24/71 (33.8%) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- NRCT: non-randomized controlled trial RCT: randomized controlled trial P: prospective R: retrospective NR: not reported T: treated C: controls MU: million units ::: In most studies, virological and/or biochemical response are associated with a lower risk of development of HCC, which is less clear in nonresponders. In the study of Imai et al. \[[@B198]\], patients with sustained biochemical response after interferon therapy were at low risk for development of HCC (risk ratio versus controls 0.06; 0.95 in nonresponders). Also in the study of Mazzella \[[@B184]\], a statistically significant effect of interferon treatment was demonstrated when biochemical responders were compared with controls but not when compared with nonresponders. In the study of Benvegnù *et al*\[[@B188]\], the beneficial effect of interferon treatment on development of HCC was independent of the type of response. In the study of Yoshida *et al*\[[@B194]\] the risk for HCC was reduced especially among patients with sustained virological but also merely biochemical response that tested positive for HCV RNA. Okanoue *et al*\[[@B200]\] studied 1,148 patients with chronic hepatitis C treated with interferon-alfa, 40 of them having cirrhosis (fibrosis stage F4). They were followed for 1--7 years after therapy. The cumulative incidence of HCC was significantly decreased in sustained biochemical responders, compared to nonresponders and transient responders, in patients with stage F2 fibrosis, but not in the more advanced stages F3 and F4. In the study of Testino *et al*\[[@B204]\] HCC did also develop in sustained biochemical responders. Tanaka *et al*\[[@B205]\], however, demonstrated in 55 patients with HCV-cirrhosis that long-term administration of interferon prevented HCC in those with biochemical and virological response, whereas HCC only appeared in nonresponders. The mechanisms by which an interferon treatment might reduce the risk of HCC development in cirrhosis caused by HCV independent of virological response remains speculative. Maintenance of serum transaminases at low levels may protect against the development of HCC as hepatocyte necrosis, cell damage and increase in hepatocyte replication result in increased DNA damage, influencing hepatocarcinogenesis. Other possible mechanisms for prevention of HCC are the direct and indirect effects of interferon. It is, however, perplexing that only 6 or 12 months of therapy can produce this benefit without virological response. Because of potential biases in the published trials it is premature to advocate the use of interferon as established therapy in HCV infected patients with cirrhosis to prevent HCC. Prospective randomized controlled trials should reproduce the findings in large numbers of patients before a definitive conclusion on the long term effects of interferon in HCV cirrhosis can be established. It must also be realized that a sustained virological response to interferon-alfa monotherapy can be obtained only in 0--8% of patients with cirrhosis \[[@B206]-[@B208]\]. New treatments are now available for chronic hepatitis C, which are more performant in difficultly to treat cases as patients with cirrhosis. The combination of interferon-alfa and ribavirin results in a sustained virologic response in up to 25% of cirrhotics due to hepatitis C \[[@B207]\]. A sustained virologic response of 32% was reported after peginterferon-alfa-2a monotherapy \[[@B207]\] and of 43% after combination of peginterferon-alfa2a and ribavirin \[[@B173]\]. It should be investigated in prospective trials, taking into account the sustained virological and biochemical responses if these more performant treatment regimens will also influence favorably the incidence of HCC, as no data on the long-term effects of these treatments are available up to now. Secondary prevention -------------------- A few sudies focus on the possible role of intereron in the secondary prevention of HCC recurrence in patients with chronic hepatitis B and C after curative resection or ablation. Ikeda *et al*\[[@B209]\] showed that interferon prevented HCC recurrence after complete resection or ablation of the primary tumor depending on the clearance of HCV viremia. Kubo *et al*\[[@B210]\] reported a decreased recurrence after surgical resection independent of clearance of HCV or normalization of serum ALT. Another study demonstrated prevention of HCC recurrence after medical ablation therapy for primary tumors in hepatitis B but not in hepatitis C patients by the use of interferon-alfa \[[@B211]\]. Conclusions =========== Chronic hepatitis B and C, mostly in the cirrhotic stage, are responsible for the majority of the hepatocellular carcinomas worldwide. The rising incidence in HCC in developed countries during the last two decades is due to the increasing rate of hepatitis C infection and improvement of the clinical management of cirrhosis. Vaccination against hepatitis B seems to protect against the development of HCC. In patients with chronic hepatitis B or C, interferon alpha treatment in a noncirrhotic stage is protective for HCC development in responders, probably by prevention of cirrhosis development. When cirrhosis is already present, the protective affect is less clear. Further prospective long-term studies should be performed on the new treatments for chronic hepatitis B and C. Some studies also suggested a favourable effect of interferon alpha in the prevention of HCC recurrence in patients with chronic hepatitis B and C after curative resection or ablation. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= PPM participated in the literature search and was responsible for the redaction of the paper. SMF participated in the redaction of the manuscript and critical review of the paper. JLV participated in the literature search and finalizing of the lay-out of the paper.

PubMed Central

2024-06-05T03:55:58.188809

2005-5-20

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166580/", "journal": "World J Surg Oncol. 2005 May 20; 3:27", "authors": [ { "first": "Peter P", "last": "Michielsen" }, { "first": "Sven M", "last": "Francque" }, { "first": "Jurgen L", "last": "van Dongen" } ] }

PMC1166581

Background ========== *Castleman\'s disease*(CD) is a rare lymphoid disorder where pathogenesis is a lymphoid tissue hyperplasia related to chronic herpes virus infection. It has been described in nearly every lymph node basin since it first description by B. Castleman in 1956 \[[@B1],[@B2]\]. Two basic pathologic types of this disease could be encountered: the hyaline vascular (HV) and the plasma cell (PC) types. The first tends to be localized in one lymph node and asymptomatic; the second has a more aggressive course and tends to be multifocal with systemic manifestations. The authors\' present two cases of localized Castleman\'s disease arising in the peri-renal space and review previous reports abdominal and retroperitoneal CD. A review of literature on Castleman\'s disease pathogenesis, clinical and radiological characteristics as well as its treatment is also included in this manuscript. Case presentation ================= Case 1 ------ A 33-years-old woman with no significant past medical history complained of abdominal right upper quadrant discomfort associated with an history of weight lost (8 kg over 2 months). Physical examination revealed a right upper quadrant mass on deep palpation. Routine hematology and blood biochemistry were normal. The patient was HIV1-2 negative. Chest and abdominal roentgenograms were considered normal. Abdominal ultrasonography (USG) revealed a large hypoechogenic mass, with regular border in the right anterior peri-renal space. Computed tomography (CT) scan showed a 10 × 8 cm mass with regular contour, containing small calcifications, which strongly enhanced with vascular contrast. The lesion was in contact with the right kidney and ureter. Surgery was planned with a preoperative diagnosis of malignant retroperitoneal tumor versus lymph node hyperplasia. Through a right transverse incision, after mobilization of the duodeno-pancreatic bloc, a tumor was found in contact of the right kidney, ureter and caval vein. While the possibility of malignancy could not be neglected, the mass was dissected *en-bloc*with wide margin in peri-renal fat. To allow free surgical margin clinically a segment of the right ureter as well as inferior pole of kidney were also excised *en bloc*. A right nephrectomy was finally performed latter on as the ureteral defect could not be repaired. Histopathological examination of resected specimen revealed localized Castleman\'s disease of the hyaline vascular type. Patient had smooth postoperative recovery and is free of disease more than 6 years after resection. Case 2 ------ A 25-year-old woman with no significant past medical history, presented with post-prandial epigastric discomfort evolving over 2 years and post-prandial vomiting since 1 month. The patient reported 2.5 kg weight loss over 2 months. Physical examination revealed a left para-renal mass on deep palpation. Blood analyses were not relevant. CEA and CA19-9 were in the normal range. The patient was HIV1-2 negative. Chest and abdominal X-ray films were considered normal. An upper abdominal barium follow through was considered normal. Abdominal ultrasonography revealed a 6 cm diameter hypoechogenic mass in the left peri-renal space. CT scan showed a 6 × 7 cm mass, containing multiple small calcifications, which was highly hypervascular and regular in shape (figure [1](#F1){ref-type="fig"}). An arteriography confirmed the presence of hypervascularity with flushing of the mass (figure [2](#F2){ref-type="fig"}). Surgery was planned with a preoperative diagnosis of malignant retroperitoneal tumor. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Computed tomography scanner showing a large left-pararenal mass (Hyalin-vascular type of Castleman\'s disease). Note the presence of microcalcification (white spot) within the mass. ::: ![](1477-7819-3-33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Arteriography showing the presence of a hypervascular mass (Hyalin-vascular type of Castleman\'s disease) with rapid flushing of the tumor. The feeding vessels originated from the aorta and left renal artery. ::: ![](1477-7819-3-33-2) ::: At laparotomy a 7 cm diameter mass was found in the left anterior peri-renal space, just inferior to the renal artery. The lesion was completely excised with what seems, clinically, to be a capsule. No organ resection was needed. Pathologic diagnosis was localized Castleman\'s disease of the hyaline vascular type. Patient had had a simple postoperative period, except of persisting lumbar pain, attributed to a small inferior renal infarct (confirmed by CT scan). The patient is free of disease 5-years after resection. Discussion ========== Castleman\'s disease (CD), also known as angiofollicular lymph node hyperplasia, was first reported by Symmers in 1921 \[[@B3]\]. This pathology was characterized by B. Castleman in 1956 as a benign lymph node hyperplasia resembling a thymoma \[[@B1],[@B2]\]. Keller *et al*identified two pathologic types of CD in 1972 \[[@B4]\]. First the hyaline vascular (HV) type which present as a pathological and extensively hypertrophied lymph node. Macroscopically it appears as an encapsulated homogenous mass with an orange-yellowish color. Microscopically, it is characterized by giant lymph follicles centered on a central vessel with marked hyalinization. Follicles are surrounded by circumferentially arranged layer, in an onion skin feature, of small polyclonal B-lymphocytes. These pathologic lymph nodes present a strong hypertrophied vascular arborescence \[[@B4]-[@B8]\]. The plasma cell (PC) type has the same macroscopic aspect as HV type, but contains much more mature polyclonal plasma cells with a less marked hyalinization and vascularization \[[@B4],[@B6]-[@B8]\]. It been shown that this two types histology are not always clearly separated and that mixed HV-PC types can also occur \[[@B7]\]. The histology of PC type is not specific of systemic CD and can be found in autoimmune disease, AIDS and in lymph nodes draining carcinoma, so it is imperious to exclude this condition before diagnosing CD of PC type \[[@B5],[@B6]\]. This implies that serologic testing for HIV should be performed whenever a diagnosis of CD is contemplated\[[@B5],[@B6]\]. The etiology of CD is related to chronic Human Herpes virus 8 (HSV8) \[[@B7]\] as HSV8 has been found in lymphoid cells in case of systemic form, or PC type, of Castleman\'s disease \[[@B9]\]. Its nature is not neoplastic as confirm by the fact that the lesion are made of a polyclonal proliferation \[[@B6],[@B7]\]. It\'s seems that CD is the result of a chronic low grade inflammatory process triggered by latent infection with HSV8, which leads to lymphoid system hyperplasia \[[@B5]-[@B7]\]. Human herpes virus 8 (HSV8), also called Kaposi\'s sarcoma-associated herpes virus (KHSV) is the initiator of this chronic inflammation by establishing a chronic or latent infection in lymph nodes \[[@B7]\]. Chronic infection by HSV8 stimulates secretion of IL-6 which in turn induces a hyperplasic reaction of the lymphoid system \[[@B6],[@B7]\]. While this lymphoid hyperplasia could be contained in one lymph node as in the localized form, which is mainly of HV type, it could also be generalized as in the systemic, or multifocal, form which is the predominant form for the PC type \[[@B6]\]. The patient\'s immunological status seems to play a major role in the development of these two forms. While localized form is encountered mainly in immunocompetent patients, the systemic form is found in patient with AIDS or other immunodeppression related either to immunosuppression or pathological state \[[@B6]\]. The localized form of CD arises predominantly in the mediastinum, where it was first described by B. Castleman \[[@B1]\]. However, it can be found in the neck, abdomen, axilla, inguinal region and in virtually all lymph node area \[[@B5],[@B7],[@B8],[@B10],[@B11]\]. Even non-nodal tissue could be involved, as is has been described in: lung, pancreas, breast, adrenal gland, muscle and other extremely rare locations \[[@B4],[@B7],[@B10],[@B11]\]. Testa *et al*\[[@B11]\] have reported the location of 315 cases of localized CD, 65% were in the mediastinum, 16% in the neck, 12% in the abdomen, 3% in the axilla and 4% in diverse locations. A literature review of the abdominal and retroperitoneal case of localized HV type of CD has been done. In 1992, 54 abdominal and retroperitoneal cases were reviewed by Seco *et al*\[[@B5]\]. Now 195 cases of localized CD have been reported, in the world literature, arising in the abdomen and retroperitoneum. Of these 195 cases, 122 (63%) were in the retroperitoneum and 73 (37%) in the abdominal cavity (Table [1](#T1){ref-type="table"}). Of the 122 lesions localized in the retroperitoneum, 24 (20%) were in the peri-renal region, as were our cases. Nearly all this lesions were derived from lymph node tissue, but 5 of these 195 cases (2%) seem to have originating in extra lymphoid organ. Three pancreatic, and one each of splenic and adrenal CD have been described \[[@B58]-[@B60],[@B65],[@B80]\]. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Abdominal and retroperitoneal case of localized Castleman\'s disease (HV type) ::: ***Location*** ***Number of cases*** ***Authors*** ---------------------- ----------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ **Retroperitoneu**m 97 Seco *et al*\[5\], Bapat *et al*\[12\], Yamakita *et al*\[13\], Morishita *et al*\[14\], Genoni *et al*\[15\], Ng *et al*\[16\], Baikovas *et al*\[17\], Johnson *et al*\[18\], Martino *et al*\[19\], Guglielmi *et al*\[20\], Ziv *et al*\[21\], Gheysens *et al*\[9\], Halvic *et al*\[22\], Herrada *et al*\[23\], Furuhata *et al*\[24\], Ebine *et al*\[25\], Sadamoto *et al*\[26\], Gravalos *et al*. \[27\], Iwamoto *et al*\[28\], Sanna *et al*\[29\], Singletary *et al*\[30\], Curciacrello *et al*\[31\], Gonzalez Sanchez *et al*\[32\], Schutz *et al*\[33\], Perez Garcia *et al*\[34\], Irsutti *et al*\[35\], Parez *et al*\[36\], Buchanan *et al*\[37\] *Peri-renal* 24 Ebisno *et al*. \[38\], Inoue *et al*. \[39\], Takihara *et al*. \[40\],,Feudis *et al*\[41\], Barret *et al*. \[42\], Okada *et al*\[43\], Present two cases Mesentery 27 Seco *et al*. \[5\], Barki *et al*. \[44\], Hung *et al*. \[45\], Schroff *et al*. \[46\], Makipernaa *et al*. \[47\], De Heer-Groen *et al*\[48\], Parez *et al*\[36\], Neerhout *et al*\[49\], Powel *et al*\[50\], Burke *et al*\[51\] Greater omentum 3 Volta *et al*\[52\], Kiguchi *et al*\[53\] Gastric 2 Kiguchi *et al*\[53\], Yebra *et al*\[54\] Peri-pancreatic 5 Kiguchi *et al*\[53\], Rotman *et al*\[55\], Brossard *et al*\[56\], Inoue *et al*\[39\], Erkan *et al*\[57\] Pancreatic 3 Chaulin *et al*\[58\], Corbisier *et al*\[59\], Lepke *et al*\[60\] Porta hepatitis 5 Rahmouni *et al*\[61\], Farkas *et al*\[62\], Peck *et al*\[63\], Cirillo *et al*\[64\] Adrenal gland 1 Debatin *et al*\[65\] Pelvis 27 Seco *et al*\[5\], Latte *et al*\[66\], Daley *et al*\[67\], Boxer *et al*. \[68\], Tsukamoto *et al*\[69\], Luburich *et al*\[70\], Isik *et al*\[71\], Ylinen *et al*\[72\], Schwartz *et al*\[73\], Mondal *et al*\[74\], Kiguchi *et al*\[53\], Calvo Villas *et al*\[75\], Mac Donald *et al*\[76\], Fields *et al*\[77\], Kkasantikul *et al*\[78\], Halvic *et al*\[22\], Murphy *et al*\[79\] Spleen 1 Taura *et al*\[80\] ***Total of cases*** ***195*** ::: The clinical presentations of CD differ greatly between the localized and the systemic forms (Table [2](#T2){ref-type="table"}). The first appears in young generally healthy patients and cause few symptoms \[[@B6],[@B7]\]. Abdominal and retroperitoneal locations, as were our cases, can be associated with mass effect symptoms related to compression of adjacent organs. This could present as: post-prandial discomfort, anorexia, vomiting, weight loss, urinary retention and abdominal or lumbar pain \[[@B80]\]. Systemic, or multifocal, CD is associated with systemic disturbance as anemia, increased erythrocyte sedimentation rate, polyclonal hypergammaglobulinemia, hypoalbuminemia and thrombocytopenia which all can be associated with a specific symptoms \[[@B6]-[@B8]\]. The clinical picture includes asthenia, fever, weight loss, generalized lymphadenopathy, hepatomegaly, splenomegaly, peripheric edema, pleural effusion, impaired renal function and sometimes polyneuropathy \[[@B6]-[@B8]\]. Rarely POEMS syndrome (Polyneuropathy, Organomegaly, Endocrinopathy, M protein and Skin change) or amyloidosis may be associated to systemic CD \[[@B6]-[@B8]\]. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Clinical forms of Castleman\'s disease ::: Localized form Multifocal form ---------------------------- ------------------------------------------------- --------------------------------------------------------- Mean age (years) 3^rd^decade 6^th^decade Clinical signs Incidental mass effect Systemic symptoms Localization Mediastinum, cervical or abdominal, etc\... Multifocal, mostly peripheric lymph nodes Histologic type HV, rarely HV-PC PC Treatment Surgical resection Corticosteroids, chimiotherapy, radiotherapy Prognosis Excellent: 100% survival at 5 years Poor: median survival of 30 months Recurrence after treatment Extremely rare, related to incomplete resection Nearly always Association Rarely lymphoma Frequent: AIDS, Kaposi\'s sarcoma, lymphoma and myeloma Table adapted according to references: \[5, 6, 7, 11\] ::: Radiographic characteristics of CD are non specific, but some features could help to suspect the diagnosis (Table [3](#T3){ref-type="table"}) \[[@B77],[@B80]\]. Plain radiographic finding includes a mass effect and in nearly 30% of localized form calcifications harboring a radial arrangement or star-shaped calcifications which is said to be characteristic of CD \[[@B77]\]. Ultrasonography (US) usually demonstrates a hypoechogenic and homogenous mass with quite clear delimitation \[[@B8],[@B30],[@B77]\]. US can show central areas of sharp acoustic shadowing due to calcification \[[@B30]\]. CT scan show a solid, homogenous and well delimited mass which enhance with vascular contrast as a results of hypervascularity \[[@B5],[@B30],[@B77],[@B80]\]. It can also show star-shaped microcalcifications which are quite specific on pre-contrast images \[[@B5],[@B27],[@B77]\]. Post-contrast IV study, while demonstrating dense enhancement of the mass, could demonstrate a central stellate scar \[[@B30],[@B80]\]. Angiography shows a strongly hypervascular lesion, which present a dense and homogenous flush during the capillary phase \[[@B5],[@B6],[@B8],[@B40],[@B53],[@B77],[@B80]\]. This flush begins in periphery to become diffuse whiting the mass and is specific for the HV type of CD \[[@B77],[@B80]\]. It can also demonstrate hypertrophied feeding vessel, an useful information when resection is plan \[[@B8],[@B53],[@B77],[@B80]\]. Magnetic resonance imaging (MRI) characteristics of CD are: hypodense mass on the T1 weighted study and hyperdense lesion on the T2 weighted image sometimes with star-shaped calcifications \[[@B60],[@B77],[@B80]\]. Gadolinium injection produces an enhancement which appear in periphery to become diffuse similarly to the flush observed during angiography \[[@B77],[@B80]\]. All these radiological finding are not specific but some like the star-shaped calcifications and the type of hypervascularisation are quite specific and should alert clinician to the possibility of CD. In summary, in front of an abdominal or retroperitoneal mass which is well delimited, homogenous and harbor star-shaped calcifications and hypervascularisation associated with typical flush, the diagnosis of CD should be strongly suspected \[[@B53],[@B77],[@B80]\]. The differential radiological diagnosis is mainly malignant neoplasm because of the hypervascularity \[[@B40],[@B77]\], and the fact that 80% of the retroperitoneal tumor are malignant \[[@B40],[@B81],[@B82]\]. The major tumors found in the retroperitoneum are soft tissue sarcoma (liposarcoma, fibrosarcoma, leiomyosarcoma, neurofibrosarcoma, undifferentiated and rabdomyosarcoma) which are frequently heterogeneous mass and show necrosis on CT scanner; vascular tumor (hemangiosarcoma and lymphangiosarcoma) which are cystic and of liquid density on USG and CT scanner; and the lymphoma which generally presents as multiple adenopathy and homogeneous mass on CT scanner \[[@B30],[@B83]-[@B86]\]. Urological tumors like seminoma, prostatic cancer and teratoma tends to give rise to metastatic disease in the retroperitoneum in the form of adenopathy which are generally multiple \[[@B86]\]. While a preoperative diagnosis of CD is difficult to obtain, fine needle biopsy is not a definitive tools because of there low specificity and the differential diagnosis with lymphoma is impossible by this approach \[[@B77],[@B88]\]. In addition to be non-specific and rarely yielding enough useful tissue, needle biopsy is associated with tumoral seeding with reported frequency of 1/40 000 to 1/1 000 biopsy \[[@B83],[@B89]\]. Thus when surgery is planned, indication for fine needle biopsy should be carefully deliberated, while an open biopsy could always be done during surgery \[[@B83]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Radiological characteristics of Castleman\'s disease ::: **Non specific signs** **Specific signs** --------------- -------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------- Radiography calcification. Star-shaped calcification. Echography Hypoechogenic and homogenous mass. Central areas of acoustic shadowing (calcification). CT scanner Tissue density, Homogenous and well delimited mass. Contrast enhancement beginning in periphery. Star-shaped microcalcifications Star-shaped scar on post-contrast study. Arteriography Hypervascularity with hypertrophied feeding vessel. Flush beginning in periphery to become diffuse and homogenous during capillary phase. MRI Hypodense on T1 and hyperdense on T2. Contrast enhancement beginning in periphery. Star-shaped microcalcification or star-shaped hypodense signal on T2. Table adapted according to references: \[5,9,27,30,34,42,48,65,\] ::: Treatment of CD differs between localized and multifocal, or systemic, forms. The standard therapy of localized form is surgical excision which is curative when resection is complete and *en-bloc*. No recurrences have been reported after total excision in the literature as it was the case with our cases \[[@B5]-[@B7],[@B11],[@B38],[@B40],[@B53],[@B82],[@B90]\]. Because these lesions are highly vascularised, embolization before surgery could be helpful to minimize blood lost during surgery \[[@B5],[@B53]\]. The problem that is faced during resection is to resolve the differential diagnose between a malignant pathology and CD. Macroscopically it is nearly impossible because CD lesions harbor dense fibrous adherences to adjacent organ and hypervascularization typically seen in malignant pathology \[[@B5],[@B38],[@B82],[@B90]\]. For this reason peroperative diagnosis by open biopsy is helpful, and necessary in case of CD suspicion. It enables one to avoid extensive resection and especially resection of nearby organ for this benign disorder which does not invade adjacent organ even in case of tight contact \[[@B38],[@B82],[@B83],[@B90]\]. The five years survival after resection is nearly 100% for the localized form \[[@B6]\]. Recurrences have rarely been reported generally when excision was incomplete \[[@B6],[@B38],[@B82],[@B90]\]. For the systemic form no curative therapies have been found yet. Corticotherapy, immunosuppressive drugs, chemotherapy and radiotherapy have been tried without any convincing results \[[@B5]-[@B7],[@B91],[@B92]\]. The prognosis of this form is poor with a median survival of 30 months \[[@B6]\]. However recently, the reports of improvement and prolonged survival after splenectomy have been reported with steroid therapy and chemotherapy this could change the prognosis of the systemic form in the future \[[@B91],[@B92]\]. Of importance from the surgical point of view is that in these cases the remission was obtained after splenectomy only \[[@B91]\]. The systemic, PC type, and extremely rarely the localized, HV type, forms are associated with malignant disorder such as Kaposi\'s sarcoma, malignant lymphoma and myeloma. This association is stronger in HIV positive patients with CD, for example Kaposi\'s sarcoma is associated to 13% of case of PC type CD in HIV negative and in 75% of case when HIV positive patients. The association with Kaposi\'s sarcoma could be explained by the fact that HSV8 is cause in the pathogenesis of these two disorders \[[@B7]\]. These neoplasms can appear as late as 8 years after the diagnosis of CD, so a long term follow-up is required in patient in whom the diagnosis of CD has been established \[[@B6]\]. Conclusion ========== Castleman\'s disease (CD) is a rare lymphoid disorder, where etiology is related to Human herpes virus 8. CD could present in two forms: the localized (hyaline vascular type) and the systemic (plasma cell type). The localized hyaline vascular form has a unique indolent lymph node hyperplasia; which can be found in the abdomen, retroperitoneum or any lymph node basin; as a solitary mass. Localized CD is radiologically nearly undistinguishable from malignant neoplasms but some characteristics are quite specific, like the type of hypervascularization and the star-shape microcalcifications. A good preoperative work-up and an open biopsy during surgery, for abdominal and retroperitoneal mass if no diagnosis has been establish, can help to avoid extensive resection when facing this benign disorder. Complete surgical excision is curative; recurrences have only been described after incomplete resection. The prognosis is excellent with a five years survival of nearly 100%. List Of Abbreviations ===================== CD = Castleman\'s Disease HV = Hyaline vascular type (of Castleman\'s disease) PC = Plasma cell type (of Castleman\'s Disease) CT scan = Computed tomography scanner US = Ultrasonography Competing Interests =================== No competing interests have to be reported for this work and manuscript by any of the authors or institution. Authors\' Contributions ======================= **PB**: Study design, literature review, patient\'s follow-up, medical charts review, and manuscript. **GC**: Literature research and review. **GZ**: Medical charts review. **FR**: Literature review and manuscript. **OH**: Patient\'s surgical management, manuscript review. **PhM**: Manuscript review. Acknowledgements ================ Patients consent was obtained for publication of these case reports.

PubMed Central

2024-06-05T03:55:58.194734

2005-6-7

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1166581/", "journal": "World J Surg Oncol. 2005 Jun 7; 3:33", "authors": [ { "first": "Pascal", "last": "Bucher" }, { "first": "Gilles", "last": "Chassot" }, { "first": "Guillaume", "last": "Zufferey" }, { "first": "Frederic", "last": "Ris" }, { "first": "Olivier", "last": "Huber" }, { "first": "Philippe", "last": "Morel" } ] }

PMC1167557

Like SARS, Ebola, and other emerging infectious diseases, antibiotic resistance in bacteria may have a zoonotic origin \[[@pmed-0020232-b1]\]. Evidence suggests that antibiotic use in agriculture has contributed to antibiotic resistance in the pathogenic bacteria of humans, but the chain from cause to effect is long and complicated. Antibiotic use clearly selects for antibiotic resistance, but how far do these effects extend beyond the population where antibiotics are used? Antibiotics and antibiotic-resistant bacteria (ARB) are found in the air and soil around farms, in surface and ground water, in wild animal populations, and on retail meat and poultry \[[@pmed-0020232-b2]\]. ARB are carried into the kitchen on contaminated meat and poultry, where other foods are cross-contaminated because of common unsafe handling practices \[[@pmed-0020232-b10],[@pmed-0020232-b11]\]. Following ingestion, bacteria occasionally survive the formidable but imperfect gastric barrier, and colonize the gut \[[@pmed-0020232-b12]\]. Patterns of colonization (asymptomatic carriage) and infection (symptomatic carriage) in human populations provide additional evidence that ARB occasionally move from animals to humans \[[@pmed-0020232-b13],[@pmed-0020232-b14]\]. The strongest evidence comes from the history of the use of antibiotics for growth promotion in Europe. After first Denmark and then the European Union banned the use of antibiotics for growth promotion, prevalence of resistant bacteria declined in farm animals, in retail meat and poultry, and within the general human population \[[@pmed-0020232-b8],[@pmed-0020232-b15]\]. Despite the evidence linking bacterial antibiotic resistance on farms to resistance in humans, the impact of agricultural antibiotic use remains controversial \[[@pmed-0020232-b16]\] and poorly quantified. This is partly because of the complex of population-level processes underlying the between-species ("heterospecific") and within-species, host-to-host ("horizontal") spread of ARB. To emerge as human pathogens, new strains of ARB must (1) evolve, originating from mutations or gene transfer; (2) spread, usually horizontally among humans or animals, but occasionally heterospecifically; and (3) cause disease. All three of these steps are complex and imperfectly understood. The emergence of a new type of resistance is a highly random event, which can\'t be predicted accurately, and may involve multiple steps that preclude perfect understanding even after the fact. Spread is equally complicated and may obscure the origins of resistance. In some cases, emergence of resistance in one bacterial species is a consequence of the emergence and spread in another species, followed by the transfer of resistance genes from one bacterial species to another. Because of the underlying complexity, mathematical models are necessary to develop theory---a qualitative understanding of the underlying epidemiological processes \[[@pmed-0020232-b20]\]. Theory helps researchers organize facts, identify missing information, design surveillance, and analyze data \[[@pmed-0020232-b26]\]. Horizontal Transmission {#s2} ======================= Theory clearly shows that the impact of agricultural antibiotic use depends on whether resistant bacteria have high, low, or intermediate horizontal transmission rates in human populations \[[@pmed-0020232-b23],[@pmed-0020232-b24]\]. The rate of horizontal transmission among humans is determined by the underlying biology of the pathogen, medical antibiotic use, and hospital infection control, but not by agricultural antibiotic use \[[@pmed-0020232-b22]\]. On the other hand, a farm where multiple antibiotics are used routinely, universally, and in low quantities for growth promotion is likely to be an excellent environment for the evolution of multiple resistance factors, including some variants that might never have evolved in humans. Thus, even very rare transmission resulting from agricultural antibiotics may have a medical impact by introducing new resistant variants to the human population. The epidemiology of spread in the human population dictates how the impact of agricultural antibiotic use should be assessed. Zoonotic pathogens, such as Campylobacter and Salmonella, are generally regarded as having low horizontal transmission rates in human populations. While resistance in zoonotic infections should be directly attributable to resistance in the zoonotic reservoir, the impact of agricultural antibiotic use remains controversial \[[@pmed-0020232-b18],[@pmed-0020232-b27]\]. Zoonotic species could acquire resistance genes from human commensal bacteria during the infection process, but this hypothesis is difficult to test. For pathogens with high horizontal transmission rates, resistant bacteria will spread rapidly once they have emerged, and prevalence will be maintained at a steady state by horizontal transmission. Thus, the impact of subsequent heterospecific transmission is limited ([Figure 1](#pmed-0020232-g001){ref-type="fig"}). Nevertheless, one or two heterospecific transmission events could be sufficient to cause the appearance of a highly successful ARB genotype in humans, affecting the timing, nature, and extent of spread within the human population \[[@pmed-0020232-b22]\]. Not only are such events difficult to trace, but their impact is impossible to measure, since there is no way to know what type of resistance would have appeared and with what temporal pattern, if transfers from animals had been prevented. ::: {#pmed-0020232-g001 .fig} Figure 1 ::: {.caption} ###### The Emergence and Spread of Antibiotic Resistance in Bacteria with High Horizontal Transmission Rates Emergence and spread begins with a honeymoon period following the approval of a new antibiotic; the honeymoon ends when resistance emerges. Prevalence increases exponentially at first, but it eventually approaches a steady state. The impact of agricultural antibiotic use must be assessed by comparing the observed situation with the counterfactual situation, an imaginary world in which antibiotics were never used in agriculture. The impact of agricultural antibiotic use is, then, the total number of cases of resistance that would not have happened without the use of antibiotics in agriculture. This is approximately the difference between the time of actual emergence and the counterfactual emergence, multiplied by the steady-state prevalence. While we don\'t know what would have happened in any particular case, we can estimate the likely magnitude of agricultural impacts. ::: ![](pmed.0020232.g001) ::: The case where horizontal human transmission rates are intermediate is particularly interesting. If each case in a population generates approximately one new case (a situation we call "quasi-epidemic" transmission), each instance of heterospecific transmission will initiate a long chain of horizontal transmission that eventually burns out. Quasi-epidemic transmission can amplify a relatively low amount of heterospecific transmission and substantially increase prevalence \[[@pmed-0020232-b23]\]. The effect is sustained as long as heterospecific transmission continues. A corollary is that banning agricultural antibiotic use would have maximal benefits if horizontal transmission is quasi-epidemic \[[@pmed-0020232-b24]\]. Moreover, the effects are most difficult to estimate because both heterospecific and horizontal transmission must be accounted for. These principles apply to bacteria associated with outpatient antibiotic use and community-acquired infections as well as those that are primarily hospital-acquired. Although quasi-epidemic transmission would seem to be a special case, it may in fact be the rule for many hospital-acquired bacteria because it is the natural endpoint of the interplay between economics and ecology \[[@pmed-0020232-b30]\]. By spending money on hospital infection control, hospital administrators can reduce nosocomial transmission rates for resistant bacteria. For example, hospitals may screen and isolate patients who are likely to be carriers (i.e., active surveillance) and implement infection-control measures, but this comes at the cost of isolating patients \[[@pmed-0020232-b31]\]. Total costs are minimized by spending just enough to eliminate (or nearly eliminate) the pathogen; thus, quasi-epidemic transmission is the economic optimum \[[@pmed-0020232-b30]\]. The Community as a Reservoir for Resistance {#s3} =========================================== Horizontal transmission is further complicated by population structure, such as the movement of humans through hospitals and long-term care facilities. Medical antibiotic use and horizontal transmission rates are high in hospitals, but this is counterbalanced by short hospital stays. An emerging view for hospital-acquired bacterial infections is that persistent asymptomatic carriage plays a key role in the epidemic of resistance. ARB can asymptomatically colonize a person for years: even if the number of other people infected during a single hospital visit is less than one, this number will exceed one when summed over several hospital visits \[[@pmed-0020232-b25],[@pmed-0020232-b32],[@pmed-0020232-b33]\]. Thus, the ecological reservoir of resistance in the community plays an important role in the increasing frequency of resistance in hospital-acquired infections. Short hospital visits and long persistence times of ARB in people guarantee that some of the costs associated with failed infection control are passed on to other hospitals---new carriers are frequently discharged from one hospital only to be admitted to another hospital later \[[@pmed-0020232-b30]\]. Thus, the harm done by these ARB is borne by the whole human population, particularly all of the health-care institutions that serve a single catchment population. In economic terms, the damage caused by the carriage of ARB is a kind of pollution. By comparing the total number of new carriers generated in the community, the impacts of agricultural antibiotic use on hospitals can be compared directly to the impact hospitals have on each other ([Figure 2](#pmed-0020232-g002){ref-type="fig"}). The rate of heterospecific transmission is intrinsically difficult to measure directly because the risk of exposure and colonization per meal is very small. Nevertheless, agricultural antibiotic use may generate as many carriers as hospitals for the simple reason that the population experiences many more meals than hospital discharges \[[@pmed-0020232-b34]\]. When agricultural and nosocomial transmission are equally rare in the population, the latter will be much easier to identify and quantify. ::: {#pmed-0020232-g002 .fig} Figure 2 ::: {.caption} ###### How Large Is the Impact of Antibiotic Use in Agriculture? Comparing the amount of antibiotics used in agriculture with the amount used in medicine means comparing fundamentally different things because they affect the emergence of medically important antibiotic resistance in different ways. For hospital-acquired infections, it is more appropriate to think about ARB carriage in the community as a kind of pollution that flows into hospitals. Thus, the appropriate way to measure impact is by counting how many new carriers are added to the community reservoir from hospital discharges versus from exposure to bacteria that originate on farms. Different formulas describe these processes. To count ARB carriers among hospital discharges, let *x* denote the proportion of patients from a hospital (or other institution) that are colonized on discharge. In some discharged patients, resistant bacteria clear quickly, but a fraction, *p*, become ARB carriers. Some proportion of patients were already carriers at the time of admission, denoted by *k*. Institutions vary by size, *H*, and average length of stay (1/*s*). Thus, the rate that new carriers are discharged from a hospital is given by the formula: *sH(px − k)*. This formula measures the contribution of a hospital to the number of ARB carriers in the community. For example, a hospital with 400 filled beds (*H* = 400 people) serves a US population of about 250,000 people. With a five-day average length of stay (the discharge rate is *s* = 0.2 per patient per day), the hospital discharges about 80 patients each day. If we suppose that 20% of patients acquire resistant bacteria while hospitalized, and one in four of these patients become carriers (*px − k* = 0.05), a hospital would discharge about four persistently colonized people per day---about 1,460 carriers after one year, or approximately 0.58% of its catchment population. A different formula characterizes heterospecific transmission, following exposure to ARB on contaminated food. We let *g* denote the daily per-capita rate that ARB are ingested with a meal. Similarly, we let *h* denote the proportion of those ARB populations that survive the gastric barrier and persistently colonize. The number of new carriers generated in the community by agricultural antibiotic use in a population of size *N* is: *ghN*. For example, if the average person consumes some ARB in 1% of meals (*g* = 0.03 per person per day), followed by colonization with probability one in 2,000 (*h* = 0.0005), agricultural antibiotic use would generate about four new carriers per day in a population of 250,000 people, *N*, approximately the same number as a hospital. The formulas illustrate a general principle: "A large number of people exposed to a small risk may generate many more cases than a small number exposed to a high risk" \[[@pmed-0020232-b34]\]. ::: ![](pmed.0020232.g002) ::: A Natural Experiment: Glycopeptides and Vancomycin-Resistant Enterococci {#s4} ======================================================================== Is the impact of agricultural antibiotic use on the emergence and spread of ARB in humans large or small relative to medical antibiotic use? Put another way, are farms or hospitals bigger polluters? A large-scale natural experiment was conducted in the United States and several European countries when each adopted different policies on glycopeptide use in animals (avoparcin) and humans (vancomycin) \[[@pmed-0020232-b16],[@pmed-0020232-b17],[@pmed-0020232-b35]\]. Many European countries approved avoparcin for animal growth promotion in the 1970s, but the US did not. In the early 1980s, demand for vancomycin in US hospitals surged because of increasing aminoglycoside resistance among enterococci and methicillin resistance in Staphylococcus aureus. Physicians in US hospitals also used oral vancomycin for some Clostridium difficile infections \[[@pmed-0020232-b37]\]. In the late 1980s and early 1990s, vancomycin-resistant enterococci (VRE) emerged and spread through US health-care systems. In Europe, hospitals used less vancomycin because most enterococci were sensitive to aminoglycosides, and oral vancomycin was seldom used. VRE still emerged and spread through European hospitals, but the problem has been less severe than in the US \[[@pmed-0020232-b40]\]. A different pattern emerges for community prevalence of VRE. VRE are rarely found outside of hospitals in the US, except for patients who have a prior history of hospitalization. Community prevalence of VRE in the US is typically less than 1%. In contrast, community prevalence of VRE was estimated at 2%--12% in Europe during the late 1990s, including carriage by people with no history of hospitalization \[[@pmed-0020232-b17],[@pmed-0020232-b41]\]. In other words, the European community reservoir generated by vancomycin use in hospitals and avoparcin use in agriculture was apparently much larger than the US community reservoir generated only by vancomycin use in hospitals. The prevalence of VRE in the community declined after the EU banned avoparcin \[[@pmed-0020232-b15]\]. Thus, avoparcin is at least partly responsible for the reservoir of VRE in the European community, but how much of that reservoir came from avoparcin and how much came from hospitals? To weigh the impact, we subtract the community prevalence of VRE in the US (\<1%) from the community prevalence of VRE in Europe (\>2%). The remainder (\>1%) is attributed to avoparcin. This analysis probably underestimates the real impact because vancomycin was used less in European than in US hospitals. Thus, avoparcin use in Europe would appear to be responsible for generating a larger reservoir of VRE in the community than US hospitals. Put another way, the impact of avoparcin use on European hospitals was larger than the impact of US hospitals on one another. Conclusion {#s5} ========== Despite the evidence that avoparcin use has had a large impact on the emergence and spread of VRE by increasing the reservoir of VRE in the EU, some uncertainty continues to surround the clinical significance of VRE strains of animal origin and of the zoonotic origins of resistance in general. Bacterial strains circulating in hospitalized populations may be genetically distinct from those circulating in the general human population \[[@pmed-0020232-b13],[@pmed-0020232-b17],[@pmed-0020232-b49]\]. Thus, bacterial populations are some combination of zoonotic, quasi-epidemic, and epidemic strains. The complexity of bacterial population biology and genetics makes it practically impossible to trace bacteria (or resistance factors) from the farm to the hospital, or to directly attribute some fraction of new infections to agricultural antibiotic use. Asymptomatic carriage of resistance factors by nonfocal commensal bacteria adds to a general risk of resistance, but transfer of resistance among bacterial species is unpredictable and difficult to quantify. Until more evidence is available, it is prudent and reasonable to consider bacteria with resistance genes a general threat \[[@pmed-0020232-b50]\]. Some part of the controversy over agricultural antibiotic use has been a disagreement about how to weigh evidence and make decisions when the underlying biological processes are complex. In this case, the effects of agricultural antibiotic use on human health remain uncertain, despite extensive investigation, and the effects may be unknowable, unprovable, or immeasurable by the empirical standards of experimental biology. What should be done when complexity makes an important public-health effect intrinsically difficult to measure? What is an appropriate "null hypothesis" or its equivalent? Should the same standards of proof be used in science and science-based policy? Where should the burden of proof fall? Scientific assessments for policy should summarize the best state of the science, recognizing that the burdens and standards of proof are necessarily softer because of the uncertainty that is introduced by biological complexity. The best decisions weigh the evidence in light of the inherent uncertainty. The EU banned the use of antibiotics for growth promotion, based on the precautionary principle. The use of the precautionary principle was criticized by some as unscientific in this context. In fact, the intrinsic problem of knowability, posed by the biological complexity of the problem, makes the use of precautionary decision making particularly suitable in this arena. The assumption that plausible dangers are negligible, even when it is known that such dangers are constitutively very difficult to measure, may be more unscientific than the use of precaution. Summary Points -------------- The emergence and spread of ARB is complex and intrinsically difficult to study; mathematical models can help with understanding underlying mechanisms and guiding policy responses.Agricultural antibiotic use may generate novel types of ARB that spread to the human population; models can help estimate how much additional disease has been caused by agricultural antibiotic use.Transmission of ARB from animal to human populations is particularly difficult to measure, as it is the product of a very high exposure rate to potentially contaminated food, and a very low probability of transmission at any given meal.Depending on the assumptions used, the model suggests that transmission from agriculture can have a greater impact on human populations than hospital transmission.A comparison of patterns of colonization of VRE in Europe and the United States, which had different patterns of agricultural and hospital antibiotic use, suggests that agricultural antibiotic use can have important quantitative effects on the spread of resistance in the community. **Citation:** Smith DL, Dushoff J, Morris G Jr (2005) Agricultural antibiotics and human health. PLoS Med 2(8): e232. The views presented in this paper represent the personal views of the authors and do not construe or imply any official position or policy of the Fogarty International Center, National Institutes of Health, Department of Health and Human Services, or the US government. ARB : antibiotic-resistant bacteria VRE : vancomycin-resistant enterococci [^1]: David Smith is a mathematical epidemiologist and infectious disease ecologist at Fogarty International Center, National Institutes of Health, Bethesda, Maryland, United States of America. Jonathan Dushoff is on the research staff in the Department of Ecology and Evolutionary Biology at Princeton University, Princeton, New Jersey, United States of America, and at Fogarty International Center, National Institutes of Health. J. Glenn Morris, Jr., is a physician epidemiologist and specialist in infectious diseases, and chair of the Department of Epidemiology and Preventive Medicine in the School of Medicine, University of Maryland, Baltimore, Maryland, United States of America. [^2]: **Competing Interests:** DLS and JGM received funding four years ago from Pfizer to conduct a risk assessment for the emergence of streptogramin resistance in Enterococcus faecium. JD declares that he has no competing interests.

PubMed Central

2024-06-05T03:55:58.197092

2005-7-5

{ "license": "Creative Commons Zero - Public Domain - https://creativecommons.org/publicdomain/zero/1.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1167557/", "journal": "PLoS Med. 2005 Aug 5; 2(8):e232", "authors": [ { "first": "David L", "last": "Smith" }, { "first": "Jonathan", "last": "Dushoff" }, { "first": "J. Glenn", "last": "Morris" } ] }

PMC1168885

Introduction {#s1} ============ "Academic medicine" might be defined as the capacity of the health-care system to think, study, research, discover, evaluate, innovate, teach, learn, and improve. As such, little could be more important---particularly as new discoveries in science offer tremendous opportunities and emergent diseases pose huge threats. Yet something is not right with academic medicine. Worse, the diagnosis is not entirely clear, although many august bodies have reported on the issue ([Table 1](#pmed.0020207-t001){ref-type="table"}), and the treatment is unknown. Moreover, these previous consultations have been based in the industrialised world, particularly the United Kingdom and the United States, and few have taken a global view. ::: {#pmed.0020207-t001 .table-wrap} Table 1 ::: {.caption} ###### Reports of Major National Academic Medicine Organisations ::: ![](pmed.0020207.t001) ::: The International Campaign to Revitalise Academic Medicine (ICRAM) was founded to give young medical academics an opportunity to think about the future of academic medicine ([Box 1](#box1){ref-type="boxed-text"}). It started with only two premises: (1) at the start of the new millennium it was necessary to think globally, and (2) "more of the same" was not the answer. Reinvention was needed. Box 1. ICRAM ------------ In 2003 the *BMJ*, *Lancet*, and 40 other partners launched ICRAM, a global initiative that is committed to fostering a debate about the future of academic medicine ([bmj.com/academicmedicine](bmj.com/academicmedicine)). The campaign arose because of a persistent concern that academic medicine is in crisis around the world. At a time of increasing health burden, poverty, globalisation, and innovation, many have argued that academic medicine is nevertheless failing to realise its potential and global social responsibility. ICRAM is composed of the following: A core working party of 20 medical academics representing 14 countriesStakeholder groups representing the areas of academia, business and industry, government and policy makers, journal editors, patients, professional associations, and students and traineesRegional groups covering the worldA facilitating committee that helps plan and execute the ICRAM work Through a series of stakeholder and regional consultations, systematic review of the available evidence, and future scenarios building, ICRAM intends to produce a series of recommendations for reform in global academic medicine, including the following: Developing a vision and values for academic medicineRecommending strategies for building capacity in academic medicine, including better career pathsProposing how academic medicine could improve its relationships with "customers," including patients, policy makers, practitioners, and others ICRAM decided to undertake some scenario planning, a technique of thinking about both the future and the present. The technique works by gathering together a team who consider the instabilities in the present and the drivers of the future and who then imagine plausible but different futures. The aim is not to predict the future, which is impossible, but to enable richer conversations through stretching thinking on what the future might bring. Once the scenarios have been created, they can be used to think more deeply about the present and the near future. The group began by considering current instabilities in academic medicine around the world ([Box 2](#box2){ref-type="boxed-text"}). None of these ideas are new, and most people would probably agree with most of them. Box 2. Current Instabilities in Academic Medicine ------------------------------------------------- Widespread, even universal, agreement that things are not right but little agreement on the exact nature of the problemLack of capacity in "translational research"---that which brings innovations directly to patientsThe substantial gap between best, evidence-based practice and what actually happensThe canyon between academics and practitionersThe growing difficulty/impossibility of a single individual being competent in practice, research, and teachingUse of citation indices in research assessment, which overemphasises the value of basic research and underemphasises the importance of applied research that may bring more immediate benefit to patientsA lack of mutual respect among different categories of researchers---basic, clinical, public health, primary care, applied, etc.Problems with career progression for academicsShortages of doctors wanting to enter careers in researchIn many, probably most, countries those doctors who enter careers in research, who ideally would be the "best and the brightest," being likely to earn much less than those who can spend at least some time working in private practiceResearch often not being concerned with the biggest health problems (particularly true in a global context)Systems of accreditation of doctors making life difficult for doctors who want to pursue an academic careerAlthough no clinicians are openly against research, clinicians being often unimpressed with doctors who concentrate on researchToo much medical research being undertaken by doctors with limited training in research methods---making the research of poor qualityResearchers often setting little store by quality improvement projects, although such projects are one of the ways of ensuring that all patients receive the most up-to-date careMuch of the teaching in medicine being done by people with very little training in medicineThe ratio of teachers to students in many medical schools being so low that the quality of teaching is reducedIn many countries academic medicine lacking a well-resourced institution to speak for itAcademic medicine relating poorly to its stakeholders---patients, policy makers, practitioners, the public, and the mediaThe great pressures on health services, such that academic medicine is often squeezed and forgotten Much of what determines the future of academic medicine will lie outside the control of medical academics themselves. The world will change around them, and they will have to follow. But there will also be change that comes from within academic medicine. [Box 3](#box3){ref-type="boxed-text"} shows some of the drivers of the future considered by the group. Box 3. Drivers of Change in Academic Medicine --------------------------------------------- New science and technology, particularly genetics and information technologyThe rise of sophisticated consumersGlobalisationEmergent diseasesIncreasing gap between rich and poorDeath of distance"Big hungry buyers" demanding more from health careSpread of the internet and digitalisationManagerialismIncreasing anxieties about securityExpanding gap between what could be done and what can be afforded in health careLack of agreement on where "health" begins and endsAgeing of societyFeminisation of medicineIncreasing accountability of all institutionsLoss of respect for expertsRise of self careRise of ethical issues24/7 societyEconomic and political rise of India and China Five Scenarios {#s2} ============== In building scenarios, the group used a time span of 20 years, but some of the scenarios are more futuristic than others. We decided to write up the scenarios as if they were in the past in order to give some idea of how they might arise. The scenarios are summarised in [Table 2](#pmed.0020207-t002){ref-type="table"}, and described more fully in the rest of this article. ::: {#pmed.0020207-t002 .table-wrap} Table 2 ::: {.caption} ###### Summary of Scenarios ::: ![](pmed.0020207.t002) ::: **Academic Inc.: "Academic medicine flourishes in the private sector".** Slowly but surely the public sector around the world realised that it could not support the costs of academic medicine. Medical students had high earnings during a professional lifetime: why shouldn\'t they pay for their education? And if researchers were doing something valuable then shouldn\'t they be able to find a market for their product---accepting that sometimes payment would come from the public sector? The process of academic medicine moving almost entirely into the private sector began with an increasing number of medical schools becoming private. The most prestigious schools went first. In an increasingly global market these schools could charge high fees, pay their staff well, and improve their facilities. They also invested a great deal in information and communication technology, bringing state-of-the-art learning to their students. This meant too that the schools could run courses for students far away from their geographical base. As these schools developed they expanded internationally, sometimes forming alliances with other prestigious schools but also taking over weaker schools. Soon the best schools were operating on all five continents. In the branches in developing countries, medical student bodies tended to comprise both students from the developed countries along with a quota from the developing countries. Competition was intense and involved both cost and quality. Schools that managed to improve quality while reducing costs---usually through clever use of technology---flourished, but a great many medical schools disappeared. The numbers of students, however, increased, and the competition for talent was intense, with schools offering generous bursaries to poor but bright students and becoming ever more sophisticated at finding high-quality students in resource-poor populations. As happens in most intensely competitive markets, medical schools also competed by occupying niches. Schools offered very different kinds of courses, specialising in older students, basic science, rural medicine, surgical skills, training doctors for poor communities (in both the country where the medical school was based and lower and middle income countries), and many other subject areas. Sometimes the students\' fees were paid by governments, local communities, or the military in order to produce students who met their needs. Many students attended schools in countries other than their own. Health research happened almost entirely in the private sector, but in a wide range of organisations: pharmaceutical companies, medical schools, biotechnology companies, small companies offering a huge range of services, and charities. Companies were founded not only by researchers but also by patients, practitioners, and others. Many of the companies founded by academics offered complex and innovative heath services. As in all business, to be successful companies had to be highly responsive to the needs of customers, including patients and governments. Those that were innovative, flexible, responsive, and relentlessly cost-conscious flourished, but many companies "failed". Little stigma was, however, attached to "failure". Indeed, as in Silicon Valley at the end of the 20th century, experience of "failure" was seen by many as an important qualification in a leader. The injection of more competitive pressure and a competitive business model into academic medicine increased not only efficiency but also "effectiveness": research was much more relevant, and the time lag between the development of new ideas and their introduction into practice was dramatically shortened. Basic science was still well funded because both governments and investors recognised the potentially high returns. Research into the health needs of poor and marginal populations also improved because public sector bodies concentrated their resources into these problems, leaving the problems of the wealthier to the market. Applying a competitive business model to academic medicine meant that efficiency and effectiveness trumped equity. On the negative side, the scenario "Academic Inc." resulted in a two-tier system, with the rich finding it easy to create careers in academic medicine and the poor finding it hard to enter the profession---despite the generous bursaries available to some. In addition, much more attention continued to be paid to the health problems of wealthier people and countries, and the brain drain from poor to rich countries accelerated. Innovation also suffered. Private academic medicine enjoyed less lead time and had more direct and immediate accountability to its shareholders than when it was publicly funded. **Reformation: "All teach, learn, research, and improve".** Twenty years ago there was increasing concern about the gap between academic medicine and practice, with important research results not being implemented, too much irrelevant research, bored students, and practitioners who stopped learning. In some medical communities and among their medical leadership, the response was not to try and strengthen academic medicine and make it more responsive but rather to abolish it and instead to bring the processes of teaching, learning, researching, and improving into the main stream of health care, and tailor these to local needs. This innovative---though not initially welcomed---response proved to be highly successful and was copied everywhere. A century of separation of academic medicine from practice was ended. Professors disappeared. The entity "academic medicine" was dead. It was akin to the destruction of the monasteries and so became known as the reformation of academic medicine. Teaching, learning, researching, and quality improvement all began to take place in the practice setting and were everybody\'s business. But importantly the fiction that an individual could be competent in practice, teaching, research, and improvement finished. It was teams that had to have all these competencies not individuals, and substantial investment was made in getting teams to work well and to communicate to a degree rarely seen before in health care. The teams were supported by advanced technology that provided online learning, decision support, answers to questions that arose during practice, and access to research results. Much of the teaching and research was done in collaboration with patients, and all teams included patients as well as practitioners, students, professional researchers, and other health professionals. Research was built around the questions that arose when doctors (and other health professionals) and patients consulted together. The questions were collected by the National Question Answering Service, which provided evidence-based answers to questions when it could. The service also organised research to answer questions that arose commonly, were unanswered at that time, but could be answered. Teams of different sizes and skills were assembled to conduct research; some of the researchers were permanently in practice but others---particularly basic researchers---were resident in research institutions and joined teams as needed. Some research was driven by discoveries made in basic science rather than questions that arose during practice; the fact that practitioners and researchers were used to working together in teams facilitated "translational" research. The teams also switched back and forth from research to quality improvement, ensuring that research developments were fed through into practice. Studies reporting the results of quality improvement projects were published just as frequently as research studies, and highly efficient information systems ensured that relevant information reached practitioners quickly---unlike in the old days when practitioners had been deluged with research results, most of which were irrelevant to them. Intellectual leadership in health care was provided by specialist societies, which included patients, citizen juries, researchers, health professionals other than doctors, health managers, and policy makers as equal members. In most countries the specialist societies were gathered together into an academy or institute of health care, and there was an international academy that was much more than a talking shop: it had a strong influence on world leaders. Health students started their training with six months in institutions that taught them how to learn. The learning professionals who staffed these institutions were available to all the practice teams. Students then learnt through attachment to practice teams, starting with a spell in general practice. Some students specialised early, some becoming competent cardiologists within five years. Becoming an independent professional depended not on university degrees or exams but on a demonstration of competencies determined by a national body dominated by patients that used the most up-to-date methods for assessment. Learning was by doing, and the divisions of undergraduate, specialist, and continuing education disappeared---as did the divisions among teaching, learning, researching, and quality improvement. One problem with the "Reformation" scenario was that it lacked stability because it required shared values and beliefs, which not all teams held. Plus decision making could be slow, and it was hard for brilliant and charismatic individuals to shine as leaders and thinkers. One result was that such individuals eschewed careers in health care, research, and education. In developing countries, in particular, the lack of an academic medicine structure meant that there were fewer opportunities to influence medical research and training. **In the public eye: "Success comes from delighting patients and the public".** Academic medicine was slow to recognise the rise of global media, "celebrity culture," and the use of public relations (or spin) to drive the political process, but once it did recognise how the world had changed it responded dramatically. Whereas it had been suspicious of the media and public appeal and rather patronising to patients, academic medicine realised that to succeed it must delight patients and the public and learn to use the media. The most successful academics became those who were very responsive to patients and the public, capturing their imaginations, and appearing regularly on their television screens. Some medical academics became as well known as film and rock stars and were feted by politicians. All academic institutions became dominated by public citizens and patients, and the public and media relations department became the most important department in any institution. Money---from both public and private sources---followed "interest," which was often influenced through game and "reality" shows on television. Academic medicine learnt from sport, and large prizes were awarded to those who won academic competitions. Although some academics were horrified by these developments, others remembered how John Harrison had been stimulated to solve the problem of calculating longitude through the promise of a large prize. Not all decisions on research priorities and resource allocation were made in the glare of television cameras. Although all decisions put the public interest first and were made by the representatives of the public, some were made by more sedate and evidence-driven bodies like citizens\' juries, where randomly selected patients and members of the public were presented with detailed evidence by "experts." Medical training was also conducted in the public eye, with students receiving much of their training from expert patients. The agenda for training was set predominantly by the public and patients, and responsiveness to patients was the number one characteristic of successful doctors and students. There was much greater diversity than at the beginning of the millennium in the form and size of academic institutions, with both huge public and private universities and smaller institutions that were often built around one charismatic individual. Competition among the institutions was intense---particularly for "celebrity" teachers and researchers. Only those institutions that could attract and keep public attention could survive. In developing countries, the academic health community linked itself with strong consumer movements, such as those focused on HIV/AIDS, and the leading nongovernmental organisations established their own medical schools. These ensured a powerful public voice, that training was tailored to local need, and a committed group for field testing new research advances. On the negative side in the "In the Public Eye" scenario, medical academics felt more anxious about their job security and ability to succeed. Even celebrity academics worried their time in the spotlight was short lived. Advances in science, medicine, and technology were shaped by popular appeal, and thus subject to fads and fashion. Some patients struggled with their new-found status as governors, and there was little regulation of health information. **Global academic partnership: "Academic medicine for global health equity".** In 2005, the world began to find the growing global gap between the rich and poor unacceptable. The concern was driven partly by the media and global travel bringing the plight of the poor in front of the eyes of the rich, but it was also driven by anxieties over global security. Terrorism was recognised to be fuelled by the gap between rich and poor. Global policy makers also understood better---particularly after the report of the Commission on Macroeconomics and Health \[[@pmed-0020207-b2]\]---that investment in health produced some of the richest returns in not only social but also economic development. Health care was a "must have" not a "nice to have". Money flowed into health in the poor world, and governments required that the investment be accompanied by learning, research, planning, and evaluation. The primary concern of much of academic medicine became to improve global health, particularly through concentrating on the health problems of the 90% who had previously received only 10% of health care resources (the 90:10 gap). Academics became excited by this kind of work not only because it was intellectually exciting and highly rewarding in personal terms but also because it was where money and prestige were most likely to be found. The result was that it was impossible for an academic institution to be a world leader without a substantial investment in global health and extensive links around the world. The view and scope of academic medicine broadened. It was increasingly concerned with human rights, justice, economics, and the environment, recognising that these are the major drivers of health. This broader view meant that academic medicine (usually referred to as "global health innovation" by 2012) became the main institution concerned with the rights of those who will be alive 50 years from now, a group that previously had nobody to speak for them. But at the same time basic science remained important because of the contribution it could make to global problems like finding vaccines and new treatments for malaria, AIDS, and emergent diseases like rapidly spread respiratory virus, which appeared in 2010 and killed millions in a global pandemic. Academic medicine, in partnership with governments (and where corruption is prevalent, with nongovernmental organisations), became a major driver towards achieving the millennium development goals. The G8 governments had signed an accord that prohibited recruitment of academic health professionals from developing countries. Medical schools and research institutions formed themselves into networks linking with local nongovernmental organisations, joining developed and developing countries and forming links among developing countries. A network was formed whereby the universities in developed countries committed 10% of their faculty members\' time to addressing problems of the developing world. Some institutions formed developed country--developing country pairs, some merged, and researchers, teachers, and students moved regularly between both settings. The net flow was to the developing world, with the 90:10 divide beginning to correct itself surprisingly rapidly. Big investments in information and communication technology meant that those in developing countries had the same access to information and modern learning methods as those in developed countries. The networks of institutions developed a global governance structure with substantial input from politicians, practitioners, policy makers, the public, and patients. Academic medicine moved from being marginal to central in global affairs, and medical academics, particularly those with experience in both the developing and developed world, became global leaders. It was a development that happened naturally because of their broad interests in human rights, justice, and the environment. The "Global Academic Partnership" scenario for academic medicine, however, was idealistic and sometimes struggled with realising its full potential---despite the best intentions of its architects and practitioners---because it required enormous political will and global cooperation. Too often nations would revert to narrow self interest. Academics as well often longed for the comforts of the developed world and sometimes felt exhausted by extensive travelling and the enormous problems of the developing world. **Fully engaged: "Academic medicine engages energetically with all stakeholders".** Early in the new millennium academic medicine became concerned that its relationships with its stakeholders were mostly poor. The public had little or no understanding of what academic medicine was or why it mattered. Its very name implied irrelevance to many. Patients often felt patronised by academics, and many practitioners---including doctors---were unconvinced of the value of academic medicine. "I wouldn\'t want a professor of surgery touching me," was a commonly heard refrain. Although some leading academics did have good relationships with politicians, policy makers found that many academics were not interested in policy problems and that the studies they produced, even if relevant, came too late to be useful. Policy makers recognised that biotechnology might be very important in future wealth creation, but it was difficult to fund because the public profile of academic medicine was both low and clouded. Most medical academics recognised that they were doing a poor job of relating to stakeholders and that it was thus unsurprising that they were misunderstood, underappreciated, and seen as largely irrelevant. This, they thought, was particularly unfortunate as the ability of the system of health care to discover, think, study, learn, and evaluate had never been more important. The medical academic community thus decided that it had to do better, and across the globe medical academics organised themselves to engage fully with their stakeholders. In many countries this meant the creation of new organisations. In others it involved the transformation of existing organisations: the gongs and gowns were abandoned, and focus groups began. Fifty prestigious universities in developed countries with medical faculties partnered with universities in developing countries to help stop the "brain drain" and replace it with a "brain gain" through incentive programs that provided resources for training and research, academic recognition, travel funds, and family support. Everywhere medical academics had to learn how to communicate with the public, patients, and practitioners. They had to stop being elitist and patronising and recognise the messiness of public discourse. Crucially, they had to be much cleverer in handling the media, telling them not only about their successes but also sharing their uncertainties and problems. But communication on its own wasn\'t enough. Academic medicine had to bring its stakeholders inside its processes. The governance of academies included patients, the public, practitioners, health administrators, and policy makers. Sometimes the president of an academy was not a distinguished researcher but a prominent patient, journalist, or community organisation leader. The medical academics discovered that their arguments were taken much more seriously when advanced clearly by a patient rather than by themselves. Patients, health administrators, and community organisation representatives became involved not only in peer review of grants and studies but also in the prioritising, designing, and conducting of research. Medical students became the main drivers of medical education rather than simply its consumers. Slowly but surely medical academics became not a group apart but a highly diverse group of people with a broad set of skills and backgrounds. They were at the centre of a vibrant community of patients, members of the public, practitioners of all stripes, policy makers, members of the media, marketing experts, and politicians, all of whom were interested in learning, studying, researching, and thinking about health care. Some academics, it must be said, found the change uncomfortable and were unconvinced of its value. Critics talked of "dumbing down" and popularisation. They fretted that in abandoning its elitism academic medicine had lost its ability to be truly original and speak independently. For most, however, academic medicine was so much more fun than it used to be. Applications to medical schools increased. Health services invested more in evaluating what they did and paid more attention to the results. More funds flowed into basic research, and there were improvements in the connections between the many diverse groups involved in research---with the result that intellectual silos were breached. Lessons from the Scenarios {#s3} ========================== None of these scenarios will come to exist as we have described them, but the future is likely to contain some elements from each of them. We have tried to identify common features in the scenarios to learn lessons for now. These features are shown in [Box 4](#box4){ref-type="boxed-text"}. Our main hope for these scenarios is that other groups may find them useful in thinking about both the present and the future of academic medicine. The scenarios will need to be adapted to the particular social, economic, and political conditions of different regional and national settings. As such, they are tools that can be used globally, modified as required. We seek not agreement but broader thinking. Box 4. Common Features Shared By All Five Scenarios --------------------------------------------------- It is likely to be important in all scenarios for academic medicine to put more effort into relating to its stakeholders: the public, patients, practitioners, politicians, and policy makers. This may demand the development of new institutions that involve all these groups.Academic institutions will need to be more globally minded.Teaching, researching, improving, leading, and providing service will continue to be important, but expecting individuals to be competent in them all will be increasingly impractical.Teamwork will become ever more important, but it will also be necessary to allow individuals to shine and flourish.Competition among academic institutions is likely to increase, and the competition will increasingly be international.Academic institutions will need to become more "business-like" in all the scenarios. They will also need to be more adept at using the media.Teaching and learning will be increasingly important---not least because dissatisfied students may go elsewhere. Learning will be lifelong and will depend heavily on information technology.It will be increasingly important to combine research, both basic and applied, with implementation and improvement. The gap between knowledge and practice will become increasingly intolerable.The range of types of academic institutions is likely to become increasingly diverse, with medical schools or academic centres just one form.Academic medicine will need to be ever broader in its thinking and skill set, combining with and learning from other disciplines such as economics, law, ecology, and humanities.Thinking about the future will become increasingly important for academic institutions but also increasingly difficult. The members of the working party of the ICRAM are Tahmeed Ahmed (Scientist, Clinical Sciences Division, International Centre for Diarrhoeal Disease Research Bangladesh, Dhaka, Bangladesh), Shally Awasthi (Professor, Department of Paediatrics, King George\'s Medical University, Lucknow, India), A. Mark Clarfield (Professor, Department of Geriatrics, Soroka Hospital, Ben-Gurion University of the Negev, Beersheva, Israel), Lalit Dandona (Director, Centre for Public Health Research, Administrative Staff College of India, Hyderabad, India), Amanda Howe (Professor of Primary Care, School of Medicine, University of East Anglia, Norfolk, United Kingdom), John P. A. Ioannidis (Chairman, Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece), Edwin C. Jesudason (Academy of Medical Sciences National Clinician Scientist, Health Foundation Leadership Fellow, and Lecturer in Paediatric Surgery, School of Reproductive and Developmental Health, University of Liverpool, Liverpool, United Kingdom), Youping Li (Director, Chinese Cochrane Centre, West China Hospital, Sichuan University, Chengdu, China), Juan Manuel Lozano (Professor, Department of Pediatrics and Clinical Epidemiology, Javeriana University School of Medicine, Bogota, Colombia), Ana Marusic (Professor, Department of Anatomy, Zagreb University School of Medicine, and Editor, Croatian Medical Journal, Zagreb, Croatia), Idris Mohammed (Outgoing Provost, College of Medical Sciences, Department of Medicine and Clinical Immunology, University of Maiduguri, Maiduguri, Nigeria), Gretchen Purcell (Pediatric Surgery Fellow, Pittsburgh Children\'s Hospital, and Adjunct Assistant Professor of Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania, United States), Karen Sliwa-Hähnle (Professor, Department of Cardiology, CH Baragwanath Hospital, University of the Witwatersrand, Johannesburg, South Africa), Sharon E. Straus (Associate Professor, Department of Medicine, Toronto General Hospital, University of Toronto, Toronto, Ontario, Canada), Tessa Tan-Torres Edejer (Scientist, Department of Health Systems Financing, World Health Organization, Geneva, Switzerland), Timothy Underwood (Medical Research Council/Royal College of Surgeons Clinical Research Training Fellow, University of Southampton, Southampton, United Kingdom), Robyn Ward (Associate Professor, Department of Medical Oncology, St Vincent\'s Hospital and School of Medicine, University of New South Wales, Darlinghurst, New South Wales, Australia), Michael S. Wilkes (Vice Dean and Professor of Medicine, School of Medicine, University of California, Davis, California, United States), and David Wilkinson (Deputy Head and Professor of Primary Care, School of Medicine, University of Queensland, Brisbane, Australia). **Note Added in Proof** Reference information for the summary of this article being published in the *BMJ*: Clark JP, for ICRAM (2005) Five futures for academic medicine: The ICRAM scenarios. BMJ. In press. **Citation:** Awasthi S, Beardmore J, Clark J, Hadridge P, Madani H, et al. (2005) Five futures for academic medicine. PLoS Med 2(7): e207. *This article is a shortened version of a report entitled "The Future of Academic Medicine: Five Scenarios to 2025" \[[@pmed-0020207-b1]\] published by the Milbank Memorial Fund ([www.milbank.org](www.milbank.org)). A summary of this article is being published in the* BMJ. ICRAM : International Campaign to Revitalise Academic Medicine [^1]: Shally Awasthi is Professor, Department of Paediatrics, King George\'s Medical University, Lucknow, India. Jil Beardmore is Research and Project Coordinator at the Centre for Global Health, University of Ottawa, Ottawa, Ontario, Canada. Jocalyn Clark is Associate Editor of *BMJ*, London, United Kingdom, and Project Manager of ICRAM, London, United Kingdom. Philip Hadridge is an organisation consultant in Cambridge, England. Hardi Madani is a fourth-year medical student at the Royal Free and UCL Medical Schools in London, United Kingdom, and a part-time pharmacist. Ana Marusic is Professor, Department of Anatomy, Zagreb University School of Medicine, Zagreb, Croatia, and Editor of *Croatian Medical Journal*, Zagreb, Croatia. Gretchen Purcell is Pediatric Surgery Fellow at Pittsburgh Children\'s Hospital and Adjunct Assistant Professor of Medicine at the University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America. Margaret Rhoads is a medical student at Imperial College London, London, United Kingdom. Karen Sliwa-Hähnle is Professor, Department of Cardiology, CH Baragwanath Hospital, University of the Witwatersrand, Johannesburg, South Africa. Richard Smith is former Editor of *BMJ*, co-founder of ICRAM, and Chief Executive of UnitedHealth Europe, London United Kingdom. Tessa Tan-Torres Edejer is Scientist, Department of Health Systems Financing, World Health Organization, Geneva, Switzerland. Peter Tugwell is Professor of Medicine and Head of the Centre for Global Health at the University of Ottawa, Ottawa, Ontario, Canada, and Leader of ICRAM. Tim Underwood is Medical Research Council/Royal College of Surgeons Clinical Research Training Fellow, University of Southampton, Southampton, United Kingdom. Robyn Ward is Professor, Department of Medical Oncology, St Vincent\'s Hospital, Darlinghurst, New South Wales, Australia, and School of Medicine, University of New South Wales, Darlinghurst, New South Wales, Australia. See Acknowledgments for the members of the working party of the International Campaign to Revitalise Academic Medicine. [^2]: **Competing Interests:** JC is an employee of the BMJ Publishing Group, which benefits from academic medicine. RS is an employee of the UnitedHealth Group, which benefits from the output of academic medicine. It\'s not easy to see how the company might be affected financially from any of the possible changes in academic medicine. He is also an unpaid governor of St George\'s, University of London, and an unpaid board member of the Public Library of Science, both of which will be affected by any changes.

PubMed Central

2024-06-05T03:55:58.199189

2005-7-5

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168885/", "journal": "PLoS Med. 2005 Jul 5; 2(7):e207", "authors": [ { "first": "Shally", "last": "Awasthi" }, { "first": "Jil", "last": "Beardmore" }, { "first": "Jocalyn", "last": "Clark" }, { "first": "Philip", "last": "Hadridge" }, { "first": "Hardi", "last": "Madani" }, { "first": "Ana", "last": "Marusic" }, { "first": "Gretchen", "last": "Purcell" }, { "first": "Margaret", "last": "Rhoads" }, { "first": "Karen", "last": "Sliwa-Hähnle" }, { "first": "Richard", "last": "Smith" }, { "first": "Tessa Tan-Torres", "last": "Edejer" }, { "first": "Peter", "last": "Tugwell" }, { "first": "Tim", "last": "Underwood" } ] }

PMC1168887

Background ========== From a public-health perspective, the major obstetric and perinatal problem in Australia is the poor health of Indigenous mothers and babies. Although Australia as a whole has one of the lowest perinatal mortality rates (PMR) of any of the established market economies \[[@B1]\], Indigenous Australians have PMRs that are at least twice as high as their non-Indigenous counterparts \[[@B2],[@B3]\]. Moreover, the relative excess of deaths among Indigenous babies has not improved over time. In Queensland in 1987--1989, the PMR was 2.2 times higher among Indigenous compared with non-Indigenous babies and in 2000--2002 the rate was still 2.1 times higher. A similar lack of progress is evident in other states \[[@B4]-[@B6]\]. The risk of perinatal death is strongly related to a baby\'s birthweight and gestational age. For example, the risk of death for a moderately preterm baby (33 to 36 weeks gestation) in Queensland is 8.2 times greater than that of a term baby and for a very preterm baby (27 to 32 weeks) the risk is 41.3 times greater. Similarly, a baby who was born at term but weighed less than 2500 g was 6.3 times more likely to die than a baby who weighed more than 2500 g. In Queensland, Indigenous babies are 1.6 (95%ci: 1.5,1.7) times more likely to be born preterm (\<37 weeks gestation) and 2.2 (2.0,2.5) times more likely to be low birthweight at term than non-Indigenous babies \[[@B3]\]. Similar disparities in birthweight and gestational age have been reported from the other states of Australia \[[@B2],[@B5],[@B7]-[@B10]\]. The aim of this paper is to estimate the number and proportion of excess deaths among Indigenous babies that can be explained by their higher risk of low birthweight and preterm birth. This analysis strategy assumes that the distribution of birthweight and gestational age among Indigenous babies reflects the prevalence of antenatal risk factors such as smoking, infection, maternal nutrition and psycho-social stress; while any excess of mortality that remains after the excess low-birthweight and preterm risk among Indigenous babies is removed might say something about the quality of medical care at the time of confinement. Although such reasoning has many proponents \[[@B11]-[@B13]\], risk factors such as smoking and infection are likely to also have at least a small effect on mortality that is independent of birthweight and gestational age. That is, adjustment of perinatal mortality rates by birthweight and gestational age is not a perfect way of assessing the quality of medical care at the time of confinement. It is similar to case-mix adjustments used in other settings to allow for differences in risk \[[@B14],[@B15]\]. Such adjustments are not expected to remove all confounding. Instead, the reasoning is that the adjusted rates, although not perfect, provide a useful way of identifying policy issues, setting agendas, and facilitating discussion. To be more specific, if a large proportion of excess deaths can be explained by the unfavourable birthweight and gestational age distribution of Indigenous babies, then that would suggest that priority should be given to implementing primary health care interventions to reduce the risk of low birthweight and preterm birth (eg, interventions to reduce maternal smoking or genitourinary infections). Conversely, if only a small proportion is explained by birthweight and gestational age, then other strategies might need to be considered such as improving access to high-quality hospital care around the time of confinement. Methods ======= Data were obtained from the population-based Queensland Perinatal Data Collection for the five years 1998 to 2002. This was the most recent five-year period for which complete data were available. The database includes information on all livebirths and stillbirths of at least 20 weeks gestation or 400 g birthweight. A perinatal death is defined as a stillbirth or the death of a liveborn baby within 28 days of birth. Our data set comprised 231,039 births and 2,255 deaths to non-Indigenous mothers and 13,920 births and 273 deaths to Indigenous mothers. Indigenous status is based on the self-reported Indigenous status of the mother and the gestational age is based on the best clinical estimate, which might be derived from the date of the last menstrual period, ultrasound in early pregnancy or maturity scoring of the neonate at birth. The method used is not recorded. In this paper we report the results of a Mantel-Haenszel procedure, which was used to determine the relationship between perinatal mortality and Indigenous status adjusted for the effect of low birthweight and preterm birth. The proportion and number of deaths that could be avoided were estimated by comparison of the crude and adjusted rates. We also used Poisson-regression models to adjust perinatal mortality rates for birthweight and gestational age. We variously fitted single week or four week categories of gestational age and 250 g and 500 g categories of birth weight. The results were the same as using the Mantel-Haenzel approach with broad categories of gestational age and birthweight, and we included these in preference to the Poisson-regression models for ease of interpretation. Results ======= The crude perinatal mortality rate among Indigenous babies was 19.6 per 1000 births, which was 2.01 (95%ci:1.77,2.28) times higher than the rate among non-Indigenous babies. When perinatal mortality rates were compared within each birthweight and gestational age strata, the point estimates suggested that Indigenous babies were only slightly more likely to die than non-Indigenous babies (Figure [1](#F1){ref-type="fig"}). The test for homogeneity of the rate ratios across strata was not significant (χ^2^(3) = 5.78, p = 0.1226) suggesting that the effect of Indigenous status is the same across the birthweight and gestational age strata (except for statistical noise) and that it is appropriate to use the adjusted combined estimate: M-H Adjusted RR = 1.13, 0.99--1.28. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Rate ratios^a^comparing Indigenous to non-Indigenous perinatal mortality stratified by preterm and birthweight status, 1998--2002. a) Rate ratios greater than 1.0 indicate higher mortality among Indigenous babies. b) Differences between stratum-specific rate ratios are not statistically significant (χ^2^(3) = 5.78, p = 0.1226). ::: ![](1743-8462-2-11-1) ::: These results suggest that if the population of Indigenous babies had the same birthweight and gestational age distribution as non-Indigenous babies, then the relative disparity would be reduced by 87% and there would be 20 fewer deaths of Indigenous babies per year. Discussion ========== Interpretation of results ------------------------- Perinatal mortality among Indigenous babies has remained twice that of their non-Indigenous counterparts for more than a decade. We found that most of this mortality excess is because Indigenous babies are at greater risk of being born too early and too small. In contrast, the case fatality rates of Indigenous babies who were born preterm or of low birth weight were similar to their non-Indigenous counterparts. Using a framework advocated by several perinatal epidemiologists \[[@B11]-[@B13]\], these results suggests that, broadly speaking, access to high quality care during confinement is adequate for Indigenous mothers and babies. That is, priority should initially be given to primary health care interventions to reduce the proportion of preterm and low birth weight babies. Policy implications ------------------- Risk factors for preterm birth and low birth weight include smoking, gentio-urinary tract infections, poor maternal nutrition and psycho-social stress \[[@B16]-[@B19]\]. Several studies have reported a higher prevalence of these risk factors among Indigenous compared with non-Indigenous mothers. More specifically, the prevalence of smoking among Indigenous women during pregnancy has been reported to be more than 60%, which is at least 3 times the prevalence for non-Indigenous women \[[@B9],[@B20],[@B21]\]. A recent Cochran review found that there are effective primary health care interventions to help and support women to stop smoking that lead to fewer preterm babies and better birthweights \[[@B22]\]. Further, we know that Indigenous women are more than two times as likely to have a urinary tract infection during pregnancy as non-Indigenous women \[[@B23]\]. In overseas studies, primary health care interventions to detect and treat asymptomatic bacteruria have been shown to decrease preterm birth by 40% \[[@B24]\]. In Australia, the best example we have of a primary health care initiative aimed at reducing risk factors among Indigenous mothers is the Strong Women Strong Babies Strong Culture program in the Northern Territory \[[@B25]\]. This program resulted in increased early attendance for antenatal care, reduced numbers of STDs and a reduced proportion of low birthweight babies \[[@B26]\]. Although such results are encouraging, if substantial progress is to be made across the whole of Australia, a properly funded national initiative is needed. Such an initiative would include funding to improve access to culturally appropriate primary health care during the antenatal period, which would deliver, inter alia, interventions for smoking cessation, screening and treatment of genito-urinary tract infections, screening for domestic violence, and programs aimed at reducing alcohol consumption and poor nutrition. It would not be a case of one strategy fits all. Instead local partnerships with possibly different types of service models would be needed to implement the national initiative. This approach will encourage creativity, innovation and risk taking, which will be essential ingredients to tackling a situation that has proved difficult to improve. Study limitations ----------------- Using vital statistics to set agendas has a long and continuing tradition in public health \[[@B27]\]. The advantages of such statistics are convenience, low cost and total enumeration. The disadvantages are insufficient and inaccurate data, which create uncertainty about the validity of the results \[[@B27]\]. This study used four variables: perinatal death, birthweight, gestational age, and Indigenous status. It is unlikely that an important number of perinatal deaths were missed because they are checked against notifications to the Registrar-General of Births, Deaths and Marriages. It is also unlikely that there are important errors in the measurement of birthweight. Consequently, the main areas of uncertainty are Indigenous status and gestational age. With regard to Indigenous status, some mothers may be reluctant to identify as Indigenous, others may be non-Indigenous with an Indigenous male partner, or midwives may not ask the mother or make an educated guess \[[@B28]\]. However, of all the types of mortality data, perinatal mortality provides the most accurate estimate of excess Indigenous mortality because the numerator (number of perinatal deaths) and denominator (number of births) for the rate can be obtained from the one data set. This is in contrast to adult death rates where identification of Indigenous people can be different in death registration data (the numerator for mortality rates) and population data (the denominator). This problem of the numerator not being appropriate for the denominator is not unique to comparisons of Indigenous and non-Indigenous Australians; it hinders interpretation of race-specific rates around the world \[[@B29],[@B30]\]. Thus, of all the routinely available mortality data, perinatal data provides the most valid estimate of the mortality excess for Indigenous people and provides robust support for policy discussions. For several reasons, gestational age is known to be less accurate among Indigenous than non-Indigenous babies \[[@B31]\]. Nevertheless, previous work in Queensland and elsewhere has shown that gestational age in combination with birthweight provides a better statistical adjustment of mortality rates than birthweight alone \[[@B32],[@B33]\]. We therefore considered it better to present birthweight and gestational age adjusted rates, rather than just birthweight-adjusted rates. Conclusion ========== Although perinatal mortality rates in Queensland have decreased over the last 16 years, the rates in Indigenous populations remain at least double those in the non-Indigenous population. Our analyses, stratified by birthweight and gestational age, suggest that the priority for reducing the excess mortality among Indigenous babies is primary health care to reduce the prevalence of risk factors during the antenatal period. A primary health care approach encompasses a much-needed component of an overall shift towards empowerment of Indigenous women and increased awareness and ownership of health which has the potential to play an important role in reducing the social inequality that has resulted in outcomes such as those found for perinatal mortality. Authors\' contributions ======================= TJ performed the statistical analysis and participated in drafting the manuscript. MC conceived of the study and its design and participated in drafting the manuscript. Both authors read and approved the final manuscript.

PubMed Central

2024-06-05T03:55:58.203197

2005-5-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168887/", "journal": "Aust New Zealand Health Policy. 2005 May 27; 2:11", "authors": [ { "first": "Trisha", "last": "Johnston" }, { "first": "Michael", "last": "Coory" } ] }

PMC1168888

Background ========== Periodic phenomena are widespread in biology, including, among others, membrane potential oscillations, cardiac rhythms, smooth muscle contraction, calcium oscillations, protoplasmic streaming, glycolytic oscillations, cAMP oscillations, oscillations in neuronal signals, insulin secretion (pancreas), gonadotropic hormone secretion, cell cycle, circadian rhythms, and ovarian cycle (see e.g. \[[@B1]\]). Consequently, there are numerous biological applications where periodicities must be detected from experimental biological data. Because the data measured from biological applications are inherently noisy and usually sparsely sampled, efficient algorithms are being developed to extract as much information as possible. In the past few years there has been an explosion of available microarray gene expression data. Detecting periodicity in gene expression is of great importance because it indicates, e.g., cell-cycle regulation \[[@B2]\] as well as the effect of circadian rhythms \[[@B3]\]. The significance of the detection of cell-cycle regulated processes is further emphasised by the linkage between cell-cycle and cancer (see, e.g., \[[@B4]\] and \[[@B5]\]). To this end, microarrays have been used to study the circadian gene expression in *Neurospora Crassa*\[[@B3]\] as well as cell-cycle regulated genes, e.g., in budding yeast \[[@B6]\], in fission yeast \[[@B7]\], and in human cells \[[@B5]\]. The task of finding periodicity in time series measured from a biological system can be viewed as a decision problem based on spectral analysis together with multiple hypothesis testing. A formal statistical testing procedure for the detection of periodic expression profiles was recently introduced by Wichert *et al.*\[[@B8]\]. It relies on the use of a so-called Fisher\'s *g*-statistic for which the exact null-distribution can be derived under the Gaussian noise assumption. Recently, a number of other methods for detecting periodic transcripts have also been proposed \[[@B9]-[@B13]\]. A major difference between the method proposed by Wichert *et al.*\[[@B8]\] and other methods is that Wichert\'s method is capable of detecting unknown frequencies whereas other methods are designed for detecting fixed frequencies. From a computational point of view, the problem of finding unknown frequencies is even more demanding since no prior knowledge of the frequency to be detected is available. In many biological applications it is more important to search for periodicities having an unknown frequency. However, in some applications, such as large-scale cell cycle studies, the period length is usually known and thus provides additional information for testing. In this paper, our goal is to tackle both of the two problems. That is, we develop two methods, one for the detection of unknown frequencies and the other for testing fixed frequencies. In many applications, including those arising from bioinformatics, the exact noise characteristics are usually unknown and can be remarkably non-Gaussian. Furthermore, the observed gene expression time series can exhibit other non-idealities, such as outliers, short length and distortion from the original wave form. Therefore, it would be useful to have a robust method for detecting periodic components, i.e., a method that also works well when the original (Gaussian) noise assumption no longer holds. A robust, rank-based, non-parametric spectral estimator was recently introduced in \[[@B14]\]. In this paper, we extend the approach of \[[@B14]\] to the detection of periodic time series. This results in a robust testing procedure which is insensitive to a heavy contamination of outliers, missing-values, short time series, nonlinear distortions, and is completely insensitive to any monotone nonlinear distortions. We also consider a permutation-based alternative to the method proposed in \[[@B8]\] and show that, when the data is contaminated with the above mentioned non-idealities, this results in a more robust method. As discussed, e.g., in \[[@B15]\], the performance of a method can be proven using either extensive simulations, analytical proofs or multiple plasmode experiments (to use a term from \[[@B15]\]). Plasmode is a real data structure whose true structure is known. Although very useful, the kinds of benchmarkings proposed in the literature (related to periodically behaving gene expression time series) so far do not belong to any of the above categories. In particular, the proposed benchmarking frameworks cannot be considered as plasmodes since the true structure (periodic genes) is not known but, instead, is based on partial biological knowledge (or other measurements, such as protein-DNA binding). Hence, the performance cannot be assessed solely on real data. Therefore, as the analytical proofs are hard to obtain in this case, we perform extensive simulations to show the superior performance of the proposed methods. However, we also apply the proposed method to real experimental data to show that the methods perform well on real data and that the results are biologically meaningful. Results ======= Computational methods --------------------- In order to be consistent with the previously published methods, we use similar notation as in \[[@B8]\] and also consider the same model for the periodic time series *y*~*n*~= *β*cos(*ωn*+ *φ*) + *ε*~*n*~, (1) where *β*≥ 0, *ω*∈ (0, *π*), *n*= 1,\..., *N*, *φ*∈ (-*π*, *π*\], and *ε*~*n*~is an i.i.d. noise sequence. To test for the periodicity, define the null hypothesis as *H*~0~: *β*= 0, i.e., time series consists of the noise sequence alone, *y*~*n*~= *ε*~*n*~. In the following we develop a method for detecting unknown frequencies and later introduce a modification which can be applied to the detection of known frequencies. We first review the Fisher\'s test for the detection of periodic transcripts as introduced by Wichert *et al.*in \[[@B8]\]. ### Wichert\'s method The method proposed by Wichert *et al.*\[[@B8]\] is based on the periodogram spectral estimator, defined as ![](1471-2105-6-117-i1.gif) where *N*is the time series length. The periodogram is further evaluated at (harmonic) normalised frequencies ![](1471-2105-6-117-i2.gif) where *a*= \[(*N*-- 1)/2\] and \[*x*\] denotes the integer part of *x*. To test for the periodicity formally, some kind of a test statistic must be chosen. The so-called *g*-statistic for one time series is given by ![](1471-2105-6-117-i3.gif) In plain words, the *g*-statistic is the maximum periodogram ordinate divided by the sum of all periodogram ordinates for *l*= 1, \..., *a*. Large value of *g*indicates a strong periodic component and leads to the rejection of the null hypothesis. Wichert *et al.*\[[@B8]\] resort to a result by Fisher that, under the Gaussian noise assumption, gives the exact distribution of the *g*-statistic under the null hypothesis (see, e.g., \[[@B16],[@B8]\]). The exact *p*-value for a realisation of the *g*-statistic is shown to be ![](1471-2105-6-117-i4.gif) where *b*is the largest integer less than 1/*x*and *x*is the observed value of the *g*-statistic. Because there are usually thousands of time series that are tested simultaneously, whether they exhibit periodicity or not, there is a possibility that a time series can have a small *p*-value by chance. To correct the *p*-values for multiple testing, Wichert *et al.*\[[@B8]\] use the method of Benjamini and Hochberg (see, e.g., \[[@B17]\]), which controls the False Discovery Rate (FDR). The FDR method controls the expected proportion of false positives (Type I errors) at a given rate *q*. The threshold of the FDR depends on the evaluated p-values. The FDR procedure for the ordered set of *p*-values *p*~(1)~, *p*~(2)~,\...,*p*~(*M*)~, where *M*is the number of time series, is as follows 1\. Let *i*~*q*~be the largest *i*for which ![](1471-2105-6-117-i5.gif) 2\. Reject the null hypothesis for the time series corresponding to the *p*-values *p*~(1)~, *p*~(2)~,\..., ![](1471-2105-6-117-i6.gif) The procedure based on the periodogram spectral estimator for periodicity detection has several well-known and important properties. For example, if the noise sequence ∈~*n*~is i.i.d. Gaussian and the true underlying frequencies are among the harmonic frequencies Ω = {*ω*~*l*~: *l*= 0,\..., \[(*N*- 1)/2\]}, then the (square root of the) periodogram is the minimum variance unbiased estimator of the frequency content at discrete frequencies Ω (see, e.g., \[[@B18]\]). However, it is also widely recognized that the standard periodogram is an inconsistent spectrum estimator (see, e.g. \[[@B19]\]). Despite this weakness, the periodogram is a standard method that is theoretically well-founded. Under the popular Gaussian working assumption, widely invoked in spectral estimation, the distributional characteristics of the periodogram are known and useful. Indeed, it is this distributional characterisation that forms the basis for Fisher\'s *g*-test for whiteness. In other words, Equation (5) provides the exact significance value for a realisation of the *g*-statistic. This provides a solid theoretical basis for the use of the method proposed in \[[@B8]\]. Concerning the problem of detecting hidden periodicities, some generalisations and improvements over the traditional methods have been proposed. A recent review of the proposed methods can be found in \[[@B20]\]. Many of the improved methods are some type of generalisations of the traditional methods, such as the correlogram or periodogram. Artis *et al.*\[[@B20]\] report that two particular methods are generally found to have a good performance: so-called mixed spectrum methods by Priestley and Bhansali (see, e.g. \[[@B19],[@B20]\]), and a modified method based on the maximum periodogram ordinate by Chiu \[[@B21]\]. The method by Priestley and Bhansali uses a certain type of windowing of the correlogram for the \"smoothing\" purposes (i.e., in order to reduce the variance). The method by Chiu in turn modifies the *g*-statistic by replacing the average spectrum in the denominator with a proper trimmed mean of the ranked periodogram ordinates. As we explain in the Discussion Section, the same modifications/improvements can be used in the proposed robust method (to be introduced shortly) as well. Indeed, the windowing can be utilized much in the same way as it is used in the method by Priestley and Bhansali. Similarly, the modification(s) of Chiu can be directly implemented in the case of the proposed robust estimator. Fisher\'s test and the proposed robust method can be considered to be analogous in the sense that neither of them incorporates any modifications for the traditional periodogram/correlogram approach. In Simulations Section we perform an extensive comparison of the unmodified, standard methods since this provides a better insight into the performance differences of the traditional methods and the proposed robust rank-based methodology. Furthermore, from an extensive simulation point of view, comparison of many different modifications together with the robust method would also result in a rather large number of simulations, especially because it is possible to use many different combinations of the modifications. Note, however, that the further modifications, such as the ones proposed by Priestly, Bhansali, and Chiu, can be implemented in both of the frameworks (the traditional and the robust rank-based) to gain further performance improvements. We discuss this issue further in Discussion Section. In many applications, particularly in bioinformatics, the noise distributions are usually unknown and can remarkably deviate from the Gaussian assumption. The methods based on Gaussian noise assumption may fail, or even produce invalid results, when the model assumptions do not hold. The goal of this paper is to introduce an alternative for the standard methods, aimed at providing a robust estimator. ### A robust alternative As a starting point we should like to remind that the periodogram *I*(*ω*) is equal equivalent to the correlogram spectral estimator (see, e.g., \[[@B25]\]) ![](1471-2105-6-117-i7.gif) where ![](1471-2105-6-117-i8.gif) is the biased estimator of the autocorrelation function ![](1471-2105-6-117-i9.gif) Note that the required values for ![](1471-2105-6-117-i8.gif) for *m*\< 0 are obtained by invoking the inherent symmetry of the autocorrelation function: *r*(-*m*) = *r*(*m*). Consequently, the *g*-statistic in Equation (4) as well as the corresponding significance value in Equation (5) can also be computed using *S*(*ω*) instead of *I*(*ω*). It is natural to try to obtain robustness by replacing ![](1471-2105-6-117-i8.gif) with a robust alternative. Before continuing to the robust method, it is important to note that, especially in the case of gene expression time series, the data is often contaminated with missing values. Therefore, the spectral estimation method must take missing samples into account. Let *I*~*m*~be the set of time indices *k*for which both *y*~*k*~and *y*~*k*+*m*~are available and *K*~*m*~= \|*I*~*m*~\|. As long as *K*~*m*~≠ 0, a missing data-adapted version of the unbiased estimate of the autocorrelation can be obtained as ![](1471-2105-6-117-i10.gif) We cover only the versions adapted to missing data in the following text since they are equal to the standard estimators in case of complete data sets. Note that *K*~*m*~equals *N*- *m*for the complete data sets. Next we consider a recently introduced rank-based autocorrelation estimator \[[@B14]\] for the problems of spectrum estimation. This estimator is a moving-window extension of the Spearman rank correlation coefficient, quantifying the association between the sequences {*y*~*k*~} and {*y*~*k*+*m*~}. The resulting quantity, *ρ*^*S*^(*m*) is actually an alternative estimator of the standard correlation coefficient *ρ*(*m*) between these sequences (see e.g. \[[@B19]\]) ![](1471-2105-6-117-i11.gif) where ![](1471-2105-6-117-i12.gif)\[·\] denotes the expectation operator and *μ*~*yi*~= ![](1471-2105-6-117-i12.gif) \[*y*~*i*~\] is the mean of the sequence. Recall that the sample correlation coefficient between two *N*length sequences {*x*~*i*~} and {y~*i*~} is defined as ![](1471-2105-6-117-i13.gif) where ![](1471-2105-6-117-i14.gif) denotes the sample mean of {*x*~*i*~}. Under the assumption of stationarity, it immediately follows from Equation (9) that the correlation coefficient *ρ*(*m*) is related to the autocorrelation function *r*(*m*) by ![](1471-2105-6-117-i15.gif), where ![](1471-2105-6-117-i16.gif) is the variance of the sequence. Since it is important to remove the mean of the sequence prior to spectrum estimation to avoid low frequency artifacts and since ![](1471-2105-6-117-i17.gif) is simply a scale factor, the problem of detecting periodic components in a data sequence may equally well be based on *ρ*(*m*) as *r*(*m*). Consequently, we consider spectral estimators of the form ![](1471-2105-6-117-i18.gif) where ![](1471-2105-6-117-i19.gif) estimates the correlation coefficient between {*y*~*k*~} and {*y*~*k*+*m*~} and *L*is the maximum lag for which the correlation coefficient is computed. More specifically, we consider the correlation coefficient between the data ranks *R*~*y*~(*i*) and ![](1471-2105-6-117-i20.gif), defined by ![](1471-2105-6-117-i21.gif) where *C*is a normalisation factor, *R*~*y*~(*i*) denotes the rank of *y*~*i*~in the set *S*= {*y*~*j*~: *j*∈ *I*~*m*~} and ![](1471-2105-6-117-i20.gif) denotes the rank of *y*~*i*+*m*~in the set *S\'*= {*y*~*j*+*m*~: *j*∈ *I*~*m*~}. By selecting either *C*= *K*~*m*~or *C*= *N*in Equation (12) yields the unbiased or the biased estimate of the correlation coefficient between the rank sequences, respectively. Because both of these rank sequences assume every value from 1 to *K*~*m*~precisely once, their average is (*K*~*m*~+ 1)/2, independent of the data values {*y*~*n*~}, and their sample variance, when scaled by the sequence length *K*~*m*~, can be shown to be (![](1471-2105-6-117-i22.gif) - 1)/12. More generally, if *C*= *K*~*m*~, and since *ρ*^*S*^(*m*) is the correlation coefficients between ranks, it is bounded by -1 ≤ *ρ*^*S*^(*m*) ≥ 1 for all *m*. In the following we shall use *L*= *N*- 2 in Equation (11) because *N*- 2 is the largest lag for which Equation (12) can be computed. We shall use the biased estimate, i.e., *C*= *N*, here because of its connection to the equivalence between *I*(*ω*) and *S*(*ω*). Moreover, the use of the biased estimate in spectrum estimation (Equations (6) and (7)) can be interpreted as triangular weighting of the autocorrelation function estimate. Windowing is usually applied to reduce the scalloping loss effect which is the reason why some frequencies are inferior to others \[[@B25]\]. As in the case of the standard autocorrelation estimate, the values for *ρ*^*S*^(*m*) for *m*\< 0 are obtained by symmetry: *ρ*^*S*^(-*m*) = *ρ*^*S*^(*m*). This also helps in computing Equation (11) since it reduces to ![](1471-2105-6-117-i23.gif) where ![](1471-2105-6-117-i32.gif)(*x*) denotes the real part of *x*. As opposed to standard periodogram or the corresponding correlogram, the proposed robust spectral estimator is not guaranteed to be non-negative. We shall hence use the absolute value of ![](1471-2105-6-117-i24.gif) in the following. ### Significance values In the same way as Wichert *et al.*\[[@B8]\] do, we use the *g*-statistic and evaluate ![](1471-2105-6-117-i25.gif) for each time series spectral estimate. However, we do not have the luxury of resorting to an exact distribution of the *g*-statistic, e.g., under the Gaussian noise assumption. To obtain the significance values we consider two common ways of computing them: simulation and permutation-based methods. This also opens up a possibility for adjusting some parameters in the proposed robust method that were previously kept fixed. In particular, we vary the number of equidistant frequencies at which the spectral estimate is evaluated and change Equation (14) accordingly by incorporating more terms in the max-operator as well as in the sum in the denominator. Instead of having a fixed set of *a*+ 1 normalised frequencies as in Equation (3), we can evaluate the spectral estimate at \[(*K*- 1)/2\] + 1 equidistant frequencies ![](1471-2105-6-117-i26.gif) Although the method is rather insensitive to the selection of *K*, we found that *K*= 2*N*generally provides a good performance. Evaluating ![](1471-2105-6-117-i24.gif) at more frequencies can be viewed as a smoothing or interpolation of the original discrete spectral estimate. From the implementation point of view it is worth mentioning that Equation (13) evaluated at frequencies shown in Equation (15) can be computed using the fast Fourier transform (FFT). As was already discussed in the Background Section, in some cases one might be interested in testing fixed instead of unknown frequencies. The proposed method can naturally be adapted to that case as well. If *ω*\' is the known frequency for which the spectral content is to be tested, then a modified *g*-statistic, *g*\', can be used ![](1471-2105-6-117-i27.gif) In the following, we mainly concentrate on the use of the standard *g*-statistic for detecting unknown frequencies. However, the same methods, such as simulation and permutation based significance values, can also be applied to the modified *g*-statistic. In Experimental results Section we apply both the standard and the modified *g*-statistics to real microarray data. When the noise satisfies the i.i.d. assumption, the decision between simulation and permutation test-based significance values is facilitated by the following observation. A statistic *T*is said to be distribution-free over a collection of distributions ![](1471-2105-6-117-i28.gif) if the distribution of *T*is the same for every joint distribution in ![](1471-2105-6-117-i28.gif). Consider the signal model shown in Equation (1) under the null hypothesis *H*~0~: *β*= 0, i.e., *y*~*n*~= *ε*~*n*~, and assume that *ε*~*n*~is again i.i.d. and has continuous distribution. It is easy to see from Equation (12) that, for any time series {*y*~*n*~}~*n*=\ 1,\...,*N*~, the *g*-statistic depends on {*y*~*n*~}~*n*=\ 1,\...,*N*~only through the rank sequence {*R*~*y*~(*i*)}~*i*=\ 1,\...,*N*~, where *R*~*y*~(*i*) denotes the rank of *y*~*i*~in the original sequence. This implies that the *g*-statistic is distribution-free over the class of all joint distributions of *N*i.i.d. continuous univariate random variables (see, e.g., \[[@B22]\]). In other words, for each *N*and independent of the type of the noise, the *g*-statistic has exactly the same null distribution as long as the noise term satisfies the continuous i.i.d. assumption. Therefore we can choose to use either the simulated distributions (must be simulated separately for each times series sequence length) or the permutation tests, depending on the circumstances. ### Simulation-based significance values The simulation-based method is simple. Given the model as in Equation (1) together with some distributional assumptions for *ε*~*n*~, generate a set of *P*random time series under the null hypothesis. Evaluate the test statistic shown in Equation (14) on each of the *P*time series. Use the obtained *g*-values to compute an estimate of the distribution of the *g*-statistic under the null hypothesis. The distribution can be estimated, e.g., using kernel density estimation methods. The testing can then be performed as explained above except that the significance values are computed/integrated relative to the estimated distribution. Note that the null distribution must be estimated for each time series length separately but, due to the distribution-free property, the null distribution is independent of the noise characteristics under the i.i.d. assumption. ### Permutation-based significance values A more flexible way of obtaining *p*-values is to use permutation tests \[[@B23]\]. Although they are a relatively old concept, permutation tests have only recently become interesting in practise because of the intensity of needed computing power. The idea is simple: 1\. Choose a test statistic. 2\. Evaluate the test statistic on the original data. 3\. Randomly permute the data and evaluate the test statistic on every permutation. 4\. Estimate the distribution of the test statistic with the help of the sample generated in point 3. 5\. Use the estimated distribution to get a *p*-value for the original test statistic computed in point 2. A sequence of random variables {*X*~*n*~}, *n*= 1, 2,\...,*N*is exchangeable, if the joint distribution of ![](1471-2105-6-117-i29.gif) is the same as that of the original sequence *X*~1~, *X*~2~,\..., *X*~*N*~for all permutations *π*. Under the null hypothesis, the elements of the time series *y*~*n*~are i.i.d. and therefore exchangeable, and hence the permutation test can be applied. Alternatively, as the application of a random permutation destroys any periodic structure that is present in the original sequence, permutation tests can be used to assess how highly structured the given time point values are in the light of the chosen test statistic versus other permutations of the given sample. As the concept of permutation tests is non-parametric, they can be applied without knowing the exact distribution of the data at hand. Instead of performing all the *N*! permutations for each time series, we have chosen to permute each of the original time series for *P*= 5000 times. As our simulations show, this seems to be quite an adequate number of iterations. The selection of *P*is always a compromise, because too high *P*makes computations too slow and too low *P*weakens the accuracy and resolution of the calculated *p*-values. For example, time series having a very periodic structure can get a *p*-values of zero due to the low value of *P*. While we have mainly applied the permutation tests to the robust estimator, it must be noted that with the help of permutation tests the robustness of the periodogram can also be improved. As we show in Results section that if we add, e.g., some impulsive noise to the simulated data, the results when using the periodogram method as in Equation (5) are not as good as when we use permutation tests to find the *p*-values. ### Correction for multiple testing In order to facilitate the comparison between the proposed and previous methods, the obtained *p*-values are corrected exactly in the same way as in the method by \[[@B8]\]. Simulations ----------- We put the presented methods to a test by first going through simulated data, where the ground truth is known, and then by finding periodically behaving genes in real microarray data. In simulations, we use exactly the same test signal model as in \[[@B8]\] for comparison purposes, namely Equation (1) with *β*= ![](1471-2105-6-117-i30.gif) and *φ*= -*π*/4, i.e., ![](1471-2105-6-117-i31.gif) where *n*= 1,\...,*N*, *ω*is uniformly randomly chosen in the interval *ω*∈ \[0.05*π*, 0.45*π*\] (we wanted to avoid frequencies near zero and the Nyquist frequency) and *ε*~*n*~is an i.i.d. noise sequence. An essential parameter is the amplitude (*β*= ![](1471-2105-6-117-i30.gif)) which affects the signal-to-noise ratio and which we would like to have the same as in \[[@B8]\]. We chose to consider three types of non-idealities, namely (*i*) pure standard Gaussian noise (zero mean and unit variance, see Figure [1(a)](#F1){ref-type="fig"}), (*ii*) standard Gaussian and impulsive noise (number of impulses equals ten percent of the sequence length, amplitude ± 6 times the standard deviation of the Gaussian noise, see Figure [1(b)](#F1){ref-type="fig"}), and (*iii*) standard Gaussian noise and *x*^3^distortion, where all values were raised to the power of three after adding the noise (see Figure [1(c)](#F1){ref-type="fig"}). In each example sequence in Figure [1](#F1){ref-type="fig"}, the normalised frequency of the original sinusoidal is *ω*= 0.1. Figures [2(a)--(c)](#F2){ref-type="fig"} show the spectral estimates for the time series in Figures [1(a)--(c)](#F1){ref-type="fig"}, respectively, using both the standard periodogram and the proposed robust method. Note that the spectra have been scaled for viewing purposes. Figures [2(b)--(c)](#F2){ref-type="fig"} already illustrate a remarkable difference between the two methods. For more details about the performance of the proposed robust method as a spectrum estimator, see \[[@B14]\]. A detailed comparison of the periodicity detection capabilities is performed next. Let us first examine the power of the test, i.e., one minus the probability of the type II error (false negative). The power of the test is estimated for the three different test cases as well as for different time series lengths and for different noise parameters using 10000 Monte Carlo runs, see Figure [3](#F3){ref-type="fig"}. The significance level is set to *α*= 0.05. In all the three cases, the case-specific noise assumptions are used for both the null hypothesis (*β*= 0) and the alternative hypothesis (*β*\> 0). In this simulation, we use the signal model shown in Equation (17) to represent a periodic signal (i.e., the alternative hypothesis). In the right column of Figure [3](#F3){ref-type="fig"}, the length of the time series is set to 40 and the power is shown as the function of varying noise parameters. Figure [3](#F3){ref-type="fig"} clearly shows that the power of the proposed robust hypothesis testing method is remarkably better than that of the Fisher\'s test, especially in the case of outliers and non-linear distortion. More interestingly, however, the proposed method is also more powerful in the case of standard Gaussian noise. Next we consider another simulation. In the same way as in \[[@B8]\], two thousand time series of length *N*= 10, 20, 40, 45, 50 and 100 were generated to test the periodicity detection. One thousand and nine hundred of the time series were plain noise and one hundred time series were generated according to Equation (17). We again consider the three aforementioned noise models. As explained in the Computational Methods Subsection, we evaluated the *g*-statistic and *p*-value for each time series and then used the FDR rule to determine which of the time series were considered to be cyclic for a certain FDR level. The FDR level, at which the expected rate of false positives is controlled, was chosen similarly as in \[[@B8]\], i.e., *q =*0.15, 0.10, 0.05, 0.01 and 0.005. For each *N*and *q*the simulation was run for 99 times for the simulation-based cases and 9 times for the permutation-based cases. Median statistics are reported for the number of found periodic components, the number of correctly identified periodic components (shown in parenthesis) and the number of truly periodic time series among the top 100 ranked sequences (*Z*). If we take a look at the results in Tables [1](#T1){ref-type="table"} to 9 we can draw some immediate conclusions. First, when the noise is plain Gaussian, Tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"} show that both methods perform approximately equally well. There are no significant differences between the two methods in terms of the number of detected genes or in terms of the number of correctly detected genes. However, the numbers of truly periodic genes among the top 100 ranked sequences (*Z*-scores) show somewhat favorable performance for the robust method, especially for the short time series *N*= 20 and *N*= 40. Indeed, this observation agrees with previous findings \[[@B14]\] where the robust method was found to have a good performance as a spectrum estimator for short time series. By comparing Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"} and Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}, it is obvious that the permutation tests do not provide any significant performance gain over the traditional approach where the significance values are computed using the simulation-based method or Equation (5), respectively. In both cases, the *Z*-scores are about the same, as expected. The only notable difference is seen in the number of found periodic genes for short time series (e.g., *N*= 40, 45, 50) and small FDR levels (*q =*0.005, 0.01, 0.05) where the numbers are slightly higher when permutation tests are used. This suggests that the permutation-based method finds a bit smaller *p*-values than the simulation-based method. Tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}, [6](#T6){ref-type="table"}, [7](#T7){ref-type="table"}, [8](#T8){ref-type="table"} clearly show the superior robustness of the proposed method over the traditional Gaussian analysis. As can be seen from Tables [1](#T1){ref-type="table"} and [5](#T5){ref-type="table"} and Tables [2](#T2){ref-type="table"} and [6](#T6){ref-type="table"}, there is only a minor performance degradation between the Gaussian case and the combined Gaussian and impulsive case. On the other hand, Tables [3](#T3){ref-type="table"} and [7](#T7){ref-type="table"} and Tables [4](#T4){ref-type="table"} and [8](#T8){ref-type="table"} clearly indicate the sensitivity of the periodogram method to fluctuations from the original Gaussian noise assumption. As discussed above, Tables [5](#T5){ref-type="table"} and [6](#T6){ref-type="table"} show that, in the case of the robust method, permutation-based significance value computation performs approximately equally well as the simulation-based computation. The only notable difference is again seen in the number of found periodic genes for short time series (e.g., *N*= 40, 45, 50) and small FDR levels (*q =*0.005, 0.01, 0.05). Tables [7](#T7){ref-type="table"} and [8](#T8){ref-type="table"} in turn show that, apart from the *Z*-scores, the permutation-based method mitigates the sensitivity of periodogram method to the fluctuations from the model (Gaussian) distribution. The robustness of the proposed method is further demonstrated by its response to combined Gaussian noise and nonlinear cubic distortion. As explained in Computational Methods Section, the robust method depends on the observed time series only through the rank sequence. Any monotone distortion preserves the ordering of the samples. Therefore, the rank-based method is completely insensitive to any monotone distortions. Consequently, the results for the third test case are identical to those presented in Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}. The results for the periodogram method are shown in Table [9](#T9){ref-type="table"}. Experimental results -------------------- The data sets that are considered here are from the following papers: \[[@B6],[@B5],[@B7]\]. For each time series experiment (13 in total), we apply the proposed robust methods for detecting genes having both fixed and unknown frequency components. For the fixed frequency we use the one that corresponds to the length of the cell cycle. Following the idea presented in \[[@B8]\], a simple method for estimating the cell cycle length/frequency is to compute the average robust spectral estimate. For each time series, we present the number of statistically significant genes that are found to be periodically behaving at a specific level of the FDR (*q =*0.05). For the cdc15 experiment by Spellman *et al.*\[[@B6]\], the sampling time was not equidistant in the beginning and at the end of the data set. Considering the missing time points as missing values would result in a large number of missing values with a regular pattern of occurrence. Although the proposed robust methods can cope with missing values, such a regular pattern of missing values can artificially cause many small significance values and hence result in an unreliably large number of statistically significant periodic genes. This can be avoided e.g. by interpolating the expression values for the systematically missing time points, in which we used simple linear interpolation. Only non-missing expression values are considered in the interpolation. If the expression values of both the previous and the next time instants are missing, then the interpolated sample is defined to be missing as well. This results in a more conservative number of detected genes for the cdc15 experiment. We chose to consider only those genes that have less than 30% missing values and decided to rule out all except the Score3 experiment in the data by Whitfield *et al.*\[[@B5]\] because of high degree of irregular sampling and short time series length. The obtained results are shown in Table [10](#T10){ref-type="table"}. The total number of genes analysed in each data set is denoted as *M*. The number of found periodic genes having fixed and unknown frequency are denoted as *P\'*and *P*, respectively. The corresponding figures from \[[@B8]\] are shown in parentheses. For the detection of periodic components having unknown frequency, we used the permutation-based method. As was shown in Simulations Section, both the simulation and permutation based approaches performed approximately equally well. Hence, for the ease of implementation, we used the simulation-based method for the detection of periodic components having a fixed frequency. If we first take a look at the numbers of detected genes having a periodic component of an unknown frequency (*P*) shown in Table [10](#T10){ref-type="table"}, we can see that generally the numbers of periodically behaving genes are lower than those of the corresponding figures in \[[@B8]\]. In other words, the proposed robust method seems to provide more conservative estimates, although the cdc15 experiments shows an exception. In the case of real data further comparison between the two methods is much more subjective than in simulation experiments since the ground truth is not completely known. Based on the simulation results shown above, one could put more faith on the robust method, especially in the cases where the Gaussian noise assumption is violated. Hence one could argue that the robust method has ignored more non-periodic time series, particularly ones where outliers and other non-idealities have caused artificial variation in the periodogram. The number of detected truly periodic genes can be increased, at the cost of detecting more false positives, by using a higher level of *q*. Let us then focus on the numbers of found periodic genes when using a fixed frequency in the robust method (*P*\'). Concerning the numbers of detected periodic genes in the data sets by Spellman *et al.*\[[@B6]\], the results are in concordance with the previously published ones \[[@B24]\]. On the cdc28 data set, the proposed method finds a slightly higher number of periodic genes. Direct comparison between the numbers *P*and *P\'*is not meaningful as the number of detected genes depends on the significance level, which may be dependent on the used method. The comparison is better done using the ordered gene lists which we will discuss shortly together with three benchmark gene sets (see below). We have not yet come across with another study, besides the original paper, that would have examined the periodicity in the data by Rustici *et al.*\[[@B7]\]. Rustici *et al.*investigate the global cell cycle control of gene expression in the fission yeast *Schizosaccharomyces pombe*using DNA microarrays. Thus, the comparison of values in Table [10](#T10){ref-type="table"}, related to Rustici *et al.*data, is not feasible. Table [10](#T10){ref-type="table"} shows that the number of detected periodic genes ranges from 673 up to 3131. In the case of very large number of detected periodic genes, more insight into the data set can be gained by looking at the ordered list of genes. Table [10](#T10){ref-type="table"} only shows the number of detected genes. Further insight can be gained by looking at the enrichment of the genes assumed to be cell cycle regulated among the top ranked genes. In particular, we resort to the three different benchmark gene sets introduced in \[[@B24]\]. In order to provide a direct comparison with the results shown in \[[@B24]\], we show similar enrichment graphs for the both robust detection methods (fixed and unknown frequency) in Figure [4](#F4){ref-type="fig"}. Note that some of the benchmark genes are ignored during the analysis since they have more than 30% missing values. There are also a few benchmark genes for which no exact match was found among the genes in the data provided by \[[@B6]\]. Hence the graphs are drawn based on slightly smaller benchmark gene sets, namely, 101--112 (B1), 309--339 (B2), and 465--505 (B3), depending on the experiment. Figure [4](#F4){ref-type="fig"} reveals some interesting results. First, let us compare how well the proposed robust method (with the fixed frequency) finds the benchmark genes when compared to previously published methods (for comparison, see the corresponding graphs in \[[@B24]\]). The results for the benchmark gene set B1 are shown on the first row in Figure [4](#F4){ref-type="fig"}. For all the data sets, the performance of the robust method is either between the best methods and the amplitude-independent method by Zhao *et al.*\[[@B9]\], or close to the method by Zhao *et al.*\[[@B9]\]. This finding is not surprising. As discussed in \[[@B24]\], this benchmark gene set is biased towards periodic genes which are strongly regulated, i.e., have large amplitudes. In general, because high-amplitude periodic genes are more easily detected from noisy expression data, the gene sets identified from such studies are likely to be biased towards high amplitude genes. An advantage of amplitude-independent methods is, however, that they detect small-amplitude periodic transcripts better, and hence may identify genes which are not yet known to be periodic. For the benchmark gene set B2, the performance of the proposed robust method is approximately the same as that of the best methods reported in \[[@B24]\], except for the alpha experiments on which the robust method performs slightly worse. Noteworthy is that the benchmark gene set B2 is obtained from a separate Chromatin IP experiment and thus is independent of the previous gene expression studies. Concerning the benchmark gene set B3, the robust method performs better than the majority of the methods. Notably good performance is seen on the data from the Cdcl5 and the Cdc28 experiments. Interestingly, the benchmark set B3 is also likely to be biased, but towards small amplitude genes \[[@B24]\]. This strengthens the assumption that the potential of the proposed robust method is especially in detecting unknown, small-amplitude, periodic genes. Yet another interesting observation can be drawn by comparing the solid and dashed curves in Figure [4](#F4){ref-type="fig"}. As can be expected, the method which detects especially cell cycle frequencies ranks the benchmark genes higher than the method which detects unknown (all) frequencies, i.e., the solid line is above the dashed line. Another expected behavior is that the method which detects unknown frequencies also detected a great number of genes assumed to be related to the cell cycle. However, from another point of view, Figure [4](#F4){ref-type="fig"} also indicates that there are some statistically significant periodic patterns which are more significant than some of the cell cycle related ones. Possible sources of those significant periodic patterns may include, among others, systematic artifacts in the array/experiment preparation, unknown periodic biological processes, or simply the considerable amount of experimental noise (false positives). The top 300 ranked genes for all the data sets analysed, obtained using the proposed robust method, are provided on our companion website. Discussion ========== As discussed above, some extensions and improvements over the traditional periodogram/correlogram approach have been proposed in the literature. Two particular modifications, namely utilisation of windowing and a trimmed *g*-statistic, were reported to provide a good performance in a recent review by Artis *et al.*\[[@B20]\]. Although we provided an extensive comparison of the unmodified traditional and the (unmodified) proposed rank-based methodologies, the further modifications can be implemented in a straightforward fashion in both of the frameworks. These extensions for the robust rank-based method will be examined in future studies but let us give an overview of the possible modifications. As discussed previously, the biased version of the robust correlation estimator can be viewed as a type of weighting or windowing. More generally the windowing is typically incorporated into the computation of the spectral estimate (see, e.g., Equations (6) and (11)). Different windows provide different properties for spectral estimators. For example, the shape of the window can be used to control the smearing and leakage effects whereas the length of the window compromises with the spectral resolution and the variance \[[@B25]\]. In general, the used windows can be chosen from a general class of windows, including, among others, Bartlett, Daniel, and Parzen windows (see, e.g. \[[@B16],[@B18],[@B19]\]). Concerning the detection of hidden periodicities, windowing can be applied much in the same way as it is used in the method by Priestley. Similarly, the modification by Chiu \[[@B21]\], i.e., the use of a proper trimmed mean of the ranked periodogram ordinates in place of the average periodogram, can be applied to the robust rank-based estimator as well. A drawback associated with the use of the trimmed *g*-statistic in the traditional periodogram setting is that only asymptotic distribution of the test statistic is available. The discrepancy between the true distribution and the asymptotic one can be remarkable in the case of small sample size typical e.g. in gene expression studies. These difficulties can be circumvented by the computer intensive simulation and permutation-based methods explained above. The proposed method has other possible extensions as well. As a periodically behaving gene may be involved in several different biological processes, its expression pattern may contain several dominant frequencies. In that regard, the testing procedure can be extended to detect several frequency peaks from the spectral estimate. See, e.g., \[[@B19],[@B21]\] for extensions of Fisher\'s test to that direction. In cell cycle related studies, a cell population is usually forced to synchrony prior to taking the measurements using an external synchronisation method. The synchronisation is achieved by arresting the cells at a specific phase of the cell cycle after which they are released. However, as time evolves, the cell population gradually loses its synchrony. Such a phenomenon can be viewed as time-varying (low-pass) filtering of the expression values where the time-varying filter kernel corresponds to the distribution of the cell population over the cell cycle. Inverse methods have been developed to correct for the effect of the loss of synchrony \[[@B26],[@B27]\]. Several interesting questions remain to be studied. First, the inverse filtering problem as such is fairly sensitive to noise and is further complicated by the fact that the accuracy level at which the filter kernel (i.e., distribution of the cell population) can be measured is limited. Therefore, the corrected time series may contain even more obscure non-idealities than the uncorrected ones. Consequently, robust methods are potentially even more important when periodic components are sought from the time series which are corrected for the loss of synchrony. Future studies are also needed to compare the robust periodicity detection method, when applied to both uncorrected and corrected time series, to see whether the inversion of the loss of synchrony brings any additional gain in the case of robust periodicity detection. In addition to the simulation results presented in the Simulations Section, we also performed preliminary simulations where the amplitude of the periodic signal was attenuated to model the loss of synchronisation. We noticed that if the average amplitude of the sinusoidal signal remained the same, the results were similar to those in the tables of the Simulations Section. Further comparisons must also be made to assess the performance differences between the proposed method (possibly combined with a proper inversion method for the loss of synchrony) and alternative methods in which a model for the loss of synchrony is incorporated into the statistical testing framework \[[@B12],[@B11]\]. Although elegant, such combined approaches have potential difficulties in that they usually result in a computationally intensive optimisation problem \[[@B11]\] and/or include several distributional assumptions \[[@B12]\]. Furthermore, the inversion of the loss of synchrony is performed blindly, i.e., without any additional measurements, which the distribution of the cell population could be estimated with. Future experiments are needed to address these questions. Conclusion ========== The presented method yields a robust way of finding periodicity in short time series data. As illustrated in Simulations Section, the proposed robust detection method is remarkably insensitive to different kinds of non-idealities in the data, such as heavy contamination of outliers, missing values, short time series, nonlinear distortions, and is completely insensitive to any monotone nonlinear distortions. The results also show that the proposed method has clearly better performance than the Fisher\'s test, even in the case of the standard Gaussian noise. Furthermore, the results on real data demonstrate that the proposed method performs well on real data and that the results are biologically meaningful. As illustrated in Figures [2(a)--(c)](#F2){ref-type="fig"} and more extensively reported in \[[@B14]\], the robust method serves also as a good spectral estimator. As the time series measured from biological systems are usually short and prone to contain different kinds of non-idealities, we believe that the robust detection method presented in this paper will find many important applications in this field. Authors\' contributions ======================= MA carried out an implementation of the methods, performed the computations and co-drafted the manuscript. HL developed the statistical methods, helped in computations and mainly drafted the manuscript. RP helped in developing the statistical methods. OY-H conceived of the study and participated in its design and coordination. HH helped in the implementation of the computational methods. All authors read and approved the final manuscript. Acknowledgements ================ The support of Tampere Graduate School in Information Science and Engineering (TISE) and the Academy of Finland are gratefully acknowledged. We are indebted to Dr. Korbinian Strimmer for providing an implementation of their methods, stimulating discussions, and useful suggestions. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Examples of time series**. An example of a time series composed of a sine and (a) additive standard Gaussian noise, (b) additive standard Gaussian and impulsive noise, and (c) additive standard Gaussian noise and cubic distortion. ::: ![](1471-2105-6-117-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Examples of spectral estimates**. The spectral estimates for the time series in Figures 1 (a)-(c), respectively, using both the standard periodogram and the proposed robust method. ::: ![](1471-2105-6-117-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Power of the test**. The power of the tests (*y*-axis) for the three different test cases as the function of the time series length and varying noise parameters (*x*-axis). The solid (resp. dashed) line corresponds to the proposed robust method (resp. Fisher\'s test). Three different types of non-idealities are considered, namely, pure standard Gaussian noise (the first row), standard Gaussian and impulsive noise (the second row), and standard Gaussian noise and *x*^3^distortion (the third row). The left (resp. right) column shows the results for different time series lengths (resp. different values of the noise parameters). ::: ![](1471-2105-6-117-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Benchmark results**. The fraction of the benchmark set that is identified (*y*-axis) as the function of the highest ranked genes (*x*-axis). The solid (resp. dashed) line corresponds the robust detection having fixed (resp. unknown) frequency. The dotted line shows the performance of the random gene selection. The columns from left to right correspond to the Alpha, the Cdcl5 and the Cdc28 experiment by \[6\]. The rows, from top to bottom, correspond to the three different benchmark gene sets B1, B2 and B3. for more details about the benchmark gene sets, see \[24\]. ::: ![](1471-2105-6-117-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Number of inferred periodic time series: case 1. The number of inferred periodic time series using the robust method and standard Gaussian noise in the data. *p*-values were obtained by simulating the distribution of the *g*-statistic using 10000 time series composed of Gaussian noise. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ------ --------- --------- --------- ---------- 0.15 0 2(1) 107(90) 109(96) 117(99) 115(100) 0.10 0 1(1) 96(87) 103(94) 110(98) 109(100) 0.05 0 1(0) 83(79) 95(89) 101(96) 105(100) 0.01 0 1(0) 59(59) 80(79) 90(89) 101(100) 0.005 0 0(0) 32(32) 62(61) 64(64) 100(100) *Z* 12 49 89 93 95 100 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Number of inferred periodic time series: case 2. The number of inferred periodic time series using the robust method and standard Gaussian noise in the data. *p*-values were obtained by using permutation tests. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ------ --------- --------- --------- ---------- 0.15 0 4(3) 108(92) 113(96) 111(98) 119(100) 0.10 0 1(1) 99(90) 106(94) 106(97) 112(100) 0.05 0 1(0) 88(84) 97(89) 101(95) 106(100) 0.01 0 0 65(64) 80(78) 86(86) 101(100) 0.005 0 0 46(46) 61(61) 71(71) 100(100) *Z* 15 48 91 92 95 100 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Number of inferred periodic time series: case 3. The number of inferred periodic time series using the periodogram method and standard Gaussian noise in the data. *p*-values were obtained by using Equation 5. ::: *q\\N* 10 20 40 45 50 100 -------- ------ -------- --------- --------- --------- ---------- 0.15 1(0) 17(11) 111(89) 108(86) 117(97) 122(100) 0.10 0 9(6) 100(85) 98(80) 108(96) 115(100) 0.05 0 2(1) 83(78) 79(71) 99(93) 107(100) 0.01 0 0 56(55) 54(52) 84(83) 101(100) 0.005 0 0 23(23) 20(20) 54(54) 99(99) *Z* 13 39 85 83 93 99 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Number of inferred periodic time series: case 4. The number of inferred periodic time series using the periodogram method and standard Gaussian noise in the data. *p*-values were obtained by using permutation tests. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ------ --------- --------- --------- ---------- 0.15 0 2(0) 109(94) 111(98) 114(98) 117(100) 0.10 0 2(0) 98(91) 103(96) 107(98) 112(100) 0.05 0 2(0) 87(84) 98(95) 103(98) 105(100) 0.01 0 2(0) 68(67) 88(86) 91(80) 102(100) 0.005 0 1(0) 55(55) 70(70) 80(79) 100(100) *Z* 14 40 92 94 97 100 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Number of inferred periodic time series: case 5. The number of inferred periodic time series using the robust method and standard Gaussian plus impulsive noise in the data. *p*-values were obtained by simulating the distribution of the *g*-statistic using 10000 time series composed of Gaussian noise. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ------ -------- -------- --------- ---------- 0.15 0 1(0) 73(62) 84(73) 101(84) 114(100) 0.10 0 1(0) 60(55) 76(69) 91(81) 109(100) 0.05 0 1(0) 48(46) 58(56) 77(73) 105(100) 0.01 0 0 22(21) 21(21) 49(49) 100(99) 0.005 0 0 13(13) 17(17) 34(34) 95(95) *Z* 9 35 73 78 84 99 ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Number of inferred periodic time series: case 6. The number of inferred periodic time series using the robust method and standard Gaussian plus impulsive noise in the data. *p*-values were obtained by using permutation tests. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ------ -------- -------- -------- ---------- 0.15 0 1(0) 70(57) 83(72) 93(82) 117(100) 0.10 0 0 59(51) 71(68) 89(79) 111(100) 0.05 0 0 48(45) 60(58) 77(72) 105(99) 0.01 0 0 24(23) 35(33) 54(53) 99(98) 0.005 0 0 22(21) 27(27) 41(41) 95(95) *Z* 9 38 72 78 82 99 ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Number of inferred periodic time series: case 7. The number of inferred periodic time series using the periodogram method and standard Gaussian plus impulsive noise in the data. *p*-values were obtained by using Equation 5. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ---- -------- -------- -------- -------- 0.15 0 0 16(16) 24(22) 28(28) 77(71) 0.10 0 0 13(12) 20(12) 25(24) 73(69) 0.05 0 0 11(11) 13(13) 17(17) 65(63) 0.01 0 0 3(3) 6(6) 9(9) 52(52) 0.005 0 0 0(0) 1(1) 3(3) 40(40) *Z* 11 29 44 46 52 79 ::: ::: {#T8 .table-wrap} Table 8 ::: {.caption} ###### Number of inferred periodic time series: case 8. The number of inferred periodic time series using the periodogram method and standard Gaussian plus impulsive noise in the data. *p*-values were obtained by using permutation tests. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ------ -------- -------- -------- -------- 0.15 0 6(5) 38(29) 40(33) 18(16) 82(76) 0.10 0 2(2) 34(26) 36(32) 14(13) 76(72) 0.05 0 1(1) 23(22) 32(29) 10(9) 69(69) 0.01 0 0 16(15) 18(18) 7(6) 61(61) 0.005 0 0 15(14) 17(17) 7(6) 53(53) *Z* 11 32 47 49 44 80 ::: ::: {#T9 .table-wrap} Table 9 ::: {.caption} ###### Number of inferred periodic time series: case 9. The number of inferred periodic time series using the periodogram method and standard Gaussian noise and cubic distortion in the data. *p*-values were obtained by using permutation tests. ::: *q\\N* 10 20 40 45 50 100 -------- ---- ---- -------- -------- -------- --------- 0.15 0 0 49(44) 79(64) 89(74) 107(93) 0.10 0 0 39(36) 71(62) 80(69) 98(90) 0.05 0 0 25(24) 52(49) 64(59) 90(88) 0.01 0 0 8(8) 28(28) 44(43) 82(82) 0.005 0 0 8(8) 19(19) 37(36) 67(67) *Z* 7 15 68 71 79 91 ::: ::: {#T10 .table-wrap} Table 10 ::: {.caption} ###### Number of inferred periodic genes from real microarray data. Results obtained by using real microarray data are presented here. Permutation tests were used to obtain significance values for the robust spectra. Symbols: *N*is the length of the time series, *M*is the number of genes analysed, *P*(resp. *P*\') is the number of found periodic genes having an unknown frequency (resp. frequency corresponding to the cell cycle length). Notes in the table: ^a^Since the elutriation time course did not show any significant periodic components, we did not perform the test with a fixed frequency. ^b^The average spectrum showed several major peaks. The one in the vicinity of the assumed cell cycle frequency was chosen manually. ::: Cell type Experiment *N* M *P\'* *P\'/M* *P* *P/M* Source -------------- ----------------- ----- ------- --------- ---------- ------------ ------- -------- S.cerevisiae cdc15 24 5287 981 18.6 946(766) 17.9 \[6\] S.cerevisiae cdc28 17 6103 363 6.0 32(105) 0.5 S.cerevisiae alpha 18 6056 346 5.7 139(468) 2.3 S.cerevisiae elutriation 14 6074 \_a \_a 4(193) 0.07 Human HeLa Score3 48 41508 1285^b^ 3.1^*b*^ 3580(6043) 9.0 \[5\] S.pombe cdc25-1 19 4373 1431 32.7 759 17.4 \[7\] S.pombe cdc25-2 36 4422 2605 58.9 2197 49.7 S.pombe cdc25-sep1 20 4700 2624 55.8 2295 48.8 S.pombe elutriation1 20 4229 1948 46.1 551 13.0 S.pombe elutriation2 20 3961 1453 36.7 384 9.7 S.pombe elutriation3 20 4236 673 15.9 355 8.4 S.pombe elutr.-cdc10-br 22 4647 3131 67.4 2431 52.3 S.pombe elutr.-cdc25-br 21 4272 2405 56.3 767 18.0 :::

PubMed Central

2024-06-05T03:55:58.204502

2005-5-13

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168888/", "journal": "BMC Bioinformatics. 2005 May 13; 6:117", "authors": [ { "first": "Miika", "last": "Ahdesmäki" }, { "first": "Harri", "last": "Lähdesmäki" }, { "first": "Ron", "last": "Pearson" }, { "first": "Heikki", "last": "Huttunen" }, { "first": "Olli", "last": "Yli-Harja" } ] }

PMC1168889

Background ========== The introduction of the Expectation value (E-value) in the Blast program in 1990 \[[@B1]\] was a major milestone in the development of sequence search algorithms. For any sequence match with a score *S* obtained in a given database, the E-value is the number of hits of same or higher score that can be expected by chance. E-values tell biologists how likely they are to encounter a specific sequence match in a database search, which is a more useful view of biological significance than a mere similarity score. Except in some special cases, low E-values are commonly accepted as an evidence of sequence homology. The recent years have seen a growing interest for RNA motif searches, driven by the discovery of important new classes of regulatory RNA genes and motifs in all organisms. Non-coding RNAs are characterized in a large part by long range base pair interactions, whereas linear sequence constraints are not as important as in protein coding genes. As a result, standard sequence alignment programs such as Blast are not suited to RNA motif search. Computational biologists have addressed this problem in several ways, notably through descriptor-based systems in which the topology of base paired regions is user-specified \[[@B3],[@B5],[@B4]\], Stochastic Context Free Grammars (SCFG) which use a complete statistical model of RNA elements \[[@B6]\], and Secondary Structure Profiles, which use position weight matrices describing stems and single strands in the RNA motif, as found in the ERPIN program \[[@B8]\]. Although the last two methods compute alignment scores, the complexity of the underlying models and search algorithms has hampered the estimation of an E-value to date. When the behavior of the score distribution is known, such as in sequence alignment scores, E-values can be computed either directly, or by empirically fitting a histogram of scores from a sample of random sequences to the assumed distribution function \[[@B1],[@B2]\]. Unfortunately, we will show here that a search algorithm such as the one used in ERPIN does not produce predictable score distributions. A possible workaround for this practical limitation is to run simulations on randomized sequences and use the observed hit count at score *S*as the E-value for this score. However, since interesting high scoring solutions can be extremely rare (commonly less than one random occurrence per genome) simulations often require days of calculations. Can we then estimate score distributions without having to run such lengthy simulations? In this article, we show that ERPIN score distributions can be estimated a priori through a discrete convolution analysis of score profiles, based on a random model of nucleotide frequencies. This led us to develop a computational procedure that estimates the score distribution of ERPIN profiles in a very short time, before the actual search begins, so that each solution can be automatically assigned an E-value. ERPIN profiles ============== ERPIN is an RNA motif search software using as input a training set of aligned RNA sequences, and a target database in which the motif is to be identified. The training set contains both RNA sequence and their common secondary structure, specified as shown in Figure [1](#F1){ref-type="fig"} in the form of single strands and helices. Importantly, gap characters (insertions or deletions) are allowed in single strands but not in helices. Helices are composed of two distinct strands of equal length. A region is defined as a continuous stretch of complete single strands or helical elements. When only one strand of a helix is included in a region, this strand is considered as a single strand. A mask is a subset of a region constituted of single strands and/or complete helices. ERPIN converts each helix and single strand in the alignment into a lod-score profile. This involves two steps. First, columns in the alignment are converted into frequency profiles, recording the frequencies of bases or base-pairs in column *c*as: ![](1471-2105-6-118-i1.gif) Here *n*~*ic*~is the number of bases or base-pairs of type *i*in column *c*, and *N*is the total base or base-pair count in a column. There is one frequency profile for each single strand or helix in the alignment. In the case of a single strand, *i*∈ {*A*, *T*, *G*, *C*}, whereas for a helix, *i*∈ {*AA*, *AT*, *AG*, *AC*, \..., *CC*} and each column in a helical profile actually refers to two positions in the initial alignment. Single strand profiles thus have 4 rows while helix profiles have 16 rows. The special case of gap-containing single strands, where a fifth character is added to the profile is discussed later on. Frequency profiles are then converted into lod-score profiles, where values for column *c*are defined as: ![](1471-2105-6-118-i2.gif) where *b*~*i*~is the background frequency of base or base-pair *i*in the target database. Base-pair frequencies are considered as the product of individual base frequencies. The score of a helix or single-strand element is obtained by presenting a target sequence to this element\'s profile and summing the scores obtained for every profile column. For ungapped elements, the calculation is straightforward. For gap-containing single strands, a dynamic programming matrix of the profile and target sequence is constructed, which provides the best possible alignment score \[[@B8]\]. In a first stage, let us ignore this alignment procedure and focus on ungapped elements. Exclusions and pseudocounts --------------------------- We define as an exclusion a profile element for which no base or base pair is observed in the training set, and thus *P*~*ic*~= 0. Exclusions may be due either to an unsufficient size of the training set, or to a true avoidance of this particular base or base-pair at this position in the RNA molecule. In any case, the log-odd ratio formula would produce a value of -∞ for such cases, thus requiring a special treatment. Exclusions are dealt with either by using arbitrary low values (e.g. -30) or by introducing pseudocounts in the frequency matrix that simulate what could have been observed in a larger training set. Pseudocounts are based on some prior knowledge of \"typical\" substitution frequencies in RNA molecules, as observed in a model RNA sequence alignment. The pseudocount calculation procedure used in ERPIN is the same spirit as that of Henikoff and Henikoff \[[@B9]\], but we use a different definition of pseudocounts, as explained below. Let us first reformulate the values in any column *c*of a frequency profile: ![](1471-2105-6-118-i3.gif) Where *n*~*ij*~is the number of {*i*, *j*} couples in column *c*, *P*(*i*, *j*) is the joint probability of finding *i*and *j*in this column, and ![](1471-2105-6-118-i4.gif). This develops into: ![](1471-2105-6-118-i5.gif) Where *P*(*i*\|*j*) is the conditional probability of observing *i*, knowing that *j*is observed in the column. This conditional probability amounts to the observed frequency of *i*→ *j*substitutions. To introduce pseudocounts in ERPIN frequency profiles, *P*(*i*\|*j*) is replaced with the average substitution frequencies observed in a model RNA sequence alignment, expressed in the form of a substitution matrix *M*. Pseudocount-based frequencies can be expressed as: ![](1471-2105-6-118-i6.gif) where *M*is a square matrix whose columns *j*are normalized, ∑~*i*~*M*~*ij*~= 1, so that ![](1471-2105-6-118-i7.gif). See Methods section for construction of *M*. The relative ratio of pseudocounts to true counts in the final frequency matrix is then controlled by a user-defined weight parameter *α*∈ \[0, 1\], such that: ![](1471-2105-6-118-i8.gif) Since *M*~*ij*~are generally ≠ 0 in the substitution matrices (either for single strand or helices), most exclusions in the frequency profile are replaced by nonzero values as soon as *α*≠ 0. Not only the resulting lod-score profiles are basically devoid of arbitrary low values, but they better reflect \"natural\" base and base-pair substitution frequencies observed in real RNA alignments. This is especially interesting in helical regions, since the substitution matrix for helices represent natural exchanges between frequent base-pairs such as Watson-Crick, G:U or even G:A, while incurring strong penalties for exchanges involving rare base pairs. This maintains a large fraction of strongly negative values in helix profiles, which is desirable for the sake of search specificity. Shapes of score distributions: finite and non-finite scores =========================================================== We define as a \"finite\" score the score obtained for a sequence that does not contain any match to a profile exclusion. When pseudocounts are used, almost all scores are finite, but when pseudocounts are not in use (*α*= 0), many scores, especially in helix profiles, are \"non-finite\", although in practice they are replaced by arbitrary low values. Let *S*denote a finite score obtained at a given site in a random sequence. For any *x*\> -∞, the conditional probability formula reads: *P*(*S*\> *x*) = *P*(*S*\> -∞).*P*(*S*\> *x*\|*S*\> -∞) (7) In this decomposition, it is noteworthy that: • The first factor is independent of *x*as soon as *x*is finite, • The second factor can be computed based on profile elements that contain no exclusion. Let *S*~*c*~(*c*= 1, 2, .., *w*) denote a score for column *c*of a profile, and *S*= *S*~1~+ *S*~2~+ \... + *S*~*w*~the score obtained by presenting a given sequence to this profile. If *w*is large enough (*w*≳ 10) and the distributions of random variables *S*~*c*~are (i) independent and (ii) identically distributed, the sum *S*follows a normal distribution (central limit theorem \[[@B11]\]). Due to exclusions arising with different frequencies in different columns, condition (ii) is generally not fulfilled, but, using the decomposition given by formula (7) rewritten for column *c*, we can write: *P*(*S*~*c*~\> *x*) = *P*~*fs*,*c*~.*P*~*f*,*c*~(*x*) (8) where *P*~*fs*~is the probability that a score is finite, and *P*~*f*~(*x*) is the probability that a finite score is higher than *x*. This gives, for a complete profile: ![](1471-2105-6-118-i9.gif) In many cases, the probability distributions of finite scores *P*~*f*,*c*~are similar enough so that their sum *S*is normally distributed. If this behavior was always observed, the score distribution could be fully determined by computing the mean value *μ*and the standard deviation *σ*that characterize the normal law, which would enable a direct calculation of the E-value \[[@B2]\]: ![](1471-2105-6-118-i10.gif) Figure [2](#F2){ref-type="fig"} shows the score distributions obtained using the tRNA region spanning the anticodon and *T*Ψ*C*loop (two helices + three single-strands) at each position of a 100 Mb random sequence database, with pseudocounts switched off (*α*= 0). As expected, finite scores (Fig [2a](#F2){ref-type="fig"}) follow a normal distribution, while total scores (Fig [2b](#F2){ref-type="fig"}) are unevenly distributed. So called \"non-finite\" scores may be biologically relevant, since many valid substitutions are potentially absent from the training set. However, non-finite scores are detected only when high enough to fall into the extreme end of the distribution. This part of the distribution is composed mainly of finite scores and should thus behave like that of finite scores. Unfortunately finite scores may also deviate from a Gaussian distribution, for instance when score distributions in successive columns are too different from each other. Pseudocounts are used to inject missing substitutions into frequency profiles, resulting in most \"non-finite\" scores becoming \"finite\". But what is the behavior of finite scores when pseudocounts are in use? Fig [3a--d](#F3){ref-type="fig"} and [4a--b](#F4){ref-type="fig"} show finite score distributions for a variety of ungapped RNA motifs (shaded bars), obtained using a typical level of pseudocounts (*α*= 2.10^-4^) in frequency profiles. Although some training sets (tRNA, SECIS) have nearly gaussian distributions, others (let-7 miRNA, snoRNA, polyA sites) are more erratic. This is due to the larger number of exclusions in the later sets -- only partially compensated for by pseudocounts -- and/or their non-uniform distribution over profile columns. If we aim to address true biological problems with such imperfect or sparse training sets, we necessarily have to deal with this type of score distribution that cannot be approached with classical methods. Nonetheless, we will still be using the decomposition formula (7), as it provides an important reduction of the range of values of the random variables involved. Score distributions of helices and single-strands ================================================= Ungapped helices and single strands ----------------------------------- How can we estimate score distributions such as those in Fig [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"} ? Let us admit that profile columns are independent. After matching the profile to a purely random sequence, the resulting scores for each column would thus behave as independent random variables, say *X*~1~and *X*~2~for columns 1 and 2. Therefore, the final score for two columns would be: *S*= *X*~1~+ *X*~2~ Then, the probability of obtaining a score *S*= *x*is: ![](1471-2105-6-118-i11.gif) The last formula defines the *discrete convolution product*\[[@B11]\] of two distributions. The overall score distribution can be obtained by doing the calculation for every possible values of *u*and *v*. Using the separation formula (7) between \"exclusions\" and \"finite scores\", equation (10) can be written: ![](1471-2105-6-118-i12.gif) These operations can easily be extended to *N*columns by iterating the products on successive columns in the same single-strand or helix profile. At each successive iteration, scores are discretized on a predefined grid so that the number of possible scores increases linearly with the number of columns (see Methods section for algorithm). We performed such an analysis on a variety of helix and single strand profiles, with grid intervals set at Δ*x*= .05. In Figures [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}, score distributions estimated from discrete convolution (solid lines) are compared to scores obtained through simulation on a random database of variable size (shaded bars). There is a very good agreement between the discrete convolution and simulation. Gap-containing single strands ----------------------------- The score of a gap-containing single strand in the ERPIN program is computed from the dynamic programming alignment matrix. Therefore, it is the maximum of several values, and could be expected to comply with an extreme value distribution. However, gapped single strands in ERPIN are very diverse entities that may include oddities such as single-nucleotide strands, or strands mostly filled with gaps. This results in very uneven distributions that we were not able to model satisfyingly. Therefore, the score distribution of a gapped single strand in ERPIN is currently estimated based on a short simulation performed on a random sequence (see Methods section for details). Score distributions of complete regions with gaps ================================================= Score of a configuration ------------------------ When presenting a sequence to a whole region, the presence of gaps in single-strands results in multiple allowed positions for flanking helical elements. A configuration is a specific arrangement of helix elements determined by the number of intervening gaps (Fig [1](#F1){ref-type="fig"}). There is one score for each allowed configuration, which is the sum of scores for all helices and single strands in this configuration. We therefore need to compose the different score distributions to obtain the distribution of the total score for one configuration. This is again done using a discrete convolution of these distributions, with the same procedure and grid parameter as above. This provides the score distribution of a single configuration. Note that, although a configuration may contain gapped single strands of which score distribution was not produced by a discrete convolution, such distributions can now be treated by this second convolution round applied to whole profiles. Score of a complete region -------------------------- The number of gaps in a single strand is bounded by the maximum number of gaps observed for this strand in the training set: *mxgaps*. For a simple hairpin-loop motif with *mxgaps*possible gaps in the loop, there are (*mxgaps*+ 1) possible configurations. For a whole region containing *N*strands with gaps (*i*= 1, 2, \..., *N*), the number of configurations is: ![](1471-2105-6-118-i13.gif) Erpin evaluates all possible configurations without any construction rule or strategy. A combinatorial explosion is avoided by implementing multi-stage searches, where search at each stage is limited to a defined mask, or subset of the region under study. The score of a region or mask at in given site is the maximum score obtained for all possible configurations. Since a motif is identified only after all possible configurations are evaluated at a given site, our estimation of motif scores requires taking into account this additional complexity. Let *K*denote the number of configurations for a given motif, and *S*~*i*~, the score obtained for the *i*th configuration. As the ERPIN program does not permit the of addition gaps relative to those present in the training set, *K*is necessarily bounded but its value can be relatively large. We are now interested in the maximal score obtained for all configurations at each site. This is the extreme value distribution, or the distribution of a random variable *M*defined as: ![](1471-2105-6-118-i14.gif) If *P*~*fs*~= *P*(*S*~*i*~\> -∞) and *p*~*i*~(*x*) = *P*(*S*~*i*~\> *x*\|*S*~*i*~\> -∞), and the random variables *S*~*i*~are statistically independent (*s.i*) and identically distributed (*i.d*), then: ![](1471-2105-6-118-i15.gif) ![](1471-2105-6-118-i16.gif) The most \"interesting\" scores are expected to be of the same order of magnitude as those obtained by training set sequences. For any realistic training set, these scores should be very high compared to scores obtained on random sequences and, therefore, their probability should be very low. In this case *P*(*M*\> *x*), given by formula (17), behaves at the first order approximation as *K.P*~*fs*~.*p*(*x*). For a database of size Ω, considering that individual sequences in the database are large enough compared to the search motif so that border effects can be ignored, the final E-value, is : *E*(*x*) = *P*(*M*\> *x*).Ω (18) Figure [5](#F5){ref-type="fig"} compares these computed E-values (solid lines) to simulations performed on a random database (circles), for complex RNA regions encompassing multiple helices and singles strands (gapped or ungapped). Overall there is a very good agreement between E-value and simulation, consistent with our hypothesis that configuration scores are independent and equally distributed. Importantly, E-values remain accurate for RNA regions containing large gapped single strands, such as snoRNA (Fig [5b](#F5){ref-type="fig"}), Let-7 miRNA (Fig [5c](#F5){ref-type="fig"}) and SECIS (Fig [5d](#F5){ref-type="fig"}), and over a wide range of scores. This last point is also important, since \"borderline\" solutions with an E-value around 0.1 or 1 are potentially more interesting biologically as low E-value solutions. Moreover, computing times for overall E-value calculations in all our tests motifs remained insignificant relative to database scan times. In the case where pseudo-counts are switched-off, profiles contain multiple \"non-finite scores\" which are excluded from the convolution process. This may imply a lack of accuracy in the left-hand side of the estimated score distribution, where scores have low values, but should have little effect in the region of biologically interesting scores. Therefore we do not expect E-values to deteriorate significantly in practise when pseudocounts are switched off. Conclusion ========== We have presented a method to estimate the score distributions of RNA helices or single strand profiles and of their combinations into larger motifs. This method is based on discrete convolutions. The computing time of the discrete convolution algorithm increases quadratically with profile size and remains in any case negligible relative to database scan durations. This procedure is implemented in the last release of the ERPIN software (V. 4.2) and provides accurate estimates of E-values for practical applications. Interestingly, the discrete convolution approach can be applied as well to others sequence scoring models -nucleic acids or proteins -- based on ungapped profiles with independent columns. Methods ======= ERPIN program and utilities --------------------------- The ERPIN program (sources and executables) is available at <http://tagc.univ-mrs.fr/erpin/>. Simulated score distributions of independent helix and single-strand profiles (Fig [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}) were obtained using the -*hist*(histogram) option of ERPIN and utility programs *epnstat*, *convhstat*, *convsstat*and *mstat*provided in the distribution. For Fig [5](#F5){ref-type="fig"}, full motif searches were performed using ERPIN Version 4.2.5 with pseudocount weight *α*= 2.10^-4^. Graphical outputs for figures [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"} were produced using the Matlab \[[@B13]\] package. Training sets ------------- Training sets for profile statistics and ERPIN runs in Fig [2](#F2){ref-type="fig"}, [3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"} are available on the ERPIN web site and were obtained as follows: • tRNA: 903 type I tRNA sequences (all species) from the 1997 version of M. Sprinzl\'s nuclear tRNA alignment \[[@B17]\]. • SECIS (Selenocystein Insertion Sequence): 117 metazoan SECIS sequences, from our own compilation \[[@B7]\]. • snoRNA (Small nucleolar RNA): 217 archaean C/D box snoRNA sequences, compiled and aligned by Fabrice Leclerc at CNRS Nancy (Personal communication). • Let-7 miRNA: 27 animal miRNA precursor sequences from our previous compilation \[[@B15]\]. • PolyA sites: 2327 human polyadenylation sequences from our previous compilation \[[@B16]\]. RNA substitution matrices for pseudocounts ------------------------------------------ Pseudocount calculation requires substitution matrices obtained from a model RNA sequence alignment or \"training set\", annoted with secondary structure information (helix or single strand). Klein and Eddy have developed RNA substitution matrices previously \[[@B18]\], but we use a different type here. Training set columns are converted into profiles containing raw nucleotide or base-pair counts. Let *Q*denote a nucleotide or base-pair count profile of width *w*, produced by a concatenation of all single strand or helix profiles from the pseudo-count training set. *Q*is either a 4-line matrix for single strands (*h*= 4) or a 16-line matrix for helices (*h*= 16). The substitution matrix *M*introduced in Section \"Exclusions and Pseudocounts\" is then a square matrix of size *h*× *h*defined as: ![](1471-2105-6-118-i17.gif) ![](1471-2105-6-118-i18.gif) ![](1471-2105-6-118-i19.gif) The square matrix *N*is actually a \"correlation matrix\" of the profile lines since element *N*~*ij*~is the scalar product correlation of lines *i*and *j*. Coefficients *λ*~*j*~make this matrix normalized, so that: ![](1471-2105-6-118-i20.gif) Probability conservation is verified for *P*and therefore it is also verified for *P\'*: ![](1471-2105-6-118-i21.gif) Finally, it is obvious from formula (6) that *P\"*also verifies verifies probability conservation, hence: ![](1471-2105-6-118-i22.gif) Substitution matrices can be generated from any RNA sequence alignment using the utility program *mksum*of the ERPIN distribution. Default matrices provided with distribution (*SUM.dat*file) were obtained using a 16S/18S rRNA training set from R.Gutell (\[[@B10]\]) containing 6310 sequences from all three phylogenetic domains. We used the secondary structure of E-coli 16S rRNA as the consensus structure, resulting in 481 columns of helix profile and 7512 columns of single strand profile. Option -*pcw*is used to set pseudo-count weight *α*in the ERPIN program. Default internal value is 2.10^-4^, but this has been rescaled for users by a factor of 2.10^-3^giving a default user value of 0.1 and a practical maximal value that should not exceed 1. Effects of pseudocounts and of the *α*parameter on profiles can be visualized using the utility program *pview*. Score distribution of gap-containing single-strands --------------------------------------------------- The score distribution of gap-containing single strands is evaluated by repeatedly computing profile scores with a random sequence of same length *L*as the profile and same composition as the target sequence database. The calculation is repeated *C.L*^2^times, with *C*a constant, and *L*\<*L*~*max*~in order to limit CPU time for unusually large strands. Default values are *C*= 300 and *L*~*max*~= 12. Histograms and discrete convolution product ------------------------------------------- Although discrete convolutions can be computed using iterated Fast Fourier Transforms, this approach is subject to numerical approximations in practice. A direct calculation is more accurate and proved fast enough in all cases tested. Time complexity of the discrete convolution algorithm is *O*(*N*^2^) where *N*is the total number of profile columns. This value remains tractable even for the largest RNA motifs. The convolution algorithm was adapted from those found in the Octave \[[@B12]\] and Matlab \[[@B13]\] packages. The linear sampling interval was set at Δ*x*= .05. CPU time for the complete E-value calculation (including profile construction, convolution of independent profiles and convolution of configurations) for motifs in Fig [5](#F5){ref-type="fig"} ranged from 10^-4^s to 0.8s on a 2.6 GHz Intel Pentium workstation with 1 Gb of RAM. Extreme value distribution -------------------------- Formula (17) used for calculating the extreme value distribution is of the type (1 - (1 - *x*)^*N*^). If *x.N*\> 1 the result is obtained with the C library function *x*![](1471-2105-6-118-i23.gif)*x*^*y*^which lacks precision when *x*≪ 1. Otherwise we compute (17) using the binomial formula for (1 - *x*)^*N*^. Authors\' contributions ======================= AL conceived the discrete convolution approach, programmed the software and participated in drafting the manuscript, ML and JFF performed program runs and statistical analyzes of outputs, DG participated in the design and coordination of the study and wrote most of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We thank the ACI IMPBio program for their support to the development of the ERPIN software and Web server. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### An example of ERPIN training set containing two double helices (noted 2 and 4), and three single strands (noted 3, 5 and 7). Due to gaps in the alignment, helix 2 spans 9 to 11 nt, and helix 4 spans 6 to 7 nt. Combinations of these allowed ranges give 6 possible configurations for the whole RNA motif. ::: ![](1471-2105-6-118-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distributions of finite and total scores obtained from the motif encompassing the anticodon and *T*Ψ*C*loop of tRNA, at each position of a 100 Mb random sequence database. This region covers three single strand profiles and two helix profiles and spans a gap-containing single- strand profile that is not included in score calculation. ERPIN results were processed by the *epnstat*utility program. A: finite scores. B: total scores (both finite and non-finite). ::: ![](1471-2105-6-118-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Comparision of finite score distributions obtained from discrete convolution of helix profiles (solid lines) and simulation (shaded bars). The various helices in the region under study were combined into a larger 16 × *W*profile, where *W*is the total number of base-pairs in the region. Lod-scores were computed based on a uniform nucleotide composition, by the *convhstat*utility program. ::: ![](1471-2105-6-118-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Comparision of finite score distributions obtained from discrete convolution of single-strand profiles (curve) and simulation (shaded bars). The various strands in the region under study were combined into a larger 4 × *W*profile, where *W*is the total number of nucleotides in the region. Lod-scores were computed based on a uniform nucleotide composition, by the *convsstat*utility program. ::: ![](1471-2105-6-118-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Comparision of computed E-values (solid lines) and number of solutions obtained from simulation on a random database of uniform nucleotide composition (circles), for different RNA motifs. Numbers following \"region\" refer to secondary structure elements in the corresponding training set available from <http://tagc.univ-mrs.fr/erpin/>. E-values were computed using the *mstat*utility program. (a) tRNA region covering the anticodon and *T*Ψ*C*stem-loops; (b) C/D box snoRNA region covering the major stem and C+D boxes; (c) Let-7 miRNA region covering the complete precursor hairpin; (d) SECIS element covering the large 14 bp stem and apical stem+loops. ::: ![](1471-2105-6-118-5) :::

PubMed Central

2024-06-05T03:55:58.210379

2005-5-13

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168889/", "journal": "BMC Bioinformatics. 2005 May 13; 6:118", "authors": [ { "first": "André", "last": "Lambert" }, { "first": "Matthieu", "last": "Legendre" }, { "first": "Jean-Fred", "last": "Fontaine" }, { "first": "Daniel", "last": "Gautheret" } ] }

PMC1168890

Background ========== A diverse range of applications that impact on clinical diagnosis and prognosis rely on the real-time detection and quantification of Polymerase Chain Reaction products such as bacterial diagnostics \[[@B1],[@B2]\], viral load \[[@B3]-[@B6]\], cancerous cell burden \[[@B7],[@B8]\] and parasite quantification \[[@B9]\]. Each particular assay dictates the requirement for instrument capacity, data reproducibility and run length times. There is an expansive collection of different fluorescent chemistries and instruments available for real-time amplification reactions \[[@B10]-[@B12]\]. These platforms represent a technological advance from the traditional thermal cycler and typically offer faster, smaller (reduced volume) and more (high-throughput) reactions. Whilst the number of instruments with different software and fluorescent dyes is considerable, the obvious drawback is that the resultant data may not be repeatable or even directly comparable between instruments or even within instruments on different runs. A statistically appropriate number of samples and runs need to be conducted in order to have confidence in the results or to assign mathematical significance to any of the factors affecting real-time amplification. In addition, there is a shortage of reliable, universal DNA standards at present. Increasing diversity through technological advances in DNA quantitation instruments and software may further challenge the standardisation of results. It should be ensured that proper quality controls under-pin both existing and emerging technologies. The aim of this study was to evaluate the repeatability of three different real-time PCR platforms including the effect of PCR reaction mix. Previous work carried out has shown that fluorogenic 5\' exonuclease (\"TaqMan^®^\") probe chemistry was more repeatable than Scorpion probe or SYBR^®^Green I chemistries for genetically modified soya amplification on the LightCycler^®^(Roche Diagnostics, Lewes, UK) and the ABI PRISM^®^7700 (Applied Biosystems, Warrington, UK) \[[@B13]\]. Hence, for the amplification of a \'model\' plasmid dilution series we have selected \"TaqMan^®^\" probes to assess inter-platform repeatability of the LightCycler^®^, the ABI PRISM^®^7700 and the Rotor Gene 3000™ (Corbett Research, Cambridge, UK). These three real-time PCR platforms have different temperature control mechanisms, specifically a 96 well heating block (ABI PRISM^®^7700), an air fan and/or coil with 32-position carousel (LightCycler^®^), and ambient air with 72-position rotor (Rotor Gene 3000™) \[[@B10]\]. In addition the performance of two commercially available master-mixes were compared. Statistical analysis was performed on the data to assess the levels of inter-platform repeatability and dynamic range of the instruments with different master-mixes. Results ======= Repeatability Estimates ----------------------- Precision estimates were determined based on each platform with PCR reaction mix combination at every plasmid copy number across the three days using ISO guidelines \[[@B14]\]. Precision estimates were calculated as repeatability standard deviations of the Ct (threshold cycle number) values from the amplification of each plasmid dilution. In addition, the mean value and standard deviations of these repeatability standard deviations were also presented in Table [1](#T1){ref-type="table"} to provide a measure of the variance of these estimates. Percentage coefficients of variance (%CV) were not used to represent precision estimates as a low mean Ct value in this experiment is indicative of good precision and therefore percentage CV would generate misleading results. Although the assay was not designed to assess biological sensitivity of the system, and the dilution levels did not approach the limit of detection for the system, the 10^8^plasmid copy number data set had the best overall precision associated with it. The most probable reason for this is the availability of the target analyte in the sample. The ABI PRISM^®^7700 had the lowest mean repeatability estimate of all the platforms in this study when combined with the FastStart reaction mix, as shown in Table [1](#T1){ref-type="table"}. However, the means of the repeatability estimates for the two PCR reaction mixes when combined with this platform were different (0.202 for the FastStart mix and 0.362 for the Excite™ mix). With the exception of the 10^4^plasmid copy number, every repeatability estimate value for the reactions performed on the ABI PRISM^®^7700 using FastStart was lower than the corresponding Excite™ reaction mix. Indeed, consistently across the three platforms, the FastStart mix had the lowest average repeatability estimates in this trial. The value of the mean and standard deviations of the repeatability estimates were comparable when the FastStart PCR reaction mix was employed on the ABI PRISM^®^7700 and on the LightCycler^®^(0.202 and 0.219 respectively and 0.047 and 0.082 respectively). The lowest repeatability estimate calculated in this study was found to be 0.100. This value was generated on the LightCycler^®^using the FastStart reaction mix and a plasmid copy number of 10^4^. Conversely, the highest repeatability estimate (1.378) was also produced on the LightCycler^®^but the Excite™ PCR reaction mix was employed with a plasmid copy number of 10^6^. Unexpectedly, the highest repeatability estimates for each data set are not necessarily associated with the low plasmid copy numbers \[[@B15]\]. The lowest repeatability estimates and the highest repeatability estimates for each platform and PCR reaction mix data set show no discernible trend with respect to plasmid copy number. The least repeatable system in this trial based upon the data in Table [1](#T1){ref-type="table"} was the combination of the LightCycler^®^with the Excite™ PCR reaction mix. Comparison with the results of other published studies was not possible, as the statistical methods used were not clearly reported in the literature. Analysis of Variance -------------------- An ANOVA approach using the threshold cycle number was used to identify sources of variability within the data. Table [2](#T2){ref-type="table"} shows the significance levels of the four main factors of day, plasmid copy number, reaction mix, and platform, as well as the significance levels of the interaction terms. The factors and interaction terms were compared to the Error MS (mean square) in order to assess their significance. All four individual factors were highly significant. From the calculated F-variance ratio the factor of day had the least significant effect of the four factors, followed by PCR reaction mix (approximately 1.5 times the effect of day), then platform (approximately 7 times the effect of day), and then, as expected, plasmid copy number (approximately 45 times the effect of day). Even though the effect of day had a significant effect upon the results, this effect was very small when compared to the effect of platform and plasmid copy number. The interaction of the four main factors was also investigated, this demonstrated that the majority of interaction terms were significant. However, in general the magnitude of the effect of these interaction terms was very small compared to the effect of the four main factors. ### Day Effect Day had a statistically significant effect upon the Ct value of the raw data (P\<0.00001). The average Ct value in the full data set decreased from 19.76 to 18.54 across the 3 days (data not shown). ### Plasmid Dilution Effect As Figure [1](#F1){ref-type="fig"} illustrates, the number of plasmid copies was inversely proportional to the average Ct value \[[@B6],[@B12]\] within the instrument dynamic working range and above the LOD (limit of detection). Our results concur with Sanchez *et al*\[[@B5]\] who reported that cytomegalovirus quantification was linear between 10^1^and 10^6^copies, and with Moberg *et al*who used 10^2^to 10^7^DNA copies in their real-time assay \[[@B16]\]. At extremely low or high plasmid copy numbers the linear relationship with Ct value may be affected. At high copy number it may be difficult to set an accurate baseline because threshold values will be reached after only a few amplification cycles. In addition, reactions with very low numbers of target molecules may be disproportionately affected by pipetting errors and by stoichiometric effects. This highlights the need to work within the linear dynamic range of the instrument and the potential inaccuracy introduced by extrapolation of calibration curves. Fresh plasmid dilutions were utilised to counteract the previously observed progressive loss of intact plasmid DNA, particularly at low plasmid copy number dilutions. This is probably due to physical adhesion of the DNA to the tube walls, although precautions, such as the use of siliconised tubes, were undertaken. Fresh plasmid dilutions were made each day. Figure [2](#F2){ref-type="fig"} illustrates how the average Ct value alters across the three days (which have three different dilution sets) for each plasmid copy number. ### PCR Reaction Mix Effect The results from reactions containing FastStart mix had significantly lower average Ct values than their Excite™ 2x mix reaction counterparts (data not presented) -- the difference is approximately 1 cycle per plasmid dilution. The smallest interaction term is that of PCR reaction mix with platform (P = 0.019), indicating good consistency of results when using the reaction mixes across different platforms. ### Platform Effect The platforms have different heating mechanisms and therefore it was expected that their performance characteristics such as repeatability might differ significantly. The largest significant interaction term (F = 150.8) was that of platform with day, indicating that each platform may not be behaving consistently between the three days. Closer inspection of the data shows that the majority of the variation associated with the platform by day interaction is attributed to the Rotor Gene day 2, as illustrated in Figure [3](#F3){ref-type="fig"}, which appears inconsistent compared to the other two platforms. This interaction effect of platform with day was not significantly different from the factor of day or PCR reaction mix. The interaction of PCR reaction mix and platform (shown in Table [2](#T2){ref-type="table"}) was found to be statistically significant at the 5% level. The full raw data set is available \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Discussion ========== Use of a \'model\' system that contained little or no contaminants, coupled with a balanced experimental design enabled accurate assessment of repeatability. Other factors such as extraction methods and complex matrices can potentially increase analytical variability. As demonstrated by the present study, a range of factors can affect performance of each of the instruments. In addition, results are influenced by the threshold value settings applied to the machine by the analyst. To underpin meaningful quantitative PCR measurement, internationally accepted guidelines should be established to standardise the analysis and interpretation of real-time PCR results. PCR reference standards could also be used as indicators of instrument reliability and amplification efficiency \[[@B17],[@B18]\], while internal reaction controls could additionally monitor effects of PCR inhibitors and normalise for any instrument non-uniformity. A recent publication has looked at the plateau phase and the effect of enzyme inhibition during this stage of amplification \[[@B19]\]. With regard to analysis and software compatibility, a study conducted by Kuhne and Oschmann \[[@B20]\] has demonstrated the feasibility of importing standard curves from different LightCycler runs. Historically, standard curves have been the method of choice for determining absolute DNA concentration. However, the quantity of the standards used should be accurately calculated by gravimetry or other primary methods, as absorbance at a wavelength of 260 nm is not sufficient alone. Our present study has shown the potential inaccuracies introduced by importing standard curves, considering the significance of the factor \'day\' on the data presented. Even under the controlled conditions used in the present study, significant variability was observed, suggesting that transferring standard curves between runs could be unreliable unless adequate controls are used. Recent research using more sophisticated software approaches has demonstrated the ability to use grand mean calibration curves between runs following compensation for differences in amplification efficiency \[[@B21]\]. Amplification efficiency changes throughout the course of a PCR reaction and is dependent on enzyme kinetics \[[@B22],[@B23]\]. The potential pitfall of assuming that the sample analyte has the exact same kinetics and amplification efficiency of the standards has been highlighted recently \[[@B15]\]. A Galton-Watson branching process has been proposed which involves considering the number of molecules present during PCR cycles and uses Michaelis-Menten enzyme kinetics together with a population growth approach \[[@B23]\]. One other source of uncertainty associated with using amplification curves is the issue of handling outlier data points. In a recent publication Bar *et al*discussed this subject and presented a new statistical method \[[@B24]\]. This may have applications for clinical samples that have been extracted from complex matrices or any analysis with high levels of variability between results. It is not surprising that difficulties in data comparison arise due to the many different mathematical and software approaches applied to real-time quantitative PCR, despite sharing the same mathematical fundamentals \[[@B25]-[@B28]\]. The data generated in this study is not intended to indicate any preference for a particular piece of equipment or PCR reaction mix. The instrument specification, reliability, operator training, servicing and comparability all contribute to overall assay precision. In choosing a real-time platform, we would recommend that users consider the statistical outcomes of our results and perform their own critical assessments to ensure that an instrument is fit for their purpose. Generally any such judgement would involve factors such as the volume of samples for processing, the speed at which results need to be obtained and the reproducibility or repeatability of the results. In addition to choice of appropriate real-time platform, accurate quantitative measurement is dependent upon rational experimental design, optimisation and careful data interpretation. Conclusion ========== We have demonstrated that in this trial the most repeatable system was the ABI PRISM^®^7700 with the FastStart PCR reaction mix. However the mean repeatability estimate of the LightCycler^®^combined with the FastStart mix was comparable. In addition, we have shown that the plasmid copy numbers used in this trial did not display a distinct relationship with instrument repeatability. An ANOVA established that plasmid copy number, platform, PCR reaction mix and day significantly affected the results in a descending order of magnitude. Inherent differences in the data handling between instruments still presents a challenge to the normalisation of real-time PCR assays between laboratories and users of different instruments. We have confirmed that standardised real-time PCR detection instruments and data analysis approaches are imperative in order to guarantee the integrity and the repeatability of DNA measurement experiments. Methods ======= Plasmid Composition and Dilution -------------------------------- The target DNA was a pPCR-Script Amp SK(+) plasmid (Stratagene, Amsterdam, The Netherlands) with an insert at the *Srf*I site in the multiple cloning site (supplied by N.I.S.T, Gaithersburg, MD, USA). A portion of the insert sequence follows with the primer binding sites indicated in bold face type and the probe binding site underlined: 5\'-***CAGACCGCTCGACGATAGG***T[CAGCACTGTCTCGTTGACAGGCGTG]{.underline}GTCAATCAGCCTGAAATCCTCGATCAGAGTGTGC*CG****ATCTCTGGTCCACGTCCT***-3\'. From a stock solution of plasmid (3 × 10^13^copies) in molecular biology grade water (Sigma, Poole, UK), 1 in 10 serial dilutions were carried out to a copy number of 3 × 10^5^per mL using the same batch of water. Fresh dilutions were made every day from the same stock and using the same conditions and were kept at 4°C before and between use. At low copy numbers particularly, a reduction in target copies available for amplification was observed on prolonged storage, possibly due to DNA absorption to the walls of the tube (data not shown). Fresh dilutions were prepared each day in order to minimise this source of variation. Real-time PCR Conditions ------------------------ A 100-bp region of the insert sequence was amplified using HPLC purified oligonucleotide primers (Sigma-Genosys, Cambridge, UK) and a fluorogenic 5\' exonuclease probe (Sigma-Genosys, Cambridge, UK) diluted in molecular biology grade water (Sigma, Poole, UK). The primers and probe used were P1A (forward primer): 5\'-ACAGACCGCTCGACGATAGG-3\', P1B (reverse primer): 5\'-AGGACGTGGACCAGAGATCG-3\' and TAQ1 (probe): 5\'-{FAM}CAGCACTGTCTCGTTGACAGGCGTG{TAMRA}-3\' (FAM -- 6-carboxyfluoresceine; TAMRA -- 5\'-carboxytetramethylrhodamine). Primer Express^®^software (Applied Biosystems; Warrington, UK) was used to design the primers and probe. The thermal cycling conditions used were dependent on the PCR reaction mix employed and were optimised protocols based on the manufacturer\'s guidelines. Amplification conditions for those reactions that contained \'FastStart Master mix\' (Roche Diagnostics, Lewes, UK) were a 10 minute hold at 95°C followed by 55 cycles of 95°C for 15 seconds and 60°C for 30 seconds. For reactions with the Excite™ 2x kit (Biogene, Kimbolton, UK) the amplification conditions were a 2 minute hold at 50°C followed by a 10 minute hold at 95°C and then 55 cycles of 95°C for 15 seconds and 60°C for one minute. Instrumentation & PCR Reaction Mixes ------------------------------------ Three platforms compatible with \"TaqMan^®^\" probes were assessed -- LightCycler^®^(Roche Diagnostics, Lewes, UK), ABI PRISM^®^7700 (Applied Biosystems, Warrington, UK) and Rotor Gene 3000™ (Corbett Research, Cambridge, UK). All of these instruments were found to be compatible with FastStart (Roche Diagnostics, Lewes, UK) and Excite™ 2x (Biogene Kimbolton, UK) mixes following optimisation. The amplification performances of both PCR reaction mixes were investigated on each of the three machines using the plasmid dilution series as target. The probe was added at a final concentration of 200 nM per reaction for all experiments. The Primers P1A and P1B were used at a final concentration of 1500 nM each in the Excite™ mix reactions whereas the FastStart mix reactions contained the primers at a final concentration of 500 nM each. A supplement of magnesium chloride (final concentration of 3 mM) was added to those reactions containing Excite™ mix that were run on the LightCycler^®^. In addition, magnesium chloride at a final concentration of 3 mM was required for all those reactions containing the FastStart mix except for the equivalent amplification reactions run on the LightCycler^®^, where a final concentration of 2 mM of magnesium chloride was optimal. A solution of BSA (BSA 100x; New England Biolabs, Hitchin, UK) at a final concentration of 0.1 mg/mL was added to the Excite mix for reactions run on the LightCycler^®^. To every reaction except the negative controls, 1 μL of plasmid DNA was added. Additional molecular biology grade water was added to a final volume of 20 μL and 25 μL for the FastStart and Excite™ mixes respectively. Reactions were set up in 20 μL glass capillaries, 0.1 mL plastic tubes and 96 well optical plates for use on the LightCycler^®^, Rotor Gene 3000™ and ABI PRISM^®^7700 respectively. Experimental Design ------------------- Six experiments (two per platform) were run, and repeated for three consecutive days. Three replicates of each plasmid DNA dilution plus three negative controls (24 reactions in total) were amplified per experiment. One specific analyst prepared the complete master-mixes containing primers, probes, water and other supplements as already outlined, in a pre-PCR \'clean\' room using the same pipettes each day. A second specific analyst added the plasmid DNA in a dedicated template addition area and used the same instruments for all template additions. Those amplification reactions containing the Excite™ 2x mix were run at the same time each morning on the Rotor Gene 3000™, ABI PRISM^®^7700 and LightCycler^®^and the reactions containing FastStart mix were run at the same time each afternoon on the Rotor Gene 3000™, ABI PRISM^®^7700 and LightCycler^®^. For each run the same instrument, rotor and carousel (where applicable) were employed to minimise intra-run variability. Baseline Settings & Data ------------------------ The Ct (threshold cycle number) was used as the measurand in order to assess the level of product in proportion to the level of starting target. As recommended by the manufacturers\' guidelines, the Ct values were calculated from the mid-point of the exponential phase of amplification. However, each real-time platform required some manual intervention to set analysis parameters. In addition specific analysis software was associated with each instrument and inherent differences in data manipulation may introduce further inter-platform variability. As amplification efficiency has been shown to change over the course of the reaction \[[@B29]\], we have based our analysis on the mid-point of the exponential phase to measure threshold cycles irrespective of platform, run or PCR reaction mix. Statistical Analysis -------------------- The results were interpreted and analysed according to the instrument manufacturer protocols. For each platform, graphs of cycle number versus arbitrary fluorescence units were generated and from this raw data, Ct values were automatically calculated by the software provided with each instrument \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. The statistical package Statistica 6.0 (Statsoft; Tulsa, OK, USA) was used to analyse the data for all the instrument runs by ANOVA (Analysis of Variance) and then repeatability estimates were carried out. Abbreviations ============= FAM: 6-carboxyfluoresceine TAMRA: 5-carboxytetramethylrhodamine BSA: bovine serum albumin Ct: threshold cycle number \%CV: Percentage coefficient of variance MS: Mean square Authors\' contributions ======================= CED carried out the experimental work and drafted the manuscript; FQ carried out the experimental work; MJB participated in the design of the study and performed the statistical analysis; MJH conceived of the study; JRB produced the plasmid material; AJW conceived and designed the study. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 **Raw Data Table**. Table 3 consists of the raw data results comparing the effect of day, plasmid copy number, PCR reaction mix and platform on the mean Ct value. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The work described in this study was supported under contract with the UK Department of Trade and Industry as part of the national measurement system valid analytical measurement (VAM) program. We would like to thank the Biotech Division at NIST, (USA) for providing us with the plasmid DNA. The authors wish to thank Jacquie Keer for editing and helpful discussion. In addition, David McDowell and Gavin Nixon are thanked for their help with primer and probe design. We would like to thank Corbett Research UK for the loan of the Rotor Gene 3000™ instrument. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Average threshold cycle number (Ct) plotted against plasmid copy number**. The data have been pooled from each platform, PCR reaction mix and day to give an average Ct value. Linear regression performed on the ploted data values from this graph determined that the gradient was -2.92 and that the amplification efficiency \[21\] was greater than two. However the amplification efficiency cannot be greater than two since PCR is an exponential reaction where the amount of product doubles at each cycle. ::: ![](1472-6750-5-15-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Average threshold cycle number (Ct) plotted against day for each plasmid dilution illustrating the interaction of day and plasmid copies**. The data have been pooled from each platform, PCR reaction mix and plasmid dilution. ::: ![](1472-6750-5-15-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Average threshold cycle number (Ct) plotted against day for each platform illustrating the interaction of day and platform**. The data have been pooled from each PCR reaction mix and plasmid dilution. ::: ![](1472-6750-5-15-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Comparison of repeatability estimates of the PCR measurements using different reaction mixes and platforms over three days (using standard deviations of the Ct value). ::: Repeatability Estimates (standard deviations of the Ct value) based on Individual Plasmid Copy Number ------------------------------------------------------------------------------------------------------- ------------------------------- ------------- ----------- ----------- -------- ----------- Plasmid Copy Number Platform and PCR Reaction Mix ABI 7700 LightCycler RotorGene Excite FastStart Excite FastStart Excite FastStart 10^2^ 0.543 0.182 0.568 0.233 0.888 0.186 10^3^ 0.452 0.204 1.177 0.190 0.184 0.445 10^4^ 0.223 0.305 1.123 0.100 0.256 0.592 10^5^ 0.243 0.168 0.947 0.224 0.209 0.221 10^6^ 0.362 0.183 1.378 0.281 0.772 0.333 10^7^ 0.473 0.189 0.919 0.156 0.319 0.247 10^8^ 0.236 0.184 0.454 0.352 0.322 0.166 Degrees of Freedom 6--8 6 6--8 5--6 6--8 6 Mean 0.362 0.202 0.938 0.219 0.421 0.313 Standard Deviation 0.130 0.047 0.331 0.082 0.286 0.156 (Note that one replicate is missing for the LC platform coupled with the Fast Start reaction mix -- the capillary broke). ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Variation attributable to day, plasmid copy number, PCR reaction mix and platform following an ANOVA test using threshold cycle number (Ct) as the variable factor. ::: **Univariate Tests of Significance for Ct value. Sigma-restricted parameterisation. Effective hypothesis decomposition.** ------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------- -------- -------- -------- ----------- **Effect** **SS** **DF** **MS** **F** **P** Day 96.0 2 48.0 221.4 \<0.00001 Plasmid Copies 12919.2 6 2153.2 9932.0 \<0.00001 PCR Reaction Mix 75.0 1 75.0 346.2 \<0.00001 Platform 700.5 2 350.2 1615.5 \<0.00001 Day with Plasmid Copies 53.2 12 4.4 20.5 \<0.00001 Day with PCR Reaction Mix 1.9 2 0.9 4.3 0.01442 Plasmid Copies with PCR Reaction Mix 8.5 6 1.4 6.5 \<0.00001 Day with Platform 130.8 4 32.7 150.8 \<0.00001 Plasmid Copies with Platform 51.0 12 4.2 19.6 \<0.00001 PCR Reaction Mix with Platform 1.7 2 0.9 4.0 0.01901 Day with Plasmid Copies with PCR Reaction Mix 14.6 12 1.2 5.6 \<0.00001 Day with Plasmid copies with Platform 18.3 24 0.8 3.5 \<0.00001 Day with PCR Reaction Mix with Platform 49.6 4 12.4 57.1 \<0.00001 Plasmid copies with PCR Reaction Mix with Platform 4.5 12 0.4 1.7 0.06337 Day with Plasmid Copies with PCR Reaction Mix with Platform 11.8 24 0.5 2.3 0.00091 Error 54.4 251 0.2 (Where SS = sum of squares; DF = degrees of freedom; MS = Mean Square; F = ratio of variances; p = probability of obtaining a specific result, given by the null hypothesis). :::

PubMed Central

2024-06-05T03:55:58.213023

2005-5-25

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168890/", "journal": "BMC Biotechnol. 2005 May 25; 5:15", "authors": [ { "first": "Carol E", "last": "Donald" }, { "first": "Fizza", "last": "Qureshi" }, { "first": "Malcolm J", "last": "Burns" }, { "first": "Marcia J", "last": "Holden" }, { "first": "Joseph R", "last": "Blasic" }, { "first": "Alison J", "last": "Woolford" } ] }

PMC1168891

Background ========== Gene therapy for treating cancers has been intensively investigated. Therapeutic gene transfer using viral or nonviral vectors has been shown to be clinically feasible and safe. However, the therapeutic outcome appears not to be high as expected, presumably because the in vivo gene delivery efficiency is low. In this sense, a more effective modality is needed to improve the therapeutic benefit. Combined transfer of genes with different mechanisms of action may potentate the therapeutic outcome. Indeed, combination cancer gene therapy using cytokine and suicide genes has been shown to be more effective than single gene transfer in animal models \[[@B1]-[@B9]\]. Interleukin (IL)-12 is a heterodimeric cytokine composed of 35- and 40-kDa subunits, which are covalently linked by a disulfide bond \[[@B10],[@B11]\]. IL-12 stimulates interferon-γ (IFN-γ) production by natural killer (NK) cells, enhances the cytolytic activities of NK cells and cytotoxic T lymphocytes (CTLs), stimulates the differentiation of Th1 cells and inhibits angiogenesis \[[@B12]-[@B14]\]. These biological activities have provided a rationale for the use of IL-12 in cancer immunotherapy. In fact, recombinant IL-12 (rIL-12) exhibits significant antitumor activity in various animal models \[[@B12]-[@B14]\]. Subsequent phase I and II studies on various cancer types using intravenous or subcutaneous administration of human rIL-12 \[[@B15],[@B16]\] showed some beneficial responses, but there were also adverse effects, which interrupted the clinical trials \[[@B17],[@B18]\]. Recently, it has been reported that intratumoral administration of rIL-12 shows dose-limiting toxicity and results in measurable locoregional immunological responses in head and neck squamous cell carcinoma patients \[[@B19]\]. Since intratumoral IL-12 gene transfer could minimize the adverse effects of systemic rIL-12 administration \[[@B17]\], the effectiveness against tumors of cell vaccines expressing IL-12, or viral or nonviral vectors encoding IL-12, has been evaluated in various animal models \[[@B18],[@B20]-[@B23]\]. The results indicate that, compared to IL-2, IL-4, interferons, and granulocyte-macrophage colony-stimulating factor, IL-12 is one of the most effective immunomodulators for cancer treatment \[[@B13],[@B18],[@B20],[@B21],[@B23]\]. Pilot clinical trials have been conducted in patients with advanced primary and metastatic liver cancer on the basis of these results \[[@B24]\]. Intratumorally administered adenoviral vectors expressing IL-12 had low efficacy but no cumulative toxicity \[[@B24]\]. Suicide gene therapy deploys genes such as cytosine deaminase (CD), herpes simplex or varicella zoster virus thymidine kinase (TK), or nitroreductase, all of which convert nontoxic prodrugs into cytotoxic agents \[[@B25]-[@B27]\]. Systemically administered prodrugs are converted to the active agents only in cells expressing these suicide genes, so the therapeutic effect is maximized while systemic toxicity is reduced. CD, which normally catalyzes the deamination of cytosine to uracil during RNA biosynthesis, is found in many bacteria and fungi but not in mammalian cells \[[@B28]-[@B30]\]. It also deaminates 5-fluorocytosine (5-FC) to 5-fluorouracil (5-FU), which is highly toxic to cells. 5-FU has been used clinically since the late 1950s to treat colorectal cancer \[[@B31]\]. A cellular metabolite of 5-FU, 5-fluorouridine triphosphate, is incorporated into RNA in place of UTP and consequently the multiple functions of RNA are impaired. Another metabolite of 5-FU, 5-fluorodeoxyuridine monophosphate, blocks thymidylate synthase activity and thus inhibits DNA synthesis. No currently available gene delivery system can deliver therapeutic genes including suicide genes to all the target cells *in vivo*. However, tumor cells expressing a suicide gene can kill neighboring tumor cells that do not express this gene by the by-stander effect, which could partly overcome the gene delivery problem \[[@B32]-[@B35]\]. A number of suicide gene therapies have been shown to be effective against various cancer cell lines in culture and in animal models. Subsequent phase I, II or III studies using CD or TK genes were performed in colorectal cancer patients with liver metastases and in brain tumor patients \[[@B36]-[@B47]\]. The antitumor activities appeared less than expected \[[@B46]\], but the results showed that suicide gene therapy is safe to apply in the oncology clinic \[[@B36]-[@B47]\]. A combination of IL-2 and TK genes exerted a more potent therapeutic outcome in a metastatic carcinoma model, and induced antitumor immunity \[[@B1]\]. Since then, combination gene therapies using various cytokine and suicide genes have been intensively investigated \[[@B2]-[@B9]\]. In the present study, we investigated the effectiveness of combined transfer of IL-12 and CD genes against renal cell carcinoma (Renca)-tumors in mice. Methods ======= Cells and mice -------------- HeLa, 293, Renca (a cell line spontaneously derived from BALB/c renal cell adenocarcinoma), and NK cell-sensitive YAC-1 cells were cultured in DMEM or RPMI-1640 medium (Gibco-BRL) supplemented with 10% fetal bovine serum (FBS) and penicillin/streptomycin. Cultures were maintained in a 5% CO~2~atmosphere at 37°C. Male or female wild-type BALB/c mice, 7--8 weeks of age, were purchased from the animal breeding laboratory at the Korea Research Institute of Bioscience and Biotechnology. They were kept (five mice per cage) in isolation under specific pathogen-free conditions, exposed to 12-h light/12-h dark cycles, and provided with standard feed and water *ad libitum*. Construction of replication-defective adenoviral vectors -------------------------------------------------------- cDNAs for *Esherichia coli*cytosine deaminase, p35 and/or p40 murine IL-12 subunits \[[@B48]\] or green fluorescent protein (GFP) were inserted by a standard cloning procedure into pX.dCMV.pA, an adeno-transfer plasmid harboring the immediate early gene promoter from human cytomegalovirus \[[@B49]-[@B51]\]. The shuttle plasmids were cotransfected into 293 cells together with pJM17, which harbors the genomic DNA of adenovirus 5 (Microbix, Canada), by a standard calcium phosphate method. Primary plaques were isolated and screened by PCR to ensure the presence of the inserts. Positive clones were plaque-purified twice and a single plaque was chosen for further amplification in 293 cells. Recombinant adenoviruses were purified by cesium chloride gradient centrifugation as previously described \[[@B49]-[@B51]\], dialyzed to remove the cesium chloride and frozen at -70°C.The viral titer was determined by a 293 plaque assay and expressed as plaque-forming units (pfu). Cell viability assay -------------------- HeLa cells were plated on 60-mm dishes at 5 × 10^5^cells/plate and mock-transduced, or transduced with Ad.CD, for 2 h at a multiplicity of infection (MOI) of 50 in serum-free DMEM. Cultures were continued for a further 12 h. Transduced and non-transduced cells were then mixed in different ratios and incubated in 24-well dishes for 24 h. The culture media were replaced with fresh media containing various concentrations of 5-FC, and incubation was continued for 4 days. Cell viability was assessed by the trypan-blue exclusion assay. Western blotting ---------------- HeLa cells were plated on 60-mm dishes at 5 × 10^5^cells/dish and transduced with recombinant adenoviruses in serum-free DMEM for 2 h. The culture media from mock- or virus-transduced cells were retained for measurement of released IL-12 at 48 h postinfection, and the cells were harvested in a lysis buffer (10 mM Tris-HCl, pH 7.5, 10 mM NaCl, 1.5 mM MgCl~2~, 10 mM β-mercaptoethanol). Cell extracts were prepared by three cycles of freezing (-70°C) and thawing and precleared by centrifugation. Fifty micrograms of precleared cell lysates were subjected to SDS-10% polyacrylamide gel electrophoresis (PAGE). The separated proteins were transferred to PVDF membranes, which were blocked with 3% bovine serum albumin (BSA) in phosphate-buffered saline (PBS) and incubated with rat anti-mouse IL-12 antibody (PharMingen), then with horseradish peroxidase-conjugated rabbit anti-rat lgG antibody (Sigma). Proteins were detected according to Amersham\'s ECL protocol. The culture supernatants were incubated with anti-IL-12 antibody. A suspension of protein G-Sepharose beads (100 μl) was added and the incubation mixtures were further incubated at 4°C for 12 h. The immunoprecipitates were washed three times with 40 mM N-2-hydroxyethylpiperazine-N\'-2-ethanesulfonic acid, pH 7.4, 100 mM KCl, 0.1% NP-40, 0.1 mM DTT, and once with PBS, then immunoblotted with the rat anti-mouse IL-12 antibody described above. Animal model and treatment -------------------------- Tumors were generated on the flanks of BALB/c mice by subcutaneous (s.c.) injection of 1.5 × 10^5^of Renca cells in 0.1 ml of PBS. After visible tumors had developed for 7--9 days after inoculation, the mice were injected once intratumorally with recombinant adenoviruses at 1 × 10^9^pfu in 10 mM Tris-HCl, pH 7.4, 1 mM MgCl~2~, 10% glycerol. Animals undergoing 5-fluorocytosine therapy received intraperitoneal administration of 5-FC (400 mg/kg) daily for 10 days. Tumors were measured prior to virus injection and subsequently at intervals of 3 days. Linear calipers were used to measure the longest diameter (a), width (b), and depth (c). The tumor size was calculated by the formula: (a × b × c/2). ELISA for IFN-γ --------------- ELISAs were performed using essentially the reagents and procedures described in the Endogen (USA) product literature. Flat-bottomed 96-well microtiter plates (Nunc Maxisorp) were coated with rat anti-IFN-γ antibodies (1 μg/ml) at 4°C overnight and blocked with 4% BSA in PBS for 2 h at 37°C. Renca cells were treated with mitomycin C at 50 μg/ml for 20 min at 37°C. Splenocytes (2 × 10^6^cells/ml) obtained from the tumor-bearing mice 21 days after virus treatment were stimulated with the inactivated Renca cells in U-bottomed 96-well plates at an effector-to-stimulator ratio of 10:1. Supernatants (50 μl) from 24 h and 48 h cultures were added to the ELISA plate wells and incubated for 1 h. After several washing steps, rat anti-mouse IFN-γ antibody conjugated with biotin was added. Color was developed using avidin-conjugated horseradish peroxidase and OPD substrate (Sigma). The reaction was terminated by the addition of 2N H~2~SO~4~. The absorbance was measured at 490 nm with an ELISA plate reader. The amounts of IFN-γ were quantified by interpolation of a standard curve generated using known amounts of standard mouse rIFN-γ (Endogen). The biological activity of the IL-12 produced from the recombinant adenoviruses was assessed on the basis of its ability to induce IFN-γ secretion from murine splenocytes. Murine splenocytes (1.5 × 10^6^cells/well) were washed with cold medium and incubated with the culture supernatants. IFN-γ in the incubation media was then determined by an ELISA, as described above. Cytotoxicity assay of NK cells ------------------------------ The cytolytic activity of NK cells was determined by a standard 4-h ^51^Cr-release assay. The spleens from dead mice obtained 21 days after tumor inoculation were removed aseptically and single-cell suspensions, prepared by passing the spleens through a metal mesh, were suspended in 5 ml of 0.83% ammonium chloride to lyse the red blood cells. The splenocytes were then used as effector cells to assay NK cell activity. YAC-1 cells (2 × 10^6^) in 0.5 ml of RPMI-1640 with 20% FBS were labeled with 50 μCi of Na^51^CrO~4~(Amersham) for 90 min. The labeled cells were washed three times with serum-free DMEM and mixed (1 × 10^4^per well) with the effector cells in a U-bottomed 96-well plate for 4 h at 37°C at three different effector-to-target ratios (100:1, 33:1, and 11:1), each in triplicate. The mixtures of target and effector cells were incubated for 4 h at 37°C. The percentage of specific ^51^Cr release was calculated as \[(cpm experimental release - cpm spontaneous release/cpm maximum release - cpm spontaneous release)\] × 100. The maximum ^51^Cr release was determined from the supernatant when labeled cells were lysed by adding 0.1 ml of 3% SDS. Spontaneous release was determined from the supernatants after labeled cells were incubated in 0.1 ml of medium without effector cells. The standard deviations for each triplicate sample and for spontaneous release were less than 10% and 20%, respectively. Statistics ---------- Statistical analysis was carried out using the Student\'s *t*test (unipolar, paired) and the chi-square test (two-tailed). The values were considered statistically significant when the *P*value was less than 0.05. Results and discussion ====================== Recombinant adenoviruses encoding the p35 and/or p40 subunits of murine IL-12 express bioactive IL-12 ----------------------------------------------------------------------------------------------------- We constructed replication-defective adenoviruses containing the p35 and/or p40 subunits of murine IL-12 under the transcriptional control of the immediate early gene promoter of human cytomegalovirus (Figure [1](#F1){ref-type="fig"}). Ad.IL-12 contains murine cDNAs of both IL-12 subunits, which are linked by an internal ribosome entry site sequence from encephalomyocarditis virus. To test whether Ad.p40 and Ad.IL-12 express the p40 subunit and the p35-p40 heterodimer respectively, we transduced HeLa cells with the two constructs at the indicated MOI. Replication-defective adenovirus encoding GFP (Ad.GFP) was used as a control. Lysates of cells transduced with Ad.p40 or Ad.GFP were resolved by SDS-PAGE followed by immunoblotting with anti-IL-12 antibody, which reacts specifically with the p40 subunit of mouse IL-12 and with free p40 (Figure [2A](#F2){ref-type="fig"}, left panel). Ad.p40 was found to express p40. The culture media from cells transduced with Ad.IL-12 or Ad.GFP were immunoprecipitated with anti-IL-12 antibody. The immuno-complexes were resolved by SDS-PAGE under nonreducing conditions and immunoblotted with anti-IL-12 antibody. The two protein bands detected might be a p40 homodimer (p80) and a p35-p40 heterodimer (p70) (Figure [2A](#F2){ref-type="fig"}, right panel, indicated by arrows). Under nonreducing conditions, Ad.IL-12 produced a doublet of 79/75 kDa that was sensitive to a reducing agent \[[@B52]\]. Ad.p40 produces a band migrating with an apparent molecular mass of approximately 90 kDa, which on a reducing gel reveals a band consistent with the p40 monomer \[[@B53]\]. Thus, our result appeared to be consistent with previous findings. We could not assess p35 expression because no antibody specific for this subunit was available. We next examined whether cells transduced with adenoviruses encoding the p35 and/or p40 subunits release bioactive IL-12. HeLa cells were mock-transduced, or transduced with a 1:1 (pfu ratio) mixture of Ad.p35 and Ad.p40, Ad.IL-12, or Ad.GFP at the indicated MOI, and the culture supernatants were collected after 24 h incubation. Splenocytes from naïve mice were incubated with equivalent volumes of the culture supernatants, and the IFN-γ contents of the medium were measured using ELISA (Figure [2B](#F2){ref-type="fig"} and [2C](#F2){ref-type="fig"}). The culture supernatants from cells transduced with a mixture of Ad.p35 and Ad.p40 or Ad.IL-12 exhibited much higher IFN-γ inducing-activity than the controls. These results suggest that cells transduced with Ad.p35 plus Ad.p40 or Ad.IL-12 release bioactive IL-12. Recombinant adenovirus encoding cytosine deaminase (Ad.CD) expresses catalytically active cytosine deaminase ------------------------------------------------------------------------------------------------------------ We constructed a replication-defective adenovirus encoding cytosine deaminase (Figure [1](#F1){ref-type="fig"}). To examine whether the Ad.CD virus expresses cytosine deaminase capable of converting 5-FC to cytotoxic 5-FU, we transduced HeLa cells with Ad.CD at an MOI of 50. Naïve HeLa or HeLa/CD cells, or mixtures of the two at the indicated ratios (Figure [3](#F3){ref-type="fig"}), were incubated in culture media without or with the indicated concentrations of 5-FC for 4 days, and cell viability was assessed by the trypan-blue exclusion assay (Figure [3](#F3){ref-type="fig"}). In the presence of 5-FC, the viability of the HeLa/CD cells decreased markedly but the naïve HeLa cells were unaffected, suggesting that cytosine deaminase was expressed in the virus-transduced cells and metabolized 5-FC to 5-FU. The viability of the mixtures of naïve and HeLa/CD cells appeared to decrease more than the proportion of naïve cells in the mixtures, particularly with 500 μg/ml 5-FC. Thus, the Ad.CD virus appeared to exert a bystander effect, though not a drastic one. The results suggest that Ad.CD expresses catalytically active cytosine deaminase. Transfer of the IL-12 and *cytosine deaminase*genes in combination, followed by 5-FC treatment, exerts potent antitumor activity in Renca-tumor bearing mice ------------------------------------------------------------------------------------------------------------------------------------------------------------ Renca-tumor bearing mice were established as described in the Methods section, randomly divided into 4 groups, and injected once intratumorally with Ad.GFP, Ad.CD/5-FC, Ad.p35 plus Ad.p40, or Ad.p35 plus Ad.p40 plus Ad.CD. Mice receiving Ad.CD were treated with 5-FC. Tumor growth was measured at the indicated times (Figure [4A](#F4){ref-type="fig"}). Mouse survival until day 42 was recorded (Figure [4B](#F4){ref-type="fig"}). The treatment outcomes are summarized in Table [1](#T1){ref-type="table"}. Inhibition of tumor growth was most marked in mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC (*p*\< 0.001 for Ad.p35 plus Ad.p40 plus Ad.CD/5-FC vs Ad.GFP and *p*\< 0.0053 for Ad.p35 plus Ad.p40 plus Ad.CD/5-FC vs Ad.CD/5-FC). The mean survival times were for 33.2 ± 10.4 days for Ad.p35 plus Ad.p40 plus Ad.CD/5-FC group and 27.2 ± 10.9 days for Ad.p35 plus Ad.p40 group; the difference is statistically significant (*p*\< 0.0018). Among the 15 mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC, four became tumor-free and four exhibited partial responses (Table [1](#T1){ref-type="table"}). Among the 15 mice treated with Ad.p35 plus Ad.p40, one became tumor-free and two showed partial responses (*p*\< 0.0464 for comparison of the two treatments). Ad.p35 plus Ad.p40 treatment appeared to be more effective than Ad.CD/5-FC, which had a modest antitumor effect compared to the control (Ad.GFP). IL-12 and *cytosine deaminase*genes in combination, followed by 5-FC treatment, enhance the cytolytic activity of NK cells and increase IFN-γ production -------------------------------------------------------------------------------------------------------------------------------------------------------- We examined whether the increased antitumor activity of Ad.p35 plus Ad.p40 plus Ad.CD/5-FC (Figure [4](#F4){ref-type="fig"}) was associated with the induction of tumor-specific CTLs, activation of NK cells, and induction of Th1 cytokines such as IL-2 and IFN-γ. Effector cells were prepared from the splenocytes of tumor-bearing mice treated with the recombinant adenoviruses. Renca cells labeled with ^51^Cr were used as target cells. A standard 4-h ^51^chromium release assay was carried out. However, we did not detect specific lysis of the labeled Renca cells (data not shown). We performed an NK cell assay using the splenocytes as effector cells and YAC-1 cells labeled with ^51^Cr as target cells (Figure [5A](#F5){ref-type="fig"}). The effector cells from mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC lysed the target cells about 10- and 5-fold more effectively than Ad.CD/5-FC and Ad.GFP, respectively, at a 100:1 effector-to-target cell ratio. Although Ad.p35 plus Ad.p40 treatment also increased NK cell activity compared to Ad.CD/5-FC or Ad.GFP, the cytolytic activity of splenic NK cells from mice treated with this mixture was lower than that from mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC. The effector cells used in the NK cell assay were mainly splenocytes from mice in which the tumors were completely regressed or small. This may account for the increased NK cell activity in the Ad.p35 plus Ad.p40 plus Ad.CD/5-FC treatment group compared to the Ad.p35 plus Ad.p40 treatment group. We next examined whether the virus treatments induced Th1 cytokines, which may represent innate and/or adaptive immune responses. Splenocytes from tumor-bearing mice treated with the viruses were incubated with inactivated Renca cells. The IL-2 and IFN-γ contents of the culture supernatants were measured by ELISA 24 and 48 h after incubation, respectively. We detected no IL-2 induction in splenocytes from any of the treatment groups (data not shown). However, IFN-γ production was about twice as great in splenocytes from mice treated with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC as in those treated with Ad.CD/5-FC or Ad.GFP (Figure [5B](#F5){ref-type="fig"}, *p*\< 0.0179 and \< 0.0394 for Ad.p35 + Ad.p40 + Ad.CD/5-FC vs Ad.GFP and Ad.CD/5-FC, respectively). Ad.p35 plus Ad.p40 treatment also increased IFN-γ, but its effectiveness was significantly less than that of Ad.p35 plus Ad.p40 plus Ad.CD/5-FC (*p*\< 0.0076). The results suggest that Ad.p35 plus Ad.p40 treatment, but not Ad.CD/5-FC, may enhance NK cell activity and stimulate IFN-γ production. Thus, the enhancement of NK cell activity may be partially responsible for the inhibition of tumor growth by treatment with Ad.p35 plus Ad.p40 plus Ad.CD/5-FC or with Ad.p35 plus Ad.p40. The rapid tumor cell killing effect induced by suicide gene products induces a strong antitumor immune response, such as induction of tumor-specific CTLs and increased NK cell activity \[[@B54]-[@B57]\]. However, we detected neither a tumor-specific T cell immune response nor an enhanced NK cell activity after Ad.CD/5-FC treatment alone. It is possible that the antitumor activity of the Ad.CD constructed here was so weak that the tumor cells were not killed rapidly (Figure [4](#F4){ref-type="fig"}). However, this is unlikely, because Ad.CD/5-FC treatment increased the antitumor activity of Ad.IL-12 (see Figure [6B](#F6){ref-type="fig"} and [6C](#F6){ref-type="fig"}). Alternatively, the bystander effect, which kills neighboring tumor cells not expressing suicide genes, may have exerted a negative effect on the generation of tumor-specific CTL and/or active NK cells. Immune cells or antigen-presenting cells located close to the tumor cells could have taken up, and been killed by, the toxic metabolites generated by the suicide gene products. This may explain why the NK cell activity in Ad.CD-treated mice was significantly lower than in Ad.GFP-treated mice (Figure [5A](#F5){ref-type="fig"}, *p*\< 0.001). Transfer of the *cytosine deaminase*gene followed by 5-FC treatment increases the antitumor effectiveness of IL-12 gene transfer in Renca-tumor bearing mice ------------------------------------------------------------------------------------------------------------------------------------------------------------ Since a mixture of Ad.p35 and Ad.p40 was used in previous experiments (Figure [4](#F4){ref-type="fig"}), we examined whether Ad.IL-12, containing p35 and p40 in a single adenoviral vector (Figure [1](#F1){ref-type="fig"}), was more effective against the tumors when combined with Ad.CD/5-FC. Renca-tumor bearing mice were divided randomly into four groups, treated intratumorally with (1) Ad.IL-12 plus Ad.CD/5-FC, (2) Ad.IL-12, (3) Ad.CD/5-FC, or (4) Ad.GFP (control). Tumor sizes were measured at the indicated times (Figure [6A](#F6){ref-type="fig"}). There was a significant inhibition of tumor growth in mice treated with Ad.IL-12 plus Ad.CD/5-FC compared to Ad.GFP (*p*\< 0.0456). The effectiveness of Ad.IL-12 (5 × 10^8^pfu) plus Ad.CD/5-FC (5 × 10^8^pfu) appeared to be similar to that of Ad.IL-12 (1 × 10^9^pfu) alone. This contrasts with the effectiveness of Ad.p35 plus Ad.p40 plus Ad.CD/5-FC as compared to Ad.p35 plus Ad.p40 (Figure [4](#F4){ref-type="fig"}). Presumably the in vivo gene delivery of the Ad.p35-Ad.p40 mixture was not so efficient as that of Ad.IL-12. We further examined whether CD gene transfer followed by 5-FC treatment increased the antitumor effectiveness of IL-12 gene transfer. Sixteen Renca tumor-bearing mice were divided into 2 groups. Since a higher dose than 1 × 10^9^pfu of adenoviral vector appeared to be toxic to the mice, the total dose given was 1 × 10^9^pfu. Mice in each group received Ad.IL-12 (5 × 10^8^pfu) with either Ad.CD (5 × 10^8^pfu) or Ad.GFP (5 × 10^8^pfu) as a control. Both groups were treated with 5-FC for 10 days. Tumor sizes were measured at the indicated times (Figure [6B](#F6){ref-type="fig"}). Combined transfer of the IL-12 and CD/5-FC genes resulted in significant inhibition of tumor growth compared to IL-12 gene transfer alone (*p*\< 0.0301). Six of the 8 mice treated with Ad.IL-12 plus Ad.CD/5-FC, but only 4 of the 8 mice treated with Ad.IL-12 plus Ad.GFP/5-FC, became tumor-free and survived at 88 days. The mean survival times were 80.5 ± 20.8 days for the Ad.IL-12 + Ad.CD/5-FC group and 67.3 ± 28.3 days for the Ad.IL-12 + Ad.GFP/5-FC group; the difference is statistically significant (*p*\< 0.0368). Conclusion ========== Among various cancer gene therapies, immunogene therapy is a promising but challenging modality. Recent clinical trials using Ad.IL-12 have shown that intratumoral IL-12 gene transfer is a feasible and well-tolerated procedure but exerts only mild antitumor effects \[[@B24]\]. An increased Ad.IL-12 dose may increase the antitumor efficacy but might entail a systemic, toxic effect that would limit the use of Ad.IL-12 alone. In this study we have attempted to improve the antitumor efficacy in Renca tumor-bearing mice by using adenoviral vectors to transfer the IL-12 gene in combination with *cytosine deaminase*, followed by 5-FC treatment. The results show a marked inhibition of tumor growth and significant prolongation of survival time, suggesting that IL-12 plus CD gene transfer followed by 5-FC treatment may be an alternative modality for the human cancer treatment. Abbreviations ============= Ad: adenovirus; IL-12: interleukin-12; CD: cytosine deaminase; 5-FC: 5-fluorocytosine; DMEM: Dulbecco\'s Modified Eagle\'s Medium. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Cho W-K and Yoo J constructed the adenoviral vectors and analyzed the functionality of the expressed gene products. Hwang K-S, Yun H-J, and Kim S performed the tumor growth studies in the animal model, and the immunological assays with murine splenocytes. Im DS drafted the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/51/prepub> Acknowledgements ================ We are grateful to Yong J. Lee at University of Pittsburgh School of Medicine, Pittsburgh, PA, USA, for providing us with *cytosine deaminase*. This study was supported by a grant from the KRIBB Initiative Program, a grant from the 21C FrontierFunctional Genome Project from the Ministry of Science and Technology, Republic of Korea, and a grant from Samyang Genex Biotech Research Institute. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Schematic illustration of replication-defective adenoviral vectors constructed in this study. The E1 region of adenovirus type 5 (Ad 5) was replaced by the indicated genes under the transcriptional control of the immediate early gene promoter from human cytomegalovirus (pCMV). IRES; internal ribosomal entry sequence. ::: ![](1471-2407-5-51-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The recombinant adenoviruses encoding murine p35 and/or p40 subunits express biologically active IL-12. (A) HeLa cells were transduced with Ad.GFP, Ad.p40, or Ad.IL-12 at the indicated MOIs. Cells or culture media were harvested 48 h after transduction. Cleared lysates of cells transduced with Ad.GFP or Ad.p40 were immunoblotted with anti-IL-12 antibody (left panel). The culture media from HeLa cells transduced with Ad.IL-12 or Ad.GFP were immunoprecipitated with anti-IL-12 antibody. The immunoprecipitates were dissolved in SDS sample buffer without reducing agent and subjected to SDS-PAGE, followed by immunoblotting with anti-IL-12 antibody (right panel). Molecular size markers in kilodaltons are shown at the left. The positions of p40, p35--p40 heterodimer (p70) and p40 homodimer (p80) are indicated by arrows. (B and C) HeLa cells were mock-transduced or transduced with either a 1:1 (pfu ratio) mixture of Ad.p35 and Ad.p40, or Ad.IL-12, or Ad.GFP, at the indicated MOI. Splenocytes from naïve mice were incubated with the culture supernatants for 24 h. ELISA was performed to determine the IFN-γ levels after incubation. Data are means ± SD of two independent experiments, each performed in duplicate. ::: ![](1471-2407-5-51-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Ad.CD virus expresses functional *cytosine deaminase*. HeLa cells were transduced with Ad.CD at an MOI of 50 and incubated for 12 h. These cells (HeLa/CD) were then mixed with naïve HeLa cells at the indicated ratios. The HeLa/CD, HeLa, or mixed cells were incubated on 24-well dishes for 24 h, then treated with 5-FC at the indicated concentrations for 4 days. Cell viability was assessed by the trypan-blue exclusion assay. Data are means ± SD of two independent experiments, each performed in duplicate. ::: ![](1471-2407-5-51-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The antitumor activity effected by IL-12 gene transfer increases *in vivo*when combined with *cytosine deaminase*/5-FC. (A) Renca tumors were established in BALB/c mice by s.c. inoculation of 1.5 × 10^5^cells. When the tumors reached 5--6 mm diameter, the animals were randomly divided into four treatment groups (n = 15 per group) and injected once intratumorally with (1) Ad.p35 (2.5 × 10^8^pfu) plus Ad.p40 (2.5 × 10^8^pfu) plus Ad.CD (5 × 10^8^pfu), or (2) Ad.p35 (5 × 10^8^pfu) and Ad.p40 (5 × 10^8^pfu), or (3) Ad.CD (1 × 10^9^pfu), or (4) Ad.GFP (1 × 10^9^pfu). Mice treated with Ad.CD were given 5-FC (400 mg/kg) intraperitoneally daily for 10 days. Graphic representations of mean tumor growth rates are shown. Bars indicate SD. (B) Mouse survival until day 42 was recorded. Mice were sacrificed and considered as death when tumor size exceeded 1.5 cm in long and short axes. The mean survival times (± SD) were 33.2 ± 10.4 days for (1) group, 27.2 ± 10.9 days for (2) group, 21.9 ± 9.1 days for (3) group, and 19.7 ± 7.1 days for (4) group. The difference between (1) and (2) groups was statistically significant (*p*\< 0.0018) by Student\'s *t*test. The treatment outcome is summarized in Table 1. One animal in each of the groups treated with Ad.CD and Ad.GFP succumbed to toxicity, and two mice in the group treated with Ad.p35 plus Ad.p40 were excluded owing to peritoneal metastasis. ::: ![](1471-2407-5-51-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### IL-12 gene transfer combined with Ad.CD/5-FC increases splenic NK cell activity and IFN-γ production. (A) Cytotoxicity of NK cells in Renca tumor-bearing mice treated with various adenoviral vectors (see Figure 4 legend). Mice in which tumor growth was completely or partially inhibited were chosen from the treatment groups on day 21 after tumor cell inoculation. Splenocytes were isolated from the treated mice (n = 3 per group) and used as effector cells. A standard 4-h ^51^Cr release assay was performed using YAC-1 cells labeled with ^51^Cr as target cells. The percentage specific lysis was determined. Data are means ± SD. (B) Mouse splenocytes (n = 3 per group) were isolated from the treatment groups on day 21 as described above, and incubated with Renca cells inactivated by pretreatment with 50 μg/ml mitomycin C for 20 min at 37°C. Culture supernatants were obtained from the incubated cells at the indicated times. Amounts of IFN-γ in the supernatants were measured using ELISA. Bars represent mean ± SD. ::: ![](1471-2407-5-51-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### CD gene transfer followed by 5-FC treatment increases the antitumor effectiveness of IL-12 gene transfer. (A) Mice with tumors were established by subcutaneous inoculation of 1.5 × 10^5^cells of Renca. The experimental animals in which tumors reached a diameter of about 5 mm were randomly divided into 4 treatment groups (n = 7 per group) and injected once intratumorally (day 0) with Ad.IL-12 (5 × 10^8^pfu) plus Ad.CD (5 × 10^8^pfu), or 1 × 10^9^pfu of Ad.IL-12, or Ad.CD, or Ad.GFP. Mice treated with Ad.CD were given intraperitoneal 5-FC (400 mg/kg) daily for 10 days. Tumor growth was measured at intervals of three days until day 21. Graphic representations of mean tumor growth rates are shown. (B) and (C) Renca tumor-bearing mice were divided into two groups (n = 8 per group), which received single intratumorally injections (day 0) of Ad.IL-12 (5 × 10^8^pfu) plus Ad.CD (5 × 10^8^pfu) or Ad.IL-12 (5 × 10^8^pfu) plus Ad.GFP (5 × 10^8^pfu). Mice in both groups were given intraperitoneal 5-FC (400 mg/kg) daily for 10 days. Tumor growth was measured at the indicated times until day 25. Graphic representations of mean tumor growth rates are shown (B). Mouse survival until day 88 was recorded (C). The mean survival times (± SD) were 80.5 ± 20.8 days for the Ad.IL-12 plus Ad.CD/5-FC group and 67.3 ± 28.3 days for the Ad.IL-12 plus Ad.GFP/5-FC group. The difference between these mean survival times was statistically significant (*p*\< 0.0368) by Student\'s *t*test. ::: ![](1471-2407-5-51-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Tumor regression following intratumoral injection of Ad.p35 plus Ad.p40, or Ad.CD/5-FC alone, or a combination of both ::: Treatment Response (%) ------------------------------ -------------- -------- -------- --------- --------- Ad.p35 + Ad.p40 + Ad.CD/5-FC 15 4 (27) 4 (27) 0 (0) 7 (46) Ad.p35 + Ad.p40 15 1 (7) 2 (14) ‡2 (13) 10 (66) Ad.CD/5-FC 15 1 (7) 0 (0) †1 (7) 13 (86) Ad.GFP 15 0 (0) 0 (0) †1 (7) 14 (93) *N*= number of Renca tumor-bearing mice injected once intratumorally with adenoviral vectors at the doses indicated in Figure 4 legend. \"Complete\" refers to animals undergoing total and permanent regression of the injected tumor. \"Partial\" refers to animals undergoing partial tumor regression followed by regrowth, or a significant delay in growth \"Death or peritoneal metastasis\" refers to animals that died or had peritoneal metastasis during the period of investigation. None refers to animals whose tumors have continued to grow at a rate comparable to controls. †Death probably by toxicity. ‡Excluded due to peritoneal metastasis during the experiment. :::

PubMed Central

2024-06-05T03:55:58.217180

2005-5-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168891/", "journal": "BMC Cancer. 2005 May 24; 5:51", "authors": [ { "first": "Kyung-Sun", "last": "Hwang" }, { "first": "Won-Kyung", "last": "Cho" }, { "first": "Jinsang", "last": "Yoo" }, { "first": "Hwan-Jung", "last": "Yun" }, { "first": "Samyong", "last": "Kim" }, { "first": "Dong-Soo", "last": "Im" } ] }

PMC1168892

Background ========== Comparison of intraspecies multi-strain bacterial genome sequence data has shown that, even over short evolutionary time scales, genome evolution is dominated by gene insertions/deletions and gene divergence \[[@B1]-[@B4]\]. Genome levels of intraspecies genetic diversity must be examined if we are to gain a better understanding of genome evolution \[[@B5]\] and if we are to maximize the practical use of bacterial genome sequence information, for instance for development of technical applications, e.g., vaccine or drug development. One of the aims of bacterial intraspecies comparative genomics is to determine the overall genetic similarity between strains. Where sequence information is available, this type of analysis relies heavily on sequence homology and centres on the determination of conserved genes, strain-specific (i.e. unique) genes and, where the sequence provides unambiguous evidence, determination of orthologous and paralogous genes \[[@B6]-[@B9]\]. Although it has become increasingly apparent that obtaining the sequence of multiple strains per species is highly desirable, currently these types of datasets are limited in number. In their absence, other methods for performing comparative genomics have been developed. Among them, microarray-based comparative genomic hybridization (M-CGH) based on genome-sequenced strains has shown enormous potential \[[@B10]-[@B12]\]. Two different microarray-based approaches have been used to study the genetic composition of unknown bacterial strains. In the first approach, a control genome-sequenced strain was used as a reference to generate the probes for a microarray \[[@B13]-[@B16]\]. In the second approach, microarray probes were derived from the tester strain, either from a tester-derived shotgun library or a library enriched for tester-specific DNAs \[[@B17]\]. With either approach, control- and tester-derived targets are co-hybridized to the microarray and control- and tester-derived signals are compared, often by computing the Log Ratio (LR) = log~2~(tester signal/control signal). Whereas genes with similar signal in either channel are expected to have LRs near zero, genes with LRs that deviate significantly from LR = 0 are likely to show copy number changes or sequence divergence between control and tester strains. The relatively small number of studies on bacterial M-CGH has demonstrated the power of the method in a comparative genomics context despite a lack of consensus in current methods for analyzing M-CGH data. Although potential methods for standardizing and improving analysis have been suggested \[[@B15],[@B18]\] in practice, M-CGH data has routinely been analyzed by categorizing genes into two groups: genes that are likely to be conserved and genes that are likely to be divergent. One notable problem with this approach is that no attempt is made to differentiate between gene divergence and gene absence, despite the significant biological and evolutionary differences implied by these two types of events. A framework for improved analysis would require empirical data on the relationship between Log Ratio (LR) from M-CGH experiments and sequence conservation levels, however, to our knowledge no studies exist that have directly examined this question. The availability of intraspecies genome data from two strains of *Campylobacter jejuni*\[[@B19],[@B20]\], has provided us with the opportunity to examine the quantitative relationship between the LR and probe/target identity (PTI) using our *C. jejuni*microarray. This experimental design allows us to directly match microarray results to the *a priori*interpretation of gene divergence and gene absence patterns. The goal of this study is to define the analytical parameters for the accurate prediction of gene conservation levels, leading to improved interpretation of M-CGH data. We present here the results of a detailed analysis of M-CGH experiments using the two genome-sequenced strains of *C. jejuni*. Results and discussion ====================== Determination of technical variation in M-CGH experiments --------------------------------------------------------- In order to examine the Log Ratio (LR) distributions where no differential signals are expected, we performed control experiments in which dual-labelled NCTC 11168 (or RM1221) DNA was tested in a series of self-self M-CGH experiments. Although our microarray is based on strain NCTC 11168 the resulting LRs should remain close to zero because, regardless of the strain used in the self-self experiment, \"Control\" and \"Tester\" targets are identical. Thus, any observed deviation from this result is likely due to technical variability in the assay and can be used to determine a threshold for statistically significant differential signals (i.e., outliers). The LR distribution in six replicates follows a normal distribution with a mean LR for six replicates of \~0.01 ± 0.22 (Figure [1](#F1){ref-type="fig"}) and, as expected, the mean and standard deviation of the various replicates were uniform regardless of the strain used. The variances observed were due to stochastic differences in the competitive hybridization of targets to the probes on the microarray and a good estimate of the technical variation in our experimental platform. Based on this data a LR = ± 1.0, used by many experimenters to identify divergent or deleted genes in similar M-CGH studies, represents a conservative threshold for divergent gene detection, since genes in which tester and control sequences are identical have a probability of less than 3.0 × 10^-6^of having a LR greater than 1.0 ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Log Ratio distribution of self-self experiments**. The LR distribution of self-self experiments was used to determine the level of experimental variability in our experimental platform. Standardized samples of genomic DNA from *C. jejuni*NCTC 11168 (or RM1221) labelled with Cy3 and Cy5 were co-hybridized to our microarray. Results from six replicates had mean LR = 0.01 with an average SD of 0.215. Because samples from the same genomic DNA were used in both channels, LRs were expected to remain close to 0 and any deviations could be attributable to experimental error. ::: ![](1471-2164-6-78-1) ::: Analysis of the Log Ratio distribution of highly conserved genes ---------------------------------------------------------------- We analyzed data from a set of M-CGH experiments comparing strain NCTC 11168 (Control) with strain RM1221 (Tester). Because the probes in our microarray were PCR-amplified from the Control strain, Control targets should have 100% probe/target identity (PTI) with the probes on the microarray, and the LR values observed should be a function of the PTI between Tester targets and the NCTC 11168-derived microarray probes. The LR distribution of genes with 100% identity between NCTC 11168 and RM1221 (n = 114) would be expected to behave much like that of self-self experiments because in both cases Control and Tester targets are identical and thus have 100% PTI. This was found to be the case although the distribution of genes with 100% PTI had larger standard deviation (σ = 0.28) than that of self-self experiments (σ = 0.21) (Figure [2](#F2){ref-type="fig"}). Genes for which the RM1221 sequence had less than 100% sequence identity with NCTC 11168 would be expected to yield LRs that deviate from 0 due to the decreased hybridization of targets that are imperfectly matched to probes on the microarray. We examined the behaviour of genes with high levels of PTI in order to determine the level of sequence divergence that would have an observable effect on LRs. We found that whereas genes with greater than 99% PTI had LR distributions that were nearly indistinguishable from those from self-self experiments, genes with as little as 2% sequence divergence (i.e., 98% PTI and below) deviated from the LR distribution of genes with 100% PTI. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Log Ratio distributions of highly conserved genes**. LR distributions from a series of M-CGH experiments comparing two genome-sequenced strains of *C. jejuni*(NCTC 11168 vs. RM1221). Genes were binned according to PTI and the LR distributions of bins with greater than 95% PTI are presented here. Because the microarray was designed based on strain NCTC 11168, LR deviations from 0 would be the result of sequence divergence or gene absence in strain RM1221. The LR distributions of genes with greater than 99% PTI do not deviate significantly from the average distribution of a Self-self experiment whereas increasingly larger deviations are observed in the range from 98 to 95% PTI. ::: ![](1471-2164-6-78-2) ::: Analysis of the relationship between PTI and Log Ratio ------------------------------------------------------ In order to examine the relationship between % PTI and LR in greater detail, we plotted the mean LR of genes according to their % PTI (Figure [3](#F3){ref-type="fig"}). As shown previously, the LR distribution of genes with greater than 99% PTI were similar. However, in lower PTI ranges a small yet noticeable decrease in average LR was observed. Although the small number of genes with less than 93% identity makes it difficult to obtain meaningful LR trends because of high variance, decreasing PTI still led to increasingly negative LRs. One caveat of these observations is that the LR of individual data points within a given PTI range show sufficient variability to make PTI predictions based on LR values potentially inaccurate across most of the range of PTIs. For example, although the difference in mean LR of genes with 95% PTI and 96% PTI is 0.12, their standard deviations are 0.42 and 0.39, respectively. Thus, although the average LR decreased with decreasing PTI, there is considerable overlap between the distributions. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Relationship between Log Ratio and PTI**. We plotted the average LR of genes with varying levels of % PTI from M-CGH experiments comparing two genome-sequenced strains of *C. jejuni*(NCTC 11168 vs. RM1221). Genes were binned according to PTI and the LR distributions of individual bins are presented here. The number of observations made within each PTI bin is shown in the lower axis. As seen in Figure 2, the average LR of genes with high levels of PTI are very similar, although a noticeable decrease in average LR is observed even at 98% PTI. Although the average LR becomes increasingly negative as PTI levels drop, given the SD observed within each group PTI category, there is considerable overlap between categories. ::: ![](1471-2164-6-78-3) ::: Analysis of the Log Ratio distribution of absent genes ------------------------------------------------------ Genes that are absent in RM1221 (ie. 0% PTI) should have highly negative LRs because they should yield detectable hybridization signal in the Control channel coupled with a lack of signal in the Tester channel. Although the LR distribution of genes with 0% PTI was shifted towards the left (Mean = -2.34 ± 1.35), it also appeared to be bimodal, with a number of genes with higher than expected LR. When these genes were examined more closely, a common feature was a short microarray probe size (\< 250 bp). We plotted the LR distribution of genes with probe sizes \<250 bp and \>250 bp separately, and found a significant difference in their respective LR distributions (Figure [4](#F4){ref-type="fig"}). Whereas the LR distribution of the former was -0.93 ± 0.82, the LR distribution for longer genes was -2.94 ± 1.04. This \"dampening\" in LR amplitude appears to be largely the effect of an overall diminished signal for short genes (results not shown), possibly due to a difference in hybridization kinetics or hybrid stability under the hybridization and washing conditions used. The effect of decreased signal is that of decreased dynamic range because a lower signal in the control channel restricts the amplitude of the LR that can be measured. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Log Ratio distribution of genes absent from strain RM1221**. We plotted the LR distribution of genes predicted to be absent from strain RM1221 based on BLAST searches of NCTC 11168 against the RM1221 genome. Because of the lack of Tester signal predicted from these genes, LRs should be expected to be highly negative. The LR distribution (A) appeared to be bimodal with a significant number of genes bearing unusually high LRs. Further examination of these genes revealed a potential bias towards genes represented by short probes on the microarray (i.e. less than 250 bp). Separate re-plotting of the LR distributions of long (B) and short (C) probes confirms a higher average LR among genes with short probes. ::: ![](1471-2164-6-78-4) ::: Determination of thresholds for highly conserved and absent genes ----------------------------------------------------------------- One of our goals for this analysis was to determine whether the observed trends would enable us to predict the PTI in M-CGH experiments based on LR alone. Although the levels of technical variability mask the subtle effect that low levels of sequence divergence have on LR, the LR distributions at the two PTI extremes (\>98% and 0%), which correspond to highly conserved and absent genes, show very little overlap. This enabled us to establish thresholds that, with high confidence, can be used to predict absent and highly conserved genes in M-CGH data (Figure [5](#F5){ref-type="fig"}). After removal of genes with short amplicon-based probes from our analysis, we established that less than 1% of the observed LRs \> -0.8 originated from absent genes (31 of n = 7910). Similarly, less than 1% of observed LRs \< -3.0 originated from conserved genes (6 of n = 636) with all false-positive observations stemming from the *pyrC*gene, which has a PTI of only 81.2%. Of 808 observed LR measurements in the range between -3.0 and -0.8, only 221 (27.3%) originated from genes with greater than 90% PTI. Although, based on our empirical data, LR values that fall between these two thresholds are likely to be from either absent or significantly divergent genes and unlikely to be from highly conserved genes, there is significant overlap between LR distributions of absent and divergent genes. At LRs ≅ -1.4, an observation has a nearly equal likelihood of stemming from an absent gene as it does from a present gene and thus the two classes cannot be distinguished in this LR range. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Determination of thresholds for absent and conserved genes**. We calculated the proportion of genes belonging to each of four PTI categories at 0.2 LR intervals in order to determine LR thresholds that could be used to predict absent and conserved genes with a high degree of certainty. Below a LR of -3.0, the false positive rate for conserved genes is less than 1%; similarly, the false positive rate for absent genes above LRs of -0.8 is also less than 1%. In the LR interval between -3.0 and -0.8, particularly approaching the -0.8 boundary, there are significant number of genes from more than one PTI category and thus there is significant risk of misclassification. ::: ![](1471-2164-6-78-5) ::: Conclusion ========== Microarray analysis, whether in the context of gene expression or M-CGH studies, is based on determining which genes have statistically significant differential hybridization signal between two samples. In M-CGH analysis, these differential signals are the result of sequence divergence or differences in copy number. Two critical issues rise to the forefront in M-CGH analysis: a) does a gene show genuine differential signal (i.e. outside the norms of variability due to experimental error); b) what is the nature of the event that gave rise to the differential signal (i.e. sequence divergence, copy number change)? Because M-CGH generates hybridization data as a proxy for sequence similarity data, it is important that it be analyzed as such. While some empirical work has been carried out on probe/target identity (PTI) and data analysis using the microarray platform, the focus has largely been on optimization of species detection and/or identification in complex samples \[[@B21]-[@B23]\]. In these applications, the primary goal is that of optimizing probe sets and hybridization conditions to maximize the specificity of species-specific probe/target interactions, possibly at the expense of decreased assay sensitivity and thus the majority of microarrays used for species identification are oligonucleotide-based. By contrast, in comparative genomics, the primary goal is that of gene detection for the purpose of characterizing gene content, and thus the focus must shift to detection sensitivity in order to minimize the likelihood of false positive calls on gene absence events. Because oligonucleotide-based arrays can lead to erroneous gene absence calls \[[@B24]\], the majority of M-CGH studies have used amplicon-based microarrays, which are more sensitive albeit at the expense of specificity \[[@B25]\]. A common thread among bacterial M-CGH studies has been the grouping of all outliers into a single category. Currently it is unclear whether divergent and absent genes can be distinguished based on LR data alone. Although the lack of distinction between these types of events does not negate the results from these studies, it can potentially restrict further analysis of the data. For example, in any pair of intraspecies genomes, sequence similarity can be used to define genes absent in one or the other strain as well as genes that are conserved in both strains. Although the \"biological interpretation\" in the case of gene absence is unambiguous, many possibilities arise when sequences share any level of similarity. For instance, single nucleotide substitutions can lead to truncated or inactive gene products. Additionally, the level of sequence similarity required for full functional homology varies from gene to gene, increasing the complexity of the analysis even when DNA sequences are directly available. The inexact nature of hybridization analysis further compounds the difficulty in interpreting signal from divergent genes by M-CGH, and thus focusing on conserved and divergent genes ignores the increased reliability of gene absence calls. In previous work, we presented data suggesting that highly negative LR values were consistent with gene deletion events, paving the way for making the distinction between divergent and absent genes based on LR data \[[@B16]\]. When M-CGH data is analyzed such that gene absence events are grouped together with all other gene divergence events (i.e. as a continuous character model), it represents a significant loss of information both from a technical and from a biological point of view. In addition to the greater ambiguity in data interpretation as LRs approach the threshold for gene conservation, the continuous character approach negates the functional distinction that can be made between gene absence and gene divergence events. Because the LR thresholds described here could be used to reliably predict gene absence and gene conservation, it would be advantageous to focus the analysis of M-CGH data on the accurate detection of conserved and absent genes. While the data between the two thresholds should not be altogether discarded, the two thresholds represent boundaries defining regions in which gene absence and gene presence can be predicted with high confidence and thus should be given greater weight in subsequent analytical steps. It is important to note that the exact value of the LR thresholds presented here is specific to our experimental platform. The prediction accuracy achieved was remarkably high because of the uniform levels of variance across the multiple replicates analysed and because of the high correlation coefficients between replicates (the average ρ ≅ 0.92). This dataset was highly idealized because the relatively small number of replicates was carried out in such a way as to minimize technical variation. Nevertheless, a previous study in which we applied the thresholds described here on a large dataset showed that LRs below our \"absence threshold\" correlated very highly with other potential indicators of gene deletion \[[@B16]\]. Given the many documented sources of technical variability that can influence microarray results (e.g. variation in handling between individual investigators, laboratory conditions, microarray print batches), thresholds for gene presence/absence detection should be calibrated to the differential levels of technical variance found in individual microarray experiments, especially in large datasets. Kim et al \[[@B18]\] have suggested a solution to array-specific variance and normalization bias by determining thresholds specific to each array based on the point at which the LR distribution deviates from its inferred normal distribution. In practice, we have found that this approach can be susceptible to \"narrow\" LR distributions, leading to relaxed thresholds that yield an increased number of false positives for gene divergence. An alternative approach to deal with unequal variances and normalization biases across a dataset is based on normalizing multiple microarrays using the Z-score transformation \[[@B26],[@B27]\], in which LR values are divided by the standard deviation of the LR data distribution. Z score-based metrics could be used to replace Log Ratio-based metrics, enabling direct comparisons that are more valid because data from each microarray is \"variance-calibrated\". Based on the higher than expected Log Ratio values obtained in the case of absent genes, the \"relative accuracy\" of Log Ratio measurements obtained from short probes is significantly compromised under the standard hybridization conditions we used. It is important to note however, that based on the average standard deviations observed (\< 250 bp = 0.68; \> 250 bp = 0.85), results obtained from short probes do not lack precision compared to those obtained from longer probes. Nevertheless, our results show that data obtained from short probes yield anomalously high Log Ratio values. It is only because our assay represents a closed system in which all components are known that we were able to determine that short probes can significantly underestimate Log Ratio measurements. These results would not have been readily apparent in a typical experiment since there would be no *a priori*knowledge on expected Log Ratio values. Although these results were obtained in a series of CGH experiments, the anomalous Log Ratio data from short probes is likely to be encountered under any type of microarray hybridization experiment, including gene expression-profiling experiments. Although longer probes performed better in our assay, this is likely a result of the higher signal intensity obtained with long probes relative to short probes. Optimal hybridization and scanning conditions for long probes would likely be sub-optimal for short probes, leading to decreased signal and a concomitant drop in Log Ratio amplitudes. Thus the problem resides not in probe length *per se*, but rather in the mixed probe lengths encountered in our microarray. These results have important implications towards microarray probe design because the adverse probe-length effect could be mitigated through standardizing probe length. Failing that, it would be advantageous to incorporate probe length effects into any analytical framework. The results presented here have been used to examine the relationship between LR and sequence conservation. The variability inherent in hybridization-based approaches makes it unlikely that LR data from M-CGH experiments can be used to accurately predict the level of sequence identity among divergent genes. In view of the considerable ambiguity in interpreting the significance of gene divergence even when sequence information is available, the focus on gene divergence in M-CGH studies must be re-assessed. We have established thresholds for the use of LR values for the accurate detection of highly conserved and absent genes, which should increase the robustness of downstream data interpretation and should extend the range of biological interpretation of M-CGH data. An accurate determination of conserved and absent genes should increase the accuracy of strain genotyping, metabolic pathway prediction, and determination of conserved targets for vaccine or drug development from M-CGH data. Methods ======= Bacterial strains and genomic DNA isolation ------------------------------------------- Strain RM1221 was obtained from Food Safety and Health Research Unit, USDA. Strain NCTC 11168 was obtained from the American Type Culture Collection (Mannassas, VA). Genomic DNA isolation was carried out as previously described \[[@B16]\] Construction of a *C. jejuni*NCTC 11168 open reading frame DNA microarray ------------------------------------------------------------------------- The DNA microarrays used in this work were previously described in \[[@B16]\]. Additional information can be obtained at \[[@B28]\]. Microarray hybridizations ------------------------- Genomic DNA was sheared to a mean fragment size of 1.5 Kb by nebulization in 35% glycerol at 15 PSI for 45 seconds as described by \[[@B29]\]. For each sample, 5 μg of sheared DNA were fluorescently labelled using direct chemical coupling with the Label-IT (Mirus Corp., Madison, WI) cyanine dyes Cy3 and Cy5 as recommended by the manufacturer. Probes were purified by sequentially passing samples through SigmaSpin (Sigma, Oakville, ON) and Qiaquick (Qiagen, Mississauga, ON) columns. Labelled DNA sample yields and dye incorporation efficiencies were calculated using a Nanodrop ND-1000 spectrophotometer (Nanodrop, Rockland, DE). Microarray hybridizations were set-up by co-hybridizing 2 ug of differentially labelled genomic DNA samples and were carried out as previously described \[[@B16]\]. NCTC 11168 versus RM1221 hybridizations were carried out in triplicate. A set of dye-swap experiments was also carried out, giving a total of 6 replicate experiments. Self-self hybridization experiments were carried out in which separate NCTC 11168 (or RM1221) genomic DNA samples were labelled with each Cy-dye and co-hybridized to the array. Data acquisition and analysis ----------------------------- Microarrays were scanned and analyzed as previously described \[[@B16]\]. Briefly, microarrays were scanned using a Chipreader laser scanner (BioRad, Mississauga, ON) according to the manufacturer\'s recommendations. Spot quantification, signal normalization and data visualization were performed using the program ArrayPro Analyzer (version 4.5; Media Cybernetics, Silver Spring, MD). Net signal intensities were obtained by performing local-ring background subtraction and spots with a signal less than 5-times above background were excluded from the analysis. Signal intensities for replicate spots were averaged and data from each channel were adjusted by sub-array normalization using cross-channel Loess regression. The ratio of tester signal to control signal for each gene was transformed to its base 2 logarithm \[[@B30]\], log~2~\[Tester Signal / *C. jejuni*NCTC 11168 Signal\], hereafter referred to as \"Log Ratio\" (LR). LRs from the two \"within slide\" spot replicates were averaged. To increase the number of observations for statistical purposes, LR data from each microarray replicates were analyzed separately. Determining level of sequence identity between probes and targets (PTI) ----------------------------------------------------------------------- We used the BLAST software package \[[@B31]\] to determine the identity between microarray probes and predicted target sequences. Complete genome sequence information for *C. jejuni*NCTC 11168 and *C. jejuni*RM1221 was downloaded from the National Center for Biotechnology Information\'s Prokaryotic Genomes Database \[[@B32]\], GenBank records AL111168 and CP000025, respectively. We created BLAST databases from the nucleotide sequences of the open reading frames in each *C. jejuni*genome strain and queried them with the nucleotide sequences of each probe in our microarray using the BLASTN program. The percent identity of the best hit for each subject/query pair was determined. List of abbreviations ===================== M-CGH: microarray-based comparative genomic hybridization; LR: Log Ratio; PTI: Probe/Target Identity; SD: standard deviation (or σ) Authors\' contributions ======================= ENT designed M-CGH experiments, carried out downstream data analysis, and drafted the manuscript. RRA assisted with downstream data analysis. WAF carried out BLAST analysis of the two sequenced genomes. CCL carried out hybridizations and performed upstream data analysis. JHEN participated in the conception and supervised the design of the study and writing the manuscript. All authors submitted comments on drafts and read and approved the final manuscript. Acknowledgements ================ Funding for this work has been provided to ENT, RRA, WAF and JHEN through the National Research Council\'s Genomics and Health Initiative, and to CCL through Health Canada.

PubMed Central

2024-06-05T03:55:58.220342

2005-5-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168892/", "journal": "BMC Genomics. 2005 May 27; 6:78", "authors": [ { "first": "Eduardo N", "last": "Taboada" }, { "first": "Rey R", "last": "Acedillo" }, { "first": "Christian C", "last": "Luebbert" }, { "first": "Wendy A", "last": "Findlay" }, { "first": "John HE", "last": "Nash" } ] }

PMC1168893

Background ========== Due to technological developments, better communication facilities and further differentiation and specialisation of professionals, patient care is increasingly provided outside hospitals. Financiers, governments and patients advocate this development, as they consider outpatient care to be more efficient and patient centred \[[@B1]\]. This shift has repercussions for chronically ill and elderly patients with complex care needs. First, it is questionable whether optimal care for these patients can be delivered in an outpatient setting. They need sufficient time to recover and are often in frail health \[[@B2]\]. Early hospital discharge can present dangers for them. Moreover, alternatives are often scarce, as the capacity of long-term care facilities is either limited or too expensive. What is known as \'bed-blocking\' occurs when patients whose medical treatment has been completed cannot be discharged because of poor continuation of care outside the hospital \[[@B3]\]. Second, these patients often need care from health professionals in multiple settings. Consequently, patient care journeys encompass multiple transitions from one setting to another \[[@B4]\]. Systems of care often fail to organise these transitions, thus influencing the quality of care \[[@B5]-[@B7]\]. Bearing these issues in mind, a plethora of care models have been developed to substitute hospital inpatient care and to improve transitional care \[[@B8]\]. One model can be described by the term \'intermediate care\'. This model refers to a range of services intended to bridge acute, primary and social care. It is considered to serve goals like reducing the length of hospital stays, preventing hospital admissions and readmissions, improving transitions from hospital to consecutive settings, and retaining people\'s independence as long as possible \[[@B9]\]. Still, the benefits and transportability of intermediate care are ambiguous \[[@B9]-[@B12]\]. This can be attributed to the blurred definition of the concept and the wide range of services labelled as such. Models differ in focus, setting, case mix, staffing, professionals involved, commissioning and context. Hence, it is quite difficult to define the concept, to identify best practices and to compare different settings. Against this background, it is argued that the \'black-box\' of intermediate care (i.e. the processes of care) should be opened \[[@B13]\]. Process evaluations are therefore warranted and promoted \[[@B14],[@B15]\]. This paper provides such a process evaluation of an intermediate care model. The aim was to provide management with information to assess and improve the quality of care. Even so, it should also provide more generalisable insights into the accumulated impact of the model. In the Dutch health care system, a distinction is made between the financing and organisation of acute and long-term care. Acute care is financed either by social health insurance (covering about 60% of the population) or private health insurance (covering about 40% of the population). Compulsory national health insurance covers the whole population for long-term care. Health care providers organise their services in networks that are covered by either scheme. Cooperation and coordination of care takes place predominately within the acute care sector. But health care delivery is fragmented in the transition from acute to long-term care. To bridge the gap, various intermediate care models are being set up in hospitals, nursing homes and residential homes \[[@B16],[@B17]\]. However, due in part to the division of the insurance schemes, financing is often ad hoc and available only for the duration of a single project. The model evaluated in this study eventually became structurally embedded in the local health system of the South-eastern Amsterdam district. The Henriëtte Roland Holst House (HRHH), a residential home, and the Academic Medical Center of the University of Amsterdam (AMC) agreed to establish a low intensity early discharge model of intermediate care encompassing transitional care as well as a transfer unit. This model focuses exclusively on all AMC patients who no longer require hospital treatment but are not healthy enough to be discharged to their home situations. It supplements two more intensive rehabilitation models of intermediate care in nursing homes for stroke and orthopaedic AMC patients. Table [1](#T1){ref-type="table"} describes the model and its local context. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of the intermediate care model ::: Focus Starting a \'transfer unit\' in a residential home for AMC patients whose medical treatment has been completed, but are unfit to be discharged to their homes. The unit should serve as a substitute hospital ward that relieves the problem of \'bed-blocking\' in the AMC and improves transitional care to the home situation. -------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Admission criteria All AMC patients are eligible for admission to the transfer unit if they meet the following criteria: \- Patient is medically stable and curative treatment has been completed; \- Patient needs care that can be delivered by one nursing assistant; \- Patient is not eligible for other regular care services and cannot go home; \- Patient is insured; \- Patient does not need daily care and/or intensive physical therapy; \- Patient is not a drug addict, terminally ill or comatose, and does not have AIDS; \- Patient does not exhibit disturbing behaviour if he or she is a psychiatric or psychogeriatric patient. \- Patient has an official indication for discharge to a consecutive setting. Transitional care Three AMC liaison nurses control, plan and coordinate all transitions of AMC inpatients to the transfer unit systematised by agreed discharge procedures. The nursing home physician, occupational therapist, the liaison nurse and the head (an RN) assess whether an AMC patient will be admitted to the transfer ward. Setting 20 transfer beds located in 10 rooms. The unit was established outside the AMC in a residential home in the South-eastern Amsterdam district. This institution accounts for 110 residential home places, 7 places for day care, 4 community health beds and 218 apartments for assisted living. Staffing Head of the transfer unit 1.0 FTE; nursing home physician 0.33 FTE, registered nurses 0.89 FTE; liaison nurses 0.5 FTE; occupational therapist 0.5 FTE; licensed practice nurses 11.61 FTE. Two physiotherapists with a practice in the residential care home are directly available for patients of the transfer unit. An AMC geriatric nursing specialist attends multidisciplinary meetings once a week. Context The AMC and Henriëtte Roland Holst House are located in the South-eastern Amsterdam district. This region accounts for approximately 85,000 residents of whom 7,000 (8%) are older than 65, and 61% belong to an ethnic minority. A number of institutions in the region provide care for the elderly: 1 AMC, 1 nursing home, 4 residential homes, 1 public home-care agency, 1 public health agency, 1 social care agency, 5 primary care centres and 1 institution for psychiatric care. Commissioning The local public insurer structurally finances the transfer unit. The annual budget is 758,205 euros. Transitional care is financed by the AMC budgets. ::: Methods ======= Aim --- The intermediate care model was evaluated at the request of the collaborating institutions. During the course of 2000, both partners expressed concerns about how the model was functioning, and needed information to make informed decisions on how to improve the quality of care. This implied that the evaluation should be conducted from a \'managerial evaluation perspective\' \[[@B18]\]. The following questions were addressed: 1) Is the patient population admitted to the model in accordance with the *ex ante*expectations of key players and staff members? 2) How does the model ensure quality of care? Design ------ A process evaluation based on quantitative and qualitative methods was considered the most appropriate and feasible way of answering both questions. Registration forms and patient questionnaires were used to quantify the patient population. Semi-structured interviews were conducted to explore the expectations and experiences of the key players and staff members, as well as to describe how the quality of care is ensured. The project proposal was reviewed by the medical ethical committee of the AMC in August 2000 and was considered not in need of formal approval according to the Dutch legislation on experiments with human beings. There were no ethical objections raised against the study. Patient questionnaires and registration forms (quantitative analyses) --------------------------------------------------------------------- ### Sample Initially, all candidates (n = 189) for admission to the transfer unit between 1 October 2000 and 31 October 2001 were in the study. RNs working in the AMC wards selected candidates after consulting the individual patients and their families. Liaison nurses at the AMC collected the submitted applications and sent them to the transfer unit for assessment. Using formalised admission criteria (Table [1](#T1){ref-type="table"}), the nursing home physician assessed whether a candidate could be admitted to the transfer unit. 162 out of 189 (85.7%) candidates met the admission criteria and 27 out of 189 (14.3%) did not. As researchers, we did not interfere in this process. During the assessments, we requested informed consent from patients participating in the study. There were two significant differences between consenting (n = 70) and non-consenting patients (n = 119). The age distribution differed significantly between both groups (chi^2^= 12.51; p = .03), and the youngest and oldest patients were less willing to participate. Moreover, consenting patients were less likely to be refused for admission than those who did not consent. Just 2 out of 70 consenting patients (3%) were refused, while this rate was 27 out of 119 (21%) for non-consenting patients. This selective non-response can be explained by health status. Patients who did not meet the admission criteria were more seriously ill and less inclined to participate in the study. Of the 68 positively assessed patients who consented, 54 were actually admitted. During the assessment period 2 patients died and 12 were discharged to another destination. The sample is shown in figure [1](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow chart ::: ![](1472-6963-5-38-1) ::: ### Data collection Two registration forms and two patient questionnaires measured process parameters. The first form, filled in before the assessment procedure was started, recorded \'medical diagnosis\', \'reason for hospital admission\', \'reason for application\' (rehabilitation, waiting for admission to a nursing home, waiting for admission to a residential home, oncology, wound therapy, other), and \'name of the submitting AMC ward\'. After the assessment procedure \'refusal or admittance to the transfer unit\', \'date of hospital admission\', \'date of hospital discharge\', \'discharge destination\' (transfer unit, home, nursing home, residential home, other) were registered. The second form was filled out at discharge from the transfer unit. In this form \'destination of discharge\' (home with or without home care, nursing home, residential home, hospital, other), \'experienced burden of care in relation to the expected burden of care based on the application\' (the head of the unit gave his or her assessment), \'whether the patient was in the appropriate place\' (the head of the unit gave his or her assessment), and \'care delivery problems experienced\' were registered. During the assessment procedure, the patients filled out two questionnaires, assisted by a researcher if necessary: the 36-item short-form health survey (SF-36) and the Groningen Activity Restriction Scale (GARS). The SF-36 includes one multi-item scale measuring eight health concepts: 1) limitations in physical activities because of health problems; 2) limitations in social activities because of physical or emotional problems; 3) limitations in usual role activities because of physical health problems; 4) bodily pain; 5) general mental health (psychological distress and well-being); 6) limitations in usual role activities because of emotional problems; 7) vitality (energy and fatigue) and 8) general health perceptions \[[@B19],[@B20]\]. The GARS aims at measuring both the ADL and instrumental ADL disability in community-based studies with respect to the aid and services provided by professional home help and district nursing agencies \[[@B21],[@B22]\]. ### Data analysis We analysed the quantitative data using SPSS 10.1. We used T-tests and chi-squared test statistics to assess significance. Semi-structured interviews (qualitative analyses) ------------------------------------------------- ### Sample We interviewed 21 key players and staff members selected \'purposively\' for their positions, disciplines and institutions. All disciplines involved were represented in the study. In addition, nursing assistants were selected on the basis of gender, age, ethnicity and work experience. It was not feasible to interview them all. This approach resulted in the sample presented in table [2](#T2){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Interviewed professionals ::: Respondent Position Institution ------------ ---------------------------------------------- ------------- Nr 1. General manager HRHH Nr 2. Director of integrated care AMC Nr 3. Head of care department HRHH Nr 4. Chair board of directors HRHH Nr 5. Chair medical specialist staff AMC Nr 6. Nursing home physician HRHH Nr 7. Liaison nurses / head discharge unit AMC Nr 8. Liaison nurse AMC Nr 9. Liaison nurse AMC Nr 10. Geriatric nurse specialist AMC Nr 11. Liaison nurse / occupational therapist HRHH Nr 12. Registered nurse internal medicine AMC Nr 13. Registered nurse / Head of the transfer unit HRHH Nr 14. Occupational therapist AMC Nr 15. Physical therapist HRHH Nr 16. Physical therapist HRHH Nr 17. Nursing assistant transfer unit HRHH Nr 18. Nursing assistant transfer unit HRHH Nr 19. Nursing assistant transfer unit HRHH Nr 20. Nursing assistant transfer unit HRHH Nr 21. Nursing assistant transfer unit HRHH ::: ### Data collection We interviewed the respondents at their places of work using an interview guide (see figure [2](#F2){ref-type="fig"}), which was developed around the research questions. During the interviews, the guide was used in an informal and flexible way in order to prevent the researchers from imposing their own preconceptions. The interviews took approximately one hour each; they were recorded and later transcribed. One researcher coded the transcripts and wrote memos to systematise the analysis. We completed the data collection after finishing 21 interviews -- all disciplines working with the model were then represented and no new findings were expected. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Interview guide ::: ![](1472-6963-5-38-2) ::: ### Data analysis We used open and framework approaches to analyse the interviews. Building on respondents\' perceptions, we conceptualised the dynamics underlying the admittance of patients to the transfer unit. To interpret quality assurance activities and their implementation, a theoretical framework was explicated. On the basis of a typology of quality systems consisting of five elements (structural assets, allocation of responsibilities, protocols, information transfer and monitoring/feedback cycles) we identified existing quality assurance activities in the model. We contrasted respondents\' notions on the implementation of the model with recent knowledge on effective implementation \[[@B23],[@B24]\]. Rigour ------ To monitor and consider the rigour of the interviewing process, we used several strategies to rule out validity threats \[[@B25]\]. First, due to our sampling strategy we were able to identify respondents who gave socially desirable answers. We noticed that three of them were much too positive about the functioning of the transfer unit and were defensive in their responses. Second, we conducted member checks by asking respondents to validate transcripts and interpretations of their interviews. Respondents had few actual corrections, which we adopted without discussion. Third, we aimed for \'triangulation\'. We verified respondents\' essential statements by contrasting them with our quantitative data, local documents and/or international and Dutch literature. Finally, we solicited feedback from a variety of senior and other researchers (*peer review*). The researchers in the team systematically monitored the data collection, analysis and emerging findings. Colleagues at the Department of Social Medicine and the Dutch National Institute of Health Services Research reviewed earlier drafts of the manuscript. Results ======= Foreseen versus actual/perceived patient population --------------------------------------------------- The target population was described in the initial intermediate care model \[[@B26]\]. AMC patients whose medical treatment had been completed but who were unfit to be discharged to their homes -- the bed-blockers- were eligible for admission to the unit. Admission criteria further specify this target population (table [1](#T1){ref-type="table"}). Quantitative as well as qualitative data were used to verify whether members of this target population were actually admitted to the unit. The quantitative profile of the patient population -------------------------------------------------- During the 13 months of the study, 189 candidates were assessed (table [3](#T3){ref-type="table"}). The majority were female, single and older than 65. Apart from this, the profile of these candidates was more heterogeneous. Although the most prevalent diseases were cardiovascular diseases and cancer, candidates suffered from a variety of different diseases. This was also shown by the diversity of the submitting clinical wards and the various reasons for application. A relatively small number of candidates were waiting for placement in a nursing home (n = 33) or residential home (n = 3) and could be identified as bed-blockers. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Characteristics of assessed candidates ::: Process parameter Outcome ------------------------- -------------------------------------------------- Gender Male n = 72 (38.3%) Female n = 117 (61.7%) Age distribution \< 65 years n = 34 (18.3%) 65--69 years n = 21 (11.3%) 70--74 years n = 19 (10.2%) 75--79 years n = 40 (21.5%) 80--84 years n = 33 (17.7%) \> 85 years n = 39 (21.0%) missing n = 3 Home situation Living alone n = 160 (84.7%) Living with a partner n = 29 (15.3%) Medical diagnoses Cardiovascular diseases n = 44 (23.3%) Cancer n = 37 (19.4%) Other n = 108 (57.3%) Reasons for application Recovery after surgery n = 30 (17.4%) Rehabilitation n = 65 (37.8%) Waiting for a nursing home n = 33 (19.2%) Waiting for a residential care home n = 3 (1.8%) Oncology therapy n = 10 (5.8%) Other n = 31 (18.0%) Missing n = 17 ::: The \'Health-related quality of life\' (SF-36) and the \'Activities of daily living\' (GARS DL-scale) were filled out by 59 of the 70 participating patients. Their scores were comparable with those of a population in a British geriatric day hospital and a population in a Dutch community GP ward for recuperating elderly people \[[@B27],[@B28]\]. The 70 patients participating in the study had an average length of stay (LoS) in the AMC of 31.8 days (range 3--109; stdv 22.2). The 54 out of the 70 patients participating in the study who were actually admitted to the transfer unit stayed there for an average of 46.5 days (range 0--340). This average lies within the maximum of three months LoS. Looking back on their stays, the head of the transfer unit considered 12 of the 54 patients (22.2%) to have needed more care than had been expected, and 5 patients (9.3%) to have needed less care than had been expected. For 6 of the 12 more serious patients (11.1%) and for 3 of the 5 less serious ones (5.6%), the head also considered the transfer unit to have been inappropriate. The final discharge destination was home (n = 18; 33.3%), home with home care (n = 18; 33.3%), hospital (n = 6; 11.1%) or elsewhere (n = 12; 22.2%). A more complex patient load than anticipated -------------------------------------------- Respondents felt that a patient population with a heavier burden of care than anticipated had been admitted to the transfer unit. Respondents experienced a limited inflow of bed-blockers: The focus of the model was to reduce the problem of bed-blocking in the AMC. Now we notice we admit hardly any bed-blockers to the transfer unit (...). In practice it seems there are no bed-blockers in the AMC. (respondent 10) This perception was difficult to verify in the quantitative data. Based on the discharge destinations, just 36 out of 189 candidates (19%) could be considered as bed-blockers. This seems to be opposed by the average LoS in the AMC of 31.8 days, which is three times the average LoS of patients in the AMC in 2001 (9.4) \[[@B29]\]. However, the high standard deviation (22.2) and the median of 24 indicate that outliers are increasing the average LoS. Moreover, average LoS is an indicator of bed-blocking, not a valid measure. To support their perception, respondents put forward two explanations. First, respondents felt the target population envisioned in the design of the intermediate care model either did not exist in the AMC or was smaller than expected. An initial assessment of the size of the target population did not take place. This implies that the relevance of the intermediate care model may have been overestimated from the outset: You instinctively know the size of this patient group (...). As far as I know, there were no data available \[during the planning of the transfer unit\]. (respondent 8) Second, respondents suggested that the assessment procedure took too long. The model is only appropriate during a short period of patients\' care episodes because their care needs vary. Therefore, the transition from the AMC to the unit must be flexible and fast, and this was not achieved in practice. The period between application and discharge took on average 8.7 days (range 1--39). Moreover, in the study group, 14 out of 68 patients meeting the admission criteria were not discharged to the transfer unit. In the meantime, patients died (n = 2), went home (n = 5) or were admitted to another institution (n = 7). This implies that 20% of the inflow was cancelled during transitional care processes. Respondents considered the main reason for delays to be the limited availability of the nursing home physician (only three afternoons a week), whose authorisation was needed for admission. Another reason mentioned for delays was discharge planning in the AMC. Time was sometimes lost waiting for the medical application/referral forms. Due to the limited inflow of targeted patients, filling transfer beds became problematic. To avoid empty beds, respondents felt that admission criteria had been applied subjectively. In their opinion, restrictive application of inclusion criteria was not in the financial interests of the collaborating institutions: There is a negative spiral nobody in health care can ignore. (\...) On the one hand you must deliver a certain volume of care. On the other hand you have your human resources. When you fail delivering this volume of care you will loose personnel, as you earn not enough money. (\...) What happens? You can admit even less patients. So, you must compromise. (respondent 6) The HRHH maximised production to prevent budget reductions. The AMC optimised the turnover of patients to be in a better position for the annual budget negotiations with insurers. These inverse incentives may have resulted in admitting patients with a heavier burden of care to the transfer unit: I think the patient population admitted to the transfer unit has a heavier burden of care than was originally expected. (respondent 8) The medical aspects become more serious. Although the patients\' medical treatment has been completed, their health is still frail. (respondent 6) This perceived pattern was partly supported by quantitative data. According to the official hospital statistics bed-blocking days increased from 2,818 days in 2000 towards 3,315 days in 2001 while bed-occupancy rates decreased from 61.5% in 2000 to 57.5% in 2001 \[[@B29]\]. This may indicate that hospital discharge of bed-blockers is postponed to maximise bed-occupancy rates, which supports the hypothesis of the respondents. However, the severity of the self-reported health status of the patients (SF-36) did not change during the study. Quality of care --------------- Almost all respondents felt there was insufficient quality of care assurance and questioned the functioning of the model: It works, but if a few things go wrong it doesn\'t work anymore -- then the quality of care goes down fast. (respondent 11) It doesn\'t function like it should, but in the past it was worse. (respondent 17) Although we cannot verify these quotes, we assume that the intermediate care model functions poorly because it was mentioned by the majority of the respondents. Two key players (respondents 6, 10) gave an overall explanation for the poor functioning of the model. They highlighted contextual differences between the HRHH and the AMC as the main stumbling blocks. Setting up and implementing intermediate care requires a certain level of know-how and expertise. In the AMC, these kinds of requirements could easily be met, while in the HRHH they could not. Initiators took this difference for granted, thereby overestimating the organisational capacity of the HRHH: I don\'t think all of the consequences \[of setting up a transfer unit\] were foreseen\...There was no experience available of caring for these patients in another setting. (respondent 6) This overall explanation is supported by respondents\' thoughts on the implementation and on the nursing staff mix. Because the interviews were planned to take place throughout the study period, we noticed during the course of 2001 that all planned working processes and quality assurance activities were put into place. The respondents interviewed early in the study reported fewer activities than those interviewed later. Table [4](#T4){ref-type="table"} presents the processes and activities categorised in the five dimensions of a quality system. Although these activities confirmed the existence of quality of care assurance practices, the rather late implementation suggested otherwise. Respondents said that people don\'t work enough according to the agreed working processes and quality assurance practices, and concluded the implementation process was flawed: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Quality assurance activities in the transfer unit ::: Structural assets \- Description of required staff ----------------------------------------- ----------------------------------------------------------------------------- Allocation of responsibilities \- Job descriptions \- Job assessment interviews Protocols \- Description of the target population \- Admission criteria \- Discharge criteria \- Routing of the patients using a flow chart \- Nursing care plans Information transfer and record-keeping \- Transfer procedures from the AMC to the HRHH \- Patient record \- Handover procedures during shifts Monitoring and feedback cycles \- Steering group meetings \- Weekly multidisciplinary meetings \- Supervision by an AMC geriatric nursing specialist \- Patient satisfaction questionnaire upon discharge from the transfer unit \- Training and education \- Management information system ::: On the transfer unit, the wheel is reinvented every day. Because people don\'t work according to the agreements, the implementation has been flawed. (respondent 6) In the implementation literature, good preparation -- involving the relevant people, developing a proposal for change and selecting a set of multifaceted strategies -- is emphasised \[[@B23],[@B24]\]. Even so, the necessity of an open culture for change and involved management is underscored. Contrasting these insights with our data revealed shortcomings in the implementation. First, the initial intermediate care model lacked a detailed implementation strategy. Various respondents who had worked in the unit from the start confirmed this. Second, it can be questioned whether all relevant people where involved soon enough. Chief executives confessed they involved the nursing home physician too late. Also, nursing assistants said they were insufficiently prepared. Third, there seemed to be a lack of communication, resulting in a \'closed organisational culture\' and resistance to change. Finally, involvement by management was considered insufficient. There was criticism that supervision and control was too lax to bring about the desired change. This lack of management was partly due to discontinuities in leadership. Although four persons headed up the unit from the start, reorganisation of the entire residential home distracted chief executives. All these shortcomings were reflected in the following quotes: At the start, there was very little idea of what might happen. (respondent 6) They told us almost nothing about what was going to happen. I had the idea they didn\'t really know, either. (respondent 18) A nursing home physician was represented in the project group, but it wasn\'t the physician who was going to do the job. I didn\'t think this was very smart.(respondent 2) Staff members act differently. Different perceptions, different realities can be observed, but they\'re not tried out on each other or debated. (respondent 11) Apart from the implementation, respondents were also concerned with the nursing staff mix. In their view, the staff was insufficiently qualified: At this moment, I do not have a positive image of their expertise. (respondent 8) Managers and staff at the HRHH attributed the lack of expertise to the shortage of skilled nurses. Local figures showed that in 2000, nursing homes and residential homes in Amsterdam had 56 vacancies per 1,000 nursing staff. This rate is the highest in the Netherlands \[[@B30]\]. The managers of the HRHH said they had enormous problems filling the vacancies. Respondents working in the AMC underscored this, but also questioned the chosen staff mix model of the transfer unit: I was and still am disappointed in the nursing staff mix. The number of staff is okay, but on a ward where discharged hospital patients are admitted, just one RN supplemented by nursing assistants isn\'t enough. (respondent 2) Limitations ----------- The process evaluation has its limitations. The large number of non-consenting patients (119 out of 189) has biased the quantitative results. Most of the non-consenting patients were unable or less inclined to participate in the study. We attribute this in part to a flawed informed consent procedure. Consenting patients had to fill out questionnaires immediately, which made various candidates less willing, especially the sicker ones. Consequently, healthier patients are over-represented in the sample. This explains why we could not verify the perception by respondents that the patient population was more seriously ill. The validity of the qualitative findings is rather high, although some weaknesses were revealed. Because of our managerial evaluation perspective \[[@B18]\], three respondents seemed to give socially desirable answers. Furthermore, we had the impression that the evaluation became a management intervention in itself (*Hawthorne effect*). During the study we noticed that our framework of analysis, which was implicitly communicated during the interviews, was translated into management actions. Another weakness is that just one researcher conducted the interviews and analysis. Still, we were able to overcome these sources of bias because of the combination of methods, the purposive sampling procedure undertaken and the comprehensive triangulation achieved. The interviews therefore provide a credible exploration of the functioning of an intermediate care model in its local context. Overall, the transportability of the findings to other settings is limited. As we conducted a process evaluation, findings only provide a detailed description of the intermediate care model in Amsterdam. Nevertheless, we assume that the accumulated impact of the model is not unique to the Amsterdam model. From this perspective, the findings provide a good starting point for developing and evaluating models elsewhere. Discussion ========== In the current climate of health care policy, many stakeholders advocate the development and implementation of intermediate care models. However, the widespread popularity of the concept is insufficiently supported by evidence. Intermediate care is still in need of evaluation, as the benefits and the deficits of the various models are ambiguous \[[@B1]-[@B4]\]. Nevertheless, available knowledge provides enough do\'s and don\'ts for initiators who want to set up and implement intermediate care. Basically, initiators should be cautious. As our study shows, the accumulated impact of setting up such models may result in a \'bad practice\'. To prevent this, initiators should adequately plan, organise and monitor intermediate care services. The following issues are important. First, the relevance of an intermediate care model must be clear at the outset and supported by sound information. This was not the case in Amsterdam. The low intensity early discharge model was set up alongside two more intensive intermediate care models for stroke and orthopaedic patients. Consequently, the transfer unit was targeted towards a remnant and heterogeneous patient population within which the number of eligible candidates was smaller than expected. The straightforward conclusion is that one should know the size and profile of the target population. Apart from that, patients may view intermediate care as unacceptable. Little is known about this topic \[[@B31]\], and this is reflected in the call for more user involvement in intermediate care development \[[@B32]\]. Even so, this study indicates that the dynamic health status of candidates requires fast and flexible transition processes. Because the health status of patients can change rapidly, intermediate care services are appropriate for short periods during patient journeys. Moreover, poorly organised transition processes may increase the average LoS in the acute care setting which undermines achieving the goal of early discharging patients. So, detailed knowledge of the care needs, desires and size of the target population is necessary to justify the relevance of an intermediate care model. Second, policy-makers and managers should pay attention to the dynamics resulting from organisational and financial incentives. These may influence a broadening of admission criteria to maximise bed occupancy rates, as shown in Amsterdam. In the literature, this phenomenon is also known as \'Roemer\'s law\': a bed built is a bed filled \[[@B33]-[@B35]\]. Perverse incentives induce improper use of intermediate care beds, which results in another patient profile than initially anticipated. The changed patient profiles reported in various studies support this hypothesis \[[@B13],[@B36],[@B37]\]. Continuous monitoring of the patient profile in relation to bed occupancy rates is necessary to detect and overcome this phenomenon. Third, intermediate care requires a minimum of nursing staff to guarantee quality of care delivery. Their skills should meet the needs of envisioned patients. This was inadequate in Amsterdam. Nursing assistants gave direct care to patients, while medical and registered nursing staff headed up the unit. This high percentage of direct care given by relatively unqualified staff is also reported in other studies. It is considered one of the main reasons for the absence of improved outcomes of intermediate care \[[@B13],[@B36]\]. Other studies as well indicate that nursing staff *does*matter: those hospitals and nursing homes with the most highly qualified staff provide better care \[[@B38]-[@B40]\]. This underscores the importance of thoroughly considering nursing staff mix in intermediate care. However, it is difficult to determine what nursing staff mix is appropriate. There is no minimum standard available. Fourth, the implementation of intermediate care needs to be given attention. It requires know-how and expertise. In Amsterdam, the initiators overestimated the ability of the staff of the residential home to develop and operate a transfer unit for early discharged patients. These homes provide minimal care services for elderly people in stable health. These homes are less prepared to deliver care to post-acute care patients. This is reflected in the less advanced stage of development of quality assurance activities in Dutch residential homes \[[@B41],[@B42]\]. Initiators of collaborative intermediate care models should be aware of this pitfall and plan a comprehensive implementation strategy. Such a strategy must contain multiple approaches at different levels, tailored to specific settings and target groups \[[@B23],[@B24]\]. Such a strategy should ensure that all requirements are met for delivering good care. From a more general perspective, these issues promote a more rational and evidence-based management of intermediate care. It is acknowledged that the uptake of evidence in managerial practice could be better \[[@B43],[@B44]\]. As intermediate care lacks a straightforward evidence base, managers run the risk of being persuaded by political willingness rather than by \'evidence\'. Uninformed decision-making is dangerous and may ultimately harm patients; future research should fill the existing knowledge gaps. Steiner \[[@B1]\] identifies three key questions for intermediate care research: 1) Which services are best for which patients at which point? 2) Which professionals should be involved, doing what at which point? 3) What is the bottom line financially? Our findings indicate that these key questions must be answered simultaneously. One cannot properly answer one of the key questions without knowing the answers to the other two. This implies that research designs should have a broad scope, which as our study illustrates, is at the expense of rigour. However, creative study designs are being promoted that try to balance rigour and validity by combining quantitative and qualitative approaches, purpose-collected and coincidental data, and multidisciplinary research perspectives \[[@B45],[@B46]\]. These evaluative approaches provide a good foundation for developing the evidence base for intermediate care. Conclusion ========== We conclude that setting up a low intensive early discharge model of intermediate care between a university hospital and a residential home is less straightforward than was originally perceived by management, and that quality of care needs careful monitoring to ensure the change is for the better. The AMC and HRHH management have taken on these lessons, and the findings of the process evaluation have been translated into management interventions: consistent use of discharge and admission criteria, increasing patient flows by working with a second general hospital in Amsterdam, increasing nursing and medical expertise on the ward, implementing specific nursing protocols and more systematic monitoring of care. A combined quantitative and qualitative evaluation approach executed in close collaboration with the actors involved was helpful in revealing the underlying mechanism leading to the shortcomings. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= TP designed the qualitative study, conducted the semi-structured interviews, analysed the qualitative data, analysed all data, and primarily wrote the paper. DMJD designed the whole study, drafted the main protocol, assisted with the quantitative data analysis, reviewed earlier drafts of the paper, and supervised the daily research activities TFK collected quantitative data, was responsible for the quantitative data entry, did the quantitative analysis, and critically reviewed earlier drafts of the paper TJ collected quantitative data, critically reviewed earlier drafts of the paper NSK was responsible for completing the study and the supervision, designed the whole study, and critically reviewed earlier drafts of the paper Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/38/prepub>

PubMed Central

2024-06-05T03:55:58.223083

2005-5-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168893/", "journal": "BMC Health Serv Res. 2005 May 24; 5:38", "authors": [ { "first": "Thomas", "last": "Plochg" }, { "first": "Diana MJ", "last": "Delnoij" }, { "first": "Tineke F", "last": "van der Kruk" }, { "first": "Tonnie ACM", "last": "Janmaat" }, { "first": "Niek S", "last": "Klazinga" } ] }

PMC1168894

Background ========== The rigorous evaluation of health care interventions requires potential users of these interventions to participate in clinical trials to assess their effectiveness. There are currently various concerns about both the small proportion of eligible people who participate in randomised controlled trials (RCTs) and the experiences of participants \[[@B1],[@B2]\]. Inadequate information provision is a contributing factor to several of these concerns. For example, some people who would like to enter RCTs are not invited to do so, and struggle to identify trials for which they would be eligible \[[@B3]\]. Problems have been identified with the quality of information given \[[@B4]\]. Thus some people who are invited to participate in RCTs do not develop an adequate understanding to make a well-informed decision to participate or not \[[@B5],[@B6]\]. People who agree to participate in trials sometimes struggle to make sense of their participation \[[@B7]\], may not know what to do if they identify questions during the course of the trial, and are often disappointed at the lack of information they are given about the progress or results of the trials \[[@B8]\]. An information resource about ongoing clinical trials designed for potential and current participants could help overcome some of these problems. It could serve as a source of additional, semi-independent information for people who have been invited to participate in trials, and it could help people to identify trials in which they might participate. If kept up to date, it could serve as a source of information about the progress and results of particular trials. A number of trial registers have been established in recent years (for example, *meta*Register of Controlled Trials <http://www.controlled-trials.com>, UKCCCR National Register of Cancer Trials <http://www.ctu.mrc.ac.uk/ukcccr/home.asp>), and there has been increasing pressure for requirements that all clinical trials be registered in the public domain. \[[@B9]\]. Trial registers are undoubtedly helpful in terms of getting information about planned, ongoing and completed trials into the public domain and reducing the problems associated with underreporting of trials. However, the current registers are not comprehensive \[[@B10]-[@B12]\]. They were not designed to be easily accessible or useful to general public audiences and none were designed specifically to meet the information needs of members of the public as potential or current trial participants. It can be difficult to identify which registers are appropriate for particular purposes \[[@B12],[@B13]\], and it is hard to search across registers because they do not organise and present information in a standardised way \[[@B3],[@B13]\]. In this paper we report on a scoping exercise undertaken for Consumers in NHS Research (now known as INVOLVE) to assess the desirable features and feasibility of developing and providing an information resource about ongoing clinical trials for potential and current participants \[[@B14]\]. Methods ======= We recognised that people who had played different roles in relation to trials would have different perspectives on information needs relating to clinical trials. Reflecting this, we sought information and opinions from a range of people from across the UK that included: people seeking to participate in or considering participating in a trial; current and recent trial participants; representatives of national patient interest/support groups; and researchers involved in the design and conduct of trials. People who were seeking to participate and considering participating in a trial were identified through condition-specific consumer networks, in particular the National Association for the Relief of Paget\'s Disease (a trial investigating the treatment of Paget\'s disease was about to start at the time we conducted this scoping exercise). Current and recent trial participants were identified at three meetings of health care consumers and advocates. Representatives of consumer groups were selected from those patient interest/support groups with a known interest in trials (identified primarily by Consumers in NHS Research). Researchers involved in the design and conduct of trials were selected from the UK MRC trial managers network and individuals known to the research team. Several of the consumer advocates and researchers had current or recent personal experience of participating (or declining to participate) in clinical trials. We accepted their contributions from the perspective of trial participants, because the time frame of our scoping exercise did not permit us to obtain ethics committee approval to identify and approach current participants via trial offices. We used semi-structured interviews, focus groups and e-mail discussion lists to explore the kinds of information that people want and need when considering whether or not to participate in a particular trial, and the ways they might access and use this information. We asked people to discuss the feasibility and appropriateness of \'star rated\' information on the quality of particular aspects of ongoing trials, and the provision of generic guidance about how to interpret and appraise information about trials. We also asked researchers and consumer representatives what they thought made for a good quality trial and what issues might arise from an attempt to provide standardised information about ongoing randomised clinical trials in a publicly accessible participant-orientated information resource. We used topic guides to support semi-structured interviews and focus group discussions. An example of the topic guide used with potential trial participants is included \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Most interviews were audio-taped and transcribed in full. If audio-recording was not possible, members of the research team took detailed notes. Analysis of transcripts, discussion notes and e-mails was carried out using a modified \'charting\' approach \[[@B15]\]: to facilitate the identification of the range of views about particular issues, the points made by each participant were summarised under key headings that reflected the main areas of questioning. Over the course of the scoping exercise we formally interviewed 25 individuals on a one-to-one basis including: • nine potential trial participants; • six representatives from patient interest /support groups; • nine researchers (including trial principal investigators, trial managers and a research nurse); and • one pharmaceutical/medical devices industry representative. In addition, the views of 24 consumer advocates were elicited in focus groups held at national meetings of consumers and advocates, and nine consumer representatives contributed comments via e-mail. Several of the consumer advocates and researchers had personal experience of participating (or declining to participate) in clinical trials. Results ======= Information elements required by potential trial participants ------------------------------------------------------------- Our informants identified a range of information needs of potential trial participants. We grouped these into the following categories: (a) the interventions being compared; (b) the implications of trial participation for participants; (c) the scientific or methodological quality of a trial; (d) the ethical probity and governance of a trial; and (e) contact details (Figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The features identified as desirable for a centralised information resource about randomised controlled trials. ::: ![](1472-6963-5-39-1) ::: ### Interventions being compared There was a consensus that potential trial participants need to know about the \'new\' intervention being tested, its potential risks, and how it is known to compare with the current \"standard\" intervention and any other interventions with which it is compared in the trial. Information about the availability of relevant interventions outwith the trial was also mentioned as important. ### Implications for participants Information needs relating to implications for participants included practical issues such as the timescales for the trial interventions and follow-up, and the rationale for, and number of extra visits or investigations associated with the trial. Several informants highlighted the need for information about whether and how people could withdraw from or join the trial at a later date. They also thought that participants should be told whether or how they could gain access to data about their own outcomes and the trial findings, and about whether and how participants would be told if a trial was stopped early for any reason. ### Scientific or methodological quality, and ethical probity and governance There was a consensus that people should be able to be confident about the quality of any trial they might agree to participate in, but it was widely recognised that relatively few people would currently be able to judge the scientific quality or ethical probity of a trial for themselves on the basis of the information they are typically given for recruitment purposes. The provision of information about key features of trial quality (and/or about quality assessment) was thought desirable for a centralised resource. ### Contact details Several respondents from all categories mentioned that potential trial participants might benefit from the option to talk to someone before deciding whether or not to participate in a trial, and possibly also during the trial. The case was regularly made that two-way interaction was key to understanding information, as any misconceptions could be corrected and personal concerns could be addressed. For people who had been invited to participate in a particular trial, the options suggested were: • having a chance to speak with a member of the trial team (preferably someone who is not related directly to a person\'s normal care) -- someone who can explain things like randomisation; • having a chance to talk to an independent person (i.e. someone not associated with the trial) e.g. GP, consumer advocate, voluntary health organisations, advice services; and • having contact details, so know whom to contact if concerned about anything while participating in the trial. People who were themselves looking for trials in which they might participate would be likely to need: • contact details for initial enquiries about the specific trial and (possibly subsequently) local recruiting centres/clinicians; and possibly • an information helpline staffed by people who could offer general information, answer general questions about trials, and help explain information provided about specific trials. However, the capacity of trialists or trial centres to handle unsolicited enquiries was limited, they might be reluctant to have their contact details \'advertised\' in the context of an information resource that promised help to people identify trials in which they might participate. Aids to information appraisal ----------------------------- As indicated in the methods section, we explored two approaches that might help people evaluate information about trials and assess their quality: (a) a quality or \"star\" rating scheme in which key features of trials were allocated one- to three- star ratings; and (b) a guide to help people appraise trials for themselves. Informants registered some strongly felt concerns about the quality ratings option. These included doubts about the feasibility of producing valid and reliable ratings across a diverse range of trials, and fears that particular ratings were likely to be contested by trial sponsors and trialists, which could create a lot of \'hassle\' and work for the resource producers. Despite the limitations of a quality ratings scheme, there was general agreement that it would be useful to indicate that key quality assurance features were in place for particular trials. The most obvious features suggested were that a trial had been through an independent peer review process (a check on scientific quality) and that it had been assessed and approved by an independent ethics committee (a check on ethical probity). There was widespread support for the provision of a general information section within any centralised resource that could: explain the purpose of clinical trials; highlight key features and indicate points to look out for in clinical trials; provide a glossary of commonly used terms; and identify links to relevant organisations. The inclusion of a \'guide\' to help people appraise the quality of a trial for themselves was also thought appropriate. This guide could be based on existing checklists developed for peer reviewers, ethics committees and consumer representatives who appraise trials, but modified for more general public use. Key design features of a centralised resource --------------------------------------------- All informants supported the basic concept of a centralised information resource about clinical trials. This could be built using a database structure with a standard set of information elements for each trial reflecting the key areas outlined above. Respondents identified a variety of design features that could potentially enhance the usefulness of the information within the resource, including: • A sophisticated search facility that would allow people to enter their personal details (e.g. health condition, age, gender, geographical location etc), and retrieve information on trials that fulfilled their search criteria. • Options to view particular types of information in varying levels of detail. • The use of a variety of media within or as a complement to the resource (for example, diagrams or video-clips to illustrate and explain trial procedures or the interventions being compared in particular trials Potential models for providing a centralised information resource ----------------------------------------------------------------- There was widespread support for the provision of this centralised information resource on the Internet, but a clear recognition that some people would not be able to access the information effectively via this route. Opportunities for supported access via outlets in health care settings, public libraries or other community centres would also be useful. Health information services might usefully serve as valuable \'intermediaries\', adding value to the resource by helping people to appraise the information and interpret it for their own use. Discussion ========== One of the strengths of our study is that we canvassed views from a range of perspectives. We identified a broad-base of support for the concept of a centralised, publicly accessible, participant-oriented information resource about ongoing clinical trials. We also identified several concrete suggestions for the content and design of such a resource. However, time constraints prevented us from exploring the views of a larger sample of people, and from considering the issues we identified in more depth. We acknowledge in particular that further research is needed to explore the opinions of a wider spectrum of the \'general public\' as potential users of a centralised resource for information about clinical trials, and to consider the potential implications of presenting particular elements of information and in particular ways. We also acknowledge that the potential trial participants interviewed for this study were mostly people with a chronic condition. Their perspectives may differ from potential trial participants with acute health problems. However, individuals with acute conditions were canvassed for their views as current or recent trial participants. We do not claim to have identified all possible information requirements, but are confident that we have identified the main ones. In theory, a centralised information resource about clinical trials would have two main audiences in terms of trial participants: those who have been invited to participate in a trial and want to find out more about it, and those who have not been invited to participate in a trial, but would like to identify any ongoing trials for which they might be eligible. If the resource included updated reports on trial progress, it may also be useful to a third group of people who are currently participating in trials who want more information about the trials in which they are participating It is unclear whether and to what extent a centralised information resource would achieve the desired aims of improving recruitment rates, and people\'s experiences of information provision relating to trial recruitment and participation, nor whether it would have any unwanted effects. It is not clear to what extent a publicly accessible resource might increase demand for participation in trials, nor what the implications of such demands might be. At the moment, some health care providers are not in a position to offer trial participation to their service users. Increased public awareness of particular trials might place a variety of extra burdens on health care systems, for example if patients start demanding access to particular trials and/or changing health care providers in order to gain access to them. Other potential disadvantages of making information about trials more easily accessible should be explored. For example, breaches of privacy might become a problem if the knowledge that someone is eligible to participate in a particular trial (which is often readily obtained by third parties who observe markers on medical records or witness the receipt of mailings from a trial team) could quickly lead to the revelation of information about personal health status or health care experiences because trial eligibility criteria would be easily ascertained. Public access to trial progress reports might, depending on the nature of information provided, influence people\'s self-reporting of outcome status and willingness to continue to participate in a trial. Issues such as these require further consideration and investigation. While the development of the database framework, search engine and the features identified above as desirable are technologically feasible, there are several potential barriers to the development of a participant-oriented central information resource about clinical trials. Current trial registers do not contain all the information elements that potential participants might find useful. The process of gathering, checking and updating the requisite information would be logistically challenging and time consuming, and would require considerable and sustained investment. Any attempt to include trial progress reports to the database would significantly add to the logistical challenges. However, recent trends towards increasing disclosure of information about trials, including commitments to openness from some pharmaceutical companies, will facilitate the acquisition of relevant information. Conclusion ========== A centralised participant-oriented information resource about randomised trials could offer many benefits. The development of such a resource would require considerable investment, but the support for the basic concept that is evident among trial participants, consumer advocates and trialists suggest that it warrants further investigation and evaluation. Abbreviations ============= UKCCR United Kingdom Coordinating Committee on Cancer Research RCT Randomised Controlled Trial Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors contributed equally to the design and execution of the study. MKC was principal investigator. ALL wrote the first draft of the manuscript. All authors contributed to the development of subsequent drafts, and read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/39/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 2001\. Example Topic Guide. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ *Consumers in NHS Research*(now known as *INVOLVE*) funded the scoping study on which this paper is based. We would like to thank the secretarial staff of the Health Services Research Unit for transcription of the interviews. Most of all we would like to extend our many thanks to the people who agreed to be interviewed or submit opinions for this study and gave generously of their time and views. The Health Services Research Unit is funded by the Chief Scientist Office of the Scottish Executive Department of Health. The views expressed are those of the authors and not necessarily those of the funders.

PubMed Central

2024-06-05T03:55:58.227513

2005-5-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168894/", "journal": "BMC Health Serv Res. 2005 May 24; 5:39", "authors": [ { "first": "Anne L", "last": "Langston" }, { "first": "Marion K", "last": "Campbell" }, { "first": "Vikki A", "last": "Entwistle" }, { "first": "Zoë", "last": "Skea" } ] }

PMC1168895

Background ========== Adult outpatient parenteral antibiotic therapy (OPAT) programs have been reported in the literature for over 20 years \[[@B1]\]. These programs have now become an accepted alternative to inpatient therapy for select patients and infection types \[[@B2]-[@B4]\]. OPAT programs have been demonstrated to be a safe, effective and acceptable alternative to hospitalization \[[@B5]-[@B7]\]. Cost analysis of OPAT programs in the U.S., Canadian and other settings have been reported \[[@B5],[@B8]-[@B13]\]. An adult OPAT program was implemented at our hospital in 1995. A cost analysis performed at our institution from 1995 to 1998 showed significant cost avoidance for both the hospital and the Ministry of Health \[[@B14]\]. Informal patient satisfaction surveys have shown that the program has been well received by patients. However, there are no published reports quantifying preference for treatment location of OPAT program patients. Willingness to pay (WTP) is a method of quantifying preference and this methodology is gaining popularity in health care \[[@B15]-[@B19]\]. WTP provides a measure of how much an individual values a particular treatment preference. The objective of this study was to elicit hypothetical infection treatment location preferences, and the willingness to pay for this preferred treatment option from OPAT program referral patients. Methods ======= This was conducted as a multidisciplinary, single centre, prospective study at a 1000-bed Canadian adult tertiary care teaching hospital. Our OPAT program receives about 250 patient referrals and provides approximately 3,000 patient-days of outpatient parenteral therapy per year \[[@B20]\]. General characteristics of patients managed by this service are described elsewhere \[[@B14],[@B20]\]. Adult patients are accepted into the program if they had a proven or suspected infection requiring one or more parenteral antimicrobials for an expected minimum duration of 5 days, are medically stable, have an acceptable venous access, demonstrate a willingness and capability to perform the necessary self-management tasks and live in a suitable home environment with access to a telephone. Once enrolled, the OPAT pharmacist and nurses provide patient teaching, insert the appropriate vascular device, liaise with community nursing personnel, coordinate delivery of drugs and supplies and arranged appropriate patient follow up. These patients are treated in their home environment and return to the hospital periodically for follow-up purposes only. Patients requiring short courses of parenteral antibiotics are typically excluded from the program and are treated as inpatients or managed in a hospital medical daycare setting. This study was approved by the University Ethics Committee and the Hospital Research Committee. Patient enrollment ------------------ Patients referred to the OPAT program during the 9-month study (September 1999 -- June 2000) were considered eligible for the patient preference assessment. For inclusion into the study, the questionnaire had to be administered to the patient prior to assessment by the OPAT clinical staff for possible inclusion into the program. Only consenting patients were administered the WTP questionnaire. This precaution was taken to avoid the possibility of biased responses from the participants in an attempt to obtain the treatment location of their choice. Willingness to pay ------------------ Contingent valuation methodology (CVM) was used to quantitatively measure patient preference (i.e. WTP) for intravenous antibiotic treatment location \[[@B15]-[@B19]\]. CVM is a survey-based approach for eliciting a consumer\'s monetary valuations for program benefits for use in cost-benefit analysis. The specific methodology employed was similar to that adopted by Donaldson et al \[[@B21]\]. The consumer utility being measured was compensating variation and the survey measured WTP in the context of program availability. Since the questionnaires were given to individuals undertaking the valuation who were already consumers of the treatment in question (i.e. were receiving parenteral antimicrobials when enrolled) and thus the primary uncertainty at the time of the questionnaire was the probability of treatment course outcomes, an ex-post perspective was adopted \[[@B15],[@B16]\]. The WTP questionnaire consisted of two hypothetical scenarios that were created based upon OPAT program data collected during the period 1995--1998 (see [Additional File 1](#S1){ref-type="supplementary-material"}) \[[@B14]\]. Each hypothetical scenario involved an infection that required a 23-day course of intravenous antibiotics (the average duration of therapy for patients enrolled in the OPAT program). The first scenario described a hospital treatment course, while the second scenario described a similar treatment regimen that was administered in the home setting. Using these historic data, the risks associated with each treatment location were also provided to the patient \[[@B14]\]. Patients were asked to specify their preference for treatment location based upon these scenarios. Utilizing open-ended questions, patients were asked to quantify their preference by stating how much they would be willing to pay to obtain treatment in their preferred location. The survey was initially designed and tested on ten patients, four clinical nurse specialists and two infectious diseases specialists. We utilized comments from these individuals to modify our survey in order to improved readability and understandability. Data collection --------------- A single investigator coordinated all patient self-administered WTP questionnaires. Partial completion of the surveys was identified in four of the first eleven patients; thereafter, the investigator examined the questionnaires for completeness at bedside and encouraged patients to provide responses if necessary. If requested by the patient, the questionnaire was read aloud and the investigator recorded the patient\'s responses. In these cases, the completed questionnaire was subsequently reviewed with the patient to ensure accuracy. Demographic information, socioeconomic data, infection details and WTP for treatment location were collected for all patients. Data analysis ------------- Means, medians and ranges for WTP by treatment preference location were determined. Differences in means and medians were testing using t-tests and Mann-Whitney U tests, respectively. To estimate an overall monetary valuation, a total WTP was calculated for patients preferring hospital treatment and a total WTP was also calculated for those preferring home treatment. A positive association between WTP and income was assessed to determine the construct validity of our questionnaire \[[@B22]\]. WTP was examined for normality using histograms to determine if a natural log transformation was necessary to fit the assumptions of linear regression. Using WTP as the dependent variable, multiple linear regression was utilized. Univariate analyses were performed between each of the possible predictor variables (gender, marital status, level of education, employment, annual household income, infection type) and the dependent variable using ordinary least-squares linear regression. Variables associated with WTP with a p-value \< = 0.10 in the univariate analyses were considered in the multiple linear regression models. Adjusted r^2^was calculated for the multivariable models to determine the amount of variance in the outcome variable explained by the predictor variables in the final models. Among significant variables, two-way interactions were investigated. No adjustments were made to p-values to account for multiple comparisons. Studentized residuals and Cook\'s distance were examined to determine if assumptions of multiple linear regression were violated. Two-sided P values are reported for all analyses. A p value of less than 0.05 was considered to be statistically significant. All analyses were conducted by using SPSS, version 10. Results ======= During the 9-month study period, 131 patients were considered eligible for enrollment in the contingent valuation analysis. Of these patients, 40 were excluded, as the investigator was unavailable to conduct an interview prior to the assessment by the OPAT team, informed consent could not be obtained due to language barriers, decreased cognitive status was evident or patients simply declined to participate. The remaining 91 patients completed the WTP questionnaire. Patient demographics and socioeconomic status are reported in Table [1](#T1){ref-type="table"}. Participants were typically married males in their sixth decade of life with a secondary school education or greater. The majority of participants were retired or were employed with an annual household income of less than \$60,000. Osteomyelitis was the most common type of infection for which parenteral therapy was required. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient demographics ::: **Parameter** **Value** ------------------------------------ ------------- **No. of Patients** 91 **Mean age, years**(range) 56 (25--81) **Gender** Male (%) 63 (69) **Marital Status (%)** Married 60 (66) Divorced 11 (12) Widowed 9 (10) Single 10 (12) **Highest Level of Education (%)** Elementary School 7 (8) Secondary School 29 (32) Trades/Technical College 34 (36) University Degree 16 (18) Post-graduate 5 (6) **Employment (%)** Retired 45 (49) Employed 29 (32) Unemployed 16 (18) Unknown 1 (1) **Annual Household Income (%)** \< \$20,000 22 (24) \$20,000--39,999 20 (23) \$40,000--59,999 22 (24) \$60,000--79,999 9 (10) \$80,000--99,999 3 (3) \$100,000--149,000 7 (8) \> \$150,000 4 (4) Unknown 4 (4) **Type of Infection (%)** Osteomyelitis 39 (43) Infected pacemaker/wires 9 (10) Endocarditis 9 (10) Wound infection 7 (8) Abscess 6 (7) Bacteremia 4 (5) Other^1^ 17 (17) ^1^Meningitis (3), pneumonia (4), infected graft of lower limb (2), line sepsis (2), septic arthritis (2), cellulites (1), pyelonephritis (1), CMV (1), discitis (1) ::: Willingness to pay ------------------ Of the 91 patients who were enrolled in the study, 87 (96%) indicated a treatment location preference while the remaining four participants had no preference. Of those 87 patients who indicated a preference, 77 (89%) preferred treatment at home while 10 (11%) preferred treatment in hospital. Seventy-one (82%) patients provided an interpretable response regarding WTP for treatment in their preferred location. Of those 16 patients (13 patients with a preference for home therapy vs. 3 patient with a preference for hospital therapy) who did not provide an interpretable response, one registered an astronomically high \"protest\" WTP far exceeding their ability to pay, while 15 indicated a treatment preference but provided no monetary value. For those 71 patients who provided an interpretable response, 64 patients preferred treatment at home with a median WTP of \$490 CDN (mean \$949, range \$20 to \$6250), and 7 patients preferred treatment in the hospital with a median WTP of \$500 CDN (mean \$1123, range \$10 to \$3000). Tests for differences in means and medians revealed no statistically significant differences between WTP values between the treatment locations at the 5% level. The total WTP for the seven patients who preferred hospital treatment was \$7,859 versus \$60,712 for the 64 patients who preferred home treatment. The natural logarithm of WTP values approximated a normal distribution, thus satisfying this assumption of linear regression (Figure [1](#F1){ref-type="fig"}). Only seventy-five patients (71 patients with an interpretable response plus those 4 patients with no treatment location preference) were included in the regression analysis. Multiple linear regression analysis revealed that income and treatment location preference were independent predictors of WTP (Table [2](#T2){ref-type="table"}). There was a trend towards respondents with lower incomes being willing to pay slightly less for their preferred treatment location than those with the highest incomes (p = 0.067). In addition, people who stated preferences were willing to pay significantly more for than those who did not state a preference (p \< 0.001). In the multiple linear regression model that included interaction terms (adjusted r2 = 0.543), there was also a significant interaction between income and treatment location preference such that patients with the lowest income were willing to pay significantly more for hospital treatment than for home treatment (p \< \<0.0001). The fact that there was a significant association between WTP and ability to pay (i.e. higher income) validates the theoretical construct of our survey. ::: {#F1 .fig} Figure 1 ![](1472-6963-5-43-1) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### WTP regression analysis^1,2^ ::: Parameter β - coefficient p-value 95% Confidence Interval --------------------- ----------------- --------- ------------------------- -------- Intercept 0.75 0.35 -0.82 2.32 Income, \$ CAN 0.067 ≤ 20,000 -0.991 0.031 -1.89 -0.093 21,000--79,000 -1.011 0.037 -1.96 -0.060 ≥ 80,000 Reference Location preference \<0.001 Home 6.10 \<0.001 4.58 7.54 Hospital 6.13 \<0.001 4.32 7.94 None Reference ^1^Dependent variable is the natural logarithm of WTP ^2^Adjusted r2 = 0.478 ::: Discussion ========== To our knowledge, this is the first published report quantifying preference for treatment location in an adult OPAT program patients using WTP. According to our WTP analysis, candidates for the program expressed an overwhelming preference for treatment in the home setting. Our results also demonstrated that the WTP values were similar between those patients who preferred to be treated at home and those who wished to remain in hospital. Accordingly, the total WTP value was greater (in fact, almost 8-fold greater) for those patients preferring treatment at home. This reflects the overall magnitude of societal preference for the management of infectious diseases that require intravenous therapy, but does not require institutionalization. There were several limitations to this study. We conducted this trial in one adult acute care institution, thus caution must be exercised when attempting to generalize the results to other health care settings involving different patient populations, and other infectious diseases which will require different treatment regimens. We relied on a hypothetical treatment scenario in our attempt to solicit a preference location and willingness to pay for this patient population. As the scenario did not necessarily reflect the treatment that they were about to receive, we must be careful in our extrapolation of the results. Although we acknowledge the potential problems with using such scenarios in CVM, we believe that this effect was minimized by surveying patients who were currently experiencing an infection that initiated a consult from the OPAT team (i.e. the ex-post perspective). We believe that most of these individuals would be able to realistically comprehend the health outcomes described in our scenarios. We also relied on open-ended technique rather than a bidding-game technique to solicit a WTP value. While the bidding game technique forces an upper and lower limit to the patient response and can be criticized for introducing a starting point bias, the open-ended technique has also been questioned. As described by O\'Brien and Viramontes, patient naivety regarding health care costs due to the Canadian universal health insurance environment may lead to an inability to quantify the value of an expected health improvement \[[@B16]\]. The broad range of WTP values provided by our participants may be a reflection of this naivety. In addition, O\'Brien and Gafni discuss that open-ended questions often elicit large numbers of non-responses or protest zero responses \[[@B21]\]. Indeed, some patients in our study expressed difficulty in placing a dollar value on their choice of treatment location. In some cases, this appeared to be a protest against the interview question and reflected a concern that their response would be used to determine a future fee for their treatment preference. In other cases, this may have been related to the fact that patients are not typically aware of, nor directly pay for, the costs of health care services in the Canadian health care system. Finally, WTP surveys measure only what a patient claims they are willing to pay for a particular treatment. The magnitude of payment is not necessarily an accurate reflection of what they would actually be willing to pay if they were to encounter the actual scenario. As expected, ability to pay was associated with WTP and this functioned as a confirmation of construct validity of our questionnaire. Unfortunately, as mentioned by Drummond et al, there is not an actual market for most health programs and, thus, there is no \"gold standard\" against which one can compare WTP values \[[@B22]\]. It is, therefore, difficult to establish criterion validity in this context. Conclusion ========== This study reveals that the majority, but not all, of adult inpatients receiving parenteral antibiotic therapy who are referred to an outpatient parenteral antibiotic therapy program prefer to be treated at home. Income and treatment location appear to independently predict their willingness to pay. Competing interests =================== This was an unfunded study. The authors declare that they have no competing interests. Authors\' contributions ======================= CM made substantial contributions to the conception, design, analysis and interpretation of the data; he was involved in the drafting of the article and revising it critically for intellectual content and has given final approval for the current version to be published. LF made substantial contributions to the conception, design, analysis and interpretation of the data; she was involved in the drafting of the article and revising it critically for intellectual content and has given final approval for the current version to be published. AG contributed to the study design, made substantial contributions to data collection, and assisted in the interpretation of the data; he was involved in the drafting of the article, revising it, and has given final approval for the current version to be published. AW contributed to the study conception, design, daily supervision of and participation in data collection, analysis and interpretation of the data; she was involved in the drafting of the article, revising it, and has given final approval for the current version to be published. LC contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. RN contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. CL contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. ST contributed to the study design, data collection and interpretation of the data; she was involved in the drafting the article and has given final approval for the current version to be published. BF contributed to the study design, data collection and interpretation of the data; she was involved in the drafting of the article and has given final approval for the current version to be published. PJ was the coordinating investigator, made substantial contributions to the conception, design, analysis and interpretation of the data; he was involved in the drafting of the article and revising it critically for intellectual content and has given final approval for the current version to be published. All authors take public responsibility for appropriate portions of the content of the manuscript and all authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/43/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 WTP BMC HSR Version 1 PJJ appendix 1.doc. This file contains the questionnaire used for this study ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The investigators are (or were) members of the Pharmaceutical Sciences Clinical Service Unit at VHHSC and this work was conducted as part of their professional and academic obligation to the institution, and to the patients we serve. We thank these patients, and our colleagues for their contributions to this effort.

PubMed Central

2024-06-05T03:55:58.230340

2005-6-7

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168895/", "journal": "BMC Health Serv Res. 2005 Jun 7; 5:43", "authors": [ { "first": "Carlo A", "last": "Marra" }, { "first": "Luciana", "last": "Frighetto" }, { "first": "Alan F", "last": "Goodfellow" }, { "first": "Amy O", "last": "Wai" }, { "first": "M Lynn", "last": "Chase" }, { "first": "Ruth E", "last": "Nicol" }, { "first": "Carole A", "last": "Leong" }, { "first": "Sally", "last": "Tomlinson" }, { "first": "Barbara M", "last": "Ferreira" }, { "first": "Peter J", "last": "Jewesson" } ] }

PMC1168896

Background ========== Transfer-messenger RNA (tmRNA), also known as 10Sa RNA or ssrA RNA, is a hybrid of a tRNA-like domain (TLD) and a mRNA-like domain (MLD) connected by a variable number of pseudoknots \[[@B1]\]. TmRNA is a stable and essential component of *trans-*translation, a quality-control process that rescues ribosomes stalled on mRNAs lacking stop codons. During *trans-*translation, ribosomes switch from a defective mRNA (lacking its translation-termination signal) to the MLD of tmRNA. Because a stop codon is provided by the tmRNA, the ribosomes can dissociate and recycle \[[@B2]\]. As an additional advantage, the tandem translation of the two templates generates a tagged polypeptide which is degraded by housekeeping proteases \[[@B3],[@B4]\]. For tagging, tmRNA has to be charged by aminoacyl-tRNA synthetases \[[@B5]\]. Assisted by protein SmpB, the charged tmRNA is delivered to stalled ribosomes as a quaternary complex with EF-Tu and GTP. Binding of tmRNA to ribosomes is facilitated by ribosomal protein S1, which interacts with the MLD and pseudoknots but not with the TLD \[[@B6]-[@B9]\]. Recently, cryo-EM revealed the shape of the tmRNA associated with SmpB and EF-Tu in its ribosome-bound form \[[@B10]\]. Despite this significant progress, high-resolution structures as obtained by NMR and X-ray crystallography are unavailable and expected to be difficult to obtain in the foreseeable future due to the relatively large size and flexibility of the tmRNA. In the present work we used a stepwise procedure for arriving at high-resolution models for the entire tmRNA molecule. First, 2-D structures were obtained by covariation analysis of a large number of tmRNA sequences. The basepairing information was submitted to the ERNA-3D modeling program \[[@B11]\] to build the helical sections. Structural motifs of the loops and turns were identified in SCOR \[[@B12]\], high-resolution data were extracted from known structures, and these data were incorporated into the models. Overall, significantly improved 3-D models were obtained which will be useful to understand the role of tmRNA in *trans*-translation. The described approach can be adapted to obtain high-resolution models of the members of other RNA families. Results ======= Identification of tmRNA sequences --------------------------------- The tmRNA sequences were identified previously and subjected to comparative sequence analysis (CSA) as described \[[@B1],[@B13]\]. New tmRNA sequences were obtained from the tmRNA website \[[@B14]\], through keyword searches of the literature and GenBank \[[@B15]\], or BLAST \[[@B16],[@B17]\], and various genome sequencing projects. The new sequences were examined iteratively as described in Materials and Methods to confirm tmRNA identity, remove sequence duplications, and create a meaningful alignment. New potential tmRNA sequences were maintained as a preliminary alignment in BioEdit \[[@B18]\], separate EMBL-formatted sequence files, and a HTML-formatted phylogenetic list. The sequences were ordered phylogenetically using the information in the Ribosomal Database Project (RDP) \[[@B19]\]. If the organism name was not listed in the RDP, the sequence was placed next to its closest relative using the NCBI Taxonomy resource \[[@B20]\]. Selection of tmRNA sequences ---------------------------- The new sequences were confirmed individually as tmRNAs by comparison with the closest relative using the pairwise alignment feature of BioEdit \[[@B18]\]. If there was a lack of obvious similarity, the sequence was inspected for evidence of biological features such as the ability to form a TLD and an open reading frame. Furthermore, the possibility of a two-part tmRNA was considered. A sequence suspected to be a new tmRNA was investigated further by CSA \[[@B13]\] as described in Materials and Methods. Potential new sequences of the alpha-Proteobacteria and some Cyanobacteria that were encoded in two separate sections of their genes \[[@B21]\], were compared to the two-part tmRNA sequence from a closest relative for effective comparison with the one-part tmRNAs. The 3\'- and 5\'-ends of each section were determined by pairwise alignment to generate a single sequence. Each of the 20 new two-part tmRNAs (14 sequences from alpha-Proteobacteria and six from Cyanobacteria) was subjected to this rearrangement. Comparative Sequence Analysis ----------------------------- Sequences were ordered phylogenetically using the RDP \[[@B19]\] as a guide or by alignment with the closest relative. Identical regions were aligned first. Subsequently, similar regions were aligned using invariant positions as signposts. Regions of biological significance, such as the resume and stop codons, were then considered. Finally, common secondary structure features were used to align regions that lacked primary structure similarity or biological features. Supported Watson-Crick basepairs and G-U interactions were indicated in the alignment by uppercase letters. Gaps were introduced to account for differences in sequence length and to avoid the alignment of dissimilar regions. Secondary structure was determined using covariation analysis as described \[[@B13]\] (see also Materials and Methods). The alignment was examined to identify compensatory base changes (CBCs) and other covariations. The numbers of CBCs and mismatches between the alignment columns were counted. CBCs provided positive evidence for the existence of a basepair; mismatches provided negative evidence. If the number of compensatory base changes was two times or greater than the number of mismatches, the basepair was considered supported. If a basepair was invariant, no evidence for or against its existence could be gained from CSA. A basepair was considered specific to a particular phylogenetic group if it was proven only in that group. Quality control --------------- To check for the proper assignment of basepairs, the alignment was sent through an automated pipeline of programs from RNAdbTools \[[@B22]\]. The output was inspected visually and corrections were made manually using the BioEdit program \[[@B18]\]. The revised alignment was resubmitted to RNAdbTools, and the review process was repeated until a satisfactory alignment was produced. TmRNA alignment --------------- The final alignment contained a total of 274 tmRNA sequences in 16 bacterial phylogenetic groups. A complete phylogenetic list is available at the tmRDB \[[@B23]\]. There was a substantial increase in the number of two-part tmRNAs for a total of 27 sequences: 20 from alpha-Proteobacteria (20 tmRNAs), one mitochondrial tmRNA, and six cyanobacterial tmRNAs. The nine organelle sequences included one from a cyanelle, six from chloroplasts, one from a plastid, and one from the *Reclinomonas americana*mitochondrion. The typical tmRNA was about 350 nucleotides long. The *R. americana*mitochondrion tmRNA contained only 189 nucleotides and, since it appeared to lack an ORF, may not be functional. Excluding this exception and any partial tmRNAs, the tmRNA of *Synechococcus*species PCC7009 was the shortest (250 nucleotides), and the longest was from *Chlamydophila psittaci*(425 nucleotides). The tmRNA alignment is provided as additional files [1](#S1){ref-type="supplementary-material"}: tmRNA-alignment.html, [2](#S2){ref-type="supplementary-material"}: tmRNA-alignment-wide.txt, [3](#S3){ref-type="supplementary-material"}:tmRNA-alignment-92col.txt, and [4](#S4){ref-type="supplementary-material"}: tmRNA-alignment.msf. Secondary structure of tmRNA ---------------------------- The tmRNA secondary structure features were extracted from the alignment and are listed in phylogenetic order in Table [1](#T1){ref-type="table"}. The representative secondary structure of *Escherichia coli*tmRNA is shown in Figure [1](#F1){ref-type="fig"}. Secondary structures of *Bacillus anthracis*and *Caulobacter crescentus*are presented in as [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf and [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf, respectively. TLD (helices 1, 2a and 12) -------------------------- Although a prominent feature of each tmRNA, the TLD was relatively weakly supported by CSA due to a high degree of sequence conservation. However, the structure of this region is well established by experimental evidence \[[@B24]-[@B26]\]. Helix 1 contained seven basepairs and was usually continuous with the exception of the *Anabaena*species tmRNA, which contained an insertion in the 3\'-portion of helix 1. The first pair (1G-C359 in *E. coli*tmRNA) was conserved with one exception in *Alcaligenes eutrophus*where there was a 1U-C345 mismatch possibly due to a sequencing error. The second (2G-C358,*E. coli*numbering, Figure [1](#F1){ref-type="fig"}) and third pair (3G-U357) of helix 1 were invariant and therefore neither supported nor disproved by CSA. The identities of the bases involved in the fourth (4G-C356) and fifth pair varied. The closing pair of helix 1 (7G-C353) was conserved with the exception of a 7U-A388 pair the *Trichodesmium erythraeum*tmRNA. The single-stranded region between helices 1 and 2a ranged from ten in *Dehalococcoides ethenogenes*to 13 nucleotides in one *Clostridium acetobutylicum*sequence. A U-A basepair (U6 in chain A with A88 in chain B) was possible in the *R. americana*mitochondrion tmRNA. Helix 2a was equivalent to the anticodon stem of tRNA and contained eight supported basepairs as well as a short variable internal loop in the 5\'half of the helix that occurred in a few sequences (e.g. *Caulobacter crescentus*, see [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). The first position in the helix was a conserved cytosine (C21 in *E. coli*) which formed a weakly-supported basepair with the conserved G333. The partial tmRNA from the chloroplast of *Pavlova lutheri*contained a uracil in the first position, but no information regarding the 3\'portion of helix 2a was available. The T-loop and helix 12 were highly conserved, although many sequences lacked information about helix 12 due to primer annealing during PCR amplification. Helix 12 contained four strongly supported basepairs and a fifth conserved G-C pair (340G-C348 in *E. coli*; Figure [1](#F1){ref-type="fig"}). The *Dehalococcoides ethenogenes*tmRNA had the potential to form a sixth basepair in helix 12. Helix 12 was almost always continuous, except for the tmRNA of *Carboxydothermus hydrogenoformans*which possessed four basepairs and a mismatched U333 and C347. A 331-GG-332 preceded U333 in *C. hydrogenoformans*and followed the conserved 328-GAC-330. Therefore, U333 was unlikely to pair. In the T-loop, the U341 and U342 (*E. coli*tmRNA) seen in most sequences were replaced by two guanines in the tmRNA from the *R. americana*mitochondrion (G79 and G80 in chainB) \[[@B21]\]. In the tmRNA from *Caulobacter crescentus*, the nucleotide corresponding to U342 in *E. coli*tmRNA was changed to G62 in chainB (see [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). Helical sections 2b, 2c and 2d ------------------------------ Overall, sections 2b, 2c, and 2d were well supported. Sections 2a and 2b were separated by a variable loop ranging from one to seven nucleotides in the 5\'portion and from one to nine nucleotides in the 3\'portion. Sections 2b and 2c had the potential to form a continuously stacked helix (e.g. in *Chlamydophila psittaci*tmRNA). Usually, a bulge of two to six nucleotides separated helical sections 2c and 2d (residues 309--311 in *E. coli*tmRNA, Figure [1](#F1){ref-type="fig"}). An asymmetrical loop was present in some sequences (for example, residues 40--41 in chainA, and 27--31 in chainB of *Caulobacter crescentus*tmRNA, see [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). Helix 2d was the most conserved of the three helical sections. The G43-U308 basepair (*E. coli*numbering) in helix 2d was only weakly supported, conserved in most phylogenetic groups, but altered in the Thermatogales, Cyanobacteria, alpha-Proteobacteria, and Gram-positive bacteria. A 46A-U334 pair was possible in the *Synechocystis*species PCC6803 tmRNA. Pseudoknot 1 (helices 3 and 4) ------------------------------ Pseudoknot 1 (pk1) was well supported. Of the three connecting regions, the two 5\'-regions were very short (no or only one residue) while the third was relatively long (one to 11 residues). All pseudoknots in tmRNA followed the same general design \[[@B27]\]. Most sequences contained helices 3 and 4, with the exception of the tmRNA from *Oenococcus oeni*and the partial sequence from the chloroplast of *Pavlova lutheri*, both of which lacked helix 4 and thus did not form a pseudoknot. Helix 3 usually contained five basepairs. However, a sixth pair was possible in some bacteria. Helix 4 could be split into helicalsections 4a and 4b by a bulge seen in 46 sequences (position57 in *B. anthracis*tmRNA; see [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf) or an internal loop seen in 52 tmRNA sequences. The adenine-rich terminal loop between the downstream halves of helices 3 and 4 ranged in length from twoto13 nucleotides. The mRNA-like region (MLD) -------------------------- The MLD consisted of an open reading frame (ORF) preceding helix 5 and varied from 48 (*Heliobacillus mobilis*) to 126 nucleotides (*Odontella sinensis*chloroplast). The resume codon usually coded for alanine, but for glycine in 30 sequences (e.g. *Bacillus anthracis*), aspartic acid in three sequences (e.g. *Staphylococcus epidermidis*), arginine in two uncultured species (FS1 and LEM2), serine in the uncultured species RCA1, and glutamic acid in *Mycoplasma pulmonis*. Although helix 5 was only weakly supported by CBCs, recent site-directed mutagenesis experiments combined with functional studies *in vivo*and *in vitro*\[[@B28]\] provide strong evidence for its existence. One to three stop codons were located within the helix 5 loop. A single UAA stop codon was present in 157 sequences. UAG (17 sequences) or UGA (10 sequences) were used less frequently. In 85 sequences there were two in-frame stop codons, where UAA was always the first codon, followed by another UAA (73 sequences), UAG (10 sequences) or UGA (2 sequences). Curiously, two sequences (*Bacillus megaterium*and *Chloroflexus aurantiacu*) were found to contain three tandem in-frame stop codons. Pseudoknot 2 (helices 6 and 7) ------------------------------ Pseudoknot 2 was well supported and similar in overall design to pk1. Helical sections 6b and 6c showed a potential to form a continuous helix in *Thermotoga maritima*. In beta-Proteobacteria, 6b was replaced by a short hairpin6d \[[@B1]\]. Helix 6d was observed also in three tmRNAs of the gamma-Proteobacteria *Acidithiobacillus ferroxidans*and *Francisella tularensis*. Pseudoknot 3 (helices 8 and 9) ------------------------------ Pseudoknot 3 was well supported but missing in Cyanobacteria and the organelles (Table [1](#T1){ref-type="table"}). Helical sections 8a and 8b were likely to be continuously stacked because a single helix was present in some species such as *Aquifex aeolicus*. The unusual purine-rich internal loop between helical sections 8a and 8b was present in most gamma-Proteobacteria suggesting a special function. Pseudoknot 4 (helices 10 and 11) -------------------------------- This feature was well supported and was similar in design to the other tmRNA pseudoknots. Helical sections 10a and 10b had the potential to stack because a single helix was present in *Prevotella intermedia*. In some Cyanobacteria sequences, however, pk4 was replaced with two smaller tandem pseudoknots. Secondary structure prediction of the MLD ----------------------------------------- Because CSA was unable to determine secondary structure for a large portion of the MLD, energy calculations were carried out aimed to predict structure for the single-stranded portion of the open reading frame. The region corresponding to residues 79--107 of *E. coli*tmRNA (Figure [1](#F1){ref-type="fig"}) was extracted from the alignment. A representative alignment of 197 sequences was submitted to Mfold \[[@B29]\]. Each sequence had the potential to form at least one helix, designated \"m\" (see Figure [1](#F1){ref-type="fig"}, [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf, and [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). Two or more adjacent helices were predicted for 17 sequences. The number of basepairs varied from two in *Chloroflexus aurantiacus*to ten in *Mycoplasma pulmonis*. Secondary structures of three representative tmRNA molecules ------------------------------------------------------------ Secondary structures were determined for all sequences in the alignment but only three were extracted, diagrammed, and processed for 3-D modeling. Secondary structure of *E. coli*tmRNA ------------------------------------- The 363-nucleotide tmRNA of the gamma-Proteobacterium *Escherichia coli*represented the typical tmRNA containing the TLD, the MLD, and four pseudoknots (pk1 to pk4) encompassing the pseudoknot domain (PKD). The 90-GCA-92 resume triplet coded for alanine. Two in-frame UAA stop codons (positions 120--125) were located within the terminal loop of helix 5 (Figure [1](#F1){ref-type="fig"}). Three basepaired regions (shown boxed) were only weakly supported by CSA. Helixm (residues 87--98) was predicted only by energy calculations. A slightly different helix involving residues 88--100 has been suggested by footprinting of *E. coli*tmRNA \[[@B30]\]. The evidence for the 112U-A133 basepair was weak, but was included due to the possibility of extending helix 5 (Materials and Methods). Helical section 5a (residues 108--113 and 134--137) was enlarged by the weakly supported 108G-C137, 110U-A135 and 111U-G134. The 109C-G136 pair was disproved. In helix 10ab, the basepair between 256G-C275 was only weakly supported. Helix 10ab (residues 248--256 and 274--283) could be extended by the boxed 257U-G274 pair. Secondary structure of *Bacillus anthracis*tmRNA ------------------------------------------------ Overall, the secondary structure of *Bacillus anthracis*tmRNA (see [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf) was similar to that of *E. coli*. One notabledifference was a bulged uridine (U57) between helical sections 4a and 4b in pk1. A three-basepair helixm was predicted. The resume triplet (residues 89--91) coded for glycine, and the UAA stop codon was located at residues 119--121. Two weakly-supported pairs (108C-G132 and 109U-A131) extended helical section 5a. Secondary structure of *Caulobacter crescentus*tmRNA ---------------------------------------------------- *Caulobacter crescentus*tmRNA ([additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf) consisted of two chains, A and B, of 213 and 83 residues, respectively. The resume codon (82-GCG-84 of chainA) coded for alanine and was followed by a UAA stop codon at residues 121--123. Helical sectionsm1 and m2 were predicted by energy calculations. There was weak support for 5a (G109-U135 and 111C-G133 in chainA), and the 106U-A138, 107U-A137, 108C-G136, and 110C-G134 in chainA were disproved. The pseudoknots were relatively small. Helix 11 corresponded to the absent pk4 (residues 1--18 in chain B). Tertiary structure modeling and visualization of tmRNA ------------------------------------------------------ ERNA-3D, a program developed to model RNA in three dimensions \[[@B11]\], was used on an SGI workstation as described in Materials and Methods. *E. coli*tmRNA was selected because this tmRNA is the subject of extensive research. *B. anthracis*tmRNA was chosen as an example of a tmRNA from a Gram-positive bacterium, and *C. crescentus*tmRNA was selected it represents a two-part tmRNA. In order to create the initial models, the sequence and basepairing information were entered into an ERNA-3D input file to automatically generate A-form RNA for the helices sections and specify the single-stranded regions using ERNA-3D\'s algorithm \[[@B11]\]. Since ERNA-3D avoided an XYZ coordinate system as reference for the user, the manipulation of the model from the viewer\'s perspective was simple and intuitive. The coordinates of each model were saved in PDB format \[[@B31]\] for compatibility with other molecular modeling programs. Motifs (listed in Tables [2](#T2){ref-type="table"} to [4](#T4){ref-type="table"}) were selected to model the loops and turns of a particular tmRNA. ERNA-3D selection files were generated to define clusters and place the motif in 3-D without disturbance to the rest of the model. The 3-D cursor box was used to manipulate a cluster in three-dimensional space, similar to the manipulation of a section of a physical model. Numerous high-resolution structures determined by NMR or X-ray crystallography represented a rich source of detailed information for defining biologically meaningful motifs. The SCOR database \[[@B12]\] provided a way to find suitabletemplates. In rare cases when a SCOR search for a motif did not result in an acceptable match (e.g. motif 9, Table [2](#T2){ref-type="table"}), the nucleotides were positioned manually in ERNA-3D. Otherwise, the coordinates were obtained from the Protein Data Bank PDB \[[@B32]\], extracted using the program Swiss-PDBViewer \[[@B33]\], and imported into ERNA-3D. The source motif and the region to be modeled were selected as separate clusters and aligned in three dimensions using common features (usually a shared basepair). Once superimposed, the coordinates of the residues in the source motif were copied onto the corresponding residues in the model. The template was then deleted, leaving a biologically meaningful structure. The backbone connections between the motif and the rest of the model were inspected visually and, if needed, manual adjustments were made to correct bond lengths and tetrahedral angles involving the phosphorous atom at the joint between the extracted motif and the helical structures generated by ERNA-3D. As an example of the motif modeling process, the purine-rich loop in *E. coli*pk3 (positions 204--206 and 223--225) was constructed using a similar loop in the 30S ribosomal subunit. First, the purine-rich loop was defined as motif 11a (Table [2](#T2){ref-type="table"}), and used to search the SCOR database. Positions 780--782 and 800--802 in the structure of the *Thermus thermophilus*30S ribosomal subunit \[[@B34]\] were found to conform to the motif. The 30S ribosomal subunit coordinates (1J5E.pdb in this case) were downloaded from the PDB and displayed using Swiss-PDBViewer. The coordinates of the loop and the closing basepairs were extracted and inspected to confirm that the structure was compatible. The clustered regions were aligned with the ends of helical sections 8a and 8b at the basepairs 203U-G226 and 207A-U222 of the *E. coli*model and 779C-G803 and 783C-G799 of the template. Template positions 780-AAA-782 and 800-GUA-802 were then copied onto 204-GGA-206 and 223-GAA-225 of the model. The template was deleted and the bond lengths and angles involving the atoms of the phosphates of residuesU203, G222, A206, and U222 were adjusted. In some instances, the tmRNA sequence alignment was reinvestigated using ideas derived from the 3-D model. For example, the alignments of pk1 in *Bacillus anthracis*tmRNA and relatives was changed from a two nucleotide bulge (56-AU-57) between helical sections 4a and 4b to a more feasible and equally well supported one-nucleotide bulge (U57, see [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf). The alignment of helix 10 in pk3 in *B. anthracis*tmRNA and relatives was altered from a 237C-A269 mismatch and an asymmetrical loop (C239 and 266-GU-267) to a single looped-out C269. The alignment of pk3 of *Caulobacter crescentus*and relatives was changed from four basepairs and a weakly supported fifth pair in helix 8 (between 174G-C196 of chainA) to the four basepair structure seen in [additional file 6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf. Information about spatial neighborhoods as obtained from cross-linking, site-directed mutagenesis, and functional testing of *E. coli*tmRNA was introduced and is described in detail below. All models were inspected visually for correct bond angles and distances around the phosphorous atoms at the joints between the extracted motifs and the helical regions generated by ERNA-3D. The coordinates are provided as [additional file 7](#S7){ref-type="supplementary-material"}: Ecoli-closed.pdb, 8: Ecoli-open.pdb, 9: Banthracis-closed.pdb, and 10: Ccrescentus-closed.pdb. 3-D model of *E. coli*tmRNA --------------------------- The model shown as a ribbon diagram in Figure [3](#F3){ref-type="fig"} consists of a compacted MLD and PKD with the TLD extending from the body of the molecule due to the near-coaxial stacking of the helix 2 sections. The coordinates for the TLD were taken from a previous model \[[@B35]\] which is based on two cross-linked sites, one formed between nucleotides U9/U10 near the 5\' end and nucleotides C346/U347 in the T loop, the other involving residues at positions 25--28 and 326--329 within helix 2a (motif 2 in Table [2](#T2){ref-type="table"}). Important features of the TLD include the non-Watson-Crick base pairs formed by 19-GA-20 and 333-GA-334 which have been confirmed by site-directed mutagenesis \[[@B36]\]. A very efficient UV-induced cross-link observed between the stop codon loop of helix 5 and pk2 of *E. coli*tmRNA (Wower et al., unpublished) introduced a considerable constraint of helices 5, 6, and 7, and, as has been shown recently, is consistent with the cryo-EM structure of ribosome-bound tmRNA of the initial stage of *trans*-translation \[[@B10]\]. Also considered was the previously-discovered covariation \[[@B37]\] between C44 and C66 (*E. coli*numbering, Figure [1](#F1){ref-type="fig"}) which determines the orientation of helix 2 in relation to helix 3 and thus the approximate angle by which the TLD protrudes. The 44/66 covariation is strongly supported (26 covariations *versus*four mismatches) in an alignment of 143 representative sequences (not shown). Since this is a non-Watson-Crick covariation, it is difficult to propose a precise structure in this region. More extensive studies will be required to better understand the nature of the 44/66 covariation. Overall, the distance between the 3\'end and A231 in pk3 was 180 Å, and 70 Å between the outside edges of pk1 and pk4. Helix 5 and pk2 were positioned in a parallel fashion. The nucleotides in the bulge between helical sections 6a and 6b (motif 9, Table [2](#T2){ref-type="table"}) were adjusted manually to allow for a close fit of helix 5 and pk2. This model is supported by the finding that mutations that destroy base pairing in helix 5 substantially decrease tmRNA tagging activity (Wower et al, in press) and abolish the long-distance interaction between helix 5 and pk2 as judged by the absence of a cross-link between the stop codon loop of helix 5 and pk2 (Wower, unpublished data). Evidence for the existence of helix 5 has been provided by the analysis of compensatory mutations which completely restore tagging \[[@B28]\]. Each of the four pseudoknots displayed the previously determined structural properties characterized by extensive helical stacking \[[@B27]\]. The MLD and pseudoknots were arranged in a central loop with the resume codon positioned near the junction between helices 2a and 2b (motif 3a, Table [2](#T2){ref-type="table"}). Our model reflects the finding that the pseudoknots are functionally interchangeable \[[@B38]\] and thus are likely to retain a considerable level of structural independence. Furthermore, data derived from cross-linking experiments showing that pk2 and pk3 are in close proximity whereas helix 5 and pk4 are further apart (Wower, unpublished data) agree with the presented model. 3-D model of *B. anthracis*tmRNA -------------------------------- The 3-D model of the *B. anthracis*tmRNA (Figure [4](#F4){ref-type="fig"}) was similar to the *E. coli*model. A sharper angle was modeled between helix 2 axis and the PKD. The dimensions were 150 Å from the 3\'end to the distant edge of pk2, and approximately 80 Å between the outer edges of pk2 and pk4, respectively. 3-D model of *C. crescentus*two-part tmRNA ------------------------------------------ Compared to the other two tmRNA models, the two-part tmRNA *Caulobacter crescentus*model (Figure [5](#F5){ref-type="fig"}) was slightly more elongated. It measured 195 Å from the 3\'-end of chainB to the single-stranded region between pk2 and pk3. The distance between helix 5 and the 3\'end of chainA was 55 Å. Discussion ========== We have compared a growing number of tmRNA sequences from all groups of bacteria to produce an alignment from which the secondary structure of any tmRNA could be easily extracted. Most basepairs were supported by phylogenetic evidence, whereas only a few helical sections required energy calculations. Uncertainties in assignment of basepairs, such as the pseudoknot region of chloroplasts and one-piece cyanobacterial tmRNAs, may be eliminated in the future when more sequences will become available. The common layout of the secondary structures indicated a similar function in all bacteria. The number and size of the pseudoknots varied, supporting the idea that the pseudoknots may only enhance the essential functions carried by the TLD and the MLD \[[@B28]\]. Differences in the secondary structure features were usually not random but occurred between groups of related organisms. For example, helix 6d was present only in the beta- and three close relatives of the gamma-Proteobacteria. Whether these group-specific features are responsible for differences in the *trans*-translation mechanism remains to be determined. However, strategies that exploit these differences, for example for developing new antibiotics targeted at a specific group of bacteria, can now be envisioned. In principle, tertiary structure models of any tmRNA in the alignment could be built using the described procedures. Here, we have shown how to generate a biologically more meaningful model of *E. coli*tmRNA which represents a significant progress from a previous model \[[@B27]\]. We also constructed 3-D model of the tmRNAs of *B. anthracis*and *C. crescentus*. Overall, the three models were similar in shape and size confirming that all tmRNAs have the potential to function similarly in *trans*-translation. The TLD mimicked the L-shape of canonical tRNA \[[@B39]\] and may be necessary for proper association tmRNA with the EF-Tu, SmpB, and subsequent binding to the ribosomal A-site. The lack of a D-stem has been suggested to confer flexibility \[[@B40]\], but SmpB may be responsible for stabilizing this region \[[@B41],[@B42]\]. Differences in the shapes of the three tmRNA models (e.g. the angle between helix 2 and the main body of the molecule) may be due to the difficulty in determining the precise arrangement of the pseudoknots. Considering that the pseudoknots are likely to constitute relatively independent structural units, conformational changes might occur around the connecting single strands, as well as in the MLD and the weakly-supported helix m. TmRNA may become less flexible when bound to proteins such as SmpB and ribosomal protein S1. EF-Tu, however, likely binds to the coaxially-stacked helices 1 and 12 \[[@B43]\], and therefore appears to have little effect on the conformation of the TLD. Protein SmpB was found to bind near helix 2a \[[@B41],[@B44]\], has two RNA binding sites \[[@B45]\], and thus could make additional contacts with other regions. Protein S1 is the largest ribosomal protein, has been shown to be close to numerous sites, and to be required for the binding of tmRNA to the ribosome \[[@B7]\]. Since S1 is a flexible, beadlike protein \[[@B46]\] it may not restrict the conformational potential of the tmRNA molecule. Instead, the protein may instill some constraint to the large central loop formed by the PKD and the MLD. Because S1 is known to melt helices in mRNAs \[[@B47]\], it is also possible that it unwinds helix m and exposes the resume codon and the preceding nucleotides U85 and A86 for efficient *trans*-translation \[[@B48],[@B49]\]. The tmRNA models show the resume codon in close proximity to the internal loop formed between helical sections 2a and 2b. This arrangement would allow the ribosome to \"jump\" a relatively short distance from the end of the broken mRNA onto the ORF of tmRNA. In a recent cryo-EM study of the initial stage of *trans*-translation \[[@B10]\], the tRNA-like region, SmpB, EF-Tu, and part of pk4 were located in the A-site of the ribosome. We suggest that this more open arrangement is made possible due to the flexibility of tmRNA, the melting of helix m, and/or a change in conformation induced by the binding of tmRNA to the ribosome (Figure [6](#F6){ref-type="fig"}). The opening of the central loop seems to be accompanied by a rotation of the TLD around the helix 2 axis (compare Figure [6A](#F6){ref-type="fig"} and [6B](#F6){ref-type="fig"}) and thus might properly align the resume codon with the 3\'-end of broken mRNA in the ribosomal decoding centre. At the later stages of the transit of tmRNA across the ribosome even more dramatic conformational changes were shown to disrupt helix 5 and the pseudoknotted regions (Wower et al, in press). These downstream alterations are likely mediated not by protein S1 but by the intrinsic helicase activity of the ribosome \[[@B50]\] and are required to maintain the ribosomal subunits in close proximity to the unfolded tmRNA in order to monitor *trans*-translation. Conclusion ========== This study significantly advances our understanding of *trans*-translation by providing biologically feasible 3-D models for the entire tmRNA molecule. Although the modeling of only three tmRNAs has been described here, 3-D models of every tmRNA can be extracted from the alignment. The models are characterized by a functionally significant features, including biologically relevant structures for the single-stranded regions and the close proximity between the TLD and the resume codon. Conformational changes induced by binding of tmRNA to SmpB, ribosomal protein S1, and the ribosome suggest a transformation of a free compact tmRNA to a more open ribosome-bound structure. The comparative modeling approach described here for tmRNA is easily adapted for other RNA classes. Methods ======= Comparative sequence analysis ----------------------------- The tmRNA sequences were arranged in phylogenetic order using information available in the RDP \[[@B19]\]. When the phylogenetic order could not be determined, the sequence was placed next to the closest relative as determined by the ClustalW plug-in of BioEdit \[[@B18],[@B51]\]. The sequences were made available at the tmRDB \[[@B52]\]. Aligning was done manually using BioEdit \[[@B18]\] with details described previously \[[@B13]\]. Briefly, closely related sequences were aligned first. Then, invariant positions were used as guides to align the dissimilar regions. Next, common secondary structure elements were identified by observing covariations and find support for basepairs, tertiary interactions, or other structural features. Compensatory base changes (CBCs) were observed if a change in one residue of a Watson-Crick or G-U pair was compensated by a second change to maintain basepairing. Two residues were mismatched if they did not form a Watson-Crick or G-U pair. CBCs and mismatches were counted to determine positive and negative evidence in order to prove or disprove the existence of a particular pair. A basepair was considered proven if there was at least twice as much positive than negative evidence. Invariant pairs provided neither positive nor negative evidence. If a basepair was proven in one phylogenetic group and disproved in another group, the basepair was considered to be specific to that group. The alignment and suggested CBCs were checked using RNAdbTools \[[@B22]\] to eliminate incorrectly-paired nucleotides, suggest extensions of helices, and determine the phylogenetic support for each basepair. Weakly supported basepairs adjacent to supported basepairs were considered an extension of the helix and usually included in the secondary structures (Figure [1](#F1){ref-type="fig"}, [additional file 5](#S5){ref-type="supplementary-material"}: Banthracis2D.pdf and [6](#S6){ref-type="supplementary-material"}: Ccrescentus2D.pdf). 3-D model building ------------------ The secondary structure information was used as input for ERNA-3D \[[@B11]\] installed on an SGI workstation running IRIX 6.5. ERNA-3D generated A-form RNA for each helix and calculated the conformations of single-stranded regions. The models were examined using CrystalEyes stereovision goggles and an StereoGraphics infrared emitter. Structural motifs were identified using SCOR \[[@B12]\], the coordinates were obtained from the Protein Data Bank (PDB) \[[@B32]\], extracted using Swiss-PDBViewer \[[@B33]\], and superimposed onto the model. Data obtained from site-directed mutagenesis, cross-linking experiments, or the literature were incorporated, and bond lengths and angles were adjusted manually to produce biologically feasible models. The final models were saved in PDB format (additional files [7](#S7){ref-type="supplementary-material"}: Ecoli-closed.pdb, [8](#S8){ref-type="supplementary-material"}: Ecoli-open.pdb, [9](#S9){ref-type="supplementary-material"}: Banthracis-closed.pdb, and [10](#S10){ref-type="supplementary-material"}: Ccrescentus-closed.pdb and viewed in iMol \[[@B53]\] to create the ribbon diagrams shown in Figures [3](#F3){ref-type="fig"} to [5](#F5){ref-type="fig"}. Authors\' contributions ======================= JB identified and aligned a large number of the tmRNA sequences, carried out the 3-D modeling experiments, and drafted portions of the text. CZ conceived the study, participated in all of its aspects, and wrote the final manuscript. FM developed the ERNA-3D program and incorporated new functions to allow comparative modeling of RNA. IW developed assays which allowed to test the biological significance of the *Escherichia coli*tmRNA model. JW participated in the coordination of the study and provided RNA-RNA cross-link data which were crucial for constraining the models in 3-D. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 tmRNA alignment. Species names are shown on the left with their tmRDB ID (see <http://psyche.uthct.edu/dbs/tmRDB/tmRDB.html>). Supported base pairing are shown with upper case letters and are indicated on the bottom. Secondary structure features are indicated on the top. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 tmRNA alignment. Species names are shown on the left with their tmRDB ID (see <http://psyche.uthct.edu/dbs/tmRDB/tmRDB.html>). Supported base pairing are shown with upper case letters and are indicated on the bottom. Secondary structure features are indicated on the top. Each sequence is shown on a single long line suitable for import into other programs. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 tmRNA alignment. Species names are shown on the left with their tmRDB ID (see <http://psyche.uthct.edu/dbs/tmRDB/tmRDB.html>). Supported base pairing are shown with upper case letters and are indicated on the bottom. Secondary structure features are indicated on the top. The alignment is shown in sections for easier viewing and printing. This arrangement is not suited for import into other programs. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 4 tmRNA alignment. tmRNA alignment in msf format suitable for import into editors which support the msf format. Can be used to convert into other sequence formats using, for example the Readseq server at <http://bioweb.pasteur.fr/seqanal/interfaces/readseq.html> ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 5 Secondary structure of *Bacillus anthracis*tmRNA. Phylogenetically-supported helices are highlighted in gray and numbered from 1 to 12. The 5\' and 3\' ends are indicated. Arrows represent connections from 5\' to 3\'. Residues are numbered in increments of ten. Weakly supported regions and basepairs are show in boxes. The star labels the first nucleotide of the resume codon. The tag peptide sequence is shown below the mRNA-like region. The stop codon is indicated with a solid arrowheads. Three domains are distinguished: The tRNA-like domain (TLD), the mRNA-like domain (MLD), and the pseudoknot domain (PKD). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 6 Secondary structure of *Caulobacter crescentus*tmRNA. Phylogenetically-supported helices are highlighted in gray and numbered from 1 to 12. The 5\' and 3\' ends of both chains are indicated. Arrows represent connections from 5\' to 3\'. Residues are numbered in increments of ten. Weakly supported regions and basepairs are show in boxes. The star labels the first nucleotide of the resume codon. The tag peptide sequence is shown below the mRNA-like region. The stop codons are indicated with solid arrowheads. Three domains are distinguished: The tRNA-like domain (TLD), the mRNA-like domain (MLD), and the pseudoknot domain (PKD). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 7 PDB coordinates of the closed form of *Escherichia coli*tmRNA. 3-D model of the closed form of *Escherichia coli*tmRNA. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 8 PDB coordinates of the open form of *Escherichia coli*tmRNA. 3-D model of the open form of *Escherichia coli*tmRNA. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 9 PDB coordinates of the closed form of *Bacillus anthracis*tmRNA. 3-D model of the open form of *Bacillus anthracis*tmRNA. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 10 PDB coordinates of the closed form of *Caulobacter crescentus*tmRNA. 3-D model of the closed form of *Caulobacter crescentus*tmRNA. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was supported by Grant GM58267 to JW from the National Institutes of Health. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Secondary structure of *E. coli*tmRNA**. Phylogenetically-supported helices are highlighted in gray and numbered from 1 to 12. The 5\' and 3\' ends are indicated. Arrows represent connections from 5\' to 3\'. Residues are numbered in increments of ten. Weakly supported regions and basepairs are shown in boxes. The disproved potential pairing of C109 with G136 is labeled with an open arrowhead. The star labels the first nucleotide of the resume codon. The tag peptide sequence is shown below the mRNA-like region. The stop codons are indicated with solid arrowheads. Three domains are distinguished: the tRNA-like domain (TLD), the mRNA-like domain (MLD), and the pseudoknot domain (PKD). ::: ![](1471-2199-6-14-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Motif modeling procedure**. Motifs, for example the nonamer-loop shown in the top-left panel, were identified in the known high-resolution structures (top-right) with the help of SCOR \[12\]. The PDB coordinates were extracted (bottom-right) and compared with the 3-D model generated by ERNA-3D (bottom-left) to deduce relevant models. ::: ![](1471-2199-6-14-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **3-D model of *Escherichia coli*tmRNA**. The 3-D model of *Escherichia coli*tmRNA is viewed as a ribbon diagram from the side in panel A, the top in panel B, and in panel C turned by approximately 90° around the y-axis in relation to A. Panel D shows a representation of the corresponding 2-D structure using the identical coloring scheme. Labeled are the 5\' and 3\' ends, the resume (R) and stop codons (S), and the three domains (TLD, MLD, PKD). The figure was produced with iMol \[53\] and the PDB coordinates of [additional file 7](#S7){ref-type="supplementary-material"}: Ecoli-closed.pdb. ::: ![](1471-2199-6-14-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **3-D model of *Bacillus anthracis*tmRNA**. The 3-D model of *Bacillus anthracis*tmRNA is viewed as a ribbon diagram from the side in panel A, the top in panel B, and in panel C turned by approximately 90° around the y-axis in relation to A. Panel D shows a representation of the corresponding 2-D structure using the identical coloring scheme. Labeled are the 5\' and 3\' ends, the resume (R) and stop codons (S), and the three domains (TLD, MLD, PKD). The figure was produced with iMol \[53\] and the PDB coordinates of [additional file 9](#S9){ref-type="supplementary-material"}: Banthracis-closed.pdb. ::: ![](1471-2199-6-14-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **3-D model of *Caulobacter crescentus*tmRNA**. The 3-D model of *Caulobacter crescentus s*tmRNA is viewed as a ribbon diagram from the side in panel A, the top in panel B, and in panel C turned by approximately 90° around the y-axis in relation to A. Panel D shows a representation of the corresponding 2-D structure using the identical coloring scheme. Labeled are the 5\' and 3\' ends, the resume (R) and stop codons (S), and the three domains (TLD, MLD, PKD). The figure was produced with iMol \[53\] and the PDB coordinates of [additional file 10](#S10){ref-type="supplementary-material"}: Ccrescentus-closed.pdb. ::: ![](1471-2199-6-14-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Conformational changes in *Escherichia coli*tmRNA**. Panel A: closed form of the *E. coli*model as shown in Figure 3. Panel B: open conformation adjusted to more closely resemble the ribosome-bound form as determined by cryo-EM \[10\] using [additional file 8](#S8){ref-type="supplementary-material"}: Ecoli-open.pdb. Panel C: coordinates extracted from the cryo-EM model \[10\]. The TLD is shown in dark purple, helix 2 in green, pk1 in yellow, helix 5 in pink, pk2 in turquoise, pk3 in red, and pk4 in dark blue. Panel D: Electron density map of the 50S subunit in light blue, the 30S subunit in yellow, and the bound tmRNA (in the absence of ribosomal protein S1) in dark blue (from \[10\]). ::: ![](1471-2199-6-14-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Phylogenetic distribution of tmRNA features ::: **Phylogenetic Group** **TLD** MLD **pk1** **pk2** **pk3** **pk4** **Other** -------------------------------------- --------- ------- --------- --------- --------- --------- ----------- Thermophilic Oxygen Reducers Thermatogales Green Non-sulfur & Bacteria elatives Flexibacter Cytophaga Bacteroides Green Sulfur Bacteria Planctomyces & elatives Cyanobacteria **1** **1,2** **3** Plastids **-** **-** Mitochondria **-** **-** **-** **-** **-** Fibrobacter/Acidobacter & elatives Spirochetes & elatives Proteobacteria, alpha **4** **5** **3** Proteobacteria, beta **6** **3** Proteobacteria, gamma **6** Proteobacteria, delta Proteobacteria, epsilon Fusobacteria Gram Positive Bacteria The tRNA-like domain (TLD), mRNA-like domain (MLD), and the four pseudoknots pk1 to pk4 are shown on the top. Other features peculiar to a phylogenetic group are in the right column. White fields indicate the presence, dashes the absence of a feature. Numbers suggest the following structural features: (1) certain Cyanobacteria lack these pseudoknots. (2) One-chain cyanobacterial tmRNAs contain two smaller tandem pseudoknots named pk4a and 4b. (3) The tmRNAs of some species in this group consist of two basepaired molecules \[14, 21\]. (4) The genus *Ricketsia*and its relatives lack pk2. (5) pk4 of the alpha-Proteobacteria has been reduced to a single helix (named helix 11). (6) Some species in this group contain an additional helix (helix 6d). ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Structural motifs used for the *Escherichia coli*tmRNA model ::: **Motif** **SCOR class** **tmRNA Res.** **Source Res.** **Coordinates** **Comments** ----------- --------------------------------------------- -------------------- ------------------------ ----------------------- --------------------------------- **1** 1--7, 353--363 1--7, 12--22 1IKD.pdb (chain W) ACCA end and G3-U357 pair **2** 8--28, 325--352 8--28, 325--352 tmx-34.pdb from tmRDB **3a** *internal loop with unpaired stacked bases* 29--33, 321--324 1775--1779, 1765--1768 1JJ2.pdb **3b** *stacked duplex with one non-WC pair* C35, A319 ERNA-3D **4** *stacked duplex with two non-WC pairs* 38--39, 315--316 2874--2875, 2882--2883 1JJ2.pdb **5** 309--311 ERNA-3D **6a** *pseudoknot* 49--78 1--33 1RNK.pdb pk1 **6b** *tetraloop* 87--98 5--8 1AFX.pdb the only YRRR tetraloop in SCOR **7** *nonaloop* 118--126 1834--1842 1JJ2.pdb **8** *one unpaired and stacked U* U131 U30 1B36.pdb **9** 171--174 ERNA-3D **10a** *stacked duplex with two non-WC pairs* 149--150, 165--166 288--289, 363--364 1JJ2.pdb **10b** *pseudoknot* 138--196 **6a** pk2 **11a** *internal loop* 204--206, 223--225 780--782, 800--802 1J5E.pdb **11b** *pseudoknot* 197--247 **6a** pk3 **12** *stacked duplex with one non-WC pair* G258, A273 A-G6, B-A27 420D.pdb **13a** *stacked duplex with one non-WC pair* C266, U296 ERNA-3D **13b** *pseudoknot* 248--299 **6a** pk4 Shown in columns one to four are the motif numbers in bold, their SCOR classification \[12\], the residue positions in the tmRNA model and the source structure. column five lists the filenames containing the atomic coordinates that were derived from the PDB \[31\], the tmRDB \[52\], or were generated by ERNA-3D \[11\]. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Structural motifs used for the Bacillus anthracis tmRNA mode ::: **Motif** **SCOR class** **tmRNA Res.** **Source Res.** **Coordinates** **Comments** ----------- ------------------------------------------- --------------------------------- -------------------------------- --------------------------- ------------------------------------- **1** 1--11, 12--27, 316--355 1--11, 13--28, 324--363 *E. coli***1** ACCA end and G3-U349 pair **2** *loop with stacked interdigitated bases* 28--33, 312--316 477--482, 451--455 1J5E.pdb **3** 300--301 ERNA-3D **4a** 43--53, 61--65, 294--299 43--53, 63--67, 302--308 *E. coli*model helix 2d, helix 3 **4b** *pseudoknot* 58--60, 66, 67--73 12--14, 20, 23--29, 1RNK.pdb pk1, helix 4b **4c** 54--56, 74--75 54--56, 77--78 *E. coli***6a** pk1, helix 4a **5** *heptaloop* 82--88 335--341 1JJ2.pdb **6a** *one looped out A* A112 210 1GID.pdb **6b** C114, C127 ERNA-3D **6c** *octaloop* 117--124 1499--1506 1JJ2.pdb **7a** 134--138, U163, A164, 165--169, 140--144, A171, A174, 175--179 *E. coli*10b helix 6a, looped out U163 and A164. **7b** *one looped out U* U158 U87 1JJ2.pdb **7c** 144--147, 148--156, 180--185 259--262, 264--272, 293--298 *E. coli***13b** helix 6c, helix 7, pk2 **8** G133, C170, 171--177, 178--179 C138, G181, 182--188, 190--191 *E. coli***10b** **9a** C190, 191--198, 208--215, 216 G200, 203--210, 219--226, U229 *E. coli***11a**, **11b** purine-rich loop, pk3 **9b** *pseudoknot* 202--206, 228--232 9--13, 28--32 1RNK.pdb pk3, helix 9 **10a** 199--201 154--156 *E. coli***6a** pk3 **10b** 217--227 229--239 *E. coli***6a** pk3 **11a** *one looped out C* C269 C6 1BVJ.pdb **11b** *stacked duplexes with four non-WC pairs* 241--245, 260--263 795--798, 815--818 1JJ2.pdb **12** 246--259, 275--290 259--277, 284--299 *E. coli***6a** helix 10c, helix 11ab, pk4 Motif numbers in bold, their SCOR classification \[12\], residue positions in the tmRNA model and the source structure are shown in columns one to four. Column five lists the filenames containing the atomic coordinates that were derived from the PDB \[31\], the *E. coli*model (see Table 2), or were generated by ERNA-3D \[11\]. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Structural motifs used for the *Caulobacter crescentus*tmRNA model ::: **Motif** **SCOR class** **tmRNA Res.** **Source Res.** **Coordinates** **Comments** ----------- -------------------------------------------------- ------------------------------------------------------------------ ------------------------------------------------------------ ----------------------------------- --------------------------------------------------------------- **1** A-1-13, A-14-21, B-50-83 1--13, 15--22, 322--355. *B. anthracis***1** \- **2a** *one non-Watson Crick pair and one looped-out A* A-22, B-48-49 14, 4--5 5MSF.pdb disruption in anticodon stem **2b** *loop with base triple* A-26-31; B-40-44 8--13, 24--28 1FMN.pdb **2c** A-C36, B-C35 ERNA-3D **3** A-40-41, B-27-31 ERNA-3D **4** A-47-61, A-62-70 196--210, 224--232 *B. anthracis***8a, 8b, 9a, 9c** pk1 **5** A-87-93 82--88 *B. anthracis***5** terminal heptaloop of helix m2 **6** *one looped out A and one or more non-WC pairs* A-78-81, A-99-101 113--116, 205--207 1GID.pdb internal loop between helices m1 and m2 **7** B-8-11 284--285, 291--292 1D6K.pdb **8a** A-112-114, A-131-132 ERNA-3D **8b** A-119-125 82--88 *B. anthracis***5** **9a** U148, C163 ERNA-3D **9b** A-156-165, A-146-153, A-167-171 157--166, 149--156, 192--196 *E. coli***10b** pk2. Single-stranded regions were adjusted to connect helices **10** A-154-156, A-170-171 54--55, 76--77 *E. coli***6a** Correction of two pairs in pk2 based on *E. coli*pk1 **11** A-211-214, A-177-182, A-184-193, A-177-178, A-192-198, A-194-209 360--363, 207--212, 213--222, 200--201, 228--234, 230--245 *E. coli***1**(ACCU tail) and 11b pk3 and ACCU tail of pk3 Motif numbers in bold, their SCOR classification \[12\], residue positions in the tmRNA model and the source structure are shown in columns one to four. Column five lists the filenames containing the atomic coordinates that were derived from the PDB \[31\], the *E. coli*or *B. anthracis*models (see Tables 2 and 3), or were generated by ERNA-3D \[11\]. :::

PubMed Central

2024-06-05T03:55:58.233010

2005-6-15

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168896/", "journal": "BMC Mol Biol. 2005 Jun 15; 6:14", "authors": [ { "first": "Jody", "last": "Burks" }, { "first": "Christian", "last": "Zwieb" }, { "first": "Florian", "last": "Müller" }, { "first": "Iwona", "last": "Wower" }, { "first": "Jacek", "last": "Wower" } ] }

PMC1168897

Background ========== The human small intestine is the longest segment of the gastrointestinal tract, approximately five to six meters long in adults. Although it is three to four times longer than the colon, small intestinal tumors are rare, forming about 1% of all gastrointestinal malignancies \[[@B1]\]. Like the colon, the small intestine is composed of multiple crypts (of Lieberkühn), which also contribute cells to finger-like projections called villi (Figure [1](#F1){ref-type="fig"}). Stem cells appear to reside near the crypt bases, producing numerous differentiated progeny that die within a week as they migrate upwards \[[@B2]-[@B4]\]. Stem cells also produce Paneth cells that migrate downwards, and subsequently survive a few weeks. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Schematic of a crypt of Lieberkühn and villus.**Stem cells probably reside within a niche just above the Paneth cells. Differentiated cells migrate out of the niche and up a villus, and die within a week. Illustrated are an isolated crypt and villus fragments. A cross-section illustrates that the villus fragments were essentially hollow tubes of epithelium. The schematic of stem cell lineages illustrates that cell division is always asymmetric (one stem cell and one differentiated daughter) in immortal stem cell lineages, and predominately asymmetric in a stem cell niche. However, niche symmetric divisions occasionally occurs (yellow region) leading to loss of one lineage and the expansion of another, such that total niche stem cell numbers remain unchanged. In this way, niches allow heterogeneous crypts to become homogenous. ::: ![](1741-7015-3-10-1) ::: Although there are likely to be many stem cells per crypt \[[@B2]-[@B4]\], intestinal stem cells have not been directly isolated or identified, and their properties and numbers are uncertain. Stem cell lineages may be immortal if at least one daughter retains stem cell properties after each division. However, mouse studies are more consistent with periodic losses of crypt stem cell lineages. The primary observation is that heterogeneous crypts created through mutagenesis or with chimeras eventually become homogenous \[[@B5]\]. If stem cell lineages were immortal, heterogeneous crypts would not become homogeneous. Stem cell turnover by a niche mechanism \[[@B5],[@B6]\] can explain how heterogeneous crypts become homogeneous (Figure [1](#F1){ref-type="fig"}). Niche stem cells divide but total niche stem cell numbers remain unchanged as daughter cells constantly leave the niche and differentiate. Usually a stem cell produces one stem cell and one differentiated daughter (\~95% of the time in mice \[[@B7]\]), but sometimes a stem cell lineage may become extinct when both daughter cells leave the niche, balanced by another stem cell lineage that expands when both daughters remain within the niche. Such a niche population-type mechanism (random stem cell loss with replacement) eventually results in the loss of all stem cell lineages except one, allowing heterogeneous crypts to become homogeneous (Figure [1](#F1){ref-type="fig"}). Human stem cell studies are difficult because the powerful visible fate-marking strategies of animal models are impractical. Most human crypts are visually homogeneous and therefore consistent with either immortal or niche stem cell mechanisms. A study of human colon crypts after therapeutic radiation was consistent with stem cell niches because the frequencies of mutant heterogeneous crypts progressively declined after radiation \[[@B8]\]. Similar studies have not been performed on human small intestines. One method that can reconstruct the past without direct serial observations infers ancestry from present day sequences. The past is encoded in sequences because errors may accumulate in a clock-like fashion (the molecular clock hypothesis \[[@B9]\]) and record relationships between cells and times since the most recent common ancestor or bottleneck. Just as the age of the human race may be inferred by comparing contemporary genomes, so the ages of crypt stem cell populations may in theory be inferred from their sequences. With immortal stem cell lineages, adult crypts will have greater sequence diversity compared to crypts maintained by stem cell niches that periodically \"bottleneck\" whenever all but one current stem cell lineage are lost. Therefore, crypt diversity measurements potentially reveal whether stem cell lineages are immortal (last bottleneck around birth) or turn over by a niche mechanism (more recent bottlenecks). Mutations are rare in normal cells, but recent studies illustrate that stem cell ancestries may be automatically recorded by random somatic epigenetic errors that frequently accumulate with aging \[[@B10]\]. Methylation increases with age at certain CpG-rich regions in the colon \[[@B11],[@B12]\]. This age-related methylation does not resemble a stereotypic developmental process because different cells within the same intestine may have different methylation patterns (5\' to 3\' order of methylated sites). Such random somatic errors resemble the drift of genomes during species evolution, and by analogy, such drift can be used to reconstruct somatic cell ancestry because methylation exhibits somatic inheritance \[[@B13]\] and all cells are related. Therefore, somatic methylation patterns potentially encode numbers of divisions since birth, and ancestral relationships among cells within a single crypt or villus. Here we infer from methylation patterns that crypt niches containing multiple stem cells maintain the human small intestine. Methods ======= Specimens --------- Crypts or villi were isolated from 1--2 cm^2^patches of fresh small intestines obtained from 13 patients undergoing surgery for non-small intestinal diseases (Table [1](#T1){ref-type="table"}), using an EDTA-containing solution as previously performed for colon crypts \[[@B10]\]. The research was approved by our Institutional Review Board and is in compliance with the Helsinki Declaration. Individual crypts or villus fragments were identified under a dissecting microscope (Figure [1](#F1){ref-type="fig"}) and placed in 0.5 ml microfuge tubes. Villus fragments were essentially hollow tubes of epithelium without stroma. Nine crypts were analyzed from 13 intestines for CSX, eight crypts from six intestines for MYOD, and 4--10 villi (average of seven) were analyzed for CSX from six intestines. The DNA was extracted and bisulfite converted as previously described \[[@B10]\]. DNA was also extracted from microdissected paraffin-embedded small intestinal tissues of an eight month old and a four year old. The bisulfite treated DNA from about half a crypt was amplified by PCR at CpG-rich segments of CSX or MYOD \[[@B10]\]. PCR products were cloned into bacteria and eight clones were sequenced from each crypt or villus. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Specimens ::: Patient Age/Sex Site Surgery Specimen Tags --------- --------- ---------- -------------------- -------------- ----------- A 17/F Ileum Constipation Crypt CSX, MYOD B 27/F Ileum Inflammatory Bowel Crypt, Villi CSX C 46/F Ileum Ovarian Cancer Crypt, Villi CSX, MYOD D 51/M Duodenum Whipple Crypt CSX E 52/M Ileum Colorectal Cancer Crypt, Villi CSX F 58/F Duodenum Whipple Crypt CSX, MYOD G 58/F Ileum Colorectal Cancer Crypt,Villi CSX, MYOD H 63/F Duodenum Whipple Crypt CSX I 67/F Duodenum Whipple Crypt CSX, MYOD J 72/M Ileum Colorectal Cancer Crypt CSX K 75/F Ileum Colorectal Cancer Crypt,Villi CSX L 76/M Ileum Colorectal Cancer Crypt CSX, MYOD M 78/M Duodenum Whipple Crypt,Villi CSX N 0.07/F Ileum Obstruction Paraffin CSX, MYOD O 4/M Ileum Trauma Paraffin CSX, MYOD Ileum specimens were proximal portions of total colectomies. ::: Stem cell dynamics from methylation patterns -------------------------------------------- Small intestinal methylation patterns were analyzed by the same quantitative approach we applied to colon crypt methylation patterns (\[[@B10]\], and its Supplementary Materials). Human crypts of Lieberkühn containing a total of \~500 cells \[[@B14]\] were simulated on a computer, modeling the processes of stem cell division (symmetrical or asymmetrical), differentiation, death, and random methylation errors. The parameters of the model (Table [2](#T2){ref-type="table"}) are numbers of stem cells per crypt, probabilities of symmetric or asymmetric stem cell division, a methylation error rate, and numbers of divisions since birth. These parameters are uncertain for the human small intestines, and the values are just guesses chosen to be consistent with the biology of the intestines, and are similar to the values we estimated for the colon \[[@B10]\]. We emphasize that they are just estimates, and further experiments are needed to define these values better. The model starts with unmethylated sites in the CpG-rich regions of CSX or MYOD, and errors are assumed to accumulate independently between CpG sites with the same rates as simulated in the colon \[[@B10]\]. The output is the methylation patterns of these regions after a given number of divisions. A total of 1000 simulations were performed for each crypt scenario. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Model Parameters ::: Parameter Stem Cell Niche Immortal Lineages ------------------------------------------------------------------------------------------ ----------------------------------------- ----------------------------------------- Stem Cells per Crypt 4--256 2 Probability of Asymmetric Stem Cell Division (P1)\* 0.98 to 0.89 1.0 Probability of Symmetric Stem Cell Division: zero (P0) or two (P2) stem cell offspring\* 0.02 to 0.11 0 Methylation Error Rate 2 × 10^-5^per CpG site per division\*\* 2 × 10^-5^per CpG site per division\*\* Stem Cell Division Rate 0.75 per day 0.75 per day \*P0+P1+P2 = 1, with P0 = P2 \*\* Equal probabilities for methylation and demethylation errors ::: A stem cell hierarchy is simulated with small numbers of stem cells that produce differentiated mitotic cells, which subsequently become differentiated non-mitotic cells. For example, if there are two stem cells per crypt, differentiated mitotic cells divide six more times before becoming non-mitotic. Differentiated non-mitotic cells remain through two more divisions with the oldest cells dying after each division cycle to maintain a constant number of crypt cells. In this scenario, there are 512 cells per crypt -- two stem cells, 126 differentiated mitotic cells, and 384 differentiated non-mitotic cells. With greater numbers of stem cells, numbers of differentiated mitotic cells and their divisions are correspondingly reduced to maintain a constant size of the mitotic compartment and 512 cells per crypt. Immortal stem cell lineages were simulated with strictly asymmetric division producing one stem cell and one differentiated daughter (P1 = 1.0). After a given number of stem cell divisions, eight alleles are randomly sampled from each simulated crypt, as in the experimental approach. Niche and immortal scenarios were identical except niche stem cells also exhibit symmetric divisions (P1 \< 1.0). Total niche stem cell numbers remain constant because divisions that produce two differentiated daughters are balanced by divisions that produce two stem cell daughters (P0 = P2). Symmetric divisions tend to reduce crypt diversity because methylation patterns can be lost through lineage extinction. The Paneth cell compartment was not specifically modeled because its dynamics are uncertain and these cells would be rarely sampled because their relative numbers are small (\~2--4 Paneth cells per crypt section \[[@B15]\]). Paneth cell methylation patterns should be similar to those of other differentiated cells because they survive only a few weeks in mice \[[@B16]\], although human Paneth cells may potentially have different lifetimes. Methylation of CSX or MYOD \"tags\" can be converted into a 5\' to 3\' binary code, with \"0\" representing an unmethylated site and \"1\" representing a methylated site. For example, an unmethylated CSX tag (eight CpG sites or 256 different possible tags) is \"00000000\" and a fully methylated MYOD tag (5 CpG sites or 64 different possible tags) is \"11111\". Tags can be summarized with three statistics that reflect numbers and types of stem cell divisions: 1\) Percent Methylation. Numbers of methylated tag sites can be summarized by percent methylation. For example the CSX tag \"01010101\" is 50% methylated. In general, percent methylation reflects numbers of divisions since birth. 2\) Unique Tags per Crypt. Diversity, or numbers of unique tags among the eight tags sampled from a crypt, reflects numbers of stem cells and stem cell lineage survival. Greater numbers of stem cells or longer-lived stem cell lineages would lead to greater crypt tag diversity. 3\) Intracrypt Distance. This is the average number of site differences between crypt tags (Hamming distance). For example, the distance between CSX tags 00000011 and 11000000 is four. Intracrypt distance is another measure of crypt diversity. Results ======= Crypts of Lieberkühn and villus fragments were isolated from fresh human small intestine (Figure [1](#F1){ref-type="fig"}) and individual allele methylation patterns were analyzed by bisulfite sequencing. Examples of crypt CSX methylation patterns (5\' to 3\' order of methylated sites) or \"tags\" are illustrated in Figure [2](#F2){ref-type="fig"}. Methylation represents somatic errors because CSX tags, like most CpG-rich sites \[[@B13]\], are initially unmethylated at birth. Although the tag patterns were complex, with differences within crypts and between crypts in the same intestine, the average percent methylation generally increased with age (Figure [3A](#F3){ref-type="fig"}). Crypt diversity (unique tags per crypt and intracrypt tag distances) also exhibited age-related changes (Figure [3A](#F3){ref-type="fig"}). Duodenum and ileum tag patterns appeared similar. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Examples of CSX crypt methylation patterns.**Each group represents eight CSX tags (5\' to 3\' horizontal orientation, each of the eight CpG sites is represented by a circle) sampled from a single crypt, with nine crypts per patient. Open circles are unmethylated sites and filled circles are methylated sites. Numbers of unique tags per crypt and crypt percent methylation are summarized below each crypt. The seemingly random methylation patterns collectively encode both numbers of divisions since birth and ancestry. ::: ![](1741-7015-3-10-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Methylation tag analysis**. **A:**Summary of CSX methylation tags sampled from crypts of Lieberkühn. Individual crypt values (\"X\" for duodenal crypts and \"+\" for ileal crypts) and average values (open circles for the duodenum and filled circles for the ileum) were plotted for 12 small intestines. Black filled circles represent the methylation measured in fixed small intestine from children. Solid green lines represent average values for a stem cell niche simulation best matching the experimental data. Consistent with the scatter of experimental crypt values, niche simulations also produced a variety of different outcomes summarized by dotted green lines representing 95% of simulated outcomes. Gray solid lines represent average simulated values consistent with the colon \[10\]. Small intestine percent methylation appeared to increase more slowly than the colon with age, consistent with fewer stem cell divisions relative to the colon. **B:**Summary of MYOD crypt of Lieberkühn methylation tags. **C:**Summary of CSX villus methylation tags. Average crypt values (black circles) for the same intestine are also plotted for comparison with average villus values (red circles). Unique tags and distances were significantly greater in villi compared to crypts whereas percent methylation was not significantly different. ::: ![](1741-7015-3-10-3) ::: Different tags within the same intestine are consistent with stochastic errors that accumulate independently in different cells. To extract ancestral information encoded by seemingly random tags within a single intestine, crypts were modeled assuming stochastic methylation errors and either immortal or niche stem cells (see Methods). The primary difference between the models is that the probability of symmetric division (yielding two stem cell (P2) or two differentiated daughters (P0)) is zero with immortal stem cells, whereas the probability of symmetric division is greater than zero with niche stem cells (Table [2](#T2){ref-type="table"}). Greater inter-crypt variability is expected with stem cell niches because both methylation errors and stem cell survival are stochastic. As in the experimental data, the simulated crypts contained tags with different patterns, summarized by averages and intervals including 95% of simulated crypt outcomes (Figure [4A](#F4){ref-type="fig"}). Average simulated crypt values were consistent with average experimental values for either two immortal stem cells per crypt, or 4 to 256 niche stem cells. However, niche stem cells produced more inter-crypt tag variation, which better matched the variability observed in the small intestines. Inconsistent with immortal stem cell lineages were the observations that unique crypt tag variances were greater than simulated average values for all 12 intestines, and only 4 of the 12 intestines were within 95% simulation intervals for immortal stem cells (Figure [4B](#F4){ref-type="fig"}). In contrast, experimental variances for 10 of 12 intestines fell within 95% simulation intervals for the niche scenarios. In addition, more than four unique sequences per crypt were often observed, which should occur only rarely with two immortal crypt stem cells (four possible alleles), but may occur more frequently with 4 to 256 niche stem cells (Figure [4A](#F4){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Niche versus Immortal Stem Cells**. **A:**Comparisons between niche (green) and immortal (red) stem cell simulations with CSX tags. Niche simulations assume a constant number of stem cell (from 4 to 256 stem cells per niche) and probabilities of asymmetric division (P1) from 0.98 to 0.89 (see Table 3). Outcomes of all niche simulations were similar and therefore one cannot be certain of the exact stem cell niche size. Immortal lineage simulations assume two stem cells per niche. Parameters (see Table 2) were otherwise identical between the simulations. Outcomes were more variable with niche simulations with wider 95% simulation intervals (dotted lines). The differences are most obvious with unique tags per crypt. **B:**Variances of unique CSX crypt tags per intestine (for example from Figure 2, Patient A had 2, 2, 2, 2, 3, 3, 3, 4, and 5 unique tags per crypt for a variance of 1.1, and Patient M had 2, 2, 2, 3, 3, 3, 4, 4, and 6 unique tags per crypt for a variance of 1.7). Simulations with the immortal lineage scenario (average is a solid red line, 95% intervals are dotted lines) were less consistent with the experimental data (circles) than the niche simulations (green lines). ::: ![](1741-7015-3-10-4) ::: Tag data from human large \[[@B10]\] and small intestinal crypts were similar, except the percent methylation appeared to increase more slowly with age in the small intestine (Figure [3A](#F3){ref-type="fig"}). Small intestine percent methylation was generally lower than published \[[@B10],[@B17]\] measurements of human colon crypts, with a trend towards a slower increase with age (Figure [5](#F5){ref-type="fig"}), but this difference was not significant (p = 0.24, F-test). Potentially methylation error rates are lower in the small intestine, but assuming equivalent error rates (a molecular clock hypothesis \[[@B9]\]), stem cell division rates are either equivalent or lower in the small intestine compared to the colon. For example, simulations with a mitotic rate 75% of the colon (0.75 divisions per day) and a methylation error rate identical to the colon (2 × 10^-5^per site per division) better fit the small intestine data (Figure [3A](#F3){ref-type="fig"} and Figure [5](#F5){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Comparisons of large and small intestine age-related CSX methylation rates**. Average percent methylation increased more slowly with age in the small intestine (black circles) relative to previously published \[10, 17\] colon measurements (red triangles). However this trend (black and red dotted lines) was not significant. A lower stem cell division rate was more consistent with the small intestinal data (green line, 0.75 divisions per day), relative to a model \[10\] postulating one division per day for the colon (grey line). ::: ![](1741-7015-3-10-5) ::: Crypt methylation patterns encode niche dynamics, but exact niche stem cell numbers and probabilities of symmetric divisions remain uncertain because different combinations yield similar outcomes (Table [3](#T3){ref-type="table"}). These niche combinations (4--256 stem cells and P1 from 0.98 to 0.89) are the same in large \[[@B10]\] and small intestines. For example, assuming that 95% of stem cell divisions yield one stem cell daughter, the crypt of Lieberkühn niche size is 64 stem cells. As in the colon \[[@B10]\], loss of all niche stem cell lineages except one or a \"bottleneck\" would recur on average every 3,000 divisions (with a 95% interquantile range of 1,000--7,000 divisions). Small intestinal crypt diversity (unique tags per crypt) did not change with age (Figure [3A](#F3){ref-type="fig"}), suggesting that niche dynamics or numbers of niche stem cells do not change with aging. MYOD tags sampled from six small intestines were also consistent with the same niche parameters inferred with the CSX tags (Figure [3B](#F3){ref-type="fig"}). For a further test of our model, CSX tags were sampled from villi of six intestines. Villi should exhibit more tag diversity than crypts because they receive contributions from four to ten surrounding crypts \[[@B18]\]. Furthermore, if methylation reflects the lifetime accumulation of stem cell mitotic errors, percent methylation should be similar between villi and crypts because crypt cells are only a few divisions older than villus cells. Consistent with these expectations (Figure [3C](#F3){ref-type="fig"}), there were significantly more unique tags per villus than per crypt (average 4.8 versus 3.0, p \< 0.001, two-tail t-test) whereas methylation differences were not significant (average 21.8% versus 22.9%, p = 0.76). Distances between tags within a villus were also significantly greater than tag intracrypt distances (2.5 versus 1.7, p \<0.001). Villus and crypt comparisons illustrate the ability of methylation tags to encode simultaneously both mitotic age (numbers of divisions since birth) and distinctly different ancestries. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Stem Cell Niche Parameters ::: Niche Size P asymmetric (P1)\* P symmetric (P0+P2) ------------ --------------------- --------------------- 4 0.98 0.02 16 0.97 0.03 64 0.95 0.05 256 0.89 0.11 \*P0+P1+P2 = 1, with P0 = P2 ::: Discussion ========== The adult small intestine is physically composed of millions of crypts, and cells that constantly divide and die. Underlying the homeostatic visual appearance of a normal small intestine is an ancestral somatic cell tree that grows with age (Figure [6](#F6){ref-type="fig"}). Starting with the zygote, millions of related cells create an intestine. Past and present cells are all connected through time by ancestral lineages or branches, which ultimately root in the zygote or first common ancestor. Although stem cell definitions may differ \[[@B2]-[@B4]\], a crypt common ancestor is equivalent to a stem cell because all cells in a clonal crypt unit are progeny of this single common stem cell ancestor. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Somatic cell trees**. **A:**A niche somatic cell tree. Starting from the zygote, millions of related cells create an intestine. Within each crypt are multiple stem cells (a niche with four stem cells is illustrated). With age, branches become longer reflecting more divisions since birth. Most niche stem cell lineages are lost (dotted lines) from random stem cell loss with replacement, which continuously creates newer more recent common crypt ancestors (circles). Such common crypt ancestors are stem cells because all current crypt cells are progeny of such cells. Any one of the four current niche stem cells may become a common ancestor, but only one will. In this way, crypts can appear to have both a single stem cell (most recent common crypt ancestor) and multiple potential stem cells. A villus tree (red lines in bottom tree) has essentially the same age as the crypt tree (same number of divisions since birth) but villi should exhibit more diversity than crypts because villus cells have older most recent common ancestors. **B:**Somatic cell tree with immortal stem cell lineages. Stem cells always divide asymmetrically and their lineages never become extinct. Unlike a niche somatic cell tree, branches are never lost or \"pruned\". ::: ![](1741-7015-3-10-6) ::: Both immortal and niche stem cell mechanisms can theoretically produce appropriate numbers of differentiated cells consistent with the morphology of normal small intestine. However, the shape of an intestinal somatic cell tree differs whether stem cell lineages are immortal or maintained by niches (Figure [6](#F6){ref-type="fig"}). Immortal stem cell lineages result in multiple long crypt branches, with the most recent common crypt ancestor present around birth. In contrast, a niche mechanism constantly creates newer more recent common crypt ancestors as most niche stem cell lineages become extinct. Methylation tags can distinguish between immortal or niche stem cell lineages because random somatic epigenetic errors will accumulate differently. Specifically, a niche mechanism produces more variability between crypts (wider variation in unique tags per crypt within a single intestine) and less intracrypt diversity (smaller average intracrypt distances) than immortal stem cells (Figure [4](#F4){ref-type="fig"}). Small intestine and colon crypt methylation patterns \[[@B10]\] were similar and more consistent with stem cell niches. The same stem cell dynamics (Table [3](#T3){ref-type="table"}) fit both colon and small intestinal niches. There were no apparent differences between ileal and duodenal crypt niches, and niche stem cell numbers (like the colon \[[@B10]\]) did not appear to change or decline with age. Niche stem cells defined by ancestry are physically intangible because common ancestors are defined by past events and no longer exist. Stem cells with immortal lineages are more readily identified because their pasts and futures are predictable. However, niche stem cell fates are unpredictable because all may potentially become common ancestors but only one will. The inability to predict niche survival may help explain why some adult stem cells have been so difficult to isolate or characterize. A niche somatic cell tree has few branches because random stem cell turnover eventually \"prunes\" all niche lineages except one (Figure [6](#F6){ref-type="fig"}). Such bottlenecks are predicted by our analysis to recur on average after 3,000 divisions. Other mechanisms may also be consistent with crypt methylation patterns and the assumptions of our model remain unverified \[[@B19]\]. For example, the observed variability in methylation patterns could also be generated if stem cell numbers were different between crypts within an intestine \[[@B19]\]. Although our bisulfite sequencing appears technically adequate (no evidence of nonconversion of C to T and only rare mutations (\<1 per 1,000 bases) at non-CpG sites), it is difficult to check the accuracy and reproducibility of the data. Repeat sampling of a crypt typically yields similar but not identical methylation patterns (data not shown), which may reflect crypt heterogeneity or experimental artifact. One strategy to test empirically a stochastic model that inherently yields scattered results is to examine similar but biologically distinct entities. Both small intestine and colon crypts are thought to be maintained by stem cell niches \[[@B5],[@B6]\] and the successful application of our model to the small and large intestines with methylation patterns from CpG-rich sequences on different chromosomes (CSX and MYOD) is empirical support for its ability to infer stem cell dynamics. In addition, small intestinal villi are physically different from crypts and represent mixtures from multiple adjacent crypts. Villi should be more diverse than crypts, yet villi and crypts should have similar numbers of divisions since birth. As one would expect if methylation records somatic cell histories, villus and crypt tag percent methylation were not significantly different, and villi contained significantly more tag diversity than crypts. Therefore, methylation patterns and our model are consistent with colon and small intestine crypt niches, and small intestine villi. Genetic alterations also appear to modulate stem cell dynamics because certain germline APC mutations are associated with significantly more diverse tags in normal-appearing familial adenomatous polyposis colon crypts, consistent with increased niche stem cell survival \[[@B17]\]. Mouse studies suggest that stem cell division rates are higher in the small intestines than the colon \[[@B2],[@B3],[@B20]\]. However, it is difficult with most assays to distinguish between stem and non-stem cell proliferation, and there are few human small intestine stem cell studies. Of interest, neoplasia, as with Min mice \[[@B21]\], is more frequently observed in the murine small intestines relative to the colon, whereas human small intestinal tumors are a small fraction of colon tumors \[[@B1]\]. In contrast to murine studies, human small intestinal stem cell division rates appeared to be slower or similar to those in the colon because age-related methylation appeared to increase more slowly in the small intestine (Figure [5](#F5){ref-type="fig"}), although the difference was not significant. Assuming a molecular clock hypothesis \[[@B9]\] or equivalent CSX methylation error rates throughout the lower gastrointestinal tract, the data are more consistent with equivalent or slightly lower small intestinal stem cell division rates relative to the colon. There is no definitive evidence supporting a somatic cell molecular clock hypothesis, but a priori a slower small intestinal stem cell division rate is more consistent with human intestinal cancer epidemiology \[[@B1]\], because fewer small intestinal errors (genetic or epigenetic) would be expected to accumulate during aging. A slower small intestinal stem cell division rate could also help explain why crypt purification is faster in the murine colon relative to the small intestine \[[@B5]\]. Moreover, an ability to count relative divisions since birth in different human tissues allows distinctions (Figure [7](#F7){ref-type="fig"}) between mitotic tissue ages (divisions since birth) and chronological ages (years since birth). The colon and small intestines within an individual have equivalent chronological ages but may have different mitotic ages. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **An intestinal tree based on mitotic ages or numbers of divisions since birth**. Each horizontal branch reflects a single crypt. Although all tissues have the same chronological ages, differences in stem cell division rates allow tissues within the same individual to have different mitotic ages. A small intestine tree has a more crypt branches (greater numbers of crypts) than a colon, but its branches may be shorter, which can possibly help explain why despite the greater physical size of the organ, small intestinal tumors are infrequent relative to colonic ones. ::: ![](1741-7015-3-10-7) ::: Conclusion ========== Methylation patterns suggest that niches containing multiple stem cells maintain crypts throughout the lower gastrointestinal tract. Small intestine stem cells appear to divide at equivalent or slower rates relative to the colon, and niche dynamics remain stable during aging. Although methylation tags only indirectly track stem cell dynamics and their exact interpretations are uncertain, they potentially allow for the systematic investigation of any intestine without prior experimental intervention. A somatic cell tree must underlie all human tissues and studies based on the hypothesis that certain somatic methylation errors record somatic cell histories may better define how cells divide and die during normal and abnormal aging. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= JYK performed most of the experiments and helped write the manuscript. KDS and ST helped with the statistical and quantitative analysis. DS analyzed the data and wrote the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1741-7015/3/10/prepub> Acknowledgements ================ We thank Mario Campuzano for technical assistance and Dr. Robert W. Beart for providing the specimens and clinical information. This work was supported by a grant from the National Institutes of Health (DK61140). ST is a Royal Society-Wolfson Research Merit Award holder and is supported in part by a grant from Cancer Research UK.

PubMed Central

2024-06-05T03:55:58.238746

2005-6-23

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168897/", "journal": "BMC Med. 2005 Jun 23; 3:10", "authors": [ { "first": "Jung Yeon", "last": "Kim" }, { "first": "Kimberly D", "last": "Siegmund" }, { "first": "Simon", "last": "Tavaré" }, { "first": "Darryl", "last": "Shibata" } ] }

PMC1168898

Background ========== Despite the advances in hospital care and the introduction of a wide variety of antimicrobial agents, *Pseudomonas aeruginosa*continues to be a major nosocomial pathogen particularly in patients who suffer from immunosuppression \[[@B1]\]. *P. aeruginosa*is a ubiquitous pathogen prevalent in the hospital environments, and can cause severe nosocomial infections \[[@B2]\]. The latter involve a broad spectrum of infections including the respiratory, gastrointestinal, and urinary tracts as well as wound infections, sepsis and others \[[@B3]\]. Various possible sources of *P. aeruginosa*infection in hospitals have been identified; such as tap water, medical equipment, hospital personnel and other patients \[[@B2],[@B4]\]. *P. aeruginosa*accounts for 10% of all hospital acquired infections, a site specific prevalence which may vary from one unit to another and from study to study \[[@B5]\]. Among data on site-specific infections, *P. aeruginosa*appears to be the major cause of ventilator-associated pneumonia with a high rate of attributable mortality \[[@B6]\]. Moreover this organism can contaminate a number of other medical equipment such as respirators, endoscopes, bronchoscopes, transvenous pacemakers, urinary catheters, and dialysis equipment, leading to site-related infections \[[@B7],[@B8]\]. During the last year, the average prevalence of *P. aeruginosa*nosocomial infections in our medical center was 18%. Such a high rate prompted us to study the *P. aeruginosa*genotypes circulating in the various units to reveal the clonal relationship between clinical and environmental isolates and to allow elucidating the source and mode of transmission of this important bacterium at this medical center. Results ======= Our data have shown that several mechanical devices were associated with *P. aeruginosa*infection in our patients with 55% due to mechanical ventilation and the remaining 45% due to polysite catheters, Foley catheters and different types of surgery. RAPD analysis have shown 31 genotypes present among the clinical and the environmental isolates (figure [1](#F1){ref-type="fig"}). Thirty eight of ninety (42%) of the clinical isolates and 10/23 (43%) of the environmental isolates showed genotype 1 to be distributed among the medical center units. Each of the genotypes 2--30 represented from 1--8% of the strains. Antimicrobial susceptibility testing has shown that susceptibility patterns between clinical and environmental isolates were similar with most drugs showing \> 90% susceptibility, with the exception of tobramycin and gentamycin which showed lower susceptibility (83%) in the environmental strains. Data is in concordance with previous reports in this institution \[[@B13]\]. Enzymatic assays determined that genotype 1 in addition to other 10 genotypes, in clinical and environmental isolates, were positive for all virulence factors tested for. All other genotypes were all positive for the production of protease, lecithinase and coagulase and positive in 83--89% of the cases for the production of the other 4 virulence factors (Figure [2](#F2){ref-type="fig"}). Table [1](#T1){ref-type="table"} shows the relationship between the various genotypes, to source, antibiograms and virulence factors. Discussion ========== The strain distribution based on RAPD analysis showed that clinical and environmental isolates distributed in one or more than one unit, included genotypes 1--8 and 9, with predominance of genotype 1 in all investigated units. This may indicate that cross contamination among patients lead to the spread of this genotype among the various units, possibly through transient hand carriage by health care personnel due to contact with contaminated surfaces or by patient contact with contaminated surfaces or medical equipment \[[@B4]\]. Our findings suggest that cross colonization may be an important means of *P. aeruginosa*spread and infection especially after identification of a potentially virulent clone (genotype 1) of this organism that had been propagated in various units over a period of 9 months. This could indicate that the patients were continuously infected with a strain originating from an exogenous source. The importance of cross colonization of *P. aeruginosa*in nosocomial infections was previously reported, by Bergmans *et al*who studied 100 patients admitted to an ICU ward; cross colonization accounted for 50% of all cases of acquired *P. aeruginosa*colonization, the other 50% of patients were probably colonized from endogenous sources \[[@B14]\]. The remaining genotypes were isolated exclusively from the clinical specimens of patients and were not detected among environmental isolates. Most of these genotypes did not harbor all the virulence factors tested for. Not being encountered in environmental sources may indicate that these strains may have been endogenously acquired. High rate of colonization with *P. aeruginosa*from endogenous sources occurs mainly in the respiratory and gastrointestinal tracts \[[@B2]\]. Hospitalization may lead to increased rates of carriage, particularly in the lower respiratory tract in patients undergoing mechanical ventilation, upper respiratory tract due to broncho- pulmonary colonization and infection, in the gastrointestinal tract of patients receiving chemotherapy for neoplastic diseases, or at virtually any site in patients treated with antibiotics \[[@B3],[@B15],[@B16]\]. Prevalence of strains with resistance to all antimicrobial agents will constitute a major risk for hospitalized patients. In our study, though, the most prevalent strain of genotype 1 was susceptible to all antimicrobial agents and does not constitute a problem in treatment. However, its potential of producing all virulence factors and being spread by various means in the hospital, unlike other genotypes that harbor all virulence factors, may render genotype 1 a high risk pathogen specially in the immunosuppressed and debilitated patients. The fact that genotype 1 remained susceptible to all antibiotics lead to the conclusion that patients could be continuously infected with a strain originating from an exogenous source. In addition to mechanical spread by personnel, this genotype may carry a number of adhesins that enhance its colonization in the hospital environment and render it more accessible to patients. Studies are underway in our laboratory to detect by PCR, genes encoding a number of adhesins and determine their transcription levels in genotype 1, in comparison to other encountered genotypes. This will shed light on the possible role of adhesins on the prevalence of this genotype in the hospital environment. The extracellular enzymes or toxins produced, on the other hand, will contribute to the breaking down physical barriers and help the organism to penetrate, impair host defenses, and render its new milieu more conducive to its physical, nutritional and reproductive requirements \[[@B17]\]. In summary, our data have shown the predominant prevalence of a potentially virulent *P. aeruginosa*genotype 1 in clinical and environmental specimens, circulating in the various hospital units. From a practical point of view, the results of our study emphasize the need to reinforce implementation of infection control measures, and to limit the transmission of *P. aeruginosa*among patients and from environmental sources to patients. Screening for *P. aeruginosa*carriage in all patients with nosocomial colonization or infection should also be done in research settings using both rectal and respiratory tract specimens to determine the source of colonization/infection and hence get informed on whether it was acquired endogenously or exogenously. Conclusion ========== In conclusion, our data have shown the predominant prevalence of a potentially virulent *P. aeruginosa*genotype 1 in clinical and environmental specimens, circulating in the various medical center units. Methods ======= Consecutive *P. aeruginosa*(90) isolates were recovered from different patients specimens (one per patient) submitted for bacteriological investigations at the Clinical Microbiology Laboratory at the American University of Beirut Medical Center (AUBMC), between September 2003 and May 2004. Patients acquiring a nosocomial infection due to *P. aeruginosa*as determined by clinical and laboratory testing and indicated in their medical records, were only considered in this study. Fever, and recovery of *P. aeruginosa*from the site of infection during stay at the medical center, constituted the most important criteria that defined patient\'s infection with this organism. Patients\' data collected from the medical records, included age (2 to 91 years), sex (Males: 43 and Females: 47), admission date, admission diagnosis, invasive procedures used on the patients, symptoms, and the date of the first positive culture for *P. aeruginosa*. The drugs used in the treatment of these patients, mainly included: amikacin, azactam, gentamicin, tobramycin, and tazocin. Patients were distributed within ten different units in the medical center, mainly the Respiratory Care Unit (RCU), Intensive Care Unit (ICU), Coronary Care Unit (CCU), the Surgery Unit, as well as other units. Twenty three isolates were also collected from different environmental sources, such as, respirators, respirators\' filters, water irrigation, tap water, basins, trays, bed side tables, side rails, and sink sides. Statistical analysis determined the sample size required to estimate the true proportion (percentage of *P. aeruginosa*infections during a given period of time) to within 0.10, with 95% confidence was calculated. Calculations estimated the minimum number of samples required to be 54 \[[@B9]\]. The presumptive identification of *P. aeruginosa*on culture based on colonial morphology, Gram stain microscopy, and oxidase test was further confirmed by the API NE Kits and growth at 42°C. Susceptibility of all isolates to a panel of antimicrobial agents (amikacin, aztreonam, ceftazidime, ciprofloxacin, gentamicin, imipenem, piperacillin/tazobactam, tobramycin) was determined according to the guidelines of the National Committee for Clinical Laboratory Standards (NCCLS) \[[@B10]\]. *P. aeruginosa*ATCC 25321 strain was used as positive control in all tests. DNA was extracted from *P. aeruginosa*ATCC strain and from all isolates of *P. aeruginosa*by the GFX™ Genomic Blood DNA Purification Kit (Amersham PharmaciaBiotech, Uppsala, Sweden) according to the manufacturers\' specifications. Random amplified polymorphic DNA (RAPD) analysis of the clinical and environmental isolates using two in-house oligonucleotide primers, Pa1 (5\'AGGGGTCTTG 3\') and Pa2 (5\' CTTCTTCAGCTCGACGCGACG 3\') was done. RAPD was carried out according to Matar et al \[[@B11]\] using the PTC-100™ Programmable Thermal Controller (MJ Research, Inc., Watertown, Mass., USA). Briefly, RAPD was carried out on all isolates in 100 μl reaction mixtures containing each: 10 μl of template DNA, 16 μl of dNTPs (0.2 mM), 10 μl of 10X PCR buffer (100 mM TrisHCl \[pH 8.3\], 500 mM KCl, 4 mM MgCl~2~), 1 μl of primer 1 (0.5 μM), 1 μl of primer 2 (0.3 μg/ μl), 2.5 U of *Taq*DNA polymerase and 61.5 μl of nanopure sterile water. The amplification program, included the following steps: denaturation at 94°C for 3s, annealing at 53°C for 1 min and extension at 72°C for 1 min, for 44 cycles. The cycles were followed by a final extension step at 72°C for 10 min. Amplicons were subjected to electrophoresis on 2% agarose gels at 107 volts for 2 hours. Patterns that had the same number of bands and similar fragments size were considered identical. The enzymatic activities of the isolates of *P. aeruginosa*were evaluated by spot inoculation containing 10^6^CFU/ml of the organisms in various media \[[@B12]\]. Media used: Brain heart infusion for the protease activity, 1% elastin (Sigma Chemical Co. St Louis, Mo. USA) for the elastase activity, human fibrinogen type 1 (Sigma Chemical Co. St Louis, Mo. USA) for the fibrinolytic activity, trypticase soy agar (TSA) supplemented with egg yolk (Difco Laboratories, Detroit, Mi., USA) for the lecithinase production, TSA with tween 80 (Sigma Chemical Co. St Louis, Mo. USA) for the lipase activity, DNase test agar with toluidine blue 0, (Difco Laboratories, Detroit Mi., USA) for the DNAase production, and rabbit plasma (Difco Laboratories, Detroit Mi., USA) for the coagulase activity. Positivity of tests was assessed as follows: clearing of opacity around the inoculum spots for the protease, elastase and fibrinolytic activities, white precipitate around or beneath the inoculum spots for the lecithinase activity, a turbid halo around the inoculum spots for the lipase activity, formation of a pink halo around the inoculum spots for the DNase activity, and gelling of rabbit plasma after 48 hours for the coagulase activity. Authors\' contributions ======================= GM supervised the study and wrote the manuscript. MC did the bench work and helped in writing the manuscript. GA provided bacterial isolates. ZS provided help on wards. GJ provided clinical support. UH provided clinical support. All authors read and approved the final manuscript. Acknowledgements ================ Thanks are due Miss Marie Risk and Mr. Elias Rahal for technical assistance. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### \(A) Genotypes of *P. aeruginosa*isolates found in 10 units. Lane 1: 50-bp ladder, lane 2: negative control, lane 3:ATCC25321 strain, lanes 4--18: Genotypes 1,2,3,4,5,6,7,8,9,10,11,12,13,14, and 15. (B) Lane 16: 50-bp ladder, lane 17: negative control, lane 18: ATCC 25321 strain, lanes 19--33: Genotypes 16,17,18,19,20,21,22,23,24,25,26,27,28,29, and 30. (C) Genotypes of *P. aeruginosa*from environmental sources found in the RCU. Lane 1: 50-bp ladder, lane 2: negative control, lane 3:ATCC 25321 strain, lane 4--11: Genotypes (1,4,17,25,27,28,30,31). ::: ![](1471-2180-5-29-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Percentage of clinical and environmental *P. aeruginosa*isolates producing the virulence factors. ::: ![](1471-2180-5-29-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Correlation between *P. aeruginosa*genotypes, source and site of infection, antimicrobial susceptibility and enzymatic activities of virulence factors ::: **Genotypes** **types of unit** **Number of clinical isolates** **Isolation site/ Number of isolates** **Antibiograms** **Virulence factors** --------------- ------------------- --------------------------------- ----------------------------------------------------------- ------------------ ----------------------- ------- ------- ------- ------- ------- ------- ---- ---- ---- ---- ---- ---- 1 RCU 15 DTA^1^/ 13 blood/ 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ ICU 3 DTA/ 2 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ CCU 1 Urine/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ PICU 4 DTA/2 Tongue/ 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ St Jude 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ Surgery 3 Wound/ 3 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 South 6 Abscess/ 1 Sputum/ 2 DTA/ 1 Bronchial washing/ 1 Urine/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 4 South 2 Sputum/ 1 Swab/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 9 South 3 DTA/ 1 Sputum/ 1 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 2 RCU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ ICU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ Surgery 2 Wound/ 2 S S S S S S S S \+ \+ \+ \+ \+ \+ Oncology 1 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 south 1 Pleural fuild/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 3 RCU 2 DTA/ 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 South 2 Sputum/ 2 S S S S S S S S \+ \+ \+ \+ \+ \+ Oncology 3 DTA/ 2 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 4 RCU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ St Jude 2 Blood/ 1 Catheter/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 South 1 Bile/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 5 PICU 1 DTA/ 1 S S S S S S S S \+ \+ \- \+ \+ \+ 4 South 1 DTA/ 1 S S S S S S S S \+ \+ \- \+ \+ \+ 6 RCU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ ICU 1 Sinus/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 7 ICU 2 DTA/ 2 S S S S S S S S \+ \+ \+ \- \+ \+ Oncology 1 Blood/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 8 ICU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \+ \- \+ Oncology 1 Blood/ 1 S S S S S S S S \+ \+ \+ \+ \- \+ 9 RCU 1 Urine/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ Surgery 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 10 4 South 2 Wound/ 2 S S S **R** **R** **R** **R** **R** \+ \+ \- \+ \- \+ 11 Oncology 2 leg ulcer/ 1 Bronchial washing/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 12 5 South 2 DTA/ 2 S S S S S S S S \+ \+ \+ \- \- \+ 13 9 South 1 Sinus discharge/ 1 S S S **R** **R** S S **R** \+ \+ \- \+ \- \+ 14 9 South 1 Wound/ 1 S S S S S S S S \+ \+ \+ \+ \+ \+ 15 ICU 1 DTA/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 16 CCU 1 Sternal tissue / 1 S S S S S S S S \+ \+ \+ \+ \+ \- 17 CCU 1 Urine/ 1 S S S S S S S S \+ \+ \+ \- \+ \+ 18 CCU 1 DTA/ 1 S **I** S S S S S S \+ \+ \- \- \- \+ 19 RCU 1 DTA/ 1 S **I** S S S S S S \+ \+ \+ \+ \+ \+ 20 RCU 1 DTA/ 1 S S S S S **R** S S \+ \+ \- \+ \- \+ 21 RCU 1 DTA/ 1 S S S S **R** S S **R** \+ \+ \- \+ \- \+ 22 RCU 1 DTA/ 1 S **I** S S S S S S \+ \+ \+ \+ \+ \+ 23 RCU 1 DTA/ 1 S **I** S S S S **R** S \+ \+ \- \+ \- \+ 24 RCU 1 Wound/ 1 **R** S **R** **R** **R** **R** **R** **R** \+ \+ \- \+ \- \+ 25 RCU 1 DTA/ 1 S S S S **R** S S **I** \+ \+ \+ \+ \+ \+ 26 RCU 1 Sputum/ 1 S **R** **R** S S **R** **R** S \+ \+ \- \+ \- \+ 27 PICU 1 DTA/ 1 S **R** **R** S S S **R** S \+ \+ \+ \+ \+ \+ 28 ICU 1 DTA/ 1 S **R** **R** S S **I** **R** S \+ \+ \+ \+ \+ \+ 29 4 South 1 DTA/ 1 S **I** S S **R** **R** S S \+ \+ \+ \+ \+ \- 30 RCU 1 DTA/ 1 S S S S **R** S S **I** \+ \+ \- \+ \- \+ NB: Genotype 31 was only found among the environmental isolates ^1^DTA: Deep Tracheal Aspirate AN: Amikacin CIP: Ciprofloxacin TZ: Tazocin ATM: Aztreonam GM: Gentamicin NN: Tobramycin CAZ : Ceftazidime IPM: Imipenem Pro: Protease Lip: Lipase Lec: Lecithinase Fib: Fibrinolysin Elas: Elastase Coag: Coagulase :::

PubMed Central

2024-06-05T03:55:58.242123

2005-5-20

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168898/", "journal": "BMC Microbiol. 2005 May 20; 5:29", "authors": [ { "first": "Ghassan M", "last": "Matar" }, { "first": "Mira H", "last": "Chaar" }, { "first": "George F", "last": "Araj" }, { "first": "Zaher", "last": "Srour" }, { "first": "Ghassan", "last": "Jamaleddine" }, { "first": "Usamah", "last": "Hadi" } ] }

PMC1168899

Background ========== *Yersinia pestis*, the causative agent of plague, causes rapidly progressing disease in humans with a high mortality rate, especially in the pneumonic form of the disease. Due to the severe nature of plague, its ability for aerosol transmission, and the potential for human to human transmission plague is considered to be a disease of high concern as an agent of biological warfare or biological terrorism \[[@B1]\]. For this reason, an improved vaccine for plague is desirable. Current efforts for vaccine development have focused on two proteins: LcrV (also known as the V antigen) and the capsular F1 antigen \[[@B2]\]. The best results to date have been obtained by using a combination of recombinant LcrV and F1 subunits \[[@B3]\] separately or as a fusion protein \[[@B4],[@B5]\]. These subunit vaccines demonstrate very good protection against both pneumonic and systemic forms of plague \[[@B2]\] in mouse models. One of the potential limitations of these subunit vaccines is that F1 is not required for full virulence of *Y. pestis*, as F1-negative strains have the same LD~50~value as F1-positive strains \[[@B6]-[@B9]\]. A second limitation that could result in undesired side-effects in immunized individuals is the demonstrated immunosuppressive effect of LcrV \[[@B10]-[@B13]\]. Additionally, serologic diversity of LcrV has been reported, in *Yersinia*species other than *Y. pestis*, that could theoretically limit the usefulness of an LcrV based vaccine. While the recombinant subunit vaccines are very effective in experimental animals and offer protection against F1 minus strains of *Y. pestis*\[[@B2]\], the inclusion of other antigens with the LcrV/F1 subunit vaccine candidates could improve the ability of the resulting vaccine to offer protection against multiple *Y. pestis*strains, or the new antigens could be developed as separate vaccine candidates. The type III secretion apparatus encoded on the low-calcium response (LCR) virulence plasmid, pCD1 in strain KIM \[[@B14]\], of *Y. pestis*is a conserved virulence mechanism that is absolutely required for virulence of *Y. pestis*\[[@B15]\]. YscF is a surface localized protein that is required both to secrete Yops and to translocate toxins into eukaryotic cells \[[@B16]-[@B19]\]. One report speculates that YscF polymerization is required for a YscF needle to puncture eukaryotic cell membranes \[[@B18]\]. Other researchers suggest that YscF and its homologs function to provide a base that a translocon complex is built upon, or that YscF builds a conduit from the bacterium to the eukaryotic membrane \[[@B20]\]. This suggestion seems more likely given that other proteins such as YopB, YopD, and LcrV are also required for translocation into eukaryotic cells \[[@B21]-[@B28]\]. Additionally, YscF needle producing *Y. enterocolitica*deficient in production of the translocators (LcrV, YopB, and YopD) do not translocate Yops into macrophages, demonstrating that the YscF-needle is not sufficient for translocation \[[@B19]\]. Most currently described pathogenesis-related type III secretion systems possess homologs to YscF. In pathogenic *Salmonella*and *Shigella*, the YscF homologs (PrgI and MxiH, respectively) have been demonstrated to form a needle structure that protrudes from the surface of bacterial cells \[[@B29]-[@B31]\]. The best-characterized homolog of YscF is EscF of enteropathogenic *E. coli*(EPEC). EscF is required for \"attaching and effacing\" (A/E) lesion formation on the intestinal mucosa and for type III secretion of effector proteins \[[@B32]-[@B34]\]. EscF is thought to be a structural component of the needle complex on the bacterial surface as it binds EspA, the major component of a filamentous surface organelle, and is required for formation of the EspA filaments. \[[@B32]-[@B34]\] However, this surface localization has never been visualized directly, as the only EscF antiserum generated was unable to recognize the native protein \[[@B33]\]. Based on the fact that YscF is thought to be a surface-expressed protein in *Y. pestis*and is required for virulence, we sought to determine if YscF could serve as a protective antigen against experimental plague infection. Immunization of mice with His-tagged YscF resulted in a high anti-YscF titer and significant protection against i.v. challenge with *Y. pestis*. The findings of this study suggest that YscF may be a potential plague vaccine candidate that could be used in conjunction with other plague antigens, or possibly alone if its efficacy can be improved by alternative delivery methods. Results and discussion ====================== Expression and purification of HT-YscF -------------------------------------- To facilitate the purification of YscF, *yscF*was cloned into the overexpression plasmid pET24b (Novagen) to yield a hexahistidine-tag on the C-terminus of YscF (HT-YscF). *E. coli*BL21(DE3) harboring the HT-YscF expression plasmid, pJM119, was grown in one liter of LB broth \[[@B35]\] containing kanamycin at 37°C. Expression of HT-YscF was induced after 2 h of growth with 0.3 mM IPTG and then incubated until the A~550~reached \~1.0. Cells were harvested by centrifugation and disintegrated by passage through a French pressure cell at 20,000 lb/in^2^. Subsequent to disintegration, the extracts were clarified by centrifugation at 3,200 × *g*for 20 min at 4°C. Affinity purification of HT-YscF was performed using Talon resin (BD Clontech) using standard methods. Purity of the recovered protein (\> 95 %) was estimated by SDS-PAGE on a 15% (wt/vol) gel followed by staining with a Coomassie blue stain. The purified HT-YscF ran as multiple bands on the gel, regardless of the presence or absence of a reducing agent, such as dithiothreitol (data not shown). A band that corresponded to the predicted size of HT-YscF was the dominant species and other larger bands could also be visualized (Figure [1A](#F1){ref-type="fig"}). Based on the sizes of these bands, it is likely that they represent dimers and other multimers of YscF. The presence of YscF multimers is not surprising as YscF and its homologs are known to form multimeric structures \[[@B29],[@B36]\]. Support for the contention that the larger bands are multimers of YscF is seen in Fig. [1B](#F1){ref-type="fig"}. Purified recombinant HT-YscF was immunoblotted with a Penta-His specific antibody (Qiagen). The results seen in Figure [1](#F1){ref-type="fig"} show that the larger bands seen in panel A are also recognized by the penta-His antibody (Fig. [1B](#F1){ref-type="fig"}) demonstrating that the larger bands are likely recombinant proteins containing a poly-histidine tag and, therefore, contain YscF. Additionally, the purified HT-YscF was immunoblotted and probed with serum from immunized mice (Fig. [1C](#F1){ref-type="fig"}) to demonstrate that the immunized mice mounted an immune response against the purified HT-YscF. The larger bands seen in the Coomassie stain and visualized on the immunoblots (with both anti-Penta-His antibody and post-immune sera) could also be contaminating *E. coli*proteins, that cross-react to the penta-His antibody, this possibility cannot be currently discounted. The presence of the higher molecular weight proteins, if they are not YscF multilmers and they are from *E. coli*, likely had little influence on this study as immunization with HT-YscF induced a specific antibody response to native YscF produced by *Y. pestis*(discussed below). Specificity of the antibody response to YscF -------------------------------------------- To confirm that the recombinant HT-YscF antibody response generated in immunized mice was directed against YscF from *Y. pestis*, *Y. pestis*protein extracts from the parental strain KIM8.3002 (pgm^-^, pla^-^) and extracts from a isogenic *yscF*deletion strain (kindly provided by G. Plano, University of Miami, Miami, Fl.) were immunoblotted and probed with pooled antiserum from HT-YscF immunized mice (Figure [2](#F2){ref-type="fig"}). The results in Figure [2](#F2){ref-type="fig"} demonstrate that the pooled mouse antiserum specifically recognizes YscF produced by *Y. pestis*. As seen in Figure [2](#F2){ref-type="fig"}, YscF is visualized on the immunoblot as a highly reactive band of the correct predicted size and the YscF band is only seen only in strains containing the *yscF*gene (Figue [2](#F2){ref-type="fig"}, lanes 1--4). Importantly, no bands are seen in lanes 5 and 6 that contain proteins derived from the *yscF*deletion strain. In lanes 1 and 2, calcium regulation of the YscF band is seen as expected for the LCR-regulated *yscF*gene. Transcomplementaion of the Δ*yscF*strain with pBAD-YscF (G. Plano, University of Miami) restored YscF reactivity to the HT-YscF serum in the Δ*yscF*strain (Figure [2](#F2){ref-type="fig"}, lanes 3 and 4). Demonstrating that the lack of YscF reactivity in the Δ*yscF*strain was due to the deletion of *yscF*. The higher molecular weight bands seen in the whole cell fraction (Figure [2](#F2){ref-type="fig"}, lanes 1--6) represent cross-reactive *Y. pestis*bands not specific to YscF, as they are present in the Δ*yscF*samples. The same cross-reactive bands are also present in whole cell samples probed with pre-immune serum (data not shown) suggesting that the reactivity seen is not induced by immunization with HT-YscF. The higher molecular weight band seen in the culture supernatant fractions could represent a multimer of YscF (possibly a trimer) or YscF in complex with another secreted protein. This higher molecular weight species is likely related to YscF as a similar band is not seen with pre-immune sera (data not shown) or in the Δ*yscF*strain (Figure [2](#F2){ref-type="fig"}). These results demonstrate that mice immunized with recombinant HT-YscF produce antibodies that specifically recognize YscF produced by *Y. pestis*. Active immunization of outbred mice followed by challenge with *Y. pestis*KIM5 ------------------------------------------------------------------------------ To examine the ability of HT-YscF to protect mice from *Y. pestis*infection outbred mice were immunized intra peritoneally with purified HT-YscF using complete Freund\'s adjuvant (CFA) for primary immunization and incomplete Freund\'s adjuvant (IFA) for booster immunizations or with a phosphate-buffered saline (PBS, \[[@B37]\]) control in CFA or IFA to control for adjuvant effects. For these studies, 6-to 8-week-old female Swiss-Webster mice were immunized i.p. with 40 μg/mouse HT-YscF in PBS or PBS (control mice) alone emulsified 1:1 with CFA. Experimental mice were boosted with 40 μg/mouse HT-YscF in IFA after two weeks and with 20 μg/mouse HT-YscF in IFA at 4 weeks post-immunization. Negative control mice were boosted with PBS emulsified with IFA according to the same schedule. Two weeks following the final booster immunization, sera were collected from the HT-YscF-immunized and the PBS-immunized mice to assay for HT-YscF reactivity. Sera from 22 mice from the HT-YscF-immunized and the PBS-immunized groups were tested for total IgG reactivity. HT-YscF immunized mice had a geometric mean titer of 1:40,000 for IgG specific to HT-YscF while PBS-immunized mice had HT-YscF titers less than the lowest dilution 1:800 (Table [1](#T1){ref-type="table"}). After establishing that the HT-YscF immunized mice had developed a strong antibody response to HT-YscF, the mice were challenged with *Y. pestis*. Two weeks after the final immunization, groups of 10 mice were challenged intra venously via the retro-orbital sinus with 10^1^to 10^6^CFU *Y. pestis*KIM5 (pgm^-^) in PBS (grown at 26°C). The mice were observed for 19 days after challenge, and the average doses required to kill 50% of the mice (LD~50~) for the treatment groups were calculated using the extrapolation method of Reed and Muench \[[@B38]\]. Mice that were immunized with HT-YscF demonstrated a 134-fold increase in the calculated LD~50~value as compared to PBS-immunized mice (Table [1](#T1){ref-type="table"}). These results demonstrate that immunization of mice with HT-YscF was able to provide a degree of protection to the immunized mice from a subsequent challenge with *Y. pestis*(Table [1](#T1){ref-type="table"}). While the protection provided by HT-YscF is not of the same magnitude as that seen with the protective antigens F1 or LcrV, the increased LD~50~value clearly shows that immunization with YscF affords a significant level of protection. Thus, HT-YscF becomes the only other reported antigen apart from LcrV or F1 to induce a significant protection in F1-positive *Y. pestis*. Immunization with YopD has been demonstrated to provide significant protection against a F1-minus mutant of *Y. pestis*\[[@B39]\]. The protection generated by HT-YscF suggests that YscF could be potentially developed as a novel subunit vaccine for *Y. pestis*or could serve as an additional antigen in a multivalent *Y. pestis*vaccine comprised of YscF, F1, and LcrV for example. Characterization of the antibody response to HT-YscF ---------------------------------------------------- Mice immunized with HT-YscF demonstrated a strong antibody response to YscF and provided protection to the vaccinated mice from lethal *Y. pestis*challenge. Due to the strong antibody response, isotyping analysis was performed to determine the predominant isotypes of antibodies produced by mice in response to vaccination with HT-YscF in CFA/IFA. Anti-YscF antibody titers were determined two weeks following the last immunization, prior to challenge. The YscF-specific antibody titers of PBS-immunized mice were below the ELISA assay baseline of 1:800 (Table [1](#T1){ref-type="table"}), as was the pre-immune serum (data not shown). However, the HT-YscF immunized mice reached a GMT (geometric mean titer) of 1:40,000 (Table [1](#T1){ref-type="table"}) for total IgG. Isotyping analysis was performed on the pooled sera obtained from the 22 mice selected for total the IgG analysis reported in Table [1](#T1){ref-type="table"}. Pooled sera were used to minimize the animal to animal variation expected from using outbred mice. The isotyping analysis demonstrated no significant IgA or IgM production (Table [2](#T2){ref-type="table"}) and a very high IgG titer (Table [2](#T2){ref-type="table"}) as expected from the data in Table [1](#T1){ref-type="table"}. Among the IgG subclasses IgG2b appeared to have the highest levels (Table [2](#T2){ref-type="table"}), although IgG1 and IgG2a levels were also very high (Table [2](#T2){ref-type="table"}). IgG3 levels were the lowest (Table [2](#T2){ref-type="table"}). Generally, immunization with CFA tends to drive a strong Th1 response. Immunization with HT-YscF in CFA induced a strong Th1 response, evident by the high IgG2a response. However, a strong Th2 response is also present as seen by the high IgG1 and IgG2b levels. Immunization with HT-YscF induced a strong IgG1 response in mice and interestingly, Titball et al. have shown that high IgG1 titers to a F1/LcrV recombinant subunit vaccine correlated very well with protection against pneumonic plague in mice \[[@B40]\]. This suggests that YscF, which also induces a strong IgG1 response, could possibly afford some protection against pneumonic challenge as well as against systemic challenge. Ability of α-YscF to effect Yop translocation --------------------------------------------- To examine one possible method that anti-YscF could be functioning to provide protection in immunized animals the ability to translocate Yops in the presence of anti-YscF antiserum was examined. Antibody to the surface-localized LcrV has been shown to block the ability of the TTSS in *Y. pestis*to translocate Yops into cultured macrophages \[[@B41],[@B42]\] but anti-LcrV was unable to block translocation into HeLa cells \[[@B21]\]. However, anti-LcrV was able to block Yops translocation by *Y. pseduotuberculosis*into HeLa cells \[[@B26]\]. Since YscF is also surface-localized the ability of anti-HT-YscF to block Yop translocation into HeLa cells was tested. Day et al have described an elegant methodology to follow the translocation of Yop effector by fusing them to a Elk reporter \[[@B43]\]. Elk, a eukaryotic transcriptional activator, becomes phosphorylated only after entering the nucleus, providing a reporter for translocation into eukaryotic cells \[[@B43],[@B44]\]. This methodology has the advantage of not requiring cell fractionation and protease protection assays to establish the intracellular localization of translocated proteins. To test the ability of anti-YscF to block translocation *Y. pestis*KIM8-3002 was transformed with plasmid pYopE~129~-Elk \[[@B43]\]. HeLa cells were infected at an MOI of 10 and infection was allowed to progress for 4 h. After the 4 h incubation infected HeLa cells were harvested and immunoblotted to analyze YopE, Elk and PO~4~-Elk production. *Y. pestis*KIM8-3002 (wt; \[[@B24]\]) and an isogenic translocation defective strain containing a *yopD*deletion (KIM8-3002.2, Δ*yopD*; \[[@B45]\]) both containing pYopE~129~-Elk were used as positive and negative translocation controls, respectively. Immunoblots of HeLa cells infected with the wt and the Δ*yopD*strains showed that only the wt strain elicited the production of PO~4~-Elk while the Δ*yopD*strain had no production of PO~4~-Elk. The wt and the Δ*yopD*strains were used to infect HeLa cells in the presence of a 1:10 or a 1:25 dilution of anti-YscF (titer for HT-YscF, 1:100,000) or in the presence of anti-PcrG (a *Yersinia*non-reactive antibody control, titer for PcrG, 1:20,000). The wt strain was capable of translocating YopE-Elk in presence of both anti-sera, demonstrating that anti-YscF was not capable of blocking Yops translocation and expectedly the Δ*yopD*strain was still defective for translocation. The experiment likely contained sufficient antibody against YscF to block translocation. In a previous report Pettersson et al used as low as a 1:100 dilution of an anti-LcrV anti-sera with a titer of 1:20,000 for LcrV and in that experiment translocation of Yops into HeLa cells was blocked \[[@B26]\]. These results suggest that antibody to YscF may not exert its protective effect by blocking Yops translocation. The results also suggest that while YscF is surface-exposed in the yersinae, antibody directed against YscF, unlike, anti-LcrV cannot block translocation. This may imply that YscF activity is shielded from neutralization by antibody, unlike LcrV activity that is blocked by antibody in some cases. However, since anti-LcrV was unable to block Yops translocation into by *Y. pestis*into HeLa cells \[[@B21]\] but could block translocation into cultured macrophages \[[@B41],[@B42]\], the possibility remains that anti-YscF also display this type of differential blockage. Conclusion ========== In this study we have determined that immunization of mice with recombinant YscF can protect mice from an i.v. challenge with *Y. pestis*. This is the first report to our knowledge that has utilized a conserved protein from the type III secretion complex of a gram-negative pathogen as a candidate for vaccine development. This result suggests that type III secretion complexes of other gram negative pathogens could also serve as vaccine targets. YscF and its homologs are obvious targets for use as vaccine candidates as they are surface exposed and are required for virulence in all the systems examined. The protective antibody response elicited by HT-YscF is evidence that YscF is not only expressed during the course of a plague infection, but is also in a location accessible to antibodies at some point in the infectious process. The mechanism of protection by the YscF antibody response is currently under investigation. Essentially three possibilities exist to explain the antibody\'s protective activity: increased opsonization of the bacteria, enhanced complement binding to the bacterial surface, or direct blocking of Yops translocation into the host cells. Cytotoxicity assays (data not shown) and the ability to translocate Elk-tagged YopE into HeLa cells (Figure [3](#F3){ref-type="fig"}) have shown that it is unlikely that anti-YscF directly blocks Yops translocation into HeLa cells. Suggesting that a blockage of Yops translocation may not be the mechanism whereby anti-YscF antibodies are protective. The degree of protection observed after immunization with YscF is not as great as that seen for the two known protective antigens, F1 and LcrV. This result suggests that YscF could be best used in combination with the other known antigens to formulate a tri-valent vaccine for *Y. pestis*. However, further work could lead to the development of YscF as a monovalent vaccine or combined with other antigens that could be efficacious not only against *Y. pestis*, but also against *Y. enterocolitica*and *Y. pseudotuberculosis*. Methods ======= Cloning of *yscF*for overexpression and HT-YscF purification ------------------------------------------------------------ Plasmid pJM119 was constructed by cloning a *Bam*HI-and *Xho*I-cleaved PCR product into pET24b (Novagen, Madison, WI). The primers used to amplify *yscF*were HT-YscF Start (5\' CGG GAT CCG ATG AGT AAC TTC TCT GGA TTT 3\') and HT-YscF Stop (5\' CCG CTC GAG TGG GAA CTT CTG TAG GAT GCC 3\'). *E. coli*BL21(DE3) (Novagen) harboring pJM119 was used for HT-YscF overpexpression according to the manufacturer\'s suggestions. HT-YscF was purified using Talon resin (BD Clontech, Palo Alto, CA) according the manufacturer\'s directions. Immunization of mice and infection with *Y. pestis*KIM5 ------------------------------------------------------- For primary immunization 6-to 8-week-old female Swiss-Webster mice were immunized i.p. with 40 μg/mouse His-tagged YscF or phosphate-buffered saline \[[@B37]\] PBS (control mice) emulsified 1:1 with complete Freund\'s adjuvant (CFA). Booster immunizations were performed the same as the primary immunization with the substitution of Incomplete Freund\'s Adjuvant for CFA. Mice were challenged with *Y. pestis*via the retro-orbital sinus using blunt-feeding needles. *Y. pestis*used to infect mice was grown overnight in HIB broth, sub-cultured to an A~620~of 0.2 absorbance units and grown with shaking to an A~620~of 1.0 absorbance unit. *Y. pestis*cells for infection were harvested by centrifugation and resuspended in PBS. Plate counts were performed to verify CFUs for the infectious doses. Infected animals were monitored for death for up to 19 days, after which survivors were euthanized by CO~2~inhalation, according to the guidelines of the Panel on Euthanasia of the American Veterinary Medical Association. All animal work for this project was reviewed and approved by UND\'s IACUC. Bacterial strains, growth and fractionation ------------------------------------------- Bacterial strains used were KIM8-3002 \[[@B24]\], Δ*yscF*an isogenic in-frame deletion of *yscF*(G. Plano, University of Miami), and KIM8.3002.2 Δ*yopD*\[[@B45]\]. *Y. pestis*strains were grown in heart infusion broth or on tryptose blood agar base medium (Difco Laboratories, Detroit, MI) at 26°C for genetic manipulations. For physiological studies, growth of *Y. pestis*was conducted in a defined medium, TMH, as previously described \[[@B46]\]. Bacterial cells were fractionated as previously described \[[@B24]\]. Briefly, bacterial cells were chilled on ice after growth, harvested by centrifugation, and washed in cold phosphate-buffered saline (PBS; \[[@B37]\]). Bacterial whole cell fractions were prepared by resuspending the washed cells in cold PBS and precipitating total proteins with 10% (vol/vol) trichloroacetic acid (TCA) on ice overnight. Secreted proteins were recovered from the bacterial growth medium by centrifuging the spent medium a second time, transferring the supernatant to a clean tube, and precipitating with 10% (vol/vol) TCA on ice overnight. The TCA-precipitated proteins were pelleted by centrifugation (20,800 × g at 4°C) for 20 min and resuspended in 2X sodium dodecyl sulfate (SDS) sample buffer \[[@B37]\]. Protein electrophoresis, visualization and immunodetection ---------------------------------------------------------- Proteins were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE), using 12.5 % or 15 % (wt/vol) polyacrylamide gels according to the method of Laemmli \[[@B47]\]. Samples were boiled 3--5 min before loading on the gels. Samples were loaded such that lanes containing different culture fractions represented equivalent amounts of the original cultures. Proteins were visualized in gels using GelCode Blue stain (Pierce Chemical, Rockford, IL) according to directions. For immunoblots, proteins resolved by SDS-PAGE were transferred to Immobilon-P membranes (Millipore Corp., Bedford, MA) using carbonate transfer buffer (pH 9.9) \[[@B48]\]. Specific proteins were visualized using mouse or rabbit polyclonal antibodies specific for YopE (rabbit α-YopE; gift from G. Plano, University of Miami, Miami, FL), YscF (mouse α-YscF, this study), Elk (rabbit α-Elk, Cell Signaling Technology, Beverley, MA) and PO~4~-Elk (rabbit α-PO~4~-Elk, Cell Signaling Technology). Hexahistidine tagged YscF was visualized using a penta-histidine specific antibody (mouse α-Penta-His, Qiagen, Valencia, CA) Alkaline phosphatase-conjugated secondary antibody (goat anti-rabbit immunoglobulin G or goat anti-mouse immunoglobulin G; Pierce) was used to visualize proteins by development with nitroblue tetrazolium and 5-bromo-4-chloro-3-indolylphosphate (NBT-BCIP; Fisher Scientific, Fair Lawn, NJ). Antibody characterization and isotyping --------------------------------------- Flat-bottom, 96-well Nunc Maxisorp immunoplates (Fisher Scientific, Pittsburgh, PA) were coated with 100 μl of HT-YscF solution (4 μg/ml in Binding solution (0.1 M NaH~2~PO~4~, ph 9.0) at room temperature for 2 h (or overnight at 4°C). The wells were blocked with 200 μl/well blocking buffer (1% bovine serum albumin in TTBS (tris-buffered saline \[[@B37]\] + 0.5% Tween 20) and washed with TTBS. Test sera were serially diluted in blocking buffer and 100 μl of each dilution was added to duplicate wells that were incubated for 2 h at RT (or overnight at 4°C). The plates were washed and incubated for 2 h at RT with alkaline-phosphatase-conjugated anti-mouse secondary antibody. For quantitation of YscF-specific immunoglobulin isotypes and subclasses the plates were coated with alkaline-phosphatase-labeled anti-mouse isotype-specific antibody (1:400 in blocking buffer; Southern Biotech, Birmingham, AL). The wells were washed and 75 μl 3 mM para-nitro phenyl phosphate (p-NPP) was added to each well. The plates were incubated for 15 min at RT and the reaction was stopped by the addition of 50 μl of 1.5 M NaOH to each well. A~405~was measured to monitor the cleavage of p-NPP. Antibody titers were determined as reciprocal numbers of the highest serum dilution that displayed values for optical density twofold higher than the value of the control serum. Infection assays ---------------- Infection of eukaryotic cells was performed as described previously \[[@B24]\]. Prior to infection, eukaryotic cells were subcultured into 35-mm-diameter six-well tissue culture plates in RPMI-FBS and incubated at 37°C under 5% CO~2~for 48 to 72 h to a density of 5 × 10^5^to 8 × 10^5^cells per well. Cells were washed twice with warm L15 lacking FBS immediately prior to infection. Bacteria were cultivated at 26°C in HIB and used at an OD~620~of \~1.0 for tissue culture infections. Bacteria were added (at a multiplicity of infection (MOI) of 5 to 10) directly to prewarmed medium in the wells of the six-well plates. Plates were then centrifuged at 200 × g at RT for 5 min to achieve contact between the bacteria and the target cells and incubated at 37°C for 4 h. Translocation of YopE --------------------- Detection of Elk-tagged YopE from pYopE~129~-Elk was performed as described \[[@B43]\]. *Y. pestis*strains carrying plasmid pYopE~129~-Elk were used to infect HeLa cells. After 4 h, the culture supernatants were removed, and the infected adherent cells were lysed by the addition of 100 μl of 2X SDS-PAGE lysis buffer containing Pefabloc (Roche Molecular Biochemicals, Indianapoli, IN) and phosphatase inhibitor (P-2850) cocktail (Sigma, St. Louis, MO). Samples were boiled for 5 min and loaded onto 12.5 % SDS-PAGE gels, immunoblotted to PVDF membranes and probed with Elk-1 (\#9182) or phosphospecific Elk-1 (\#9181) antibody preparations (Cell Signaling Technology). Anti-sera specific for HT-YscF (titer of 1:100,000, this study) or the *Pseudomonas aeruginosa*LcrG homolog, PcrG (titer of 1:20,000, Matson and Nilles, unpublished), were used at dilutions of 1:10 or 1:25 in the infection medium to assess the ability of α-YscF to effect YopE translocation. Authors\' contributions ======================= J .S. M. cloned *yscF*for overexpression, purified HT-YscF, characterized the reactivity of antiserum to YscF, assisted with immunization and infection of mice, assisted with the LD~50~calculation, and wrote the draft of the manuscript. K. A. D performed immunizations and infections of the mice and performed the antibody isotyping. D. S. B. helped to design the immunization protocol and edited the manuscript. M. L N. conceived of the study, supervised the work, calculated the LD~50~and edited the manuscript. Acknowledgements ================ The author\'s would like to thank Deanna O\'Bryant and Jennifer Lamoureux for assistance with mouse experiments, Jennifer Miller for help with antibody ELISAs and Gregory Plano (University of Miami, Miami, FL) for YopE antiserum and the Δ*yscF*strain of *Y. pestis*. This work was supported by the UND Faculty Research Seed Money program. J. S. M was supported by a pre-doctoral fellowship from ND-EPSCoR. Work in M. L. N.\'s laboratory is supported by NIAID grants R01-AI051520 and U01-AI54815. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Purified HT-YscF was run on SDS-PAGE gels (12.5% for Panels A and B, 15% Panel C) to analyze purity of the recovered protein. Panel A, separated proteins were stained with GelCode Blue stain (Pierce Chemical). Panel B, purified HT-YscF was immunoblotted to PVDF and probed with antibody specific for penta-histidine to identify which bands contained the His-tag. Panel C, purified HT-YscF was immunoblotted to PVDF and probed with post-immune serum from immunized mice. ::: ![](1471-2180-5-38-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Derivatives of *Yersinia pestis*KIM8-3002 (KIM5 pPCP1-minus, Sm^r^) were grown in a chemically defined medium at 26°C for 2 h in the presence (lanes 1, 3, and 5) or absence of calcium (lanes 2, 4, and 6). Lanes 1 and 2 contain *Y. pestis*KIM8-3002. Lanes 3 and 4 contain *Y. pestis*KIM8-3002 Δ*yscF*expressing YscF from pBAD18-YscF. Lanes 5 and 6 contain *Y. pestis*KIM8-3002 Δ*yscF*gene. After the 2 h growth, the culture was shifted to 37°C to induce expression of the Ysc type III secretion system and the Low Calcium Response. Following 4 h of growth at 37°C cultures, were centrifuged to obtain whole cell fractions and cell-free culture supernatant fractions. Total proteins from each fraction were precipitated with 10% trichloro acetic acid. Dried proteins were resuspended in SDS-PAGE sample buffer and electrophoresed in a 15% SDS-PAGE gel. Proteins were then transferred to a PVDF membrane and immuno-blotted with pooled mouse serum used at a 1:20,000 dilution. Mouse serum was obtained by bleeding mice subsequent to immunization with HT-YscF, serum from several mice was pooled to control for animal specific variation. The position and sizes for the molecular weight markers are indicated and the position of YscF is shown. ::: ![](1471-2180-5-38-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### *Y. pestis*strain KIM8-3002 (wt) (lane 1 and lanes 3--6) and KIM8-3002.2 (Δ*yopD*) (lane 2 and lanes 7--10) both containing plasmid pYopE~129~-Elk were used to infect HeLa cells at an MOI of 10. 4 h following infection the culture supernatant containing non-adherent bacteria were removed the remaining adherent HeLa cells were solubilized in 2X SDS-PAGE buffer. Following solubilization proteins were separated and immunoblotted to triplicate PVDF membranes. The triplicate blots were probed with α-YopE, α-Elk, or α-PO~4~-Elk, followed by incubation with an alkaline phosphatase conjugated secondary antibody and developed using NBT/BCIP. To some samples anti-sera specific for HT-YscF (lanes 3--4 and lanes 7--8) or the *Pseudomonas*protein, PcrG (lanes 5--6 and lanes 9--10), were added at dilutions of 1:10 (lanes 3, 5, 7, and 9) or 1:25 (lanes 4, 6, 8, and 10). ::: ![](1471-2180-5-38-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### IgG response to HT-YscF vaccination and LD~50~determination. ::: Immunogen anti-YscF GMT\* LD~50~ Fold increase in survival ----------- ----------------- -------- --------------------------- PBS \< 1:400 159 \- HT-YscF 1:40,000 21,344 134 \*Geometric mean titer for total IgG, determined from 22 HT-YscF immunized mice. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Antibody isotype titers from mice\* immunized with HT-YscF. ::: Antibody isotype Titer ------------------ -------------- Total Ig 1:100,000 IgM \< 1:800 IgG1 1:100,000 IgG2a 1:100,000 IgG2b \> 1:100,000 IgG3 1:20,000 IgA \< 1:800 \*Sera from 22 mice were pooled for this analysis. :::

PubMed Central

2024-06-05T03:55:58.244478

2005-6-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168899/", "journal": "BMC Microbiol. 2005 Jun 24; 5:38", "authors": [ { "first": "Jyl S", "last": "Matson" }, { "first": "Kelly A", "last": "Durick" }, { "first": "David S", "last": "Bradley" }, { "first": "Matthew L", "last": "Nilles" } ] }

PMC1168900

Background ========== Patients are increasingly urged to become involved in treatment choices, including watchful waiting (surveillance but no active treatment), by working through medical evidence to decide on a course of action with their physicians. Guidelines and evidence summaries are commonly available to physicians and frequently include a \"summary for patients\", as do some major clinical journals. Decision aids have proliferated in recent years yet most require high levels of literacy to access them and have not been designed according to the principles of plain language. This paper describes one of the first attempts to make a decision aid accessible to consumers with lower levels of literacy and education using principles from reading research and document design. To develop a prototype plain language decision aid, and to describe the design process, we designed a decision aid for treatment decision making in localized prostate cancer. Three standard therapies, surgery, radiation therapy, and watchful waiting, are available to patients and paid for by most insurers in the US. The 10-year mortality rate for each of the standard treatments is equivalent, but the side effects and the processes are quite different. This conundrum is made even more difficult because specialty physicians tend to recommend treatments performed by their own specialty \[[@B1],[@B2]\]. Even inter-disciplinary case conferences can expose patients to evidence and professional opinions that appear to conflict. The aim of the research reported here was to develop a plain language decision aid for localized prostate cancer initial management following a positive biopsy. Three decision aids (booklet, Internet, and audio-tape formats) were designed to contain equivalent information. Formative evaluation was designed to test acceptability to men with both highly functional and limited literacy skills. We wanted to make sure that the DAs were not perceived as \"dumbed down\" either in tone or content. A further aim was to identify missing information, and information that might not adequately represent the cancer experience, The aim was to include the perspective of community dwelling men in the appropriate age group and men who had had treatment for their prostate cancer. Evidence-based patient choice ----------------------------- \"Evidence-based patient choice\" is a term coined by Hope in 1996 and elaborated in a book by the same title \[[@B3]\] to articulate the addition of the patient perspective to the evidence based paradigm laid out by Sackett \[[@B4]\]. Adding \"patient choice\" to evidence-based medicine is an effort to locate the need for evidence squarely in a patient-centred approach to patient-provider encounters. The theoretical roots are in economics, ethics, psychology and medicine. There are also political influences, including the consumer choice movement and struggles over resource allocation and health care organization. The synthesis of ethical approaches to shared decision making, and rational choice models provoke both theoretical and political concerns, and is far from a perfect resolution. The practical task of supporting patients\' access to information they can use led us to develop this new prototype. We also desired to have the information available in real time to reach the most satisfying resolution of treatment dilemmas. By making our assumptions transparent, and testing the results with patients and scholars we hoped to improve both theory and practice. Approaches to evidence tools for patients ----------------------------------------- Decision aids exist in modest numbers, specific to particular clinical conditions. A recent meta-analysis identified 131 different decision aids across a variety of clinical problems. Sixty-five of these fit the Cochrane criteria and were evaluated in a randomized trial \[[@B5]\]. Decision aids have expanded the scope of traditional patient education materials and draw on several perspectives. These may be classified conceptually as 1) traditional medical, 2) technology assessment (TA), and 3) patient-centred. Traditional patient education materials mirror the medical model didactic elements. They include descriptions of: a) the disease, b) its epidemiology, c) the morbidity and mortality outcomes of the disease in the population, and d) screening and treatment recommendations for either public information campaigns or clinical encounters. This model has been criticized for its emphasis on diseases, rather than on patients and their concerns, and for being paternalistic \[[@B6]\]. Decision aids have addressed the limitations of the traditional medical approach by integrating evidence-based medicine with patient choice and by encouraging shared decision making. It has largely accomplished this with a technology assessment (TA) approach. TA focuses on the quality of the medical evidence, levels of rigor of evidence and providing symmetry in describing risks and benefits of all patient options. The most widely available examples of TA-based decision aids use the Ottawa Decision Support Framework and BMJ Best Treatments approach. The Ottawa Framework asks the decision-maker to choose among treatment or screening choices demonstrated to be effective \[[@B7]\]. It emphasizes clarifying values for outcomes and emphasizes weighing the pros and cons of alternatives in the choice process to arrive at a tentative choice that the patient discusses with the physician. Consistent with evidenced-based health care, it grades the medical evidence for each alternative according to its quality. The Best Treatments approach likewise grades the evidence, teaches patients about uncertainty in data, but does not consistently provide numerical data for side effects \[[@B8]\]. The TA approach is a major step forward in providing patients with accurate evidence for making comparisons among alternatives, but it may suffer from speaking to the professional health care community better than it speaks to patients. To develop a plain language prototype, we accepted most of the premises of the TA approach, but redesigned the presentation of results, using plain language and document design strategies. Translation of evidence into plain language ------------------------------------------- The rationale for \"re-constructing\" decision aids in plain language is that it works for patients with a wide range of literacy abilities. Plain language includes three major elements: 1) The use of everyday language and other clear writing strategies, 2) well-structured, logically sequenced, and focused information, 3) effective design and layout. It uses medical evidence as a base and attempts to lead readers through the relevant elements to provide an opportunity for synthesis and drawing conclusions about which therapy best suits the values, fears and expectations of the patient. A plain language approach incorporates social marketing principles and is often used by pharmaceutical companies, patient advocacy groups, and others to deliver compelling messages. The materials these groups produce, however, sometimes minimize or eliminate numbers in the interest of being non-threatening and engaging and accommodating limited quantitative literacy skills\[[@B9],[@B10]\]. Plain language approaches have also been used to communicate public health messages to patients. They frequently attempt to persuade them to undertake an action (e.g., engaging in cancer screening) rather than engage in shared decision making. Highlighted by the recent Institute of Medicine report \[[@B11]\], many groups have published plain language document design guidelines \[[@B12]-[@B17]\]. As stated by Hibbard and Peters \[[@B18]\], the challenge is \"how to present\\ldotsinformation so that it is actually used in decision making.\" They note that three information presentation strategies assist consumers, all of which are reflected in a plain language approach: lowering the cognitive effort required to use information; linking information to real-life situations to increase the emotional connection; and presenting numbers in frequencies (e.g. 10 in 100). We addressed each of these in developing the prostate cancer DA. To address the immediate world of patients in the midst of being told of a disease and making a decision, it is important to describe the relevant risks and benefits of treatments without losing the precision and realistic risk communication capacity of numbers. Unpacking technical jargon actually adds substantial detail about what will happen to the patient in non-technical, everyday terms. It also explains laboratory tests and other medical information that doctors will use in making recommendations. The purpose is to provide patients with tools and encouragement to engage with the doctor in discussing technical aspects of the information doctors use to make a decision. Along with risk and benefit data, a decision aid should also explain the treatment process to the patient, in order that she or he can anticipate the personal experience of undergoing treatment. This approach does not abandon or delete technical language. Rather it defines it and provides examples, and phonetic pronunciations of medical technical terms in order to improve communication between patients and health professionals. In the prototype, we introduce the idea that there is variability among patients in their values for potential outcomes. Unlike some decision aids that provide a rational recommendation based on formal utility assessment \[[@B19]\], this approach does not make a recommendation. The resolution of the values dilemma is left to patients, their families and friends, and health care providers. Needs assessment ---------------- Prior to developing a new decision aid, we performed a two-part needs assessment. In the first part, we searched the literature for surveys providing information about knowledge levels of newly diagnosed patients following biopsy and discussion with their doctors but before treatment. In the second part, we evaluated existing patient education materials for localized prostate cancer. Since we found no surveys of men actually facing the localized prostate cancer treatment decision, we interviewed 184 men newly diagnosed with localized prostate cancer from both community and tertiary care urology practices in Michigan \[[@B20]\]. This sample served as the historical controls for this paper. The survey included demographics, patient knowledge about their prostate cancer, treatment options, side effects and satisfaction with care. Bivariate and multivariable analyses were performed to adjust for potential confounding and to determine if sociodemographic factors affected understanding of key issues. Average age of the men was 64 years and 19% were African American. Seventy-one percent completed some college and 51% of patients were currently employed. White men reported higher rates of knowing their PSA, stage and grade (92%, 93% and 99% respectively) than African American men (69%, 88% and 85%). Ninety-three percent of all respondents stated that they had discussed treatment options with a physician. Only 62% knew that radical prostatectomy, external beam radiation therapy, brachytherapy and watchful waiting were standard options for localized prostate cancer. Significant associations between sociodemographic factors and a patient\'s understanding of prostate cancer treatment issues were identified. Despite objective gaps in knowledge, men showed high satisfaction with information received from their physicians. Prostate cancer has been the subject of several decision aids \[[@B21],[@B22]\] Considerable research has identified patient and health professional concerns about treatment. In part 2 of the needs assessment, we used our previously developed generic evidence template to provide evaluation guidelines, but made it specific to prostate cancer \[[@B23]\]. We used the template to comprehensively survey patient education materials publicly available from US government agencies, charities, drug companies and others for inclusion or absence of the evidence elements. The full report found key evidence to be missing from most patient education materials (PEMs) \[[@B24]\]. We found that standard treatment options were almost never explicitly compared to one other. PEMs frequently did not describe the chances of dying from prostate cancer or of experiencing other side effects of each treatment. We found most did not reflect consideration of the user in the way materials were written, structured or designed. Finally, most did not encourage patients to formulate a treatment preference to discuss with their doctors. Based on the needs assessment, we developed the decision aid, described in this report, in three formats (booklet, audio tape, and Internet) with particular attention to plain language and to potential concerns of African American men. Three major challenges lay at the heart of the plain language translation: 1) introducing the social and emotional context, 2) translating the outcome and side effect rates into plain language, and 3) encouraging shared decision making. The three formats used the same words and content organization to facilitate evaluation of the impact of print, audio and Internet media on patient knowledge and patients\' assessment of the usefulness of the different media in preparing to talk with the doctor. Methods ======= The decision aid prototype: design method and assumptions --------------------------------------------------------- Based upon our earlier analysis of good and bad aspects of the prostate cancer PEMs and an extensive search of the clinical literature, a draft of the decision aid was prepared in what might be called the standard medical parlance. This material was iteratively modified to emphasize shared decision making and incorporate the principles of plain language. The research team who designed the project was strongly guided at this point by two health literacy/plain language experts. Similarly, document design and web design were strongly guided by graphics experts. The prototype was presented to the Michigan Prostate Cancer Action Committee (PCAC), a sub-committee of the Michigan Cancer Consortium (MCC), a consortium of hospitals, universities and cancer centrss in the State of Michigan. Final edits were made based on the formative evaluation by patients, clinicians, and the development collaborators and the final document produced. The prototype booklet and Internet version that resulted from the evaluation process can be viewed at the Michigan Cancer Consortium (MCC) website \[[@B25]\]. Table [1](#T1){ref-type="table"} lists the design decisions made for the prostate cancer prototype. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Information and Information Architecture Issues and Solutions ::: **Concern** **Solution** -------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------- Complex information • Organize and structure document for ease of understanding and use, including use of summary tables • Create affective appeal using individual vignettes • Describe all elements of treatment process patients experience Engage the reader in the social context • Use pictures captioned with comments from men of varying ethnicities • Suggest connection with trusted others and other patients • Recognize and address feelings of fear No mandatory autonomy • Suggest that decisions may be made by the patient or delegated to the doctor, or trusted individual. • Suggest shared decision making is compatible with keeping or delegating Anticipate cultural issues • African Americans are likely to be sensitive to potential to withhold treatment in the name of watchful waiting. • African Americans frequently are diagnosed with more severe disease relative to whites. • Sources of authoritative advice vary among ethnic groups. These can be acknowledged subtly in text Translate medical words, but do not eliminate technical information • Define words and include a glossary Creating/choosing appropriate pictures • Simplify anatomical drawings to include only necessary visual information • Include realistic picture of torso, with head included to provide conceptual context • Include photographs of men who look serious, but not devastated Graphic Design and Layout: Use evidence-based document design principles • Spacing to emphasize different points and topic changes • Wide margins to make text less dense • Adequate page size to keep information on a topic on one page • Graphic design that draws the reader through the booklet • Text design that keeps related information in close proximity Plain language • Use conversational words and sentence structure, even if not standard grammar • Keep the writing tight. Eliminate unnecessary words. • Keep the tone friendly and personal, speaking directly to the reader. • Explain technical language even though it adds words. Assumptions behind technical words must be included • Make scientific references available on the website, but do not include in the text ::: Connecting to patients\' social and emotional world --------------------------------------------------- To connect with the concerns of men and their families and friends about prostate cancer treatment, the decision aids embed a number of emotional \"hooks\", showing photographs of middle-aged men of varying races. Quotes from different men state varying preferences for their preferred degree of autonomy: making the decision themselves, delegating it to family or the doctor, sharing it with clergy, friends, or health professionals. These statements are designed to legitimize a range of appropriate alternatives for the problem of \"who decides\". They are meant to avoid a \"mandatory autonomy\" position, allowing the patient to hand off the decision to the physician if he does not want to decide himself \[[@B26]\]. The central message promotes active patient participation, followed by making, sharing, or delegating the decision. Further, an attempt is made to imbed in the text items that speak to culture-specific issues. For example, the statement is made that watchful waiting does not mean that the doctor \"just does not want to treat me.\" This statement anticipates the common belief, particularly among African American patients, that doctors may withhold treatment for cost or other reasons. The emotional context of the booklet is also invoked by explicitly noting in the text that cancer is scary, and that treatment side effects may have heavy emotional impact. The emotional \"threat\" level of how the disease is discussed is planned to be at a moderate level. The decision aid acknowledges inevitable fear, supplies concrete understanding of disease and treatment, and encourages a sense that patients are competent to make the decision. Presenting quantitative evidence -------------------------------- The goal of an evidence-based decision aid is to make the information easy to read, remember, and use. The prostate cancer prototypes use specific textual, organizational, and design elements enhanced by appropriate message framing. They do not assume the reader knows either the technical meaning of words or the background context in which they are used. This filling in of the \"why\" and \"what for\" of laboratory tests, treatment regimens, and other medical information is essential for the millions of adults with limited literacy skills and limited experience with navigating the medical system. It is also preferred by most people when learning new information, especially in a highly charged emotional context. Specific guidelines for the prostate cancer decision aids were: 1) aim for low to average reading level (grade 7 level), 2) present quantitative risk information as simply as possible, both positively and negatively framed \[[@B27]\], and 3) use a minimum of technical medical terms, and provide a glossary, called \"explanation of medical terms\". The objective is to invite all patients to grapple with the decisions they face with as little intervening struggle with technical medical language as possible. The result, ideally, is sophisticated ideas in plain language. To invite the patient into the content requires attention to \"information architecture\" and the layout of information on pages to guide the reader through the text. Thus, the density of information on a page, the choice of a booklet or an electronic tool, and the length and \"feel\" of the information tool become important. The organizing principle was to describe each treatment option in a complete synopsis on its own on two facing pages of the booklet, or on sequential web pages under one button. The audiotape is organized in \"chapters\", like a \"book on tape\". For each treatment, the text describes, \"What happens\", \"How this treatment can help\", and \"How this treatment may cause problems.\" Each of the individual synopses tells the whole story about a specific treatment. The treatment options are followed by two comparison tables, one narrative and one quantitative. The summary tables explicitly compare results across treatments. Very brief directions guide the reader in how to use the tables. Additional information, not related explicitly to the choice (stage and grade of cancer, and PSA tests), is described in the same amount of detail as a single treatment option (two facing booklet pages). The objective is to demystify what will happen, show how doctors decide what is best, describe the degree of uncertainty about test outcomes, and indicate the medical information on which decisions are based. Side effects and mortality are presented as mean values, not including the ranges. This is a pragmatic solution to the simplification task, and should be tested further. Explaining that diagnostic test results are imperfect information, and that treatments have only a probability of success is a new idea to most patients. The design task was to present uncertainty, but keep it simple. Another quantitative principle that is critical to informed consent is presentation of absolute, rather than relative risks in terms of specifying a number out of 100 rather than a percentage. Absolute rates have been demonstrated to provide patients with improved estimates of their own chances of experiencing a particular outcome \[[@B28]\]. Shared decision making ---------------------- A third challenge was to encourage shared decision making without specifying the form of the clinician-patient conversation or the values discussion. The message on the cover of the DA first introduces the concept of choice, that \"there is no right answer\" to this treatment problem. The general statement, \"Your treatment decision is a shared one between you and your doctor\" is later stated directly. A \"things to think about\" section at the end of the booklet asks a patient to identify his most important goal for treatment, such as curing cancer, ameliorating symptoms, having the best possible sexual performance, or good bowel and bladder control. Patients are also asked to write down the best and worst thing for them about each treatment option. These sections are designed to facilitate an open discussion with the physician without directing it. The booklet was provided to each person with a request to provide written or verbal feedback. The booklet form of the decision aid was finished first, and provided the text and order of presentation that were duplicated in the audiotape and Internet versions. Qualitative methods were used to evaluate and revise the booklet before survey testing was performed with a new sample of men newly diagnosed with prostate cancer. Results ======= Formative evaluation -------------------- Evaluation was performed in two phases. Phase 1 was a focus group evaluation of the draft language. Phase 2 was a survey follow-up of 60 men newly diagnosed with prostate cancer, using the DA in real-time decision making. The Michigan State University and University of Michigan human subjects IRBs approved all study materials and procedures. Verbal and written consent for participation was obtained at the beginning of interviews and patient encounters. Phase 1: focus groups --------------------- ### Design and procedure Focus group methods were used to identify unacceptable or confusing language and to identify missing information or information needed to fully represent the important issues for decision making. This information was obtained from the perspective of men who had not experienced prostate cancer, and also those who had. The analytic techniques used were similar to ethnographic techniques, although cross-sectional rather than longitudinal qualitative reports were used \[[@B29]\]. Homogeneous groups by race were formed to encourage participants to feel comfortable and to express concerns about bias. A moderator\'s guide addressed the acceptability of content and language, perceived understandability, and usability of the format. We inquired specifically about text passages that described risk numbers, results of a randomized clinical trial, the meaning of laboratory tests and their implications for mortality and treatment. Open-ended questions asked men their perception of the DA main message. Four focus groups were conducted. Each session lasted between 1.5 and 2 hours and was facilitated by a physician. Booklets were mailed to participants two weeks prior to the focus group and participants were encouraged to review the booklet in detail to provide feedback during the focus group session. Confidentiality ground rules were established and agreed upon by participants. After the session, a twenty-five dollar honorarium was provided to compensate participants for time and travel. Immediately following participants\' departure, a debriefing session was held between the facilitator and assistant to develop and amend observation notes and record suggestions for subsequent focus groups. All focus groups were tape-recorded. ### Sample Post-treatment men who had undergone surgery or radiation therapy for prostate cancer were recruited from a tertiary care prostate cancer clinic. Community men at risk for prostate cancer due to age were recruited from a study of decision making in benign prostatic hyperplasia. Actual patients at the point of decision making were not included in the focus group, due to their potential vulnerability to inaccurate, misleading, or offensive content. Twenty-five men participated in four focus groups. Mean age was 64.5 years (range 49--80). Twenty-four had health insurance; approximately half were not college educated. ### Focus group results The men found the prototype booklet \"encouraging\" and \"comforting.\" The general consensus was that prostate cancer appeared to be something \"that can be dealt with.\" The DA was described as \"professional,\" the \"right size\" (8 1/2 × 11), and containing the right balance between the size of white space and of print. The summary table of treatments and risks and the comparison table were viewed as especially helpful and men stated that these clarified individual treatment descriptions. The level of detail was described as being \"just right\", with the exception that the decision aid did not include detailed description of non-standard treatments (cryosurgery and nerve-sparing surgery). These were added, as was a recommended diagram showing cancer spreading to the surrounding lymph nodes and nerves. ### Positive and negative framing of statistical information The booklet stated each side effect as both a number out of 100 chance of occurrence and that number subtracted from 100 as the chance of avoiding occurrence. However, the obvious redundancy was interpreted by some readers as talking down to them and insulting their intelligence. To make it less repetitive, the wording was changed to vary the outcome named and acknowledge that this was another way of stating the outcome. Table [2](#T2){ref-type="table"} shows major changes made based on feedback. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Booklet Improvements based on formative evaluation ::: **Information item identified in focus groups and approved by investigators** **Draft 1** **Draft 2** ------------------------------------------------------------------------------- --------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Emotional impact None Hearing that you have prostate cancer may shock or frighten you, your family, and your friends. These feelings are natural. They may change over time, as you learn about your diagnosis, make treatment decisions, deal with symptoms, and go on with your life. Men are often afraid to share their feelings or get help from a counselor if needed. If strong feelings are hurting you or your family, ask your doctor to suggest help. Positive and negative framing \"If 45 men out of 100 experience impotence, this means that 55 do not.\" \"\...about 45 men out of 100 have permanent impotence. This means that 55 men out of 100 will have their original level of sexual activity.\" African American differences None \"African American men are often diagnosed at a younger age than white men and with more advanced prostate cancer. However, treatment may be equally successful for both groups if given the same care.\" Specific drug name, \"like Viagra\" \"medicine that helps with erections\" Treatment detail No mention of cryosurgery or nerve-sparing surgery Included under \"newer treatments\". Drawing Three inset pictures Only two Watchful Waiting description Few disadvantages listed Added disadvantages regarding potential progression of disease ::: ### Perspectives of African American men vs. White men While all found the booklet professional and a good source of information, there were differences in what issues the groups discussed. For African American men, the multicultural photos were important in that they illustrated that prostate cancer affects men of any race. The participants did suggest inclusion of more specific mortality information by race. (See Table [2](#T2){ref-type="table"}). African American men discussed several aspects of fear, including fear about acknowledging health problems, and the absence of fear in the faces of the men in the photos. They characterized the overall message of the booklet as not to be afraid. They also cited fear and distrust as major obstacles preventing their acknowledging a health problem. African American men did not feel the text talked down to them and felt the words were manageable -- \"there are no big words in this booklet.\" One commented that, \"I didn\'t feel like you were treating me like a 3^rd^grader.\" Lastly, the African American men read the overall message of the booklet as one of \"don\'t be afraid,\" whereas the White men read the message as one of \"informed decision making\". Several men across groups commented that they rarely read health related materials, but that they read every word before coming to the focus group. \"I normally only read sports. But I read this straight through\". ### Post-treatment men vs. community men Men who had been treated for their prostate cancer appeared to find discussing the DA more difficult than did the men of the same age who did not have prostate cancer. While they appreciated the information, some indicated that they were not convinced that they really had a choice when deciding on actual therapy, either because of the doctor\'s presentation or their own concerns about cancer. They also indicated that the consequences of the treatment decision were more emotionally difficult than they anticipated. These men appeared to struggle with the emotional and psychological consequences of their own treatment decisions. Post-treatment White men suggested that the emotional impact of treatment be more strongly addressed as illustrated by the following comments: \"But this doesn\'t deal with\... your first issue is to stay alive and the second issue is the quality of your life afterwards and that may be whether you are incontinent or maybe whether you don\'t have erections\.....But to the fact that, uh, you need to talk to your surgeon about the psychological consequences of the expected outcome of this surgery. Maybe you only think about living first, but looking at it in retrospect, I was like\.... Whoa, why didn\'t you tell me that ahead of time? Not that I would have made a different decision. I just want to be well informed. I didn\'t get that\". Some post-treatment men felt that the booklet was biased toward watchful waiting as a treatment choice. This was addressed by adding specific negative aspects to watchful waiting in the summary. For example: \"But this book, uh, had in mind a definite cast toward watchful waiting as opposed to more active interventions and I agree that, uh, that in fact for many of us that was not a very viable option I mean intellectually or emotionally or whatever\" and \"But as I said before, I thought there was a subliminal message there that said that watchful waiting was probably a preferred choice.\" Phase 2: survey --------------- To test the usefulness of the revised decision aid to men at the point of decision, a new convenience sample of 60 men newly diagnosed with prostate cancer on the basis of biopsy was recruited to test the usefulness and influence of the DA in real-time decision making. Participants were recruited and provided a decision aid after receiving their prostate cancer diagnosis at a post-biopsy visit to their urologists. The urologist introduced a research assistant to patients at the end of the visit in which he was told of his cancer diagnosis. The patient was asked if they would like to be in a research study evaluating the effectiveness of the three different decision tool formats. Patients were recruited sequentially until 20 had been accrued to use the booklet. After the Internet and audiotapes were developed, 40 additional patients were recruited to evaluate these formats. Due to the necessity for computer access for use of the Internet version, these men were offered a choice of the audiotape or Internet decision aid until 20 men were recruited for Internet and 20 for audio-tape use. A telephone survey lasting approximately 20 minutes was conducted by trained survey researchers from the Institute for Public Policy and Social Research of Michigan State University. The survey included knowledge items taken from the previous needs assessment. DA evaluation questions assessing balance, clarity, and length of the DA were modeled after those of Barry et. al. \[[@B30]\]. Demographic (age, race, income, education and geography) and disease (prostate specific antigen (PSA), cancer stage, and grade) characteristics were ascertained during the interview. Knowledge questions included knowing one\'s own test results (PSA, stage, grade), treatment options and side effects. ### Sample The sample of 60 men recently diagnosed with prostate cancer had an average age of 62 years (SD = 7.9). Ninety percent reported their race as White (8% as African American), 5% had not graduated from high school, 20% had a high school dipoloma or GED, 32% had some college training, 19% had received a bachelors degree and 24% had received training post-baccalaureate. Sixty percent were employed, 33% were retired, and 5% were unable to work. Twenty-two percent had an income of less than \$25,000 and 77% reported Internet access. ### Survey results The responses of men using the three different media (internet, audio, booklet) were virtually identical across all survey items with the exception that those receiving the audio version were less likely to share it with family and friends (data not shown). Table [3](#T3){ref-type="table"} shows that the sample of men utilizing the DA and historical controls had similar knowledge of their own pathology results, with the exception that the DA sample were less likely to be informed of the stage of their cancer (Fisher\'s Exact Test P-value = 0.003). Knowledge of treatment options was not significantly different between the two groups. However, 12% more men believed watchful waiting to be a standard treatment in the DA group (87% vs 75% in controls, p = 0.073). In addition, 6% fewer men believed external beam radiation to be a standard treatment in the DA group (88% vs 94% in controls, p = 0.159). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Knowledge: DA sample vs. historical controls ::: **DA sample (N = 60)** **Historical controls (No./Total)** **Fisher\'s Exact Test P-value** -------------------------------------------------------------------------------------------------------- ------------------------ ------------------------------------- ---------------------------------- **Knowledge of personal pathology results** Knew PSA 56 (93%) 160/181 (88%) 0.34 Informed of stage of cancer 53 (88%) 178/181 (98%) 0.00 Knew grade of cancer 57 (95%) 178/184 (97%) 0.69 **Knowledge of treatment options** Knew all of age, grade, stage, health, and PSA are at least somewhat important for treatment decision. 52 (87%) 142/184 (77%) 0.14 Believed watchful waiting to be a standard treatment 52 (87%) 138/184 (75%) 0.07 Believed external radiation to be a standard treatment 53 (88%) 173/184 (94%) 0.16 Believed brachytherapy to be a standard treatment 47 (78%) 143/184 (78%) \>0.99 Believed surgery to be a standard treatment 57 (95%) 177/184 (96%) 0.71 Discussed watchful waiting with physician 28 (47%) 72/184 (39%) 0.36 Discussed external radiation with physician 53 (88%) 166/184 (90%) 0.63 Discussed brachytherapy with physician 38 (63%) 121/184 (66%) 0.76 Discussed surgery with physician 59 (98%) 163/184 (89%) 0.02 **Knowledge of side effects** Knew surgery associated with incontinence 59 (98%) 172/184 (93%) 0.20 Knew surgery associated with impotence 58 (97%) 172/184 (93%) 0.53 Knew surgery associated with painful bowel movements 33 (55%) 82/184 (45%) 0.18 Knew radiation associated with incontinence 48 (80%) 113/184 (61%) 0.01 Knew radiation associated with impotence 56 (93%) 145/183 (79%) 0.01 Knew radiation associated with painful bowel movements 46 (77%) 118/184 (64%) 0.08 ::: Discussion of treatment options with the physician showed a significant increase in surgery discussions (98% in DA group, 89% in controls, p = 0.019). Discussions of watchful waiting increased by 6% to 47% in the DA group, and discussions of radiation decreased by 2% (external beam) and 3% (brachytherapy). While not significant, and very small, these changes are consistent with the increase in knowledge of options and side effects. Table [4](#T4){ref-type="table"} shows that the DA was universally found to be clear and helpful. A substantial percentage of men reported that the DA influenced their decision making. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Clarity and usefulness of DA (n = 60) ::: ------------------------------------ -------------------------- ------------------------------- ------------------------- ------------------------------ ------------------------- ------------------ **Amount of information** Much less than needed 5% A little less than needed 14% About right 77% A little more than needed 4% DK or refused 7% **Length** Much too short 0% A little too short 11% About right 84% A little too long 5% DK or refused 5% **Clarity of words** All clear 44% Mostly clear 51% Some clear/ not 5% Most unclear 0% All unclear 0% DK or refused 2% **Difficulty of numbers** Very easy 64% Somewhat Easy 24% Somewhat difficult 12% Very difficult 0% DK or refused 3% **Treatment description balance** Complete balance 80% Slanted to surgery 9% Slanted to radiation 4% Slanted to WW 7% DK or refused 8% **Recommend DA to a friend?** Definitely would 78% Probably would 20% Unsure 0% Probably would not 0% Definitely Would not 2% DK or refused 2% **DA improved understanding** Definitely did 49% Probably did 42% Unsure 2% Probably did not 2% Definitely did not 5% DK or refused 2% **Numbers influenced my decision** Definitely did 12% Probably did 41% Unsure 2% Probably did not 22% Definitely did not 22% DK or refused 3% **DA helped decision making** A lot 14% Quite a bit 30% Moderate 32% A little 16% Not at all 9% DK or refused 5% ------------------------------------ -------------------------- ------------------------------- ------------------------- ------------------------------ ------------------------- ------------------ DK = Don\'t know ::: Eight-six percent of men reported that they shared the DA with a spouse or partner; 22% shared it with other family members; 14% shared the DA with friends. Seventy-two percent reported that they were more likely to take an active role in their treatment decision. Discussion ========== This attempt to marry evidence-based shared decision making and plain language appears to have been informative and helpful in decision making among a diverse set of men. The design process we used was found to translate medical language while retaining quantitative information. The resulting DA was almost universally found to be helpful in decision making. Men in the DA group were more likely to discuss surgery with their physicians and to know that radiation therapy has side effects. Men reported sharing the DA with family and reported that the DA increased the likelihood of taking an active role in decision-making with the physician. The comparisons reported are limited by the use of historical controls and by the unavailability of data about actual decisions made. However, the two cohorts were surveyed using exactly the same questions. Both cohorts were very well informed about their own test results and about side effects of surgery. Side effects of radiation where less well known and somewhat responsive to the decision aid. In qualitative analyses, patients identified several principles of the plain language approach as particularly important to them. Those included \"translations\" of medical language, the emotional difficulties of choices, and the use of what they called attractive layout and illustrations. Some men objected to numbers shown both as a percentage and 100 minus that percentage to rigorously frame each in both positive and negative. However, the provision of numbers to show frequency of side effects of treatments was considered essential. We did not find that patients perceived plain language as \"talking down\". It is important to ask what is gained and what is lost in translating evidence-based TA approaches to plain language. Decision aids have been demonstrated to improve knowledge and assist patients in coming to stable decisions earlier in the treatment process \[[@B31],[@B32]\]. In the face of that success, what improvements in decision aid effectiveness may be expected with the plain language integration and what may be lost? The major deletion we have made from the TA approach is the grading of the medical evidence studies used as source material. We presented rates of outcomes but did not indicate whether they came from randomized clinical trials or descriptive studies. Our reasoning is that grading actually distracts from the simple presentation of rates of mortality and side effects of treatments, and that absorbing what the rates mean is the core of what is needed to make an informed choice. References were made available separately. Health literacy research shows that even highly educated patients who are worried and stressed by a difficult health decision prefer simple, every-day language that is easy to read quickly. No patients asked for references. The credibility of the DA may have been enhanced by their physician\'s offering it to them and by the endorsement of the Michigan Cancer Consortium (MCC). MCC represents major research universities, the State health department, and 75 other major Michigan medical organizations. Conclusion ========== The plain language DA presenting medical evidence in text and numerical formats appears acceptable and useful in decision-making about localized prostate cancer treatment. The gains lie in the potential to improve on earlier gains in patient knowledge. If the actual content is not diluted, but the language, organization and design are more useful for patients, it is possible that mean knowledge scores may improve and that the median and the mean will converge. Parker et al \[[@B9]\] argue that health literacy is central to multiple health system priorities, including quality, cost containment, safety, and patients\' involvement in health care decisions. They suggest that without attention to literacy, the move toward increased patient participation in health care decisions will exacerbate disparities in access and outcomes. This could mean that people with poor health literacy cannot function successfully in an environment designed for active, informed consumers. It should be noted that among our historical controls, there were large and significant deficits in knowledge among African Americans compared with White patients. We did not have sufficient numbers of African Americans in the intervention sample to test the ability of the DA to improve this deficit or to gauge the effect the DA had on willingness to discuss treatment options with physicians. Further testing should evaluate the DA among men with historically low levels or information, as well as among very low literacy men and non-readers. Testing of this DA in a trial against usual care is needed, as is development of the plain language approach in other conditions. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors collaborated on conception and design of the study. MHR was responsible for drafting the article, the design of the decision aid evaluation and drafting the template. DRR provided the original expert draft of the DA. MHR, JTW, and AF contributed early changes. SS and JOF drafted the plain language translation. AF and JTW designed, drafted and performed the survey. KKB analyzed focus group data. RLD performed the statistical analyses. All contributed to editing and revising and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6947/5/16/prepub> Acknowledgements ================ The authors are indebted to Dr. W. Underwood, and M. Velade for their assistance in conducting focus groups. The University of Michigan Comprehensive Cancer Centre, Health Media Research Laboratory (HMRL) performed design and execution of the booklet. Supported in part by the Michigan Public Health Institute, the Michigan Department of Community Health, and the Agency for Healthcare Research and Quality with funding from the Centres for Disease Control and Prevention. The contents of the booklet described in this report do not necessarily represent the official views of the Centres for Disease Control and Prevention.

PubMed Central

2024-06-05T03:55:58.247038

2005-6-20

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168900/", "journal": "BMC Med Inform Decis Mak. 2005 Jun 20; 5:16", "authors": [ { "first": "Margaret", "last": "Holmes-Rovner" }, { "first": "Sue", "last": "Stableford" }, { "first": "Angela", "last": "Fagerlin" }, { "first": "John T", "last": "Wei" }, { "first": "Rodney L", "last": "Dunn" }, { "first": "Janet", "last": "Ohene-Frempong" }, { "first": "Karen", "last": "Kelly-Blake" }, { "first": "David R", "last": "Rovner" } ] }

PMC1168901

Background ========== The martial arts have their beginnings in the Orient but more specifically the common styles seen in Western society are from Japan, China, and Korea. Taekwondo, which originated in Korea more than 1000 years ago, is more sport than self-defense oriented \[[@B1]\]. In 2000, Taekwondo became recognized as an official sport at the Sydney Olympics. Taekwondo is a full contact free-sparring sport which awards points for head contact. As such, there has been increased interest in injury rates in the sport, especially relating to head injuries \[[@B2]-[@B5]\]. Although much of the research focuses on injury rates, very little examination into pre-competition habits and training has been conducted. The current authors felt that certain key areas needed to be addressed. These included training habits, injuries, dietary practices, and social support. Training habits and injury -------------------------- Most martial arts athletes practice between two to four times per week \[[@B1]\]. However, like in any sport, the frequency and hours of martial arts training can vary widely depending on athletic and competitive level of the individual. Training may be defined as a routine or process undertaken by athletes to further enhance their skill. Specific training may vary among each athlete, but there is usually a general format which is followed. Training classes often begin with a brief warm-up or stretching routine. This may then be followed by kicking drills, self-defense drills, training in patterns (forms) and sparring \[[@B1]\]. Taekwondo athletes have a wide variety of protective equipment available, although its use varies greatly, and concerns have been raised that the equipment often protects the attacker more than the defender \[[@B2]\]. Besides the regular training routine, students also participate in full-contact tournaments. At the time of data collection at the tournament in question, one point was awarded for any strike from the waist upward. The World Taekwondo Federation (WTF) rules and regulations for Olympic competition were used for this tournament \[[@B7]\]. For the purpose of this study, an athlete was considered injured if the following occurred: 1) any circumstance which forced the athlete to leave the competition or training session; 2) any circumstance for which the referee or athlete had to stop the competition; or 3) any circumstance for which the athlete requested medical attention \[[@B6]\]. Injuries occur in both training and competition, and trauma to the lower extremity and head are the most common sites reported \[[@B6]\]. A distinction can be made between overuse and traumatic forms of injury; although in reality they could be considered as points along one continuum. Overuse injuries may occur following continued or accumulated microtrauma to a structure or body area \[[@B8]\]. Traumatic injuries are the result of physical trauma or external force to a certain region leading to a diminished functional ability \[[@B8]\]. In the current pilot study no distinction was made between these forms of injury although it would be an interesting feature to examine in future research. The majority of research about Taekwondo injuries has examined injuries sustained during competition. Even so, it has been noted that up to 60% of injuries are not reported \[[@B2]\]. There are a variety of explanations for this, including poor recall, lack of importance placed on the injury, and unwillingness to disappoint trainers. Weight cycling -------------- Weight cycling is a term used to describe rapid weight loss following self-induced food limitation and/or dehydration. Both gradual (seasonal) and rapid (weekly) weight reduction cycles are used by athletes, and have been investigated for potential effects on nutrition and performance \[[@B9]\]. These cycles are used in various sports such as judo, rowing, wrestling, and boxing in order to make a weight category. Like many of these sports, Taekwondo consists of repeated-effort, high intensity physical demands. In addition to this, Taekwondo competition is structured in a similar fashion to boxing and rowing in that athletes are required to meet weight requirements in order to compete. Although there is no known reported research about weight cycling in Taekwondo, it is the primary author\'s experience (holding a fifth degree black belt in WTF Taekwondo and practicing Taekwondo for more than twenty-five years) that it is widely practiced in the sport. The WTF has various weight classes depending on competition level \[[@B7]\]. Table [1](#T1){ref-type="table"} illustrates the four WTF weight classes per gender for Olympic competition, which were also the categories used at the tournament for the present study. Although the World Taekwondo Federation has eight distinct weight classes per gender for all competitions and championships except for the Olympic Games, no rulings have been implemented to address weight cycling in the sport. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### World Taekwondo Federation Olympic Weight Classes for Men and Women ::: **Males** **Females** --------------------- --------------------- Less than 58 kg Less than 49 kg Between 58 to 68 kg Between 49 to 57 kg Between 68 to 80 kg Between 57 to 67 kg Over 80 kg Over 67 kg ::: To date there has been no research investigating the perceived benefit of weight cycling among Taekwondo athletes and this is an area in which much work should be undertaken. Due to the similarities between boxing and Taekwondo, with respect to competition weigh-ins, it may be possible to infer that Taekwondo athlete\'s perceptions of this technique may be similar. One study examining weight cycling among boxers reported that all the subjects felt it necessary to lose weight prior to competition and that it improved their performance \[[@B10]\]. Athletes using this weight control technique may be mistaken in thinking that an advantage will be gained over the opponent competing at his/her natural weight. To this point the research findings into the effect of food and fluid restriction has been equivocal \[[@B10]\]. There is also a belief that nutrients and strength can be restored by eating and drinking in the period between the weigh-in and the competition. Several authors have reported various techniques for rapid weight loss. A few strategies include dieting, restricting food and fluid intake, diuretic use, long runs, skipping, cycling, saunas, and exercising in rubber/plastic suits \[[@B10]-[@B12]\]. Psychological state / support ----------------------------- Despite athletes\' perceptions of the benefits of weight cycling, there are both physiological and psychological side effects. Using the Profile of Mood States-A (POMS-A), anger, confusion, depression, fatigue, tension, and vigor were measured among weight cycling amateur boxers \[[@B10]\]. The study reported that rapid weight loss was associated with significantly higher scores on anger, fatigue, and tension, with decreased vigor. The authors concluded that weight cycling resulted in negative mood and debilitated performance among their respondents. It is important for parents, coaches, and significant others to recognize these signs and address them appropriately. The idea of participating in competitive situations can be daunting for some individuals. Intense pressure, anxiety, and somatic manifestations may result. For those in athletic competition, it is vital to recognize and address these sources of stress in order to produce more successful outcomes. To date there is a lack of research in the areas of weight cycling and its perceived benefits among Taekwondo athletes. There is also limited research in the areas of social support and injury profiles in Taekwondo athletes outside of competition. This pilot study is an initial step towards increasing our knowledge in these areas. The purpose of the present study was to assess training characteristics, competition preparation habits and injury profiles of Taekwondo athletes. Methods ======= Subjects -------- Sixty Canadian male and female respondents were recruited for the study. Participants were Taekwondo athletes competing at a national-level tournament. A total of 28 respondents with an age range of 16 to 29 years returned the distributed questionnaire. Four females and 18 males completed the questionnaire. An additional six participants did not indicate their gender on the returned questionnaires. The mean age of the competitors was just over 22 years, with a mean height and weight of 68.6 inches and 148 pounds, respectively. Instrumentation --------------- A twenty-one-item questionnaire (See [Additional file: 1](#S1){ref-type="supplementary-material"} was used to obtain a general profile of the athletes. Areas of focus included: amount of Taekwondo practice; training satisfaction; protective gear used; pre-competition eating habits; competition preparedness; social support for the sport; and injury profiles. The questionnaire was modelled after one that had been developed for Dragon Boat racers. Neither questionnaire has been tested for validity or reliability. Procedure --------- The first author was working at the national carding tournament as a member of the health care team. As they entered the facility, potential participants were invited to fill out the survey by the current author and his assistants. Only card-carrying athletes competing in the tournament were given questionnaires. Prior to participation, informed consent was obtained by the participants or their guardians. At that time, any questions regarding the study or survey were addressed. Sixty questionnaires were distributed. When participants were given the questionnaire, they were asked to complete it immediately and then return it to the current author or assistants. Statistics ---------- The Statistica Release 6 statistical package was used for all analysis. Descriptive statistics and Pearson\'s chi-square test were used. When inputting data, it was noted that certain variables had missing responses. In these instances, the number of participants who completed the questions was used to calculate the results. Results ======= Training Habits --------------- Training time, measured by number of practices per week, and number of hours per training session, is outlined in Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}. Specific activities during training were also examined. The frequency of sparring was reported. Twenty-five percent (n = 7) of the competitors reported sparring one to two times per week. Over 53% (n = 15) of respondents sparred three to four times per week, and over 21% (n = 6) sparred five or more times per week. Pre- and post-training stretching was also reported. Over 40% (n = 11) of respondents reported only stretching prior to their training sessions, and close to 60% (n = 16) of respondents stretched both before and after training. When examining the use of warm-up and cool-down exercises, over 57% (n = 16) of participants noted always warming up prior to training, while almost 43% (n = 12) reported only warming up occasionally. Only six of 28 respondents (21%) reported they always engaged in post-training cool-down exercises, other than stretching. Over sixty-four and fourteen percent (n = 22) of respondents reported occasionally and never using cool down exercises, respectively. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Training Time -- Number of practices per week (n = 28) ::: **Practices Per Week** ------------------------- ------------- **Number of Practices** **Percent** 2 7.1 3 14.3 4 25 5--6 39.3 7 or more 14.3 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Training Time -- Number of hours per practice (n = 28) ::: **Hours Per Practice** ------------------------ ------------- **Number of Hours** **Percent** 1 17.9 2 53.6 3 17.9 4 7.1 5 or more 3.6 ::: The use of protective gear was also examined. Over 60 percent (n = 17) of respondents reported always using protective gear, while 39.3% (n = 11) only used it occasionally. Table [4](#T4){ref-type="table"} lists the type of gear and its percentage use by the respondents. When examining the frequency of missed practices, five respondents reported never having missed a practice even though they were injured. More athletes missed one to two practices when injured, although there were a few respondents who missed a substantial number of practices regardless of the injury frequency. Less time and practices were missed as injury number increased. Eight respondents reported that they had not experienced any injuries. Four cases had incomplete data and were subsequently not used in the calculation of missed practices. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Type and frequency of protective gear used by Taekwondo athletes (n = 28) ::: **Gear Used** Elbow Pads Shoes Shin Pads Gloves Head Gear Instep Pads Chest Protector Mouth Guard --------------- ------------ ------- ----------- -------- ----------- ------------- ----------------- ------------- **% of Use** 57.1 35.7 92.9 3.6 57.1 10.7 78.6 14.3 ::: Table [5](#T5){ref-type="table"} reports the number of missed practices for the first to fifth injuries. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Number of missed practices affected by injury frequency (n = 16) ::: **\# of Missed Practices** **1^st^Injury (n)** **2^nd^Injury (n)** **3^rd^Injury (n)** **4^th^Injury (n)** **5^th^Injury (n)** ---------------------------- --------------------- --------------------- --------------------- --------------------- --------------------- None 3 0 1 0 1 1 to 2 5 1 0 1 0 3 1 0 0 0 0 5 to 9 1 1 0 0 0 10 to 14 1 2 0 0 0 20 to 24 2 0 0 0 0 25 to 29 0 0 1 1 0 30 to 39 1 0 0 0 0 40 to 49 1 1 0 0 0 50 to 99 1 0 0 0 0 150+ 0 1 0 0 0 ::: Injury profile -------------- At this level of competition, 75% (n = 21) had six or more years of Taekwondo experience, with over 57% (n = 16) of individuals having eight or more years experience. Of the 28 participants, only 6 (21%) reported never having experienced an injury. At a value of 46.5% (n = 13), the lower limb was reported as being the region most injured in first-time injuries. The upper limb and back had an injury rate of 18% (n = 5) and 10.8% (n = 3), respectively. Over 3% (n = 1) of participants reported experiencing head injuries. Data for participants suffering injuries are reported in Table [6](#T6){ref-type="table"}. Seventy nine percent (n = 22) of athletes reported they had an injury. Of those, 47 percent (n = 13) had a lower limb injury, 18 percent (n = 5) had an upper limb injury, and 11 percent (n = 3) had a back injury. Thirteen athletes (46%) reported they had experienced a second injury. Of those, 29 percent (n = 8) had a lower limb injury, 11 percent (n = 3) had an upper limb injury, and seven percent (n = 2) had a back injury. Injury rates decreased substantially for those who reported experiencing a third injury. Only fourteen percent (n = 4) of the athletes reported suffering from a third injury. Of those, seven percent (n = 2) had a lower limb injury, while upper limb (n = 1) and back injuries (n = 1) each accounted for four percent of the injuries. Four athletes (14%) reported experiencing a fourth injury while 2 athletes (7%) reported they had a fifth injury. Of those, only lower limb injuries were reported. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Injury rates and location of injuries in Taekwondo athletes (n = 24) ::: **Number of Injury** **Lower Limb Injury (%)** **Upper Limb Injury (%)** **Back Injury (%)** **Other Injury (%)** ---------------------- --------------------------- --------------------------- --------------------- ---------------------- 1^st^ 13 (46.5) 5 (17.9) 3 (10.8) 1 (3.6) 2^nd^ 8 (28.5) 3 (10.8) 2 (7.2) 0 3^rd^ 2 (7.2) 1 (3.6) 1 (3.6) 0 4^th^ 4 (14.4) 0 0 0 5^th^ 2 (7.2) 0 0 0 ::: The frequency of injuries in training versus competition was also examined. Out of a total 22 responses, 13 respondents reported experiencing their first injury during training, while nine respondents experienced their first injury in competition. Training was most frequently reported as the time of injury, with eight out of thirteen respondents reporting second injuries occurring during training, versus five out of thirteen during competition. No competition injuries were reported for the third to fifth injuries. In order to better understand the injury data, injury rates were calculated using the basic rate formula: (\#injuries / \# athlete-exposures) × 1000 = \# injuries per 1000 athlete-exposures (A-E). Due to problematic data, only 24 of the 28 respondent\'s data were used. The overall rate of injuries was 520/1000 A-E. The injury rate was 354.2/1000 A-E for training, and 166.7/1000 A-E for competition. The injury rate for training per hour was 32.5/1000 A-E/ hour. A variety of care was sought by the athletes following injury. Of those respondents completing the survey, 25% (n = 7) did not seek any form of treatment. Another 10.7% (n = 3) of the athletes were treated by medical doctors, 10.7% (n = 3) were treated by physiotherapist, 10.7% (n = 3) were treated by chiropractors, and 3.6% (n = 1) received acupuncture. A variety of treatment combinations were also reported by 14.4% (n = 4) of the athletes. These combinations included chiropractic care as well as various sources of therapy listed above. Weight cycling -------------- Pre-competition habits are an important factor to examine in all sports. Due to weight classifications in Taekwondo, the athletes are very conscious of their weight. Certain trends were reported by participants in order to achieve the desired weight. Over 53% (n = 15) of participants reported fasting prior to the competition. Of these individuals, 33.3% (n = 4) neither ate nor drank, 50% (n = 6) only drank, and 17% (n = 2) ate but did not drink. Aerobic exercise was another method used by competitors in order to reach the desired weight category. In addition to dieting, 83%, or 10 of the 15 fasting participants reported doing aerobic activity prior to competition. Social support -------------- Support is often key to athletes at higher levels of competition. The current author examined athlete support for the sport by significant others. Seventy-eight percent (n = 22) of athletes reported they had parental support, while 14% (n = 4) reported no parental support was given, and 7% (n = 2) of respondents responded that this support did not apply. Spousal or partner support was reported by 32% (n = 9) of the athletes, while 18% (n = 5) did not receive this support, and 50% (n = 14) of respondents noted that the category did not apply to them. Because the survey was completed prior to competing, participants were asked to record if they felt prepared for the upcoming event. Fifty percent (n = 14) of respondents responded that they were prepared, 39% (n = 11) felt prepared but nervous, and 11% (n = 3) did not feel adequately prepared for the competition. Several comparison analyses were performed using Pearson\'s chi-square test. None of the values were of statistical significance, and thus not reported. The rationale for only reporting frequencies is due to the small sample size of the study, making the use of other analyses like Pearson\'s or Fisher\'s Exact Test erroneous. Discussion ========== The objective of this retrospective investigation was to assess training characteristics, competition preparation habits, and injury profiles of taekwondo athletes. By having the athletes complete a survey, several areas of concern regarding competition preparation and injuries were highlighted. Training and injuries --------------------- When examining the training habits of taekwondo athletes, the current study reviewed several components of performance. Respondents had significant experience in the sport, with over 75% having six or more years of involvement. Training time, measured by number of practices per week and number of hours per practice, was also high. Over 53% of practices were two hours, with over 45% of athletes practicing between two to four times per week. Of importance is the relationship of how training time and competition is affected by injury. Unfortunately, our sample size was too small to have meaningful comparisons. Other authors have reviewed this relationship. In a study by Feehan and Waller \[[@B2]\], competitive performance affected by previous injury was examined. On the day of the competition, 35% of respondents had a current injury affecting performance. Some of these required strapping or support in order to perform. Seventeen percent reported continuing to train/compete against medical advice. Even with these injury rates, the authors noted that fight outcome was not significantly associated with current or previous injuries. One conclusion which might be drawn is that the injuries reported were not severe enough to negatively impact the athletes\' performance. It can also be assumed that many of those with severe injuries would have chosen to withdraw from or not enter the competition until an appropriate level of health was reached. Practice activities among the respondents varied. A large proportion of respondents warmed up prior to kicking drills, while less than 25% cooled down. One possible explanation for warm-up participation may be that it is encompassed within the class. On the other hand, cool downs may be left to the discretion of the athlete once training is finished. In the current study, stretching was considered a separate activity from warm-ups and cool-downs. Over 60% of respondents stretched both before and after training, while just over 40% stretched prior to training only. In future studies, it would be interesting to note if stretching occurred after warm-up, which is a newer trend of thinking in the prevention of muscle injury \[[@B13]\]. Within the questionnaire, it was specified that cool down exercises did not include stretching. By doing this, the authors intended to eliminate the overlap between the stretching and cool down items. Future studies should allow subject to specify the various types of cool down activities used, such as light jogging or light-paced jumping jacks. Future studies should also examine the relationship between injury rates and the use of stretching, warm-ups, and cool downs. Due to a limited sample size the current study was not able to make these comparisons. The final training activity examined in the current study was sparring. This was an integral part of taekwondo training, with over 50% of respondents sparring three to four times per week. The current study attempted to examine the relationship between frequencies of sparring when injured, but there were no statistically significant differences. Birrer \[[@B14]\] reported that most injuries occur during sparring, thus it is an area of training which deserves specific focus. Future studies should focus on both injury type and frequency occurring during sparring, as well as limitations in sparring due to injury. In the current study, training was most frequently reported as the time of injury and relatively few injuries occurred during competition. Even so, the overall reported injury rate was quite high, at 520/1000 A-E. The injury rates calculated for both training and competition are likely to be skewed. Firstly, respondents were asked to simply report if they had been injured during competition or training. A more accurate representation may have occurred if athletes were asked to report how many competitions they had participated in during the previous year and if they had suffered injuries during any of these. Training injury rates may have also been affected by athletes returning to play prior to complete resolution of their problem. This could make athletes more susceptible to subsequent injuries. When reviewing injury location reported in the current study, it was not surprising, that the lower extremity received the most injuries. This was also true for all subsequent injuries reported (up to five per athlete). These results are consistent with those of several other studies \[[@B2],[@B5]\]. The upper limb was the second most frequently injured region, with the head being the least frequently injured. Sprains and strains were the most common injuries, followed by contusions, which is similar to other research \[[@B2]\]. Other reports have listed contusions and concussions as the most common forms of taekwondo-related injuries \[[@B3],[@B6]\]. Zemper et al. \[[@B5]\] reported that contusions were the predominant type of injury in his study of injury rates recorded during the 1988 US Olympic team trials for taekwondo. In general, the number of practices missed decreased with subsequent injuries. Also, there were relatively few athletes who missed a large number of practices. This is perhaps attributable to injury severity. One explanation for fewer missed practices could be that once an athlete experienced one injury, s/he was more likely to increase the use of protective gear, thus avoiding or decreasing future injury severity. Future studies should examine if there is a relationship between increased uses of protective gear following an injury. Recently, there has been increased concern regarding head injuries in taekwondo. Koh and Watkinson \[[@B4]\] reported that when compared to other contact sports, competition Taekwondo had the highest incidence rate of concussions. This might be explained by the fact that athletes are awarded points for head contact. Disturbingly, it was also found that over 30% of concussed athletes suffered more than one significant head blow in the same match. Also, among 99% of the head blows, no evasive manoeuvres were attempted. This would suggest that athletes are poorly trained in blocking skills. Pieter and Zemper \[[@B3]\] also reported that contusions and cerebral concussions were the leading injury types among young male and female Taekwondo athletes. Again, unblocked attacks were a frequent occurrence in these injuries. Widespread safety education on head injuries, and more specifically concussions, is needed among Taekwondo athletes, trainers, and referees. Improved blocking skills and headgear are a priority in order to help avoid serious injury. Following injury, a variety of care was sought by the current study\'s athletes. Interestingly, one quarter of the athletes chose not to seek any form of treatment. This could perhaps be accounted for by the athlete\'s perception of the injury being relatively minor, or being able to manage it without medical advice. Several health professional were consulted by the injured athletes. These included medical doctors, physiotherapist, chiropractors, acupuncturists, and massage therapists. In addition, several athletes consulted multiple therapists. Athletes are generally anxious to return to their pre-injury status, and often become impatient with long-term therapy. This may explain why multiple health professionals were consulted. Also, some health professionals realize the benefit to a multidisciplinary approach, and use a network of referral sources when necessary. Weight cycling -------------- Not surprisingly, more than half of the competitors in the current study dieted prior to competition in order to make their weight class. Although the questionnaire did not specifically define fasting, the subsequent question provided several categories of fasting, such as \"did not eat and drink\", \"did not drink but eat\", and so on. Even with the lack of a clear definition for fasting, fifty percent of the participants reported having completely restricted food intake, while 33% fully restricted food and liquids. Because of the nature of this tournament setting, it was not feasible to weigh athletes prior to their competition. As such, the authors were not able to report actual weight loss occurrence among the athletes. Future studies should focus on intended and actual weight loss among Taekwondo athletes in order to better capture the occurrence of weight cycling in the sport. Rapid weight loss is a common practice among athletes in weight class sports. Hall and Lane \[[@B10]\] reported that their boxing subjects lost an average of 5.16% of their body weight within one week. Along with the weight loss, subjects reported higher anger, fatigue, and tension, as well as reduced vigour. Participants were able to maintain their baseline performance of circuit training when at the reduced weight, although the scores were significantly lower than the athletes expected. It can be postulated that athletes have a misplaced sense of improved strength and performance capabilities when weight cycling for competition. Unfortunately, these views may be reinforced if a weight cycling athlete wins a competition, thus increasing the likelihood of using the strategies in the future. In a study by Alderman et al. \[[@B11]\] examining the prevalence of and weight loss techniques used by high school wrestlers, more successful wrestlers engaged in rapid weight loss (RWL) versus less successful wrestlers. This further reinforces the use of RWL among young competitors. What is particularly striking are the methods used to induce rapid weight loss. Among high school wrestlers, excessive running was used by almost 92% of individuals practicing RWL. Exercising in rubber/plastic suits and using saunas are prohibited in American high school wrestling, but they continued to be used by 40--60% of wrestlers to achieve RWL \[[@B11]\]. Thirty-six percent of respondents in the current study did aerobic exercise in addition to dieting to make their weight, but specific activities were not asked for in the survey. Many short term and long term side effects have been reported with rapid weight loss. Alderman et al. \[[@B11]\] reported multiple symptoms experienced by collegiate weight cycling wrestlers. Over 46% of participants experienced headaches, while over 44% and 42% experienced dizziness and nausea, respectively. Other symptoms included hot flashes, nosebleeds, feverish sensations, disorientation, and increased heart rate. Wenos and Amato \[[@B15]\] reported that college-level wrestlers also experienced an increased perception of effort as muscle strength and endurance decreased with rapid weight loss. Fogelholm et al. \[[@B9]\] studied the effects of gradual versus rapid weight loss in national wrestlers and judo athletes on nutrient intake, micronutrient status, and physical performance (sprint, jump height, and anaerobic performance). A 5% to 6% reduction in body weight was reported in the gradual and rapid loss groups. Nutrient intake was significantly decreased in both groups in B1, B2, K, Ca, Mg, Fe, and Zn values, compared to baseline measures. Speed, vertical jump, and anaerobic performance were not impaired by either rapid or gradual weight loss. Other studies have also reported that despite nutrient depletion, performance of Olympic level amateur boxers during rapid weight loss was not significantly different versus times of normal dietary behavior. These authors concluded that despite reduced carbohydrate intake, there were other sufficient energy sources to meet performance demands \[[@B16]\]. In contrast, Filare et al. \[[@B12]\] reported that all mean micronutrient intakes were below recommended values, while triglyceride levels and free fatty acids were increased in weight cycling judo athletes. Left hand grip values and 30-second jump test output were decreased after seven days of food restriction. By reviewing the literature, some might argue that the evidence of health risks from weight cycling is equivocal. Even so, there are several possibilities that may help explain the lack of supporting data. One possibility is that there may be no effect. Another proposed by Waslen, McCargar, and Taunton \[[@B17]\], is that the duration, frequency, and severity of food restriction among the judo athletes in their study may not have been sufficient to have an effect. Even with a lack of strong support to illustrate the ill effects of weight cycling, monitoring dietary habits of athletes in weight class sports is recommended. It is more prudent to assume that larger weight losses and more frequent dieting could potentially result in negative physiological and performance consequences. Widespread regulations need to be implemented to control weight cycling practices among weight class sports. Athletes need to be educated regarding the negative effects of the practice on both their health and performance. Psychological state/support --------------------------- Support is often key to athletes at higher levels of competition. The current authors examined athlete support by significant others. The majority of athletes reported receiving support from either their parents, or spouse/partner. Unfortunately, the questionnaire used in this study did not delve into the various aspects of psychological state or support. In this pilot study, respondents were simply asked \"Are your parents supportive of your involvement in Taekwondo?\" and \"Is your spouse or significant other supportive of your involvement in Taekwondo?\" It is obvious that neither of these questions addresses the various components involved in support. Future studies need to be more specific in questioning the types and level of support provided to athletes, whether it be emotional, financial, or other various forms of support. As such, these results are of little contributive value. It should be noted that although a large percentage of the athletes felt prepared for the competition, they also reported being nervous. The significance of anxiety and other personality traits in competitive sport has long been studied. It has been reported that winning Taekwondo athletes had lower cognitive and somatic anxiety and higher self-confidence then their losing counterparts \[[@B18]\]. Others found no support for the relationship between competition trait anxiety and Taekwondo performance \[[@B19]\]. Even so, for ultimate personal success, athletes often require a strong support base. This encompasses a sense of understanding, trust, and support from the trainer, and significant others. In weight class sports, the potential effects of weight cycling must also be kept in mind. As noted above, several studies have reported deleterious effects associated with rapid weight loss. These effects may involve one\'s mental status. Filaire et al. \[[@B12]\] reported that confusion, anger, fatigue, and tension were significantly higher after weight loss. Vigor was also significantly lower after food restriction. Thus, when considering the psychological preparedness of an athlete, multiple factors must be measured. There are a few limitations in the present study which need to be addressed. The most obvious methodological issue in this study is that the questionnaire has not been validated. There is very little reported research regarding precompetition habits among Taekwondo athletes. As such, the authors felt it necessary to develop the questionnaire, knowing that there would be issues with its validity. Because this is a pilot study, the results from this study should be used with caution and as a means to enhance future studies in this area. In addition, the small sample size significantly affected statistical analysis. No correlations were significant and thus specific conclusions regarding associations between training behaviors and injuries could not be made. The response rate was low, likely due to the fact the participants were asked to complete the surveys upon entering the tournament building. Athletes may have neglected to complete or return the surveys because of lack of time or feeling that it was not a priority prior to their match. Also, a self-report retrospective survey may be affected by poor recall and perception bias. For example, the recall of more severe and painful injuries would likely be better than that of minor injuries/trauma. The survey was also completed at a competition, thus those who were injured and not participating were already selected out. With respect to the information gained regarding weight cycling, actual weights were not taken. It would have been more informative to weigh the athletes at the mat just prior to their match and compare the result with that of their tournament weigh-in. As mentioned previously, the questionnaire used in this pilot study was vague regarding several concepts. Key definitions were not provided on the questionnaire. Future studies should ensure that all concepts are clearly defined in order to reduce subject confusion and hopefully avoid missing responses or poor response rates. Conclusion ========== The results of this pilot study are primarily descriptive. Even so, they highlight specific training habits and injuries among Taekwondo athletes. Although this pilot study examined a variety of pre-competition habits, it is evident that there are several specific areas which require more in-depth investigation. In order for safety recommendations to be implemented, it is likely that clear relationships will need to be demonstrated. Specifically, the physical effects of weight cycling on performance and improved training to avoid injury need to be examined. Athlete\'s perceptions and belief systems surrounding weight cycling, social support, and injury reporting are all topics in need of further investigation. As such, follow-up research on the relationship of pre-competition habits and injuries in Taekwondo athletes is necessary. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MK analyzed the data, wrote the abstract and various components of the manuscript. HS wrote various components of the manuscript, as well as helped to edit it. YSC helped to develop, distribute and collect the questionnaire, and edit the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2474/6/26/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 There is also an additional file attached titled, \"**Figure 1. Questionnaire**\". This is the exact Questionnaire which was distributed to the athletes. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We would like to thank Dr. Judith Waalen for reviewing the manuscript and Dr. Cameron McDermaid for his help with the statistical analysis. We would also like to thank the Canadian Memorial Chiropractic College for its support.

PubMed Central

2024-06-05T03:55:58.252395

2005-5-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168901/", "journal": "BMC Musculoskelet Disord. 2005 May 27; 6:26", "authors": [ { "first": "Mohsen", "last": "Kazemi" }, { "first": "Heather", "last": "Shearer" }, { "first": "Young", "last": "Su Choung" } ] }

PMC1168902

Background ========== Pregnancy-induced hypertension (PIH), especially preeclampsia, is a major cause of maternal and perinatal morbidity and mortality worldwide \[[@B1],[@B2]\]. The impact of PIH on birth outcomes has been extensively studied. However, the potential long-term effect of PIH on infants born to PIH mothers has been less studied. PIH has been confirmed to increase significantly the risk of low birth weight by both increasing preterm birth as well as reducing fetal growth. On the other hand, PIH has been found to be associated with an increased rate of high birth weight and large-for-gestational age babies \[[@B3],[@B4]\]. These findings suggest that PIH, more specifically preeclampsia, is a heterogeneous syndrome and that preeclampsia may appear in two forms: restricted fetal growth preeclampsia and normal fetal growth preeclampsia \[[@B4]-[@B7]\]. PIH may have different short and long term effects on infant growth between these possible two types of preeclampsia by intrauterine growth restriction (IUGR) \[[@B5],[@B6]\]. Low birth weight or IUGR babies have been associated with the occurrence of several chronic diseases such as cardiovascular diseases in later life \[[@B8]-[@B10]\]. However, the hypothesis of the fetal origins of adult disease is still the subject of debate \[[@B11],[@B12]\]. One argument is that maternal risk factors in pregnancy (such as PIH) and environmental risk factors in the postpartum period can contribute to this life-long development of the chronic disorders. A change in infant growth of the IUGR baby itself (e.g. catch-up growth) such as in the critical early infant period may also have long-term effects on health later in life, and this change of postpartum growth may be influenced by PIH. Therefore, it is important to study postpartum infant growth patterns of babies born to mothers with PIH, and to determine if there are differences in infant growth between babies with and those without IUGR. However, to date, there have been few studies on the effects of PIH on infant growth. The objective of this study was to examine the effects of the various types of PIH on infant weight gain at 28 and 42 days postpartum. Methods ======= Database -------- We conducted a retrospective cohort study based on a population-based perinatal database from Suzhou, China. The database was previously described by Xiong et al \[[@B3],[@B13]\]. In brief, data used in this analysis were population-based, collected in all 10 hospitals in the city, including16,936 pregnant women from January 1, 1989 through December 31, 1990. We excluded patients with multiple pregnancies (212 cases), chronic cardiovascular disease (36 cases), chronic renal disease or a history of renal disease (303 cases), chronic hypertension or history of hypertension (59 cases), and abnormal blood pressure (diastolic pressure ≥ 90 mmHg or systolic pressure ≥ 130 mmHg) at the first prenatal visit before 21 weeks (850 cases) and abnormal blood pressure after 42 postpartum days (988 cases). There were 2,567 women with no or missing information on infant weight at 28 days or 42 days postpartum, including 105 women who had stillbirths, 138 women who had infant deaths up to 28 days postpartum and 2,324 women who were lost to follow-up. After exclusions, a total of 11,921 pregnancies were retained for the analysis. Definition of pregnancy-induced hypertension -------------------------------------------- PIH was classified as gestational hypertension, preeclampsia, severe preeclampsia or eclampsia. According to the Chinese criteria \[[@B14]\], gestational hypertension was defined as a blood pressure equal to or greater than 130/90 mmHg on more than two occasions greater than six hours apart without proteinuria after 21 weeks of gestation. Preeclampsia was diagnosed as hypertension of equal or greater than 130/90 mmHg but less than 160/110 mmHg with proteinuria of 1+ or 2+ on dipstick in two samples 6 hours apart or greater than 0.3 grams in a 24-hour urine collection. Severe preeclampsia was diagnosed when preeclampsia was complicated by a systolic pressure of ≥ 160 mmHg or diastolic pressure ≥ 110 mmHg and/or if proteinuria was greater than 2+ on dipstick or 5 grams in 24-hour urine collection. Eclampsia was defined as seizure occurred in patients with preeclampsia. Patients with eclampsia were grouped with the severe preeclampsia group for analysis. We restricted our focus to PIH, therefore, pregnancies complicated by chronic hypertension and preeclampsia superimposed on chronic hypertension were not studied. Definition of outcomes and confounding variables ------------------------------------------------ Birth weight was measured in grams at birth. Infant weight was measured at 28 and 42 days postpartum. Gestational age was determined by the obstetricians on the basis of the information on menstrual history, physical examination or early ultrasound examination. IUGR was defined as birth weight below the tenth percentile of expected weight for gestational age \[[@B15]\]. Infant weight at 28 days was measured as the standardised medical procedures during the third postpartum home visit by personnel from the community clinic near the newborn\'s home. Infant weight at 42 days postpartum was measured in hospital during the last day of perinatal care examination while registering for the beginning of the infant care. We developed an index of Infant Growth Percentage (IGP) to measure the infant growth rate. IGP at 28 days is defined as \[(infant weight 28 days - birth weight) / birth weight\] × 100% and so is the IGP at 42 days by \[(infant weight 42 days - birth weight) / birth weight\] × 100%. Infant complications were defined as at least one of respiratory distress, newborn fever, birth injury, jaundice, sclerema, cord infection and intra-cranial bleeding. Other potential confounding variables include maternal age, body mass index \[weight (kg)/height(m)^2^\], infant\'s sex, pre-pregnancy or gestational diabetes mellitus and maternal anemia. Statistical analysis -------------------- Mean birth weight and infant weights at 28 and 42 days postpartum were compared by a variance of analysis between different PIH groups. First, the three groups with PIH disorders were compared to the normotensive referent group. Then each PIH group was further divided into two subgroups according to the existence of IUGR and the same analysis performed using normotensive pregnancies without IUGR as the referent group. Post hoc pair-wise multiple comparisons, which test which specific means differ significantly from others, were performed by Tukey and Bonferroni procedures \[[@B16]\]. In order to assess the independent effect of PIH on infant weight at 28 and 42 days postpartum respectively, multiple linear regression analysis was performed to adjust for confounding variables \[[@B16],[@B17]\]. Infant weight at 28 days and 42 days were the dependent variables. Dummy variables of severe preeclampsia with and without IUGR, preeclampsia with and without IUGR, gestational hypertension with and without IUGR, normotensive status with and without IUGR, and other confounding variables were independent variables. The regression coefficients (β) are estimated by the method of least squares \[[@B16]\]. The statistical significance (p-value) of β was also tested. All statistical analyses were performed with SPSS 10.0 for Windows (SPSS Inc., Chicago, IL). Results ======= Table [1](#T1){ref-type="table"} summarizes the demographic and reproductive characteristics of the women classified in each of the categories of the three PIH groups. Among the11,921 pregnant women, PIH occurred in 1,252 women (10.5 %). Mean maternal age at delivery was 25.8 years, 95.3 % of women were nulliparous. Almost all women gave birth at hospital (99.8 %). Mean (± SD) gestational age at delivery was 39.30 weeks (± 1.55 weeks), with a shorter gestational age for severe preeclampsia. The incidences of preterm birth (\<37 weeks of gestation) and low birth weight (\<2,500 g) were significantly higher among babies born to mothers with preeclampsia and severe preeclampsia than normotensive women. Infant complications occurred in 7 % of babies and were more frequent in preeclampsia and severe preeclampsia groups. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic and reproductive characteristics of study population by pregnancy-induced hypertension, Suzhou, China, 1989--1990 ::: Subjects ------------------------------ ---------- ------ ------- ------- ------- ------- ------------------------------ Maternal age ≤30 11,230 94.2 94.6 90.3 91.5 94.3 0.000 \> 30 691 5.8 5.4 9.7 8.5 5.7 Parity Nulliparous 11,365 95.3 95.3 94.9 97.8 97.1 0.100 Multiparous 556 4.7 4.7 5.1 2.2 2.9 Sex Male 6,026 50.5 50.8 48.6 47.4 55.2 0.290 Female 5,895 49.5 49.2 51.4 52.6 44.8 Preterm birth (\<37 weeks) No 11,527 96.7 96.6 98.2 96.7 93.3 0.023 Yes 394 3.3 3.4 1.8 3.3 6.7 Low birth weight (\<2,500 g) No 11,620 97.5 97.6 97.3 95.1 92.4 0.000 Yes 301 2.5 2.4 2.7 4.9 7.6 Mean gestational age (Weeks) 39.30 39.29 39.44 39.30 38.69 0.000 (Analysis of variance) Place of delivery Hospital 11,900 99.8 99.8 99.7 100 100 0.981 Home 11 0.1 0.1 0.1 0 0 BMI \< 24 11,012 92.8 93.1 90.0 89.6 93.3 0.005 24--28 426 3.6 3.4 5.3 5.8 4.8 \> 28 432 3.6 3.5 4.7 4.6 1.9 Infant complications No Yes No 10,989 93.1 93.1 90.9 87.0 88.5 0.000 Yes 810 6.9 6.4 9.1 13.0 11.5 \* Excluding cases with missing information. ::: Table [2](#T2){ref-type="table"} presents the overall effects of PIH on infant growth at 28 and 42 days postpartum. The infant weights were statistically significantly lower in infants born to mothers with both preeclampsia and severe preeclampsia than infants born to normotensive mothers. However, IGPs were not different between groups. For example, at 42 days postpartum, weights were significantly lower for the preeclampsia and severe preeclampsia groups than for normotensive group (mean weight = 4,763 g and 4,679 g vs. 4,869 g, p \< 0.001), but no evidence of catch-up growth in infants born to mothers with preeclampsia and severe preeclampsia (IGP = 47.7% and 48.3% vs. 49.7%, p = 0.618). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Pregnancy-induced hypertension and infant weight at 28 and 42 days postpartum, univariate analysis ::: Infant growth 28 days postpartum Infant growth 42 days postpartum ------------------------- ------------------- ------- ---------------------------------- ---------------------------------- ------ ------------------- ------- ------ Normotensive (N=10,669) 3252.4 (Referent) 416,1 4088.8 (Referent) 496.2 25.7 4869.1 (Referent) 609.9 49.7 GH\# (N=782) 3288.8 448.9 4084.1 540.4 24.2 4878.6 696.2 48.3 PE¶ (N=365) 3224.9 465.8 4018.6\* 513.7 24.6 4763.8\*\* 631.8 47.7 Severe PE (N=105) 3170.9 502.9 3941.9\* 564.9 24.3 4679.9\*\* 727.3 47.6 Total (N=11,921) 3253.2 420.9 4085.0 500.7 25.6 4864.8 618.2 49.5 \* p \< 0.05, \*\* p \< 0.01 ^1^\[(Infant weight 28 days - birth weight) / birth weight\] × 100 %, ^2^\[(Infant weight 42 days - birth weight) / birth weight\] × 100 % † SD: standard deviation, \#GH: gestational hypertension, ¶PE: preeclampsia. ::: Table [3](#T3){ref-type="table"} presents the effect of PIH on growth at 28 and 42 days postpartum, stratified for the presence or absence of intrauterine growth restriction (IUGR). Infant weights were not different between infants born to mothers with any types of PIH *without IUGR*and normotensive babies without IUGR, and IGPs were not different between any types of PIH and non-IUGR normotensive control. Conversely, infant weights were markedly lower in all infants *with IUGR*than normotensive controls without IUGR. Infant weight gains and IGPs were higher in any types of PIH and normotensive control *with IUGR*than in their counterpart groups *without IUGR*. For example, at 42 days postpartum, *in the absence of IUGR*, there were no differences in infant weight between any of the types of PIH (ranging from 4,801 g to 4,929 g) and non-IUGR normotensive group (4,905 g). The IGPs were not different between any types of PIH (ranging from 46.2 % to 47.0 %) and non-IUGR normotensive control (48.6%). However, *in the presence of IUGR*, infant weights were markedly lower in the three types of PIH (ranging from 3,882 g to 4,231 g) and in normotensive (4,255 g) groups than in non-IUGR normotensive control group (4,905 g), p \< 0.001. The IGPs were significantly higher (ranging from 59.8% to 71.3%) as compared to the reference group (48.8%), suggesting a significant catch-up growth. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Pregnancy-induced hypertension and infant weight at 28 and 42 days postpartum by newborn IUGR status, univariate analysis ::: Infant growth 28 days postpartum Infant growth 42 days postpartum ------------------------------------- --------------------------- ---------------------------------- ---------------------------------- ----------------- --------------------------- -------- ----------------- Normotensive / non-IUGR‡ (N=10,067) 3296.6 (382.0) (Referent) 4126.9 (471.3) (Referent) 830.3 25.2 (Referent) 4905.8 (592.9) (Referent) 1609.2 48.8 (Referent) Normotensive / IUGR (N=602) 2513.2 (219.9)\*\* 3451.1 (466.9) \*\* 937.9 37.3\*\* 4255.8 (563.3)\*\* 1742.6 69.3\*\* GH\# / non-IUGR (N=725) 3353.1 (395.4)\*\* 4143.6 (500.4) 790.5 23.6 4929.4 (665.8) 1576.3 47.0 GH / IUGR (N=57) 2469.8 (230.9)\*\* 3327.4 (454.8)\*\* 857.5 34.7\*\* 4231.6 (754.3)\*\* 1761.8 71.3\*\* PE¶/ non-IUGR (N=326) 3314.9 (401.8) 4091.3 (463.8) 776.4 23.4 4853.1 (572.9) 1537.2 46.4 PE / IUGR (N=39) 2472.1 (215.6)\*\* 3410.4 (512.9)\*\* 938.3 37.9\*\* 4025.4 (624.2)\*\* 1553.3 62.8\*\* Severe PE / non-IUGR (N=91) 3285.2 (434.8) 4037.4 (518.9) 752.2 22.9 4801.8 (682.8) 1516.6 46.2 Severe PE / IUGR (N=14) 2428.6 (171.8) \*\* 3321.4 (459.0) \*\* 892.8 36.8\*\* 3882.1 (463.5)\*\* 1453.6 59.8 Total (N=11,921) 3253.2 (420.9) 4085.0 (500.7) 831.8 25.6 4864.8 (618.2) 1611.6 49.5 \* p \< 0.05, \*\* p \< 0.01 ^1^\[(Infant weight 28 days - birth weight) / birth weight\] × 100 %, ^2^\[(Infant weight 42 days - birth weight) / birth weight\] × 100 % \#GH: gestational hypertension, ¶PE: preeclampsia, ‡ IUGR: intrauterine growth restriction. † SD: standard deviation ::: Table [4](#T4){ref-type="table"} presents mean birth weight differences, infant weight differences at 28 days and 42 days postpartum between various types of PIH with and without IUGR and the normotensive control group without IUGR, as well as the corresponding regression coefficients (i.e., the weight differences after adjustment for confounders). The results of multivariate linear regression were consistent with those of univariate analysis (Table [3](#T3){ref-type="table"}). After adjustment for confounding variables, IUGR babies born to any of the PIH groups had significantly lower infant weights at 28 and 42 days postpartum. For example, there was a 903 g (1,023 g before the adjustment for confounders) reduction in infant weight at 42 days postpartum in IUGR babies born to severe preeclamptic mothers relative to non-IUGR babies born to mothers with normal blood pressure, although the IGP was higher in the IUGR babies. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Pregnancy-induced hypertension and infant weight at 28 and 42 days postpartum, multivariate analysis ::: Mean birth Infant weight 28 days postpartum Infant weight 42 days postpartum ------------------------------------- ------------ ---------------------------------- ---------------------------------- ------------------- ------------ ------------------- Normotensive / non-IUGR‡ (N=10,067) Ref. Ref. Ref. Ref. Ref. Ref. Normotensive / IUGR (N=602) 783.4\*\* 776.2 (15.4)\*\* 675.8\*\* 640.0 (19.4)\*\* 650.0\*\* 619.4 (25.2)\*\* GH\# / non-IUGR (N=725) -56.6\*\* -34.0 (14.0) -16.7 -4.4 (18.6) -23.7 -12.8 (22.8) GH / IUGR (N=57) 826.7\*\* 825.4 (49.0)\*\* 799.5\*\* 815.1 (61.8)\*\* 674.2\*\* 678.9 (80.1)\*\* PE¶/ non-IUGR (N=326) -18. 4 -6.97 (20.5) 35.6 38.0 (25.8) 53.6 55.4 (33.5) PE / IUGR (N=39) 824.5\*\* 795.9 (60.3)\*\* 716.5\*\* 654.1 (76.0)\*\* 880.4\*\* 835.2 (98.7)\*\* Severe PE / non-IUGR (N=91) 11.4 -20.0 (38.8) 89.5 39.4 (48.9) 104.0 69.9 (63.5) Severe PE / IUGR (N=14) 868.0\*\* 796.8 (92.4)\*\* 805.5 674.7 (116.5)\*\* 1023.6\*\* 903.0 (151.2)\*\* \* p \< 0.05; \*\* p \< 0.01 §β, coefficient, SE, standard error; adjusted for maternal age, body mass index, diabetes, infant\'s sex, maternal anemia, gestational age \#GH: gestational hypertension, ¶PE: preeclampsia, ‡ IUGR: intrauterine growth restriction ::: Discussion ========== To our knowledge, there is no previous study that has examined the effect of PIH on infant growth in the early infant period. Our study found that when babies were born with IUGR, all PIH groups had lower infant weight and significant \'catch-up\' growth at 28 and 42 days postpartum. However, if babies were not born with IUGR, any type of PIH did not have detrimental effect on infant growth at 28 and 42 days postpartum. Our study suggests that PIH per se does not affect early infant growth. The finding of different effects of PIH on early infant growth according to the presence of IUGR further supports the hypothesis that preeclampsia is a heterogeneous disorder and present at least two subtypes: IUGR vs. normal fetal growth preeclampsia \[[@B4]-[@B6]\]. These two subtypes of preeclampsia may have different pathogeneses, clinical manifestations and short and long term effects on both infants and mothers \[[@B4]-[@B7]\]. Patients with preeclampsia who have IUGR babies may have placental dysfunction such as reduced utero-placental perfusion as the common pathogeneses. The present study shows this group of infants has sustained lower weight in the early infant period. In contrast, patients with preeclampsia who have babies with normal fetal growth may not have the significant placental dysfunction, thus have normal intrauterine as well as early infant growth. Newborns with poor intrauterine growth often demonstrate postnatal catch-up growth \[[@B18],[@B19]\]. Our study suggests such a catch-up growth is initiated soon after birth. The impact of catch-up growth on infant\'s long-term health remains unclear. There is a dilemma as to whether this catch-up growth, either inherent to IUGR babies or through growth promotion programs as promoted in many developing countries, is good or bad. In short term, such a catch-up growth enables infants to accelerate growth in order to reach their normal growth curve, resulting in reduced infant morbidity and mortality \[[@B19]\]. However, it has been suggested that catch-up growth could have a detrimental influence, leading to the onset of certain chronic diseases in later life \[[@B20]-[@B23]\]. The crucial time for the development of long-term consequences is the early postnatal period when catch-up growth occurs in around 80 % of IUGR babies \[[@B18],[@B22]\]. Some studies \[[@B8],[@B12],[@B20],[@B22]\] found that hypertension, cardiovascular diseases and mellitus diabetes in adult life were correlated with the existence of a low birth weight and a catch-up growth in the early postnatal period. The hypothesis of the fetal origins of adult disease \[[@B10],[@B24],[@B25]\] may be redefined to consider the impact of the early infant period on the risk of adult diseases \[[@B22],[@B23]\]. Our study indicates that for the infants born to mothers with various types of PIH, only those babies with IUGR showed a catch-up growth. Future long-term follow up studies are needed to examine if IUGR babies born to PIH mothers are at higher risk to develop certain adult diseases later in life than those non-IUGR babies born to PIH mothers. Several limitations of this study have to be addressed. First, our database does not have information on important factors such as maternal smoking and breastfeeding. The prevalence of smoking is high for men in China, however, reported smoking is rare among women (\<0.5%) \[[@B26]\]. Therefore, our results are not likely to be biased by not controlling for smoking. To our knowledge, almost all women breast-fed their infants at that time. However, it is possible that the presence of PIH or IUGR affected breastfeeding and thereby early infant growth. Second, the database has many cases with missing information on weight at 28 days and/or 42 days postpartum. We compared these missing cases with those pregnant women remained for analysis. The rate of preterm birth and low birth weight were higher in missing cases than in the remained study population. This is partly due to women who had perinatal deaths up to 28 days postpartum (most of those babies were preterm, IUGR, or low birth weight) and thus had no information on infant weight at 28 and 42 days postpartum were classified into missing cases. This is the reason why the study population has fewer preterm birth, low birth weight and intrauterine restriction infants. However, there were no significant differences in demographic and reproductive characteristics such as pregnancy-induced hypertension, maternal age, parity, number of prenatal visits, infant\'s sex and gestational age. Third, weight can be a good growth indicator but, ideally, should be correlated to height and head circumference \[[@B27]\], in order to differentiate between symmetrical and non-symmetrical IUGR. We only considered weight as a measure of growth pattern because it was the only available variable. Conclusion ========== The effect of PIH on early infant growth is dependent on whether or not a baby is intrauterine growth restricted. IUGR secondary to PIH is associated with significant catch-up growth in early infant period. Therefore, we speculate that IUGR babies born to mothers with PIH may be at higher risk to develop certain chronic diseases in their later life. More studies in North American or European populations are needed to confirm these results. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Each author has contributed to the concept, the design and the analysis on the study and to the writing of the draft. All authors read and approved the final version. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2393/5/10/prepub> Acknowledgements ================ We thank Dr. Zirong Guo, Department of Epidemiology of Suzhou University Medical College (China) for data preparation.

PubMed Central

2024-06-05T03:55:58.256392

2005-5-20

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168902/", "journal": "BMC Pregnancy Childbirth. 2005 May 20; 5:10", "authors": [ { "first": "Emmanuelle", "last": "Baulon" }, { "first": "William D", "last": "Fraser" }, { "first": "Bruno", "last": "Piedboeuf" }, { "first": "Pierre", "last": "Buekens" }, { "first": "Xu", "last": "Xiong" } ] }

PMC1168903

Background ========== Asthma continues to be a key challenge in the field of pediatric health. With nine million children diagnosed with asthma in the United States, it is the leading cause of school absences, emergency room visits and hospitalizations among American youth \[[@B1],[@B2]\]. Studies of asthma prevalence however, have tended to neglect the Asian-American population, leaving the burden of asthma faced by this population largely unknown \[[@B3],[@B4]\]. Epidemiological studies of asthma have tended to utilize written questionnaires in English or Spanish, focusing on symptoms \[[@B1],[@B5]\]. In the few studies that have included Asian-American children, lower prevalence of diagnosed asthma than in comparable populations has usually been found \[[@B1],[@B3],[@B4]\]. When included as part of \"other races\" in the Center for Disease Control report, Asian-American children, were found to have an asthma prevalence of 11.4% \[[@B1]\]. Such results come into conflict with results obtained using other, non-written methods. For example a study published in 2002 found 47.1% of Asian-American third-graders in New Jersey to have abnormal spirometry \[[@B6]\]. Recent studies suggest that lack of general asthma knowledge as well as a misunderstanding of the word \"wheeze\" may affect detection of asthma in Asian children \[[@B3],[@B7]\]. A study of parents of asthmatic children in Nanjing, China found 54.7% of parents had \"poor\" knowledge of the disease. Parents were unfamiliar with the triggers of asthmatic attacks, the pathology of wheezing and proper management of the disease in general \[[@B7]\]. In a large international study comparing written questionnaire responses and reaction to a video demonstrating wheezing \[[@B5]\], Cantonese speakers showed less correlation between the written and video questionnaire then did English or Spanish speaking subjects, with the inclusion of the video increasing the number of positive responses for Cantonese speakers. The \"gold standard\" in asthma diagnosis consists of history, physical examination and spirometry. As the history portion relies heavily on questions utilizing the word \"wheeze\" and other symptoms of asthma, language could interfere with proper diagnosis. The word \"wheeze\" has multiple Cantonese translations, each conveying a slightly different meaning. To our knowledge, previous studies of asthma among Chinese speakers have not reported which translation(s) they used. The translations that we have encountered are: 1) ![](1471-2458-5-48-i1.gif) which means a sound from the throat, 2) ![](1471-2458-5-48-i2.gif)(![](1471-2458-5-48-i3.gif)) which means a sound from difficulty breathing, 3) ![](1471-2458-5-48-i4.gif) which means a special sound from asthma and is the more professional expression which is used in medical books, and 4) ![](1471-2458-5-48-i5.gif) which literally means a gasping sound made after crying. A five-question instrument, the Brief Pediatric Asthma Screen (BPAS) has been validated as an additional means of screening for asthma. The BPAS also relies heavily on the use of the word \"wheeze,\" thereby being subject to the same limitation \[[@B8]\]. Two previous studies have used Chinese translations of the BPAS in attempts to determine the prevalence of asthma in Asian-American immigrant children, yielding conflicting results \[[@B3],[@B4]\]. Both studies offered the survey in English and Chinese. In the first study \[[@B4]\], conducted in 2002, parents of kindergarten through fifth grade students completed a self-administered written version of the BPAS. Using, ![](1471-2458-5-48-i4.gif), the professional term for wheezing, this study found the prevalence of diagnosed and possibly undiagnosed asthma to be 16% and 3%, respectively. In the second study, the survey was administered orally, with surveyors physically demonstrating what they believed to be wheezing \[[@B3]\]. It should be noted that the demonstration more resembled labored breathing and therefore was not specific for asthma. As such, the prevalence of diagnosed asthma in the second study remained statistically the same at 15%, however the prevalence of possible undiagnosed asthma increased over six-fold to 18.6%. This study seeks to improve the methodology for asthma screening in Chinese-American immigrant children. By use of the BPAS, a video demonstration of wheezing, and spirometry; the study compares the use of current methods to detect asthma in this population. As there were approximately one-half million hospitalizations for asthma in the US in 1995, effective detection of asthma is critical in patient care \[[@B9]\]. Methods ======= Sample ------ A convenience sample of 152 children ages 5--18, was recruited in the waiting room of the pediatrics department at South Cove Community Health Center in Boston, Massachusetts between June 16-July 23, 2004. Children with cough or fever, who were outside the specified age range, who did not speak either English or Cantonese or who had previously taken the screen were excluded from the study. Procedures ---------- The Tufts-New England Medical Center Institutional Review Board as well as the South Cove Community Health Center Board of Directors approved the study protocol. Consent was given orally at time of entry into the study from parents/grandparents/guardians of children less than 11 years of age and the child him/herself for years thereafter. Collected data was anonymous and de-identified. Families were provided with a written description of the study in their choice of English or Cantonese. The protocol was expanded mid-way through the study to include additional data collection. Data from both the \"basic\" and \"expanded\" protocol were analyzed and are presented in this paper. Questionnaire ------------- A written questionnaire was administered to all participants in the study as our primary tool for assessing asthma status. It was administered to the parent/grandparent/guardian if the child was younger than 11 years of age. Children ages 11--18, were asked to complete the questionnaire him/herself. The survey was written in English and translated into Chinese by one translator using traditional characters. A second translator then translated it back into English in order to check for accuracy. Respondents had the option of taking the survey in their choice of English or Cantonese. A bilingual English/Cantonese speaker (author ACL) orally administered the questionnaire and answered questions when prompted by the participant. The questionnaire consisted of 19 questions. The majority of questions were demographic in nature. Questions were asked regarding the child\'s age, sex, place of birth, and length of residence in the United States. Preferred language was inferred from choice of Chinese or English surveys. For children under 11, this meant that it was the parent\'s preferred language, whereas for children over 11, it was the child\'s preferred language. Information was collected on risk factors for asthma, including family smoking habits, highest level of education for parents (as a measure of socioeconomic status), allergies, asthma medication use and family history of asthma. The expanded questionnaire included questions relating to other medical conditions and non-asthma medication use. The remaining five questions in both the basic and expanded questionnaire were from the Brief Pediatric Asthma Screen (BPAS) \[[@B8]\] and were as follows: 1\. Have you/your child ever been diagnosed *by a doctor*as having asthma? 2\. Have you/your child ever had episodes of wheezing (whistling in the chest) in the *last 12 months*? 3\. In the *last 12 months*, have you had/heard your child wheeze or cough during or after active play? 4\. Other than a cold, in the *last 12 months*, have you/your child had a dry cough at night? 5\. In the *last 12 months*, have you/your child been to a doctor, an emergency room or a hospital for wheezing? This study used the translation ![](1471-2458-5-48-i5.gif) for wheeze. Though this version literally means a gasping sound made after crying, it was chosen by a bilingual speaker with extensive experience in the Boston Chinatown community to represent wheezing when used in the context of asthma. Categorization of asthmatic status ---------------------------------- All completed questionnaires were categorized for asthmatic status as described in Wolf et al \[[@B8]\]. An affirmative answer to the first question was automatically categorized as \"diagnosed asthma\". An affirmative answer to the last question or two or more of questions 2--4, was categorized as \"possible undiagnosed asthma.\" All other responses were considered \"probably not asthmatic.\" For analysis of association between asthma and demographic factors, persons categorized as \"possible undiagnosed asthma\" were grouped with \"probably not asthmatic\" to create the non-asthmatic group. The BPAS was administered prior to the respondent\'s viewing of the video and thus, children were categorized based on the respondent\'s prior understanding of the word \"wheeze.\" Video ----- We asked all parents/grandparents/guardians or children to watch the \"international\" version (AVQ 3.0) of the International Studies of Asthma and Allergies in Childhood (ISAAC) Phase One study \[[@B5]\]. Our goal was to provide participants with a visual representation of wheezing on the assumption that they might not understand the concept of wheezing. This brief, one minute, video consists of the following five sequences: 1\. \"A young person seated with clearly audible wheezing, but without breathlessness and no evidence of airway obstruction 2\. Exercise-induced wheezing 3\. Waking at night with wheezing 4\. Nocturnal coughing 5\. Severe attack of asthma\" Respondents were asked if their previous understanding of wheezing on the BPAS corresponded to what they had viewed in the video. An answer of \"yes\" or \"no\" was recorded, as were any qualitative responses in regards to the concept of wheezing. Patients screened with the expanded protocol were then asked to re-answer questions 2--5 of the BPAS, basing their answers on the depiction of wheezing in the video. These patients were than categorized again based on their second set of answers to the BPAS. Detailed analyses of the second BPAS scoring are not presented. Cantonese speaking participants taking the expanded protocol were then asked which of the four Cantonese translations best matched the video\'s depiction of wheezing. Spirometry ---------- Pulmonary function testing was performed on all participants willing to attempt it using the ndd Model 2000 *EasyOne*™ Frontline Spirometer (ndd Medical Technologies, Andover, MA), an instrument whose performance correlates well with office-based spirometry \[[@B10]\]. We sought to use spirometry as an objective measure of asthma status. Standard, calibrated scales and stadiometers were used to determine height and weight. Each child was then categorized as underweight (\<5^th^percentile), healthy weight (5--85^th^percentile), at risk for overweight (85^th^-95^th^percentile) or overweight (\>95^th^percentile) using the CDC BMI growth charts for boys and girls ages 2--20. Categorization was performed twice, to ensure accuracy. Spirometry was performed using standard procedures by trained technicians. The children\'s noses were sealed manually or by use of pediatric size spirometry nose clips. For analysis, we used only spirometry reported by the *EasyOne*™ software as having at least 2 acceptable maneuvers and FEV~1~readings within 150 ml and FVC within 150 ml (score of \"A\" or \"B\"; *EasyOne*™, manual, undated). The parameters recorded were: percent predicted forced expiratory volume during the first second (FEV1), forced vital capacity (FVC), and FEV1 to FVC ratio (FEV1/FVC). Percent predicted values are based upon the results of the NHANES III study as described in Hankinson *et. al*. without adjusting for ethnicity \[[@B11]\]. Criteria for diagnosis of obstructive lung disease were: FEV~1~\< 80% predicted and FVC \< 80% predicted FEV~1~/FVC \< 75% Numerical values correspond to those recommended by the National Asthma Education and Prevention Program of the National Institutes of Health: National Heart, Lung and Blood Institute \[[@B12]\]. Results of the spirometry were given to the patient and his/her physician so that they could be incorporated into his/her pediatric appointment. Data management & analysis -------------------------- Data was double entered into SPSS version 11.5 and crosschecked for errors by reference to the original hard copies of the surveys and data forms. Chi Square tests were used to generate odds ratios (OR), to demonstrate the likelihood of being diagnosed as asthmatic versus being non-asthmatic for the following variables: birthplace, preferred language of respondent, paternal education, maternal education, smoking in the home, overweight status, allergies and family history of asthma. Due to small numbers of values below 80%, Fisher\'s exact test was used to examine the relationship of being diagnosed with asthma versus being non-asthmatic with FEV~1~and FVC values. We then performed a forward step-wise binomial logistic regression to create a model predicting the likelihood of being diagnosed with asthma based on the following independent variables: preferred language of respondent, foreign born, paternal and maternal education, sex, age, country of origin, smoking in the home, allergy, family history of asthma and overweight as possible predictors of diagnosis of asthma. We again used Chi Square tests to elicit odds ratios for having an understanding of wheezing consistent with the video\'s representation versus inconsistent for the following independent variables: Birthplace, preferred language of respondent, asthmatic status, paternal and maternal education. We performed a forward step-wise binary regression to create a model predictive of consistent or inconsistent understanding of the word wheeze with that portrayed in the video using the following variables: preferred language of respondent, sex, age, overweight, country of origin, foreign born, smoking in the home, allergy, family history of asthma and diagnosis of asthma. For those participants of the expanded protocol we used Mc Nemar\'s test to examine pre-video BPAS scores with post-video BPAS scores. Statistical significance was set at the p = 0.05 level and borderline significance was set at p ≤ 0.10 for all tests. Results ======= Participant demographics ------------------------ Table [1](#T1){ref-type="table"} presents the demographic characteristics of the full study population (n = 152) as well as three subsets, those children completing the basic protocol (n = 63), those completing the expanded protocol (n = 89) and those with acceptable spirometry results (n = 67). Based on BPAS scores, diagnosed asthma prevalence for the full sample was 27.0% with another 5.3% having possible undiagnosed asthma. Children completing the basic and expanded protocols were similar in most ways, however the expanded protocol subset had statistically higher prevalence of family history of asthma (23.9% vs. 11.1%) and lower prevalence of paternal (50.0% vs. 77.1%) and maternal (52.1% vs. 74.5%) education at high school level or above. The subset of children that completed acceptable spirometry did not differ from the total study population demographically. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of the study population and subsets of the population used in the analyses in this study. ::: **Basic Protocol (n = 63)** **Expanded Protocol (n = 89)** **Total (n = 152)** **Spirometry (n= 67)** --------------------------------- ----------------------------- -------------------------------- --------------------- ------------------------ **Age** Less than 11 39.7% (25) 37.1% (33) 38.2% (58) 40.3% (27) 11 and older 60.3% (38) 62.9% (56) 61.8% (94) 59.7% (40) **Sex** Female 50.8% (32) 50.6% (45) 50.7% (77) 58.2% (39) Male 49.2% (31) 49.4% (44) 49.3% (75) 41.8% (28) **Birthplace** Native-born 57.1% (36) 50.6% (45) 53.3% (81) 59.7% (40) Foreign-born 42.9% (27) 49.4% (44) 46.7% (71) 40.3% (27) **Preferred Language** Cantonese 61.9% (39) 53.9% (48) 57.2% (87) 55.2% (37) English 38.1% (24) 46.1% (41) 42.8% (65) 44.8% (30) **Smoking** Smoker in the home 38.1% (24) 35.2% (31) 36.4% (55) 33.3% (22) Smoke-free home 61.9% (39) 64.8% (57) 63.6% (96) 66.7% (44) **Allergies** Yes 37.1% (23) 37.5% (33) 37.3% (56) 37.9% (25) No 61.9% (39) 62.5% (55) 62.7% (94) 62.1% (41) **Family Asthma** Yes 11.1% (7)\* 23.9% (21)\* 18.5% (28) 19.4% (13) No 88.9% (56) 76.1% (67) 81.5% (123) 80.6% (54) **Paternal Education** Did not complete high school. 22.9% (11)\* 50.0% (34)\* 38.8% (45) 39.0% (23) Completed high school or above 77.1% (37) 50.0% (34) 61.2% (71) 61.0% (36) **Maternal Education** Did not complete high school 25.5% (13)\* 47.8% (33)\* 38.3% (46) 41.4% (24) Completed high school or above 74.5% (38) 52.1% (36) 61.7% (74) 58.6% (34) **BPAS Score** Diagnosed asthma 22.2% (14) 30.3% (27) 27.0% (41) 26.9% (18) Possible undiagnosed asthma 7.9% (5) 3.4% (3) 5.3% (8) 3.0% (2) Probably not asthmatic 69.8% (44) 66.3% (59) 67.8% (103) 70.1% (47) \* p \<.05 ::: Descriptive analysis -------------------- Table [2](#T2){ref-type="table"} lists frequencies, ORs and p-values for associations between key characteristics of the study population along with prevalence of diagnosis of asthma. Being foreign born as compared to native born was associated statistically with a lower prevalence of diagnosed asthma (11.3% vs. 40.7%, p = 0.001). Reporting a family history of asthma was associated with a higher prevalence of asthma (53.6% vs. 20.3%, p = 0.001). Borderline statistical associations with diagnosis of asthma were found for lower maternal education, which was associated with lower asthma prevalence (17.4% vs. 31.1%, p = 0.09) and the presence of allergies in the child, which was associated with higher prevalence of asthma (35.7% vs. 21.3%, p = 0.05). There was no statistical association for preferred language of respondent (a variable that combines the parent/grandparent/guardian for children less than 11 and the child him/herself for those over 11 years of age), paternal education, smoking in the home, body mass index, or both FEV~1~and FVC below 80% of predicted. We could not test associations between asthma diagnosis and abnormal FEV1/FVC ratio because there were only 3 FEV1/FVC ratios below 75%. Associations between FEV1 and sex, foreign born, preferred language of respondent, smoking in the home, allergies, family history of asthma, BPAS score and paternal and maternal education were not statistically significant (not shown). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of prevalence of diagnosed asthmatics and non-asthmatics for the total sample (n = 152) by demographic categories. ::: **Diagnosed Asthmatics (n = 41)** **Non-asthmatics (n = 111)** **OR & 95% CI** **P value** --------------------------------- ----------------------------------- ------------------------------ ---------------------- ------------- **Birthplace** Native-born 40.7% (33) 59.3% (48) 5.41 (2.29 -- 12.80) **\<0.001** Foreign-born 11.3% (8) 88.7% (63) **Preferred Language** Cantonese 24.1% (21) 75.9% (66) 0.72 (0.35 -- 1.47) 0.36 English 30.8% (20) 69.2% (45) **Paternal Education** Completed high school or above 25.4% (18) 74.6% (53) 0.75 (0.33 -- 1.72) 0.66 Did not complete high school. 31.1% (14) 68.9% (31) **Maternal Education** Completed high school or above 31.1% (23) 68.9% (51) 2.14 (0.86 -- 5.31) 0.09 Did not complete high school 17.4% (8) 82.6% (38) **Smoking** Smoker in the home 27.3% (15) 72.7% (40) 1.01 (0.48 -- 2.13) 0.98 Smoke-free home 27.1% (26) 72.9% (70) **Body Mass Index (BMI)** ≥ 85 percentile 34.1% (15) 65.9% (29) 1.61 (0.75 -- 3.51) 0.22 \<85 percentile 24.2% (24) 75.8% (75) **Allergies** Allergic 35.7% (20) 64.3% (36) 2.06 (0.98 -- 4.29) 0.05 Non-allergic 21.3% (20) 78.7% (74) **Family Asthma** Asthma 53.6% (15) 46.4% (13) 4.52 (1.91 -- 10.72) **\<0.001** No asthma 20.3% (25) 79.7% (98) **FEV~1~and FVC** \>80% 25.8% (16) 74.2% (46) 0.89 (0.27 -- 2.95) 1.00\*\* ≤80% 40.0% (2) 60.0% (3) \* Odds ratios were computed using Chi-Square tests and are reflective of the likelihood of being diagnosed with asthma for each of the stated demographic variables. \*\*Fisher\'s exact test. ::: Multivariate analysis --------------------- We considered preferred language of respondent, foreign born, paternal and maternal education, sex, age, country of origin, smoking in the home, allergy, family history of asthma and overweight as possible predictors of diagnosis of asthma in a forward stepwise binary logistic regression. Only being native born and having a family history of asthma were statistically significant in the model (p = 0.002 and p = 0.003 respectively), both being associated with higher prevalence of asthma. Because of substantial missing data on parental and maternal education, we also ran the regression leaving these variables out of the analysis. Doing so did not qualitatively change the result (not shown). Video screening --------------- Participants were asked to indicate whether or not the video portrayal of wheezing was different from their prior understanding (Table [3](#T3){ref-type="table"}). About thirty-six percent (35.6%) indicated that the video was different from their conception of wheeze. Cantonese speakers were more likely to report an inconsistency than were English speakers (47.6% vs. 20.0%, p = 0.001). In a forward stepwise binary logistic regression that included preferred language of respondent, sex, age, overweight, country of origin, foreign born, smoking in the home, allergy, family history of asthma and diagnosis of asthma, Cantonese speakers and overweight children were more likely to say that the video portrayal of wheezing was different from their prior understanding (p \< 0.001 and p = 0.045 respectively). For those completing the expanded protocol, post-video BPAS scores were compared to pre-video BPAS scores using Mc Nemar\'s test, the results were not statistically significant. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Consistency of understanding wheezing between the video and questionnaire for the total sample (n = 149; 3 missing). ::: **Consistent** **Inconsistent** **OR & 95% CI\*** **P value** --------------------------------- ---------------- ------------------ --------------------- ------------- **Total** 64.4% (96) 35.6% (53) **Birthplace** Native-born 68.4% (54) 31.6% (25) 1.44 (0.73 -- 2.82) 0.288 Foreign-born 60.0% (42) 40.0% (28) **Preferred Language** Cantonese 52.4% (44) 47.6% (40) 0.28 (0.13 -- 0.58) **\<0.001** English 80.0% (52) 20.0% (13) **Asthma Diagnosis** Asthmatic 60.0% (24) 40.0% (16) 0.77 (0.37 -- 1.63) 0.494 Non-asthmatic 66.1% (72) 33.9% (37) **Paternal Education** Did not complete high school. 58.1% (25) 41.9% (18) 1.33 (0.61 -- 2.88) 0.756 Completed high school or above 64.8% (46) 35.2% (25) **Maternal Education** Did not complete high school 54.5% (24) 45.5% (20) 1.54 (0.72 -- 3.30) 0.130 Completed high school or above 64.9% (48) 35.1% (26) \*Odds ratios were computed using Chi-Square tests and are reflective of the likelihood of having an understanding of the word wheeze consistent or inconsistent with the video\'s depiction for each of the stated variables. ::: Translation of \"wheeze\" ------------------------- The forty-seven (47) Cantonese-speaking respondents completing the expanded protocol rated four choices of Cantonese translations of the English word \"wheeze\" (Figure [1](#F1){ref-type="fig"}). \"Sound from the throat\" was preferred by 46.8%, \"sound from difficulty breathing\" was preferred by 25.5%, the \"professional term for asthma\" was preferred by 10.6% and \"gasping sound made after crying\" was preferred by 17% of respondents. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Preferred translation of \"wheeze\" for respondents in the expanded protocol (n = 47). ::: ![](1471-2458-5-48-1) ::: Discussion ========== Limitations ----------- There are several limitations to our study. We did not ask about smoking by the children themselves, but limited tests of exhaled CO (not shown) suggested that at least a couple smoked. Our measure for socio-economic status (SES) was imperfect, as we had a high non-response rate for the questions of paternal and maternal education level. In most instances this was due to inability of children to recall their parents\' education level, but interpretation of educational level may also be limited by the fact that educational systems vary between the US and China. Our classification of preferred language by the parent\'s choice rather than the child\'s for children under 11 did not necessarily reflect the preferred language of the child. Our choice of Cantonese wording for \"wheeze\" turned out to be one of the less popular options, despite our having introduced it during the BPAS screenings. It is possible that if we had used the most popular choice that it would have improved our BPAS screening. In addition, qualitative reaction of respondents was that the asthma events depicted in the video were more extreme than what the respondents thought of as wheeze. We would suggest that the video might be best described as depictions of \"asthma attacks\" rather than solely wheezing. Inclusion of spirometry was not an effective tool for assessing asthma status. Possibly the addition of a bronchodilator would have improved its utility. Our results are not necessarily generalizable because we had a convenience sample from a health clinic waiting room and screened only families that spoke English or Cantonese. Interpretation -------------- The population of Chinese-American children that we enrolled in the study reported high prevalence of diagnosed asthma, but appeared to be clinically stable at the time of the screening. By screening out children who were coughing on the day of the screening, we may have underestimated overall asthma prevalence and prevalence of active asthma at screening. Few of their spirometry results were below 75% for FEV~1~/FVC and few were below 80% for FEV~1~and FVC. It is worth noting that abnormal spirometry is relatively uncommon, even among children diagnosed with asthma \[[@B13]\]. Only 41.5% of the children reported to have diagnosed with asthma would have been categorized as asthmatic by the other four questions in the BPAS screening. We cannot distinguish between the possibilities that there could be misdiagnosis of asthma as compared to prevalent mild or well-controlled asthma. For those children with diagnosed asthma, who would have been picked up by the BPAS, only 7.3% reported use of asthma medication on the day of their screening. Because we did not distinguish intermittent from persistent asthma and we asked only about medication use on the day of screening, we cannot state with certainty what percentage are not receiving optimal care. We found that asthmatic children were more likely to have allergies and more likely to have a family history of asthma. Both associations are well supported in the literature \[[@B14]\] suggesting that our study population was similar in this way to other populations. Our strongest finding was that being born in the US was highly predictive of having diagnosed asthma, which suggests that with the Asian-American immigrant population there may be a more vulnerable subpopulation visible only when data is disaggregated. We were unable to determine whether this is a function of true asthma prevalence or differential diagnosis. The BPAS screen showed only 5.3% \"possible undiagnosed asthma\" cases, suggesting that there is little undiagnosed asthma consistent with the study of Lee et al. \[[@B4]\]. However, difficulty in translating of the word \"wheeze\" into Cantonese and its centrality in the BPAS leaves ample room for uncertainty about the true prevalence of undiagnosed asthma. When screening Cantonese-speaking children for asthma we first would suggest using the translation of wheeze that literally means \"a sound from the throat\". However, we would urge caregivers to also try other translations, as there was not a single clear choice among our study population. Second, and related to the first, we would suggest that the concept of wheezing is not well-understood or popularized among Cantonese-speakers. This, combined with a low level of knowledge about asthma in general, \[[@B7]\] means that the provider needs to be careful when taking a medical history and rely less on the obvious symptom, wheezing. Even for providers treating diagnosed asthmatic children from Cantonese speaking families, we would suggest extensive follow-up to ensure proper understanding and compliance. Future studies -------------- There is a need to conduct asthma studies in the Chinese-American population that are free of the limitations of our data collection. This would include asking about children\'s smoking behavior, use of a better surrogate for socio-economic status (possibly parental occupation, which might be more easily recalled by children than education level), use of the more popular term for wheeze (\"sound from the throat\") in written or oral screening, and use of a video demonstration with less severe depictions of wheezing. Additionally, studies are needed to confirm and explain the dramatically different prevalence of asthma between foreign and US born Chinese-Americans that we found. Potential hypotheses for the difference include: differential diagnosis due to language and cultural differences and differences in environmental exposures. Interestingly, exposure to Hepatitis A, a virus more common in China, has been suggested to be protective against the future development of asthma via interaction with the TIM1 gene \[[@B15]\]. If this turns out to be the case, it would suggest a biological cause could explain at least part of the difference in asthma prevalence. Conclusion ========== We hope that this paper raises interest in and prompts examination of asthma among Chinese-American populations. There is a prevailing assumption that asthma is not a problem in this population. If nothing else, our study should bring the validity of such an assumption under scrutiny and raise awareness of the lack of general asthma knowledge in this community. In future studies examining prevalence among Chinese-Americans it is important to consider that prevalence among US born Chinese-Americans appears to be significantly higher than prevalence among foreign-born Chinese-Americans. Failure to desegregate data along these lines may mask the high prevalence of asthma within the sub-group. For both research and clinical practice purposes, the emphasis of cultural understanding needs to be explored. Because of asthma\'s commonplace stature within American culture at this time, it is easy to take one\'s level of understanding for granted, thereby missing a potentially life-threatening condition. While we raise this concern with respect to the Chinese-American population, it worth considering that it may hold true for other population groups as well. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= ROG participated in the design of the study, helped supervise the fieldwork, conducted data management, conducted the statistical analysis and took the lead in writing the manuscript and assisted in editing it. ACL conducted the field data collection, participated in data management, participated in the analysis and assisted with writing of the manuscript. RT assisted with design of the study, helped supervise the fieldwork, contributed to interpretation of findings and assisted with writing and editing the manuscript. DB provided overall direction to the study, supervised data collection, management and analysis, and participated in writing and editing the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/48/prepub> Acknowledgements ================ The authors wish to thank South Cove Community Health Center, Eugene Welch and the doctors, nurses and staff including: Sherrie Zhang, MD., Vivian Tsuei, MD., Ingrid Henar, MD., MPH, Chia-Mei Lu, Irene Chin, Yanty Leung, Chung He, Qi-Long Fun, Shu Lin, and Wendy Wong. We are indebted to the parents and children who participated in the surveys. We thank Weibo Lu for her expertise in translating our questionnaires and Karen Lee for providing oral translations. We thank Lois Doerr, PNP, of Boston Medical Center for instructing the research team in spirometry coaching techniques. We would also like to thank Cato Hiu for his assistance. William Rand, Ph.D., of Tufts University School of Medicine provided helpful advice on analytical approach. Suzanne Steinbach of Boston Medical Center made useful comments on an early draft of the manuscript. This project was funded by Tufts University College of Citizenship and Public Service and Tufts University School of Medicine, Department of Family Medicine and Community Health.

PubMed Central

2024-06-05T03:55:58.258939

2005-5-17

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168903/", "journal": "BMC Public Health. 2005 May 17; 5:48", "authors": [ { "first": "Robyn O", "last": "Greenfield" }, { "first": "Angela C", "last": "Lee" }, { "first": "Roland", "last": "Tang" }, { "first": "Doug", "last": "Brugge" } ] }

PMC1168904

Background ========== Polychlorinated dibenzo-*p*-dioxins (PCDDs), polychlorinated dibenzofurans (PCDFs), and polychlorinated biphenyls (PCBs) are persistent, lipophilic, halogenated aromatic chemicals that are developmental and male reproductive toxicants in laboratory animals \[[@B1]-[@B3]\]. These persistent chlorinated compounds are worldwide environmental pollutants that have been detected in areas as remote as the Arctic \[[@B4]\]. They are biologically concentrated and stored in human adipose tissue with prolonged half-lives. Current estimates of the half-life for 2,3,7,8-tetrachlorodibenzo-*p*-dioxin (2,3,7,8 TCDD) range from 3 to 10 years depending on age, gender, and serum concentration, with faster elimination in men, younger individuals, and those with higher peak exposures \[[@B5]\]. The general population is primarily exposed to these compounds through ingestion of contaminated food (fish, meat, milk, and their by-products), as well as through water sources, dermal contact with soil and house dust, and inhalation \[[@B6]\]. Human studies on the relationship of in utero and childhood (peri-pubertal) exposure to dioxin-like compounds (furans and PCBs) with growth and sexual development in boys are limited. A cross-sectional study of Belgian teenagers (15.8--19.6-year-olds) living in polluted and non-polluted communities demonstrated earlier male pubertal development (assessed by Tanner pubic hair and genital staging) in association with both living in polluted areas and higher serum PCBs as compared to boys living in clean areas and/or with lower PCB levels \[[@B7]\]. Lower testicular volume was associated with living in a polluted area but not with serum PCBs. Dioxin exposure assessed with a bioassay (CALUX) was not associated with male pubertal development in this study. Several additional epidemiologic studies have been conducted to investigate growth and development in relation to halogenated aromatic compounds. For example, higher prenatal exposure to DDE has been associated with greater male height and weight adjusted for height at puberty and increased weight for height in peri-pubertal girls \[[@B8]\]. In Taiwan, boys exposed *in utero*to PCBs and PCDFs from maternal ingestion of contaminated rice oil had a shorter penile length than unexposed children \[[@B9]\]. Chapaevsk, Russia, is a town of approximately 83,000 residents, located in central Russia (1200 kilometers south-east of Moscow) on the bank of the Chapaevsk river, a tributary to the Volga. The town occupies an area of 187 km^2^, half of which is occupied by industries that are mostly of the military-industrial complex. In 1989, these industries were responsible for the production of the vast majority of the manufactured products from Chapaevsk, and almost half of the city\'s work force was employed there. One of the largest chemical factories in Chapaevsk is the Khimprom Chemical Plant (Middle Volga chemical plant), which before 1949 produced chemical warfare agents (such as lewisite and mustard gas). After 1949, there was a transition to the production of industrial and agricultural chemicals, such as gamma-hexachlorocyclohexane (lindane), and other chlorine-containing products such as liquid chlorine, dichloropropionic acid, methyl chloroform, vinyl chloride, and pentachlorophenol \[[@B10]\]. These processes produced PCDDs and PCDFs as industrial contaminants, which subsequently polluted the air, soil, water and food supply in the city \[[@B10]-[@B12]\]. Revich and coworkers \[[@B10]\] have found elevated levels of dioxin-like compounds in soil less than two kilometers from the Khimprom Chemical Plant (141 ng TEQ/kg). The plant has reduced production of these chemicals since 1991; however, since PCDDs and PCDFs persist in the environment, continued human exposure from contaminated air, soil, drinking water, as well as consumption of locally grown vegetables and locally raised animals remains a concern \[[@B10]-[@B12]\]. Moreover, a large proportion of the population lives in close proximity to the Khimprom complex. Given that children may be particularly susceptible to reproductive and developmental effects of organochlorines, and the paucity of information available regarding childhood exposures to dioxins in particular, we undertook a pilot study to describe the distribution of, and identify potential predictors of exposure to, dioxin-like compounds and dioxins among adolescent boys in Chapaevsk, Russia. The pilot study was also designed to guide the development of a large prospective cohort study on the relationship of exposure to PCDDs, PCDFs, and PCBs with growth and pubertal development in peri-pubertal Chapaevsk boys. Methods ======= Informed Consent ---------------- The study was approved by the Human Studies Institutional Review Boards of the Chapaevsk Medical Association, Harvard School of Public Health, University of Massachusetts Medical School, Brigham and Women\'s Hospital, and Centers for Disease Control and Prevention (CDC). All subjects signed informed consent prior to blood draw and study participation. Study Population ---------------- From 2579 boys, aged 10--16 years in 1999, enrolled in an earlier pilot study to generate growth and maturation curves for boys in Chapaevsk \[[@B13]\], a subset of 246 older boys (14.0 to 16.9 years) were identified for a sub-study in which blood samples and questionnaire information were obtained. Older boys were chosen for study because blood samples were required and participation rates were expected to be higher than among younger children. Of the 246 boys, 221 had blood samples collected, and of these samples, 30 bloods were initially sent to the CDC for chemical analysis of dioxins, furans and PCBs. By design, of the 30 blood samples, 15 were from children with cryptorchidism or hypospadias, and 15 were from children with neither condition (controls). The selection of the 15 cases and 15 controls was done blindly in relation to factors that may predict dioxin levels. Each of the 30 boys, with his mother, was asked to complete a nurse-administered detailed questionnaire on medical history, diet, and lifestyle. The diet questions were used to measure the current and lifetime consumption of locally grown or raised foods. The question was worded, \"Does your child eat any of the following foods from local sources (i.e. your own garden or farms or lakes in the Chapaevsk area)? Yes/No\". There were separate questions for current intake and lifetime intake of each food item. The distances the boys lived from the Khimprom factory at the time of the study and during pregnancy were assessed by questionnaire based on maternal self-report as \<2, 2--6, or \>6 kilometers, and the distance at the time of the study was also estimated using ArcView GIS 3.0 mapping of addresses. Dioxin, furan and PCB chemical analysis --------------------------------------- After blood collection and serum separation, the serum samples were stored at -20°C until shipment on dry ice to the US. The chemical analysis was performed by the laboratory at the National Center for Environmental Health, Centers for Disease Control and Prevention (NCEH, CDC), Atlanta, GA. Target analytes included polychlorinated dibenzo-p-dioxins (PCDDs), polychlorinated dibenzofurans (PCDFs), non-ortho substituted (coplanar) polychlorinated biphenyls (co-PCBs), mono-ortho substituted PCBs and other PCBs (nondioxin-like PCBs). Serum samples were spiked with a mixture of ^13^C~12~- labeled PCDDs/PCDFs and coplanar PCBs as internal standards and the analytes were isolated from serum by a C~18~solid phase extraction (SPE) followed by a multi-column automated cleanup and enrichment procedure \[[@B14]\]. Samples were processed in batches of 10, which included a method blank and two quality control samples that were aliquots from pooled bovine sera spiked with PCDDs, PCDFs and co-PCBs. The analytes were separated on a DB-5 MS capillary column and quantified using selected-ion-monitoring (SIM) high resolution (10,000 resolving power) mass spectrometry (HRGC-ID/HRMS) by the method described by Patterson et al. \[[@B15]\]. Quantification was by isotope dilution mass spectrometry using calibration standards containing ^13^C labeled, and unlabeled analytes. For specific PCDDs, PCDFs, and co-PCBs that lack their own labeled standard, a labeled congener with the same degree of substitution and a similar retention time was used. Mono-ortho and nondioxin-like PCBs were extracted from an aliquot (1 gram) of sample by SPE extraction \[[@B14]\]. Total Lipids were determined for each serum sample using a \"summation\" enzymatic method, which is based on the individual measurements of the four lipid groups (free or non-esterified cholesterol, cholesterol esters, triglycerides and phospholipids) \[[@B16]\]. The limits of detection (LODs) were calculated based on long-term standard deviations that were estimated from multiple measurements of low-level standards \[[@B17]\]. For an analyte with a positive blank, the average blank observed during the study was subtracted from individual calculated results \[[@B18]\]. The lipid adjusted detection limits (DLs) were based on observed LODs and corrected for individual sample weight and recovery. Since values below the DL are expected to have greater variability and uncertainty than values above the DL, some have recommended censoring such data. In the case of our serum analyses, a substantial proportion of our results would be censored at the DL. However, when analyzing group data (as opposed to individual data points), censoring data below the detection limit may lead to the loss of useful information \[[@B19]\]. Where comparisons between uncensored and censored datasets have been performed, the use of DL cut-offs compromised the statistical power of analyses and led to biased exposure estimates \[[@B20]\]. This bias was a greater limitation than any increased variability of low values \[[@B18],[@B21]\]. Thus, in the present study, we retained measurable values below the DL and assigned a value of zero only to samples in which the analyte was not detectable or was at or below the level of contamination in procedural blanks. The concentrations of PCDDs, PCDFs and co-PCBs were reported on a lipid-adjusted basis as picograms of analyte per gram of lipids (ppt). The concentrations of mono-ortho and nondioxin-like PCBs were reported as nanogram of analyte per gram of lipids (ppb). Toxic equivalents (TEQs) were also reported for PCDDs, PCDFs, co-PCBs and other \'dioxin-like\' PCBs based on Toxic Equivalency Factors (TEFs) for individual congeners as specified in the 1998 World Health Organization (WHO) system \[[@B22]\]. Statistical Analysis -------------------- Wilcoxon ranksum tests were used to compare dioxin and PCB levels between controls and boys with cryptorchidism or hypospadias. Generalized linear models were used to identify predictors of the log (base 10) of the sum of dioxin concentrations (PCDDs + PCDFs + co-PCBs) and log sum of dioxin TEQs (PCDDs + PCDFs + Co-PCBs) for the 30 boys. Predictors of dioxin concentrations were first modeled univariately and then adjusted for age. Univariate predictors with p-values below 0.15 were considered for inclusion in the multivariate models. Multivariate model selection was based on adjusted R-square values, Mallows C(p) criterion, and forward and backwards selection procedures; final models used a significance level of 0.05 for inclusion of predictors. Two final models were selected: one which considered all potential covariates, and a second model which considered only those covariates collected on all pilot study participants (i.e., excluding PCB measurements), in order to predict dioxin levels in the larger pilot study cohort. Dietary exposures were coded as \"Yes\" or \"No\" for lifetime consumption of food from local sources (own garden or livestock, friend\'s garden or livestock, or grown/obtained in Chapaevsk) within the following categories: (1) fruits and vegetables; (2) chicken; (3) non-chicken meats which included goats, cows, pigs, and other non-chicken meat; (4) milk and dairy products, including eggs; and (5) fish. Exploratory models included factors considered a priori to have a potential relationship with serum dioxin levels such as the child\'s age, body mass index (weight in kg/height in m^2^), and local food consumption habits; reproductive factors (gestational age at birth of the index son, maternal parity \[number of live births prior to the index son\], gravidity \[number of prior pregnancies\], and breastfeeding history \[cumulative number of weeks of breastfeeding previous children\]), residential information (duration in years that the index boy resided in Chapaevsk and distance that mother lived from the factory during pregnancy with the index son and at the time of the study), and socio-demographic factors including parental education, parental occupation, and monthly household income (1 = 300--1000 rubles \[\$10--\$35 U.S.\], 2 = 1000--1500 rubles \[\$35--\$50 U.S.\], 3 = \>1500 rubles \[\>\$50 U.S.\]). Results ======= Per our selection criteria, of the 30 boys with serum samples analyzed for dioxin-like compounds, fifteen boys had congenital abnormalities (six hypospadias and nine cryptorchidism). No boy had both malformations. The 15 control boys did not have either hypospadias or cryptorchidism. The distances the boys lived from the Khimprom factory at the time of the study, along with other demographic characteristics, are presented in Table [1](#T1){ref-type="table"}. The mean (sd) distance the boys lived from the Khimprom factories was 4.45 (1.76) km. During the pregnancy with her son, of the 29 mothers, 8 (27%) lived less than 2 km, 8 (27%) lived 2--6 km, and14 (45%) lived greater than 6 km from the Khimpron factory. One subject was missing questionnaire data, including distance from Khimprom, medical history, and local food consumption. Only three of the boys\' fathers worked at the Khimprom factory during the year prior to the child\'s birth. While pregnant, two of the boys\' mothers reported alcohol intake and one smoked. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of adolescent boys from Chapaevsk, Russia (N = 30) ::: **Mean** **SD** **N** **%** ----------------------------------------------------------- ---------- -------- ------- ------- **Age (years)** 16.1 0.81 **Body Mass Index (kg/m^2^)** 19.9 3.09 **Weeks of Gestation**^a^ 39.2 2.37 **Breastfeeding duration of index son (weeks)**^b^ 36.9 25.2 **Prior maternal breastfeeding (weeks)**^b^ 18.1 38.3 **Birth weight (gm)**^c^ 3433 439 **Current residential distance to Khimprom factory (km)** 4.45 1.76 **Time boy resided in Chapaevsk(yrs)** 14.0 3.70 **Gravidity**^d^ **0** 14 48 **1** 8 28 **2--3** 7 24 **Parity**^d^ **0** 17 59 **1** 10 34 **2--3** 2 7 **Monthly family income level** **300--1000 rubles (\$10--\$35 U.S.)** 8 28 **1000--1500 rubles (\$35--\$50 U.S.)** 7 24 **\>1500 rubles (\> \$50 U.S.)** 14 48 **Maximum parental education level** **8--11**^th^**grade** 2 7 **Junior College** 17 59 **University** 8 28 ^a^One subject was missing data on weeks of gestation of the index son. ^b^One subject was missing data on breastfeeding. Prior maternal breastfeeding is the cumulative number of weeks of breastfeeding children born prior to the index son. ^c^Two subjects were missing data on birth weight of the index son. ^d^Maternal gravidity is the number of pregnancies prior to the index son and parity is the number of live births prior to the index son. ::: Since the PCDD, PCDF and PCB distributions were skewed, percentile distributions for serum concentrations and WHO-TEQs for 7 PCDD congeners, 9 PCDF congeners, 3 non-ortho (coplanar) PCBs and 6 mono-ortho PCBs are shown in Table [2](#T2){ref-type="table"}. The median (25^th^, 75^th^percentile) concentrations for total PCDDs, PCDFs and coplanar PCBs were 95.8 pg/g lipids (40.9, 144), 33.9 pg/g lipids (20.4, 61.8), and 120 pg/g lipids (77.6, 157), respectively. The median (25^th^, 75^th^percentile) concentration for total mono-ortho PCBs was 47.7 ng/g lipids (39.2, 78.4); as expected more than two orders of magnitude larger than for the other compounds. For WHO-TEQs, the median (25^th^, 75^th^percentile) for total PCDDs, PCDFs, and coplanar PCBs were 0.29 (0.1, 9.14), 7.98 (5.27, 12.3), and 7.39 (4.51, 11.9), respectively. For total mono-ortho PCBs, the median (25^th^, 75^th^percentile) WHO-TEQ was 8.80 (7.16, 15.5). The total WHO-TEQ for dioxins, furans and PCBs had a median (25^th^, 75^th^percentile) of 30.9 (18.4, 46.8). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distribution of concentrations and WHO-TEQs of PCDD/Fs and PCBs in adolescent boys from Chapaevsk, Russia ::: **Congener** **Average DL** **\# of samples above DL**^a^ **Concentrations** **WHO-TEF** **WHO-TEQ (pg TEQ/g lipid)** ------------------------------------------- ---------------- ------------------------------- -------------------- ------------- ------------------------------ --------- ------ ------ ------ ***PCDDs***(*pg/g lipid*) 2,3,7,8-TCDD 4.1 2/29 0.00 0.00 0.00 1.000 0.00 0.00 0.00 1,2,3,7,8-PeCDD 4.8 12/30 0.00 0.00 8.30 1.000 0.00 0.00 8.30 1,2,3,6,7,8-HxCDD 7.1 1/30 0.00 0.00 0.00 0.100 0.00 0.00 0.00 1,2,3,7,8,9-HxCDD 6.9 3/30 0.00 0.00 0.00 0.100 0.00 0.00 0.00 1,2,3,4,6,7,8-HpCDD 14.6 15/30 9.40 14.9 25.1 0.010 0.09 0.15 0.25 1,2,3,4,6,7,9-HpCDD 9.0 0/30 0.00 0.20 1.70 0.0000 0.00 0.00 0.00 OCDD 73.1 15/29 32.9 75.0 100 0.0001 0.00 0.01 0.01 ***PCDFs***(*pg/g lipid*) 2,3,7,8-TCDF 3.9 0/30 0.00 0.00 0.00 0.1000 0.00 0.00 0.00 1,2,3,7,8-PeCDF 4.7 1/30 0.00 0.00 0.00 0.050 0.00 0.00 0.00 2,3,4,7,8-PeCDF 4.5 29/30 8.50 12.6 17.4 0.500 4.25 6.30 8.75 1,2,3,4,7,8-HxCDF 4.2 27/30 5.50 9.15 20.1 0.100 0.55 0.92 2.01 1,2,3,6,7,8-HxCDF 4.5 21/30 0.00 5.70 8.70 0.100 0.00 0.57 0.87 1,2,3,7,8,9-HxCDF 4.2 0/30 0.00 0.00 0.00 0.100 0.00 0.00 0.00 2,3,4,6,7,8-HxCDF 4.5 1/30 0.00 0.00 1.80 0.100 0.00 0.00 0.18 1,2,3,4,6,7,8-HpCDF 16.8 0/30 3.00 5.30 10.5 0.010 0.03 0.05 0.11 1,2,3,4,7,8,9-HpCDF 6.0 0/30 0.00 0.00 0.00 0.010 0.00 0.00 0.00 ***Coplanar PCBs***(*pg/g lipid*) 3,4,4\',5-TCB 81 11.0 2/30 0.00 0.00 1.60 0.0001 0.00 0.00 0.00 3,3\',4,4\',5-PeCB 126 25.3 30/30 43.3 68.4 116 0.1000 4.33 6.84 11.6 3,3\',4,4\',5,5\'-HxCB 169 7.7 28/30 27.3 44.2 54.3 0.0100 0.27 0.44 0.54 ***Mono-ortho PCBs***(*ng/g lipid*) 2,3,3\',4,4\'-PeCB (105) 11.0 9/27 5.6 7.1 11.8 0.0001 0.56 0.71 1.18 2,3\',4,4\',5-PeCB (118) 11.0 27/27 20.7 28.0 51.6 0.0001 2.07 2.80 5.16 2,3,3\',4,4\',5-HxCB (156) 11.0 10/27 7.5 9.1 15.9 0.0005 3.75 4.55 7.95 2,3,3\',4,4\',5\'-HxCB (157) 11.0 0/27 2.3 2.6 4.3 0.0005 1.15 1.30 2.15 2,3\',4,4\',5,5\'-HxCB (167) 11.0 0/27 1.6 2.0 3.7 0.00001 0.02 0.03 0.04 2,3,3\',4,4\',5,5\'-HpCB (189) 10.9 0/25 0.4 0.6 0.8 0.0001 0.04 0.06 0.08 *Total PCDDs (pg/g lipid)* 40.9 95.8 144 0.10 0.29 9.14 *Total PCDFs (pg/g lipid)* 20.4 33.9 61.8 5.27 7.98 12.3 *Total coplanar PCBs (pg/g lipid)* 77.6 120 157 4.51 7.39 11.9 *Total PCDD/F/coplanar PCBs (pg/g lipid)* 154 273 397 12.1 17.1 38.1 *Total Mono-ortho PCBs*(ng/g lipid) 39.2 47.7 78.4 7.16 8.80 15.5 *Total TEQs* 30.9 46.8 Abbreviations: DL, detection limit. Note: 1,2,3,4,7,8-HxCDD, OCDF, and 3,3\',4,4\'-TCB 77 were not reported in the present study due to interferences in method blanks. ^a^Results for some samples could not be reported because one or more of the analytical QA/QC criteria were not met. Therefore, the denominator may be less than 30. ::: Although means and standard deviations do not appropriately describe skewed distributions, to allow for comparisons with other studies we briefly present them. The mean (sd) of total PCDDs, PCDFs, and coplanar PCBs were 104 pg/g lipids (78.0), 63.8 pg/g lipid (89.9) and 123 pg/g lipid (53.1), respectively. The mean of the TEQs for PCDDs, PCDFs, and coplanar PCBs were 5.51, 13.9 and 8.35 pg TEQ/g lipid, respectively. PCDDs, PCDFs, and coplanar PCBs accounted for a mean percent of total serum dioxin concentrations of 34.3% (range 8.4 to 80.1%), 18.0% (range 4.4 to 57.5%), and 47.7% (range 15.5 to 73.9%), respectively. The relatively high concentrations of mono-ortho PCBs contributed significantly to TEQs and were therefore also included in the total TEQs. PCDDs, PCDFs, coplanar PCBs, and mono-ortho PCBs accounted for a mean percent of total TEQs of 11.9% (range 0.14 to 59.0%), 30.4% (range 3.9 to 71.3%), 25% (range 8.1 to 43.0%), and 32.8% (range 0 to 60.1%), respectively. Figure [1](#F1){ref-type="fig"} displays the percent contribution of each PCDD, PCDF and PCB congener to the total TEQ/g lipid for the thirty boys. Based on serum concentrations, the predominant dibenzo-*p*-dioxin congeners were OCDD and 1,2,3,4,6,7,8-HpCDD (mean percents of 73.8% and 22.2%, respectively), while 1,2,3,7,8-PeCDD and 1,2,3,4,6,7,8-HpCDD were the predominant congeners based on mean TEQs (mean percents of 56.2% and 37.0%, respectively). For PCDFs, 2,3,4,7,8-PeCDF and 1,2,3,4,7,8-HxCDF accounted for the majority of the total PCDF concentrations (means of 35.5% and 32.4%, respectively), and these same two congeners accounted for the majority of the total PCDF TEQs (75.2% and 16.7%, respectively). For coplanar PCBs, 3,3\',4,4\',5-PeCB (\#126) and 3,3\',4,4\',5,5\'-HxCB (\#169) accounted for the majority of the total coplanar PCB concentrations (means of 63.4 and 35.3%, respectively), and 3,3\',4,4\',5-PeCB (\#126) accounted for 94.1% of the total coplanar PCB TEQs. For mono-ortho PCBs, 2,3\',4,4\',5-PeCB (\#118) and 2,3,3\',4,4\',5-HxCB (\#156) accounted for the majority of the total mono-ortho PCB concentrations (means of 56.0% and 19.7%, respectively), while 2,3,3\',4,4\',5-HxCB (\#156) accounted for 48.3% of the total mono-ortho PCB TEQs and 2,3\',4,4\',5-PeCB (\#118) for 30.3%. The concentrations of six prevalent PCB congeners, or so-called indicator PCBs \[[@B23]\], are presented in Table [3](#T3){ref-type="table"}. Here the largest contributors to the sum of the 6 congeners were PCB 153 (34.6%) and PCB138 (20.4%). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Percent contribution of PCDDFs and PCBs to TEQs (pg TEQ/g lipid) among boys from Chapaevsk. ::: ![](1476-069X-4-8-1) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Concentrations (ng/g lipid) of six PCB congeners^a^in blood of adolescent boys from Chapaevsk, Russia ::: **PCB congener (IUPAC \#)** **Average DL** **\# of Samples above the DL**^b^ **Median Concentration (ng/g lipid)** **Mean (SD) Concentration (ng/g lipid)** ----------------------------- ---------------- ----------------------------------- --------------------------------------- ------------------------------------------ PCB 28 20.0 1/24 10.6 11.9 (6.33) PCB 74 11.0 14/27 10.8 14.8 (10.8) PCB 99 11.0 27/27 27.7 34.1 (22.5) PCB 138 18.0 21/27 29.6 39.4 (22.2) PCB 153 19.8 27/27 51.2 64.5 (30.7) PCB 180 11.0 23/27 17.7 22.8 (14.9) Sum of 6 congeners 136 168 (99.3) Abbreviations: DL, detection limit ^a^PCB congeners include 6 prevalent or indicator congeners (DIN 51 527, 1987) ^b^Results for some samples could not be reported because one or more of the analytical QA/QC criteria were not met. Therefore, the denominator may be less than 30. ::: Only two of the 30 boys had levels of 2,3,7,8-TCDD greater than the detection limits; levels for these two boys were 17.9 and 21.7 pg/g lipid. Because all samples had different weights (4.1 to 12.1 g), different percent lipids, and percent recoveries, the DLs for TCDD in our study samples ranged from 1.5 to 16.5 pg/g lipid. We were unable to identify unique characteristics of the two boys with detectable levels of 2,3,7,8-TCDD. There were no differences in sum of PCDDs, PCDFS, and PCBs between the group of boys with cryptorchidism or hypospadias and the comparison boys. The smallest difference we could detect however, in the sum of dioxin concentrations given the small size of this study is 175 pg/g lipid (with 80% power) or 200 pg/g lipid (with 90% power). In other words, we can only detect differences that are larger than the observed standard deviation in sum of dioxin-like compounds. Of the 29 boys with lifetime food consumption data, 86% (n = 25) had consumed local meat products (in particular, 83% had consumed non-chicken meat and 52% had consumed local chicken), 83% (n = 24) had consumed local fish, 93% (n = 27) had consumed locally produced dairy products or eggs, and 97% (n = 28) had consumed locally grown fruits and/or vegetables. Using regression models adjusting for age, associations were found between the log sum of dioxin concentrations and several characteristics of the thirty boys (Table [4](#T4){ref-type="table"}). Increasing age, local non-chicken meat and fish consumption, and log of lipid-adjusted PCB congener 118 were associated with increases in the sum of the dioxin concentrations. For instance, a one- year increase in age increased the mean sum of dioxin-like compounds (291 pg/g lipids) by 30% and consumption of local non-chicken meat increased dioxin-like compounds by 75%. Gestational age in weeks was inversely associated with the sum of the dioxin concentrations, multiplicative factor of 0.92. Although the coefficient for the relationship between log sum of dioxin-like compounds and current residential distance from the Khimprom factory was less than 1.0 (which would support a decrease in dioxin levels with increasing distance from the factory), it was not statistically significant. We also explored the relationship between the sum of dioxin-like compounds and categories of GIS distances from Khimprom plants (\<2 km, 2--6 km, \>6 km) and found an inverse association, though it too was not statistically significant. Duration of breast-feeding, distance from Khimprom during pregnancy, maternal parity, and parental occupation were unrelated to sum of dioxin-like compounds or sum of dioxin TEQs. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Individual predictors of log sum of dioxin-like compounds^a^, adjusted for age in years ::: **Predictor** **Estimate** **p-value** **Multiplicative factor on dioxin (95% CI)** **R-Square Value** -------------------------------------------------- -------------- ------------- ---------------------------------------------- -------------------- Age of index son (years) 0.26 0.057 1.30 (1.00--1.72) 0.13 BMI of index son (kg/m^2^) 0.03 0.40 1.03 (0.95--1.13) 0.15 Consumption of local non-chicken meat (y/n) 0.56 0.042 1.75 (1.05--2.92) 0.26 Consumption of local chicken (y/n) 0.13 0.54 1.14 (0.75--1.72) 0.14 Consumption of local fish (y/n) 0.48 0.079 1.62 (0.97--2.71) 0.23 Consumption of local eggs (y/n) 0.20 0.36 1.23 (0.80--1.88) 0.16 Consumption of local dairy (y/n) -0.19 0.65 0.82 (0.36--1.89) 0.13 Consumption of local fruits/vegetables (y/n) -0.12 0.84 0.89 (0.28--2.77) 0.13 Current residential distance from Khimprom (km) -0.06 0.37 0.94 (0.82--1.07) 0.15 Years resided in Chapaevsk 0.009 0.75 1.01 (0.95--1.07) 0.13 Monthly family income level^b^ -0.09 0.50 0.92 (0.71--1.18) 0.14 Maximum parental education^b^ 0.21 0.26 1.23 (0.86--1.76) 0.17 Log (PCB 118) (ng/g lipid) 0.64 \<0.001 1.90 (1.37--2.63) 0.46 Congenital abnormality of index son -0.015 0.95 0.99 (0.64--1.51) 0.13 Parity^b,c^ -0.011 0.94 0.99 (0.74--1.32) 0.13 Weeks of gestation -0.08 0.072 0.92 (0.84--1.00) 0.23 Weeks of breast-feeding index son (per 12 weeks) -0.004 0.94 1.00 (0.90--1.10) 0.13 Prior breastfeeding duration^d^(years) 0.09 0.53 1.10 (0.82--1.47) 0.14 Abbreviations: Log, logarithm base 10; BMI, body mass index; PCB 118, polychlorinated biphenyl 118 ^a^Sum of dioxin-like compounds includes PCDDs, PCDFs, and coplanar PCBs. ^b^Each variable was modeled using three ordinal levels. ^c^Parity refers to the number of live births prior to the index son. ^d^Prior maternal breastfeeding refers to the cumulative number of weeks of breastfeeding children born prior to the index son. N = 30 for age, BMI, and distance from Khimprom at time of blood draw; N = 29 for all food consumption and medical history predictors; N = 27 for model with PCB118. ::: Models predicting the log of the sum of dioxin TEQ\'s (Table [5](#T5){ref-type="table"}) were generally similar in magnitude and direction for the predictors shown in Table [4](#T4){ref-type="table"}, but suggested weaker effects of age (multiplicative factor = 1.30, p = 0.14) and local fish consumption (multiplicative factor = 1.43, p = 0.32), and a stronger effect of maximal parental education (multiplicative factor = 1.50, p = 0.086) and log of PCB 118 TEQ (multiplicative factor = 2.61, p \< 0.001). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Predictors of log dioxin TEQs^a^among boys in Chapaevsk, Russia (adjusted for age). ::: **Predictor** **Estimate** **p-value** **Multiplicative factor on dioxin (95% CI)** **R-Square Value** ----------------------------------------------------- -------------- ------------- ---------------------------------------------- -------------------- Age (years) 0.26 0.14 1.30 (0.93--1.81) 0.08 BMI (kg/m2) 0.04 0.41 1.04 (0.95--1.13) 0.10 Consumption of local non-chicken meat (y/n) 0.68 0.059 1.97 (1.01--3.87) 0.20 Consumption of local chicken (y/n) 0.06 0.82 1.07 (0.62--1.83) 0.08 Consumption of local Fish (y/n) 0.36 0.32 1.43 (0.71--2.88) 0.11 Consumption of local Eggs (y/n) 0.12 0.68 1.13 (0.64--1.98) 0.09 Consumption of local Dairy (y/n) -0.43 0.44 0.65 (0.22--1.89) 0.10 Consumption of local Fruits/Vegetables (y/n) 0.35 0.65 1.41 (0.32--6.17) 0.09 Current residential distance from Khimprom (km) -0.09 0.32 0.91 (0.77--1.09) 0.12 Years resided in Chapaevsk -0.014 0.71 0.99 (0.92--1.06) 0.09 Monthly family income level^b^ -0.14 0.41 0.87 (0.63--1.21) 0.10 Maximum parental education^b^ 0.41 0.086 1.50 (0.96--2.36) 0.18 log (PCB118 TEQ) 0.96 \<0.001 2.61 (1.81--3.75) 0.57 Sexual delay/congenital abnormality 0.21 0.46 1.23 (0.71--2.13) 0.10 Parity^b,c^ 0.06 0.77 1.06 (0.73--1.54) 0.09 Weeks of gestation -0.11 0.078 0.90 (0.80--1.01) 0.19 Breast-feeding duration of index son (per 12 weeks) -0.002 0.97 1.00 (0.87--1.14) 0.08 Prior breastfeeding duration^d^(years) 0.07 0.72 1.07 (0.73--1.57) 0.08 Abbreviations: Log, logarithm base 10; BMI, body mass index; TEQ, toxic equivalents ^a^Sum of dioxin TEQs include TEQs for PCDDs, PCDFs, coplanar PCBs and other \'dioxin-like\' PCBs ^b^Each variable was modeled using three ordinal levels. ^c^Parity refers to the number of live births prior to the index son. ^d^Prior maternal breastfeeding refers to the cumulative number of weeks of breastfeeding children born prior to the index son. N = 30 for age, BMI, and distance from Khimprom at time of blood draw; N = 29 for all food consumption and medical history predictors; N = 27 for model with PCB118. ::: The best fitting multivariate models for predicting the sum of dioxin concentrations are summarized in Table [6](#T6){ref-type="table"}. Even after accounting for the strong relationship between PCB 118 concentration and the sum of dioxin concentrations on the log scale, consumption of local non-chicken meat showed a significant association with increased dioxin levels. Similarly, the best fitting multivariate model for predicting the sum of dioxin TEQs also included PCB 118 and consumption of local non-chicken meat (R-square = 0.63). The best-fitting model for sum of dioxin concentrations, excluding log PCB 118, included age, weeks of gestation, income level, and consumption of local non-chicken meat and local dairy products, with an R-square of 0.56. The best fitting multivariate model for predicting the sum of dioxin TEQs, excluding log PCB 118 TEQ, included the same five predictors as those shown in Table [6](#T6){ref-type="table"} with similar magnitudes and significance of effects (R-square = 0.56). ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Multivariate models for predicting log sum of dioxin^a^concentrations among adolescent boys in Chapaevsk, Russia ::: **Model \#** **Predictor** **Estimate** **p-value** **R-square value** **Adjusted R-square value** -------------- ------------------------------------------------------- -------------- ------------- -------------------- ----------------------------- 1 Consumption of local non-chicken eat (y/n) 0.59 0.017 0.56 0.52 log (PCB 118) (ng/g lipid) 0.63 \<0.001 2 Age (years) 0.40 \<0.001 0.56 0.46 Weeks of gestation -0.11 0.005 Consumption of local dairy (y/n) -0.76 0.037 Consumption of local non-chicken meat (y/n) 0.90 \<0.001 Monthly family income level (ordinal, low, med, high) -0.27 0.020 Abbreviations: Log, logarithm base 10 ^a^Sum of dioxin-like compounds includes PCDDs, PCDFs, and coplanar PCBs N = 26 for model 1; N = 29 for model 2. Missing data for 3 boys on lipid-adjusted PCB 118, and one boy on questionnaire items (medical history, diet). Model 1: best fit multivariate model including PCB 118 data. Model 2: best fit multivariate model (forward and backward selection) not including PCB 118. ::: Discussion ========== Among adolescent boys in Chapaevsk, Russia, higher serum levels of sum of dioxin-like compounds and sum of dioxin TEQs were positively associated with increased age, consumption of fish, local meats other than chicken, and inversely with weeks of gestation. The age association was found despite a narrow age range of slightly over two years in our study. Although not statistically significant, the distance the boy lived from the Khimprom factory at the time of blood draw was inversely associated with serum levels of sum of dioxin-like compounds and sum of dioxin TEQs. As expected, serum PCBs, specifically PCB 118, were strongly associated with both sum of dioxin-like compounds and sum of dioxin TEQs. There was no association between the distance of the residence from the Khimprom plant during the pregnancy and subsequent serum dioxin levels. One potential explanation for the lack of association may include misclassification of distance since the mother was asked to recall a time period more than 14 years prior to the study. However, we would expect that the mother would be able to recall residential history at the time of the birth of their son. Mother\'s self-reported estimates of current residential distance from Khimprom was generally accurate; twenty-one of twenty-nine mothers correctly categorized their current residential distance from the Khimprom plants based on cross-referencing using GIS mapping. Other explanations include that prenatal exposure 14 or more years prior to the current serum sample is not as strong a predictor as are exposures resulting from present residential location. Although there was no association of case status (cryptorchidism or hypospadias) with dioxin levels, we did not have sufficient power to definitively assess this relationship. Perinatal history (e.g. weeks of breastfeeding) was generally not associated with exposure measures in this population; this may be a function of older age of the children. However, there are limited data on the relationship of perinatal factors with organochlorine exposures in this age group so a null finding is of interest given reports that differences in organochlorine levels among breastfed and non-breastfed are generally no longer discernable by early school age \[[@B24],[@B25]\] and, furthermore, that dietary intake after this age contributes significantly to total dioxin intake \[[@B26]\]. In prior studies, substantial emphasis has been placed on pre- and early postnatal (via breastfeeding) exposures because of particular vulnerability during fetal and early infant development. The exposure risk factors during peri-adolescence, another period of potential developmental vulnerability, has not been studied in-detail, therefore, the identification of exposure risk factors specific to this period will enhance our understanding of this critical period. Although data on levels of PCDD/PCDFs in children is limited, our results suggest that the mean total TEQs among Chapaevsk adolescents were higher than most values previously reported in non-occupationally exposed populations of comparable or even older ages. Figure [2](#F2){ref-type="fig"} shows a comparison of the mean PCDD/PCDFs TEQ levels in Chapaevsk boys with other populations (TEQ from dioxin-like PCBs was not included, since some of these studies did not report them). The mean TEQs of pooled blood samples from 10 year-old German boys in rural and urban settings was 8.2 pg TEQ/g lipid for an urban industrial area, 9.0 pg TEQ/g lipid for an industrial area within a rural setting, and 10.1 pg TEQ/g lipid for a rural area \[[@B27]\]. In comparison, the mean TEQ in the Chapevsk boys was 19.3 pg TEQ/g lipid. With the exception of children described by Wuthe et al. \[[@B27]\], subjects in the other studies in Figure [2](#F2){ref-type="fig"} were significantly older than the Chapaevsk boys. Despite age differences, the mean TEQ in Chapaevsk boys was comparable to or even higher than the mean TEQ in older populations from other countries. For example, they were higher than mean TEQs of 18.4 pg TEQ/g lipid in adults (40.6 years old average) from South Germany \[[@B27]\] or 16.4 pg TEQ/g lipid from 20 year old Japanese women \[[@B28]\] or pooled samples from randomly selected males and females 18--69 years of age from the Spanish city of Mataro \[[@B29]\], which were 12.5 and 14.7 pg TEQ/g lipid respectively. The mean levels in adult female (mean age 41 years) non-factory workers in the Russian city of Shelekhovo were also lower at 14.5 pg TEQ/g lipid \[[@B30]\]. Mean TEQs for the general population (mean age 44.2 years) in Germany, collected in 1997--98 \[[@B31]\], and median TEQs for long-term workers of pulp and paper mill and non-workers in the U.S. in 1996 \[[@B32]\] were similar to levels found in the Chapaevsk boys. The mean levels in the study in Germany were 20.71 pg TEQ/g lipid and in the U.S. the median levels were 19.1 pg TEQ/g lipid for community residents and 21.2 pg TEQ/g lipid for low exposure workers. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Mean PCDD/PCDFs TEQ levels in Chapaevsk boys in comparison with other populations. ::: ![](1476-069X-4-8-2) ::: Although TCDD was largely below the detection limits in this small pilot sample, the two boys with detectable values had high TCDD levels (17.9 and 21.7 pg/g lipid), suggesting that exposure for at least some portion of this population is substantially higher than typical of this age group. In comparison, in a cohort of adult (mean age of 58 years) fishermen from a polluted region of Finland, the mean TCDD concentration was 19 pg/g lipid \[[@B33]\]. In adult (mean age of 53 years) residents from Calcasieu Parish, Louisiana, which is near a chemical industrial complex, the mean TCDD level was 7.6 pg/g lipid \[[@B34]\]. Not only were the dioxin levels in these two children higher than those found in these studies, but the adults in the previous studies were several decades older and therefore would be expected to have higher dioxin body burdens than younger children \[[@B35]\]. Potential explanations for the large number of non-detectable samples for TCDD include the small sample volume and young age of the subjects. In our future studies in this population, we will collect larger volumes of serum for dioxin analysis. In one of the few studies on dioxin-like compounds in which children were included, Eskenazi and coworkers \[[@B36]\] evaluated the relationship between serum TCDD concentrations and age at exposure of female residents of Seveso, Italy. Residents near the ICMESA chemical plant in Seveso were exposed to some of the highest known residential levels of 2,3,7,8-TCDD as a result of an explosion at the plant. Archived serum collected near the time of the accident was used to measure exposures. Residents closest to the plant had a median 2,3,7,8-TCDD level of 272 ppt (IQR 92 -- 883 ppt). Residential proximity to the plant and younger age (up to 13 years old) were the strongest predictors of an individual\'s serum 2,3,7,8-TCDD level. Other predictors included being outdoors at the time of explosion and consumption of homegrown food. The higher levels found in children were most likely a result of increased exposure as a result of activity patterns and a greater proportionate consumption of food, water and air than adults \[[@B37]\]. Although the exposure scenario (an acute high exposure event) is different than the chronic low/moderate exposure occurring in Chapaevsk, the results from Seveso suggest that children may be at increased risk for high dioxin exposure from environmental contamination. In our study, although distance from the Khimprom plants was a weak predictor of serum dioxin-like compounds, consumption of local foods (specifically meat and fish), as in the Seveso study, was a strong predictor of sum of dioxin-like compounds and dioxin TEQs. This finding is notable given concerns regarding environmental dioxin contamination in the community and suggests that food may be one of the more, if not most, important routes of environmental contaminant exposure for residents in this setting. In other settings, contaminated food generally contributes much more substantially to human organochlorine burden than air or soil (which may be related to residential proximity to pollutant sources) \[[@B6]\]. We will investigate this issue in more detail in our ongoing study. Conclusion ========== Recently, data were published on dioxin levels among adult residents of Chapaevsk, Russia \[[@B38]\]. Twenty-four self-selected volunteers (12 men and 12 women) provided blood samples in 1998 for analysis of dioxin-like compounds. These serum samples were analyzed by the same laboratory at the CDC, which analyzed the serum samples for our study. None of the adult subjects worked at the Khimprom plant in Chapaevsk. The mean age of subjects was 44 years for men and 45 years for women. Among all subjects, the mean TEQ of total dioxins was 61.2 pg TEQ/g lipid (range 16.4 to 168.1 pg TEQ/g lipid). The adults had the same congeners as major contributors to TEQ of dioxin, furan and co-planar PCBs as the boys in our study. We could not compare the levels of mono-ortho-PCBs in the children to levels in the adult Chapevsk residents since they were not reported. In the adults in Chapaevsk, there were positive relationships of TEQs with age and BMI and an inverse relationship of TEQs with residential distance from the plant \[[@B38]\]. In both this adult and our pediatric Chapaevsk population, PCB congeners contributed significantly to TEQs and therefore it is important to assess these congeners vis-à-vis toxicological implications mediated through dioxin-like mechanisms. We have recently begun recruiting eight- and nine-year old boys and their families in Chapaevsk into a prospective cohort study on the relationship between dioxin exposure and somatic growth and pubertal development. Results of this exposure assessment study demonstrate relatively high TEQ exposures among Chapaevsk children compared to similar aged populations elsewhere and, as evidenced by the importance of local food consumption in determining this exposure, corroborate our hypothesis that local environmental contamination likely contributes to exposure risk in this setting. Our planned additional studies of the potential health consequences of Chapaevsk children\'s exposure is an important sequel to having characterized organochlorine exposure and exposure risk among this potentially vulnerable age group. List of Abbreviations ===================== IUPAC International Union of Pure and Applied Chemistry mono-*ortho*PCBs Includes PCB congener numbers 105, 114, 118, 123, 156, 157, 167, 189 non-*ortho*PCBs Includes PCB congener numbers 77, 81, 126, 169 OCDD Octachlorodibenzo-*p-*dioxin OCDF Octachlorodibenzo-furan PCBs Polychlorinated biphenyls PCDDs Polychlorinated dibenzo-p-dioxins PCDFs Polychlorinated dibenzofurans pg/g Picogram per gram, 10^-12^g. TEF Toxic equivalency factors TEQ Toxic Equivalents (WHO~98~) I-TEQ International Toxic Equivalents (NATO-CCMS, 1988) WHO World Health Organization Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= RH conceived the design of the study and took primary responsibility for drafting the manuscript. PW performed the statistical analysis, participated in the design of the study and the drafting of the manuscript. LA participated in the drafting of the manuscript and the interpretation of the dioxin results. SK participated in the design of the study and the drafting of the manuscript. LP participated in the statistical analysis. DGP carried out the dioxin analyses. WET carried out the dioxin analyses. MML participated in the design of the study and the drafting of the manuscript. BR participated in the design of the study. OS oversaw subject recruitment and data collection, participated in the design of the study and the drafting of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was funded by the US EPA, Grant number R-82943701-0 and NIEHS, Grant Number ES00002.

PubMed Central

2024-06-05T03:55:58.262895

2005-5-26

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168904/", "journal": "Environ Health. 2005 May 26; 4:8", "authors": [ { "first": "Russ", "last": "Hauser" }, { "first": "Paige", "last": "Williams" }, { "first": "Larisa", "last": "Altshul" }, { "first": "Susan", "last": "Korrick" }, { "first": "Lynne", "last": "Peeples" }, { "first": "Donald G", "last": "Patterson" }, { "first": "Wayman E", "last": "Turner" }, { "first": "Mary M", "last": "Lee" }, { "first": "Boris", "last": "Revich" }, { "first": "Oleg", "last": "Sergeyev" } ] }

PMC1168905

Introduction ============ Obesity is of major primary care concern and is targeted as an international health objective in Healthy 2000, which seeks to reduce the prevalence of obesity to less than 20%. In the last 10 years, the number of overweight people has increased from 26 to 34% \[[@B1]\]. Conventional dietary and behavioral treatment have failed in long-term management. Dietary strategies used to manage obesity include the use of high fibre, low carbohydrates and fats diet \[[@B2],[@B3]\]. Beneficial effect of dietary fibre in the management of obesity is not well established, since their mechanism of action is not known. The discovery of new medicinal plants has led to the creation of potential drugs that modify feeding behavior and metabolism and may therefore have application in weight management. The aim of the present study is to investigate the effect of *Irvingia gabonensis*extract on body weight. Subjects and methods ==================== A total of 40 obese subjects aged between 19 and 55 years were selected from a group responding to a radio advertisement. After physical examination and laboratory screening tests, diabetics, pregnant and lactating women were excluded. None of these subjects took any weight reducing medication and none was following any specific diet. The purpose, nature and potential risks of the study were explained to all patients and a written informed consent was obtained before their participation. The local research ethics committee approved the experimental protocol. Study design ------------ The study was as a randomised, double blind placebo-controlled crossover design, and consisted of a 4-week treatment period. Subjects were given two different types of capsules containing 350 mg of *Irvingia gabonensis*seed extract (active formulation) or oat bran (placebo). Three capsules were taken three times daily, one-half hour before meals (a total daily amount of 3.15 g of *Irvingia gabonensis*seed extract) with a glass of warm water. Capsules were identical in shape, colour and appearance, with neither patients nor researchers knowing what capsule they received. During the experimental period, subjects were examined weekly, with their body weight, body fat, waist and hip circumferences recorded each time. Subjective findings such as increased or decreased appetite, feeling of lightness and gastrointestinal pains were individually noted. Side effects of the active extract, if any were also solicited and noted during each visit. The subjects were also interviewed about their physical activity and food intake during the trial, and were instructed to eat a low fat diet (1800 Kcal) as well as keep a record for seven consecutive days (using household measurements). Anthropometric measurements --------------------------- Anthropometric measurements were done at each visit, with body weight and body fat (impedance measurement using a TANITA™ monitor Scale) measurements on fasted (12 hour) subjects wearing light clothing. Waist and hip circumferences were measured by soft non-stretchable plastic tape on the narrowest and the widest parts of the trunk. Laboratory methods ------------------ Blood samples were collected after a 12 h overnight fast into heparinized tubes at the beginning of the study, after two weeks and at the end (4 weeks) of treatment. The concentrations of total cholesterol, triacylglycerol, HDL-cholesterol, and glucose, in plasma were measured using a commercial diagnostic kit (Cholesterol infinity, triglycerides Int, EZ HDL™ cholesterol, EZ LDL™ cholesterol, Glucose Trinder, respectively) from SIGMA Diagnostics Statistical Analysis -------------------- Results are expressed as mean ± SEM except for anthropometric measurements. Paired Student\'s t-test was carried out on the start and end values of placebo and *Irvingia gabonensis*capsules and also on the differences between the placebo and *Irvingia gabonensis*crude extract. Results ======= Effect on body composition -------------------------- *Irvingia gabonensis*induced a decrease in weight of 2.91 ± 1.48% (p \< 0.0001) after two weeks and 5.6 ± 2.7% (p \< 0,0001) after one month. Although the percentage of body fat was not significantly reduced with both placebo and IG, the waist circumference (5.07 ± 3.18%; p \< 0.0001) and hip circumference (3.42 ± 2.12%; p \< 0,0001) were significantly reduced by IG. A reduction of 1.32 ± 0.41% (p \< 0.02) and 2.23 ± 1.05% (p \< 0.05) was observed with the placebo after two and four weeks respectively of treatment. Effect on blood pressure ------------------------ From the second week of experimentation, the systolic blood pressure was significantly reduced by the active extract (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Effect Irvingia gabonensis on systolic (SBP) and diastolic (DBP) blood pressure ::: Treatment period (weeks) ------------ -------------------------- ---------------- ------------------ ------------------ SBP (mmHg) Active 136.41 ± 19.57 132.66 ± 18.48\* 132.83 ± 17.97\* Placebo 134 ± 5.05 121.5 ± 5.89 123.83 ± 2.92 DBP (mmHg) Active 98.5 ± 19.52 97.5 ± 22.80 94.08 ± 11.07 placebo 93.50 ± 10.31 93.83 ± 7.41 91.5 ± 6.53 Values are means ± sem. Significant differences were at \*P \< 0.001 by comparison to the placebo group. ::: Effect on blood lipids components --------------------------------- The plasma total cholesterol cencentration was reduced by 39.21%, triglyerides by 44.9% (p \< 0,05) and LDL by 45.58%. This was accompanied by a significant increase in HDL-cholesterol of 46.852%. The CT/HDL ratio (p \< 0.05) and the blood glucose level were also reduced (32.36%; p \< 0.05). No significant change was observed in the placebo group. Discussion ========== The soluble fibre of the seed of *Irvingia gabonensis*like other forms of water-soluble dietary fibres, are \"bulk-forming\" laxatives. *Irvingia gabonensis*seeds delay stomach emptying, leading to a more gradual absorption of dietary sugar. This effect can reduce the elevation of blood sugar levels that is typical after a meal \[[@B4]\]. Controlled studies have found that after-meal blood sugar levels are lower in people with diabetes given glucomannan in their food \[[@B5]\] and overall diabetic control is improved with soluble fibre-enriched diets according to preliminary \[[@B6]\] and controlled \[[@B7],[@B8]\] trials. One double-blind study reported that glucomannan (8--13 grams per day) stabilized blood sugar levels in people with the insulin resistance syndrome \[[@B9]\]. Like other soluble fibers, *Irvingia gabonensis*seed fibre can bind to bile acids in the gut and carry them out of the body in the faeces, which requires the body to convert more cholesterol into bile acids \[[@B10]\]. This can result in the lowering of blood cholesterol as well as other blood lipids. Controlled double-blind \[[@B11],[@B12]\] studies have shown that supplementation with several grams per day of soluble fibre significantly reduced total blood cholesterol, LDL cholesterol, and triglycerides and in some cases raised HDL cholesterol, these being comparable with effects noticed with *Irvingia gabonensis.* Considering the wide use of *Irvingia gabonensis*in the preparation of various dishes in Cameroon, its use should be further encouraged for the purposes of control of dietary lipids as well as for weight reduction. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Effect of *Irvingia gabonensis*crude extract on body weight, body fat, waist and hip circumferences ::: Treatment period (weeks) -------------- -------------------------- ---------------- ---------------- ---------------- Weight (kg Active 105.10 ± 16.98 102.3 ± 17.06 101.01 ± 16.63 Placebo 79.43 ± 9.83 79.43 ± 9.83 79.33 ± 10.63 Active 46.11 ± 4.4848 46.5 ± 3.68 45.34 ± 3.52 Body fat (%) Placebo 40.58 ± 3.49 40.58 ± 3.9 40.3 ± 3.8 Active 112.76 ± 20.5 109.7 ± 20.4 106.6 ± 20.79 Waist (cm) Placebo 81.1 ± 7.1 81,91 ± 7,91 81.25 ± 7.52 Active 125.69 ± 11.34 122.92 ± 10.67 121.15 ± 10.39 Hip (cm) Placebo 122.2 ± 10.7 122.2 ± 10.7 121.5 ± 10.9 ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The effect of Irvingia gabonensis on blood total cholesterol (TC), triglyceride (TRI), high density lipoprotein cholesterol (HDL-c), low density lipoprotein cholesterol (LDL-c) and glucose. ::: T-cholesterol TRI HDL-c LDL-c LDL/HDL T-cho/HDL GLUCOSE --------- --------- ---------------- ---------------- --------------- ---------------- -------------- ------------- ------------- Active Initial 215 ± 55.12 162 ± 33.15 61.23 ± 20.36 121.37 ± 36.3 1.98 ± 1.78 3.51 ± 2.70 3.8 ± 1.92 Final 130.68 ± 39.5 89.22 ± 55.63 89.9 ± 28.44 66.08 ± 34.27 0.735 ± 1.20 1.45 ± 1.38 2.57 ± 1.03 Placebo Initial 163.70 ± 25.32 130.65 ± 37.82 31.38 ± 25.21 105.06 ± 11.86 5.05 ± 3.94 6.44 ± 3.37 3.6 ± 0.41 Final 158.36 ± 30.46 100.52 ± 32.55 41.20 ± 19.53 98.55 ± 27.99 3.19 ± 1.85 4.51 ± 2.07 3.9 ± 0.74 :::

PubMed Central

2024-06-05T03:55:58.269865

2005-5-25

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168905/", "journal": "Lipids Health Dis. 2005 May 25; 4:12", "authors": [ { "first": "Judith L", "last": "Ngondi" }, { "first": "Julius E", "last": "Oben" }, { "first": "Samuel R", "last": "Minka" } ] }

PMC1168906

Introduction ============ The use of biomarkers in medicine lies in their ability to detect disease and support diagnostic and therapeutic decisions. New research and novel understanding of the molecular basis of the disease reveals an abundance of exciting new biomarkers who present a promise for use in the everyday clinical practice. The initial evaluation of a serum biomarker concerns its expression in patients with the disease and in normal individuals in order to define sensitivity and specificity. The sensitivity of a test is defined as the proportion of patients with disease having a positive test whereas the specificity is the proportion of patients without the disease who have a negative or normal test. Consequently the serum level of an ideal marker should: 1) increase pathologically in the presence of the disease (high sensitivity), 2) not increase in the absence of the disease (high specificity), 3) relate to the disease burden and extent, 4) change in accordance with the clinical evolution, reflecting the current status of disease, or better 5) anticipate clinical changes, i.e. indicating the presence of relapse before it becomes obvious at a clinical level and finally 6) possess constant serum levels (no major fluctuation) \[[@B1]\]. Additionally, a clinically suitable biomarker should fulfil the following requirements \[[@B2]\]: 1\. add independent information about the risk or prognosis 2\. account for a large proportion of the risk associated with a given disease or condition 3\. be reproducible (as determined by the low coefficient of variation) 4\. be sensitive, specific and should present with a high predictive value 5\. be of easy and cheap determination Very few markers present a threshold at which the risk suddenly rises. The interplay between sensitivity and specificity and the nature of the disease under prediction assigns suitable cut-off points. Sensitivity and specificity calculated at various cut-off points give rise to a receiver-operating-characteristic (ROC) curve \[[@B2]\]. A clinically useful biomarker will be one with the largest area under the ROC curve. A number of novel blood biomarkers of lung disease including cytokines, enzymes, adhesion molecules, collagen relevant products and products of type II epithelial cells, have been studied for their clinical applicability. The scope of this review is based on the fact that although there are numerous published papers investigating the utility of biomarkers in the clinical research field the number of review articles summarizing the current state of knowledge about the clinical applications of these molecules as diagnostic and prognostic tools in the research field relevant to acute respiratory distress syndrome (ARDS) and acute lung injury (ALI) still remains inadequately small. Serum biomarkers in Acute Respiratory Distress syndrome ======================================================= ARDS is a clinical and pathophysiologic entity characterized by severe acute injury, directly or indirectly via the blood, to the endothelial and epithelial surfaces of the lung leading to respiratory failure. The main characteristics of the syndrome are diffuse inflammation and increased microvascular permeability that cause diffused interstitial and alveolar oedema and persistent refractory hypoxemia \[[@B3]\]. Although a variety of insults may lead to ARDS, a common pathway may probably result in the lung damage \[[@B4],[@B5]\]. A complex series of inflammatory events have been recognized during the development of ARDS but the exact sequence of the events remains elusive. Immunological studies investigating bronchoalveolar lavage fluid (BALF) have shed further light into the pathogenetic mechanisms of ARDS \[[@B6]\] and formed the basis of concepts of its immunopathogenesis. A large variety of inflammatory mediators (Table [1](#T1){ref-type="table"}) have been found to be elevated in the early phase of ARDS, including lung-specific proteins, endotoxin binding proteins, tumor necrosis-alpha (TNFa), interleukins -- (ILs) -- 1, 2, 6, 8, 15, chemokines, ferritin, markers of endothelium activation (adhesive molecules and von-Willebrand factor antigen-VWF) as well as markers of neutrophil activation such as matrix metalloproteinases (MMPs) and their inhibitors and leukotrienes \[[@B5]-[@B7]\]. The majority of these molecules have features to recommend them as biologic markers in ARDS. Biomarkers have attracted a lot of attention in both ALI and ARDS since they can shed further light into the pathogenesis and pathophysiology of lung injury. Additionally, from a practical point of view, a clinical useful biomarker for ARDS must add information regarding the development of syndrome in at-risk patients that is not apparent from routine examination and investigation. The latter, could help the intensivist to monitor the disease and evaluate or modulate treatments before they have failed. Driven by this perspective idea, many studies have estimated their usefulness as early predictors of ARDS and accurate markers of lung injury before clinical changes can be detected. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### List of studied serum biomarkers in ARDS ::: --------------------------------------- -------------------------------------------------------- **Lung epithelium-specific proteins** **Surfactant-associated proteins** • SP-A • SP-B • SP-D **Mucin-associated antigens** • KL-6/MUC1 **Cytokines** • IL-1 • IL-2 • IL-6 • IL-8 • IL-10 • IL-15 • TNFa **Other serological parameters** **Markers of endothelium activation** • Adhesion molecules (E, L-selectin, I-CAM-1, V-CAM-1) • VWF **Markers of neutrophil activation** • MMP-9 • LTB4 **Ferritin** --------------------------------------- -------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, CC16: Clara-cell protein 16, IL: Interleukin, KL-6: Krebs von den Lungen-6, LTB4: Leukotriene B4 MMP-9: Metalloproteinase-9 MUC: Mucin, I-CAM-1: Intercellular Adhesion Molecule-1, sIL-2R: soluble interleukin-2 receptor, V-CAM-1: Vascular Adhesion Molecule-1, VWF: von Willebrand factor antigen ::: Cytokines (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}) ============================================================================ ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Studies measuring cytokines in patients with or at risk for ARDS ::: ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Investigator** **Patients Controls** **Biomarker / Summary** **ROC curve analysis Cut-off values** **Specificity -- Sensitivity PPV -- NPV** **Limitations** ------------------------- ----------------------- --------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ------------------------------------------------ ----------------------------------------------------------------- Pinsky et al. ^13^ 52 at risk Relation of IL-6 and TNF plasma levels to multiple-system organ failure and mortality No Not estimated Small number of patients\ No ROC curve analysis / cut-off levels\ No serial measurement Takala et al. ^14^ 20 at risk\ IL-6 and IL-8 plasma levels predict organ failure in community-acquired septic shock No Not estimated Small number of patients\ 56 controls No ROC curve analysis / cut-off levels\ Poor discriminative value of serum biomarkers per se Calandra et al. ^15^ 70 at risk IL-6 plasma levels do not predict the outcome in at risk patients for ARDS No Not reported Small number of patients\ No ROC curve analysis Kiehl et al. ^16^ 19 at risk TNFa, IL-6, IL-8 plasma levels fail to associate with severity and course of ARDS in leukocytopenic patients No Not estimated Small number of patients\ No ROC curve analysis / cut-off levels\ Leukocytopenic patients Meduri et al. ^18^ 27 ARDS TNFa, IL-1β, IL-2, IL-4, IL-6, IL-8\ Yes\ TNFa: 89-50-85-57%\ Small number of patients\ Superiority of IL-1β and IL-6 plasma levels in monitoring disease activity over commonly applied clinicophysiologic parameters. TNFa: 400 pg/ml\ IL-1β: 78-83-83-78%\ Perspective study\ IL-1β: 400 pg/ml\ IL-2: 89-83-90-80%\ Overlap of cytokine levels between survivors and non-survivors\ IL-2: 200 pg/ml\ IL-4: 89-50-85-57%\ Heterogeneity of studied population\ IL-4: 200 pg/ml\ IL-6: 77-75-81-70%\ Definition criteria of ARDS IL-6: 400 pg/ml\ IL-8: 66-50-66-50% IL-8: 400 pg/ml Agouridakis et al. ^19^ 8 ARDS^a^\ Association between increased levels of IL-2 and IL-15 and outcome in patients with early ARDS Yes\ IL-2: 100-100-100-100% IL-15: 100-100-100-100% Small number of patients\ 26 at risk IL-2: 173 pg/ml\ Limited number of studied molecules IL-15: 250 pg/ml ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, BALF: Bronchoalveolar lavage Fluid, IL: Interleukin, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, sIL-2R: soluble interleukin-2 receptor, TNFa: Tumor Necrosis Factor-alpha ^a:^Use the American European Consensus Conference definitions ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Studies measuring cytokines in patients with or at risk for ARDS ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Investigator** **Patients Controls** **Biomarker / Summary** **ROC curve analysis Cut-off values** **Specificity -- Sensitivity PPV -- NPV** **Limitations** --------------------- ----------------------- --------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ------------------------------------------- ----------------------------------------------------------------- Lesur et al. ^20^ 19 ARDS^a^\ Association of early low serum\ No Not estimated Small number of patients\ 14 at risk\ IL-2 levels with the patients\' survival No ROC curve analysis / cut-off levels\ 20 controls No serial measurement\ Discrepancies in serum and BALF IL-2 levels\ Limited number of studied molecules Parsons et al. ^23^ 77 at risk Association of serum IL-1ra, IL-10 levels with the disease outcome No Not estimated No ROC curve analysis / cut-off levels\ Poor predictive value for ARDS development\ Limited number of studied molecules Takala et al. ^24^ 52 at risk\ IL-8, IL-6, sIL-2R, E-selectin, procalcitonin\ No Not estimated Small number of non-survivors\ 9 ARDS^a^\ Persistent elevation of inflammatory markers in patients with ALI precedes its clinical diagnosis No ROC curve analysis / cut-off levels\ 45 controls Poor predictive value for ARDS development Bouros et al. ^25^ 32 ARDS^a^\ IL-4, IL-6, IL-6r, IL-8, IL-10\ Yes\ IL-4: 78-100-81-100%\ Small sample size\ 27 at risk High prognostic value of all the inflammatory markers in assessing the outcome in patients with or at risk for ARDS IL-4: 84 pg/ml\ IL-6: 59-96-69-94%\ No serial measurement\ IL-6: 160 pg/ml\ IL-6r: 78-76-76-78%\ Grouping and definition criteria IL-6r: 18 pg/ml\ IL-8: 93-96-92-96%\ IL-8: 2340 pg/ml\ IL-10: 96-92-96-93% IL-10: 98 pg/ml Schutte et al. ^26^ 30 ARDS^a^\ IL-6, IL-8, TNFa\ No Not estimated Small number of patients\ 44 at risk\ Serum levels of IL-6 and IL-8 in\ Weak correlations with clinical variables\ 17 controls ARDS and/or severe pneumonia, differentiate these entities from cardiogenic pulmonary oedema No definitive predictive value for outcome\ Overlap of cytokine levels between survivors and non-survivors\ No ROC curve analysis / cut-off levels Bauer et al. ^27^ 46 ARDS^a^\ IL-6, IL-1b, TNFa serum levels associate better with the degree of lung injury rather than clarify its specific aetiology No Not estimated Corticosteroid treatment\ 20 at risk\ Inconclusive prognostic value\ 10 controls No serial measurement\ Limited number of studied molecules ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, BALF: Bronchoalveolar lavage Fluid, IL-1ra: Interleukin-receptor antagonist, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, sIL-2R: soluble interleukin-2 receptor, TNFa: Tumor Necrosis Factor-alpha ^a:^Use the American European Consensus Conference definitions ::: Cytokines are widely recognized as mediators of an inflammatory response. Their discovery has stimulated multidisciplinary investigation to elucidate the role of these mediators in the injury and repair processes of human disease. In the lung, they are produced either by local resident cells such as alveolar macrophages, pneumocytes, endothelial cells and fibroblasts or by cells such as neutrophils, lymphocytes and platelets arriving to the lung in response to local or systemic injury \[[@B8],[@B9]\]. Cytokines are involved both in the early (TNFa, IL-1, 2, 6, 8, 15) and late phase (IL-4) of inflammation and have been shown unequivocally to be of crucial importance in the pathophysiology of septic shock, a condition frequently culminating to ARDS \[[@B10],[@B11]\]. Studies have demonstrated that in ARDS patients detectable cytokine serum levels are closely related to the disease severity and mortality \[[@B11],[@B12]\], suggesting a potential role in reflecting the severity of the lung injury. Moreover, humoral (IL-6, IL-8) and cellular markers (CD 11b) of systemic inflammation have been delineated to identify patients with septic shock at risk for organ failure, culminating to a fatal outcome \[[@B13],[@B14]\]. However, their monitoring and prognostic value in septic-shock patients still remains controversial. Calandra et al. \[[@B15]\] stated that serum cytokines could not be used as a routine laboratory test to predict the outcome in septic-shock patients. In addition, Kiehl et al. \[[@B16]\] failed to prove usefulness of plasma cytokines measurement for the evaluation of severity and course of ARDS in a small cohort of leukocytopenic septic-shock patients. In the same study, estimation of BALF levels appeared to differentiate between responders and non-responders to treatment before clinical differences become apparent. Nonetheless, it should be noted that the small sample size, the contradictive results, the lack of standardization techniques and uniform definitions for ARDS and at risk patients, together with the heterogeneity of the syndrome and the patients studied generate major concerns regarding the reproducibility and the reliability of the data presented. Consequently, there is a need for further large scale investigations in the context of appropriate clinical trials for any meaningful conclusions to be reached. Evidence from preceding studies \[[@B11]-[@B13],[@B16]\] give credence to the view that the inability of the lung to repair after ALI is the result of a persistent inflammatory stimulus that ultimately leads to an unfavorable outcome \[[@B17]\]. Fueled by this prospect, Meduri and coworkers \[[@B18]\] indicated a consistent, efficient and independent predictive value for IL-1β and IL-6 serological concentrations over time in a small cohort of patients with severe ARDS. They generated ROC curve analysis and demonstrated a clear superiority of inflammatory cytokines in monitoring disease activity over commonly applied clinicophysiologic parameters. However, this study exhibited substantial weaknesses including the retrospective analysis, the small sample size and the overlapping results between survivors and non-survivors. These observations coupled with the heterogeneity of the disease and the evidence that elevated serum cytokines may reflect the increased production or decreased clearance and not the disease activity pose major limitations to the aforementioned findings. In another study, Agouridakis et al. \[[@B19]\] evaluated both the prognostic and predictive significance of IL-2 and IL-15 for the development and outcome of patients at risk who developed ARDS or patients at risk who never developed ARDS, respectively. They applied ROC curve analysis and showed an excellent predictive value of cytokine plasma levels in terms of specificity and sensitivity compared to those observed in BALF. The most remarkable ascertainment of this study was the emergence of the discriminative usefulness of elevated IL-2 and IL-15 serological concentrations in patients with ARDS or at risk for ARDS. Nevertheless, the small number of patients enrolled combined with the lack of serial measurements throughout the clinical course of the disease and the causal diversity of the syndrome render major uncertainty to these findings. On the contrary with other analyses \[[@B16],[@B18]\], Lesur et al \[[@B20]\] found lower blood IL-2 levels in patients with ARDS compared to those that never developed the syndrome. In addition, evidence of this study regarding the strong association of early low serum IL-2 levels with the patients\' survival corroborated earlier findings \[[@B16]\]. Potential criticism of this study include the small number of patients recruited, the absence of multiple time-point evaluation of the cytokine plasma concentrations and more importantly opposite and disproportional fluctuations of IL-2 content in serum and BALF in patients with or without ARDS. The role of several inflammatory cytokines in monitoring the disease activity and predicting the survival in patients with ARDS has aroused increasing attention the past decade. One of the most intriguing aspects of the application of these biomarkers in the daily clinical practice is the early detection of patients admitted to the intensive care unit (ICU) that will develop ARDS. This approach will allow anti-inflammatory and other supportive treatments to be evaluated or eventually modified before they have failed. Predictive levels of inflammatory cytokines (IL-1, IL-2, IL-6, IL-8) for ARDS development in at risk patients have been extensively reported with controversial results \[[@B16],[@B17],[@B21],[@B22]\]. The importance of considering inflammatory constituents of serum in patients at risk for ARDS was initially raised by Parsons et al \[[@B23]\]. Authors conducted a large prospective analysis and demonstrated that although immunological parameters (IL-1ra and IL-10) were elevated in patients at risk for ARDS and exhibited a remarkable association with the disease outcome, none of these could predict the development of the syndrome. These observations were extended by the results of Takala et al. \[[@B24]\] who showed that serum levels of inflammatory mediators albeit their persistent elevation in patients with unresolving ALI, preceding its clinical diagnosis, were of poor discriminative value in patients with ALI that did or did not develop ARDS. Nevertheless, these findings provided us with useful knowledge about the inflammation marker profile on the days preceding diagnosis of ARDS, indicating a potential relation of sustained inflammatory response with a poor outcome. With this aim in mind, Bouros et al. \[[@B25]\] measured prospectively a slew of cytokines in the serum and BALF in ICU patients to identify predictive factors for the course and outcome of ARDS. The most remarkable result of this analysis was that almost all serum molecules studied showed a high prognostic value in assessing the outcome in patients with or at risk for ARDS. However, laboratory parameters failed to prove a positive correlation with the prediction of ARDS development evidence consistent with earlier studies \[[@B23],[@B24]\]. Moreover, major caveats that should be taken under consideration include the limited number of patients, the lack of sufficient follow-up serum data and the marked causal heterogeneity of the syndrome that could be a reason for the contradictive results reported in previous studies \[[@B20],[@B23]-[@B25]\]. Further prospective studies with sufficient statistical power are required to validate these results and ameliorate the predictive role of circulating inflammatory mediators in patients at risk for ARDS. Although the role of cytokines in the pathogenesis of ARDS has been extensively investigated, their importance in the differential diagnosis has not been clearly defined. It is widely accepted that numerous insults may lead to ARDS following a common pathway. Furthermore, a variety of conditions including severe pneumonia imitate clinical and radiological manifestations of ARDS and thereby it is often difficult to differentiate them. However, this would be a fruitful application because the treatment of these conditions differs considerably. Many groups of investigators have attempted to produce a discriminative systematic inflammatory profile and although much good work has been done towards this direction, the results still remain controversial. Schutte and co-workers \[[@B26]\] provided us with a really well done and heavily informative paper concerning the systemic cytokine profile in patients with ARDS, severe pneumonia and cardiogenic pulmonary oedema. Authors found remarkably and consistently elevated serum levels of IL-6 and IL-8 in ARDS and/or severe pneumonia, differentiating these entities from cardiogenic pulmonary oedema. Nevertheless, they were unable to separate the various entities of ARDS and states of severe pneumonia based solely on alterations in the immunomodulatory pattern. To streamline these observations, Bauer et al. \[[@B27]\] tested the potential of inflammatory markers (TNFa, IL-1β, IL-6) to differentiate between these two diseases. Results in harmony with the previous study \[[@B26]\] demonstrated higher TNFa serological concentrations in patients with ARDS from the remaining populations. However, they revealed the ability of immunological parameters to associate better with the degree of lung injury rather than clarify its specific aetiology. No clear relationship between serological data and patients\' survival was observed. In addition, this data exhibits major limitations, including the absence of uniform methodology (use of corticosteroid treatment in some of the patients), serial measurements and the lack of knowledge regarding serum alterations in other components of the inflammatory network (Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). Other serological parameters ============================ Markers of endothelium activation (Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}) ---------------------------------------------------------------------------------------------------- ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Studies measuring markers of endothelium activation in patients with or at risk for ARDS ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Investigator** **Patients Controls** **Biomarker / Summary** **ROC curve analysis Cut-off values** **Specificity -- Sensitivity PPV-NPV** **Limitations** ------------------------- ----------------------- -------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------- ---------------------------------------- ---------------------------------------------------------------------------- Donnelly et al. ^39^ 82 at risk\ E-selectin levels were not correlated with ARDS development and patients\' mortality. L-selectin levels exhibited a significant prognostic value No Not estimated Heterogeneity of studied population (trauma-sepsis)\ 14 ARDS^a^\ No ROC curve analysis / cut-off levels No serial measurement 62 controls Boldt et al. ^40^ 50 at risk Constantly lower E-selectin, ICAM-1 and VCAM-1 levels in survivors experiencing polytrauma than in nonsurvivors No Not estimated Small sample size\ Causal diversity of patient group\ No definitive association with patients\' mortality\ No ROC curve analysis / cut-off levels Cowley et al. ^41^ 40 SIRS\ Superiority of E-selectin plasma levels in predicting organ dysfunction and death patients with SIRS comparing to ICAM-1 No Not estimated Small number of patients\ 85 controls Causal diversity of patients studied\ No ROC curve analysis / cut-off levels Sessler et al. ^43^ 25 at risk\ Association of elevated ICAM-1\ Yes\ Not reported Small sample size\ 12 controls Sequential plasma levels with the severity of shock ICAM-1: 715 ng/ml (predicting survival) Heterogeneity of studied population\ Inconclusive association with disease Severity Kayal et al. ^44^ 32 at risk\ Cut off values of E-selectin, ICAM-1 and VWF serum levels predicted survival outcome Yes\ E-selectin: 73-80-67-85%\ Small number of patients\ 9 controls E-selectin: 128 ng/ml\ ICAM-1: 80-90-75-92%\ Most of the patients developed secondary ALI ICAM-1: 715 ng/ml\ VWF: 87-8080-87% VWF: 717% Agouridakis et al. ^45^ 23 ARDS^a^\ TNFa, IL-1, ICAM-1, VCAM-1\ Yes\ For ARDS development\ Small sample size\ 42 at risk ICAM-1 and VCAM-1 showed a high NPV for ARDS development\ TNFa: 325 pg/ml\ TNFa: 62-75-38-89%\ No serial measurement\ Correlation with the disease outcome\ IL-1: 225 pg/ml\ IL-1: 58-88-39-94%\ None of the studied markers was an independent factor for ARDS development None of the studied markers was an independent factor for ARDS development ICAM-1: 300 pg/ml\ ICAM-1: 69-75-42-90%\ VCAM-1: 260 pg/ml VCAM-1: 73-88-50-95% -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ALI: Acute Lung Injury, ARDS: Acute Respiratory Distress Syndrome, ICAM-1: Intercellular Cell Adhesion Molecule-1, IL: Interleukin, ROC: Receiver Operating Characteristic, SIRS: Systematic Inflammatory Response Syndrome, TNFa: Tumor Necrosis Factor-alpha, VCAM-1: Vascular Cell Adhesion Molecule-1, VWF: von Willebrand factor antigen, ^a:^Use the American European Consensus Conference definitions ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Studies measuring markers of endothelium activation in patients with or at risk for ARDS ::: -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Investigator** **Patients Controls** **Biomarker / Summary** **ROC curve analysis Cut-off values** **Specificity -- Sensitivity PPV-NPV** **Limitations** ----------------------- ----------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ---------------------------------------- -------------------------------------------------------------------------------------------------- Rubin et al. ^47^ 45 at risk Elevated plasma VWF is an early predictor of ALI in nonpulmonary sepsis syndrome Yes\ 77-87-80% Small sample size 25% of patients had already lung injury at the time sepsis was diagnosed\ VWF: 450% Exclusion of patients who developed\ ALI from a primary pulmonary source VWF levels measured by an old assay Ware et al. ^48^ 51 ALI/ ARDS^a^\ VWF is an independent predictor of hospital mortality in patients with ALI No\ 91-44-83-62% Inadequate sample volume\ 4 controls VWF:450% Heterogeneity of studied population\ No ROC curve analysis Ware et al. ^49^ 559\ Significant correlation of elevated VWF plasma levels with mortality, duration of unassisted ventilation and organ failures. No differences of VWF levels between septic and non septic patients No Not estimated Not definitive association with patients\' mortality\ ALI/ ARDS^a^ Lack of knowledge regarding the cellular source and the mechanisms of elevated VWF serum levels\ No ROC curve analysis / cut-off values Moalli et al. ^50^ 35 at risk\ VWF levels were higher in ARDS compared with at risk\ No Not reported Limited number of patients\ 10 ARDS\ VWF levels are not helpful in predicting ARDS development No ROC curve analysis / cut-off values 9 controls Moss et al. ^51^ 96 at risk VWF is not predictive of development of ARDS Yes\ 47-70%\ Causal diversity of patients studied\ VWF:273%\ 52-64% No definitive relation with disease severity VWF:399% Sabharwal et al. ^52^ 22 ARDS\ No significant association of VWF blood levels with patients\' mortality No Not estimated Small sample size\ 21 at risk Retrospective study\ No ROC curve analysis / cut-off values Bajaj et al. ^53^ 18 ARDS\ Serum VWF levels were non-useful markers for predicting ARDS in at risk patients Yes\ 71-62-34% Limited number of patients\ 15 at risk\ VWF: 300% No serial measurement\ 27 controls Coexisting multisystem organ failure\ Heterogeneity of studied population Moss et al. ^54^ 55 at risk\ ICAM-1, E-selectin, VWF\ No Not estimated Small sample size\ 14 ARDS\ Degree of endothelial activation varied in patients at risk for ARDS from different etiologic factors Heterogeneity of studied population 11 controls -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ALI: Acute Lung Injury, ARDS: Acute Respiratory Distress Syndrome, ICAM-1: Intercellular Cell Adhesion Molecule-1, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, VCAM-1: Vascular Cell Adhesion Molecule-1, VWF: von Willebrand factor antigen ^a:^Use the American European Consensus Conference definitions ::: The pathophysiologic sequence characterizing ALI involves apart from cytokine, free radical, proteases and aracidonic acid metabolites release, the endothelial and neutrophil activation which initiate a cascade of leukocyte-endothelium interactions and adhesions. This is followed by transendothelial migration of neutrophils and release of their cytotoxic products, ultimately resulting to microvascular and tissue injury \[[@B28]\]. Adhesion of neutrophils to the endothelium is regulated by at least three adhesion molecule families including selectins (E, L and P), integrins and the immunoglobulin superfamily (intercellular adhesion molecule- ICAM-1 and vascular cell adhesion molecule-VCAM-1) and by chemotactic signals \[[@B29],[@B30]\]. Initial interactions of leukocytes and the endothelium are mediated by members of the selectin family inducing (loose) contact with the endothelium also known as *rolling*, followed by firm adhesion requiring members of the integrin (β2) and immunoglobulin family (ICAM-1) \[[@B31],[@B32]\]. In recent years, soluble isoforms of some of these molecules {soluble-(s)-E-selectin, sICAM-1, sVCAM-1} have been detected in the circulating blood under various inflammatory conditions \[[@B33]-[@B35]\]. Mechanisms that could potentially explain an increase in circulating adhesion molecules include cytokine-induced (IL-1, TNFa) overexpression by the endothelial cells, increased proteolytic cleavage of endothelial-bound adhesion molecules secondary to endothelial damage or both \[[@B33],[@B35]\]. One attractive feature of these molecules and mostly E-selectin is that since their expression is almost restricted to stimulated endothelial cells \[[@B32]\] their presence in serum should potentially reflect the state of endothelium in disease and subsequently the disease severity in ALI. Other potential markers of endothelial cell injury that were delineated to shed further light into the pathophysiologic process of ALI include von-Willebrand factor antigen (VWF), a macromolecular antigen that is produced predominantly by endothelial cells and to a lesser extent by platelets and megakaryocytes \[[@B36]\]. Endothelial perturbation (as in at risk state) or injury (as in ARDS) results to the release of VWF from preformed stores into the circulation \[[@B37],[@B38]\]. Therefore, it appears that circulating VWF concentrations may serve as a suitable predictive marker for development of ARDS in patients at risk. So far, the potential usefulness of adhesion molecules and other markers of endothelial cell damage in reflecting the severity of endothelial damage and predicting the development or the final outcome of the disease is a subject of ongoing controversy. One of the first and most informative studies addressing this important issue was conducted by Donnelly et al \[[@B39]\]. Authors demonstrated in a large cohort of patients at risk for ARDS, that mean circulating levels of sE-selectin were not correlated with subsequent ARDS development and patients\' mortality. However, low values of sL-selectin exhibited a significant prognostic value. In contrast, Boldt et al. \[[@B40]\] studying the behaviour over 5 d of adhesion molecules (sE-selectin, sICAM-1 and sVCAM-1) in subjects experiencing polytrauma found constantly higher levels in nonsurvivors. In accordance to these findings, Cowley and colleagues \[[@B41]\] showed a superiority of sE-selectin plasma levels in predicting organ dysfunction and death in a group of patients with systemic inflammatory response (SIRS) comparing to sICAM-1 peripheral concentrations. This indicated that measurement of adhesion molecules could serve to advantage in the management of patients with sepsis. Nonetheless, the heterogeneity of patients studied (septic shock and polytraumatic) may justify these controversial results, since E-selectin expression has been found much greater in septic than in traumatic shock in experimental models \[[@B42]\]. Moreover, the relationship between the consequences of sepsis (organ failure, mortality) and blood levels of potential markers of endothelial-cell activation was strengthened by Sessler and co-workers \[[@B43]\]. Results from this study focusing on sICAM-1 sequential plasma levels, were suggestive of a strong association between the severity of shock (as determined by the presence of hypotension and the requirement of vasoactive drugs) and the circulating concentrations of the marker. The aforementioned observations were further confirmed by the study of Kayal et al \[[@B44]\]. Cut off values for three markers of endothelial activation were determined prospectively by ROC curve analysis and clearly predicted survival outcome with high sensitivity and specificity in a limited number of at-risk patients with secondary ALI. Despite that the role of soluble adhesion molecules in other inflammatory conditions strongly associated with ARDS \[[@B40]-[@B44]\] is well known, their value as markers of the disease progression and mortality has not been extensively studied in ARDS patients. To streamline these observations, Agouridakis et al. \[[@B45]\] scrutinized the role of two adhesion molecules (ICAM-1, VCAM-1) in parallel with proinflammatory cytokines in predicting the ARDS development and relating to the disease outcome. None of the studied mediators was found to be an independent factor for ARDS development, whereas both groups of molecules exhibited a considerable negative predictive value for ARDS development both in serum and BALF. Additionally, ROC curve analysis showed a clear superiority of plasma parameters in correlating with the disease outcome compared with BALF molecules. Further studies with serial BALF and serum measurements should be designed to elucidate the exact role of these markers over time in reflecting the disease behaviour and predicting the likelihood of progression. Elevated circulating concentrations of VWF in patients with ALI/ARDS were first reported in 1982 \[[@B46]\]. Their potential significance in predicting ARDS development was first demonstrated by Rubin and colleagues \[[@B47]\] who found that increased plasma levels of this marker exhibited a high predictive value both for the development of ARDS and for identifying patients with nonpulmonary sepsis who were unlikely to survive. However, authors did not use the uniform criteria for the definition of ARDS and risk state \[[@B3]\], evidence that poses major limitations to the results of the study. Furthermore, in an aforementioned study, Kayal et al. \[[@B44]\] in parallel with other findings reported a marked and independent association of circulating VWF with the disease severity as assessed by other commonly applied clinical variables. These findings were further supported by a single-center study from Ware et al \[[@B48]\]. They conducted the first comparative study of VWF concentrations in both plasma and edema fluids of patients with early ALI from a variety of causes and reported that serum VWF levels were an independent predictor of hospital mortality and were associated with longer duration of mechanical ventilation. Potential criticisms of this study was the implementation of high tidal volume ventilation that possibly increased systemic endothelial activation, the small sample size, the lack of sequential measurement and the fact that the studied biomarker appeared not to be endothelial-specific since it is produced in small amounts by platelets \[[@B36]\]. To streamline these observations and ameliorate potential hardships the same group of authors {Ware et al. \[[@B49]\]} carried out a multicenter study of 559 patients with ALI and ARDS which was recently published. In accordance with earlier studies \[[@B48]\], a significant correlation of elevated VWF plasma levels with adverse outcomes, including mortality, duration of unassisted ventilation and organ failures was pointed out. Intriguingly, authors demonstrated for the first time a negative association between markers of endothelial activation and presence or absence of sepsis, supporting the hypothesis that ALI might be an independent cause of systemic endothelial activation and injury. Finally, in the same study no modulation of plasma VWF concentrations by protective mechanical ventilation was observed. Despite the remarkable power of the presented findings, there are substantial weaknesses that deserve further investigations including the inconclusive analysis of the plasma VWF levels associated with patients\' mortality, the lack of definite knowledge regarding the source of VWF production, and the mechanisms leading to increased peripheral concentrations since the latter also could reflect decreased clearance from the circulation. Subsequent data derived from other studies \[[@B50]-[@B54]\] was rather contradictive and controversial. Even though, Moalli et al. \[[@B50]\] found a poor predictive value of serum VWF levels for the development of ARDS in a group of at risk patients, the biomarker concentrations were correlated with the disease severity,. Similarly, Moss et al. \[[@B51]\] plotted ROC curves and concluded that in patients at risk for ALI/ARDS from multiple causes, serum VWF levels failed to reliably discriminate which patients would develop ARDS. The evidence was further validated by Sabharwal et al \[[@B52]\]. Authors conducted the first study comparing plasma levels between survivors and nonsurvivors in a group of patients both at risk for and with established ARDS and observed no significant association of VWF blood levels with patients\' mortality ;predictive value of the marker was not reported. In agreement with the previous study, a study from the same group of scientists {Bajaj et al. \[[@B53]\]} using standard criteria for the definition of ARDS and at risk state \[[@B3]\] demonstrated the inability of three endothelial-specific proteins including VWF to predict the progression of ARDS in at risk patients. Nonetheless, major caveats that should be addressed include the fact that many at-risk patients had already some degree of ALI, the lack of serial measurement that could potentially show a trend towards prediction of ARDS development and the causal diversity of patients examined that could possibly affect the results of the study. The latter limitation was addressed by Moss et al. \[[@B54]\] who established that the degree of endothelial activation as determined by the plasma levels of VWF (higher in subjects with sepsis than patients with trauma) is not uniform in all patients at risk for developing ARDS. Accumulated evidence from the preceding studies suggest that the etiologic diversity of patients enrolled renders major uncertainty to the reliability of the results and highlights the necessity for further prospective studies using standard criteria for the definition of ARDS and analyzing well defined and uniform group of at risk patients in order to produce knowledge of high scientific rigidity. Direct comparison of the different studies is difficult and in a way meaningless because of the use of varying definitions for ARDS and at-risk patients as well as the inclusion of different patient populations, in which of them some degree of ALI was probably already present (Tables [3a](#T3){ref-type="table"} and [3b](#T3){ref-type="table"}). Markers of neutrophil activation (Table [6](#T6){ref-type="table"}) ------------------------------------------------------------------- ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Studies measuring markers of neutrophil activation and ferritin in patients with or at risk for ARDS ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Investigator** **Patients Controls** **Biomarker / Summary** **ROC curve analysis Cut-off values** **Specificity -- Sensitivity PPV -- NPV** **Limitations** ---------------------- ----------------------- -------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- -------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Pugin et al. ^60^ 31 at risk\ IL-8, MMP-2, MMP-9\ No Not estimated Measurement of circulating proinflammatory cytokines without the appreciation of their inhibitors or receptor antagonists is misleading mainly due to a possible neutralization\ 23 ARDS^a^ Plasma levels of inflammatory activity are not useful markers in differentiating permeability from hydrostatic pulmonary edema Small number of patients\ No ROC curve analysis / cut-off values Amat et al. ^64^ 21 ARDS^a^\ Strong association of LTB4 and IL-8 serum levels with the patients\' survival. Yes\ LTB4 + IL-8: 88-70-85-75% (markers of mortality rate)\ Small sample size\ 14 at risk LTB4: 14 pmol/ml\ LTB4 : 85-72-20-98% (marker of ARDS development) Lack of adjustment with the disease severity\ IL-8: 150 pmol/ml Inability of LTB4 plasma levels to be an independent predictive marker Connelly et al. ^66^ 75 at risk\ Serum ferritin is a sensitive and specific predictor of ARDS development Yes\ 71-83-86-67%\ Limited number of ARDS patients\ 8 ARDS^a^ Ferritin (male):270 ng/ml\ 90-60-82-75% Heterogeneity of studied population Ferritin (female): 680 ng/ml Sharkey et al. ^67^ 42 at risk\ Correlation of ferritin plasma levels with the development of ARDS, multiple organ failure and severity of lung injury Yes\ 64-73-75-62%\ Inadequate sample volume\ 16 ARDS^a^ Ferritin (male):270 ng/ml\ 92-60-95-75% Not specific cut-off values\ Ferritin (female): 680 ng/ml Elevated serum levels may reflect a systemic response to a risk factor ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ARDS: Acute Respiratory Distress Syndrome, IL: Interleukin, LTB4: Leukotriene B4, MMP-Metalloproteinase, NPV: Negative Predictive Value, PPV: Positive Predictive Value, ROC: Receiver Operating Characteristic, ^a:^Use the American European Consensus Conference definitions ::: Generally, it is strongly believed that ARDS arises as a result of tissue injury secondary to sequestration of inflammatory cells, tissue invasion, and secretion of cytotoxic products. Neutrophils have received much attention as key part of this process. Although ARDS has been described in neutropenic patients \[[@B16],[@B55]\] there is increased evidence implicating neutrophils in most cases of ARDS. They have been reported by several studies \[[@B56],[@B57]\] to exert an important role in the early phase of ALI characterized by architecture remodeling, surfactant and epithelial toxicity. They use a wide array of enzymes during the process of transmigration through biological membranes such as alveolar-capillary barrier \[[@B58]\]. These enzymes include, metalloproteinases (MMPs) such as MMP-9 also called gelatinase B which is secreted from preformed neutrophil granules in response to a variety of stimuli including proinflammatory cytokines (IL-8, TNFa). MMP-9 is secreted as a zymogen, and then activated by a variety of other proteases such as elastase, and plays a crucial role in digesting basement membranes \[[@B58],[@B59]\]. Therefore, it has been speculated that metalloproteinases probing aspects of the inflammatory response could be utilized as markers of neutrophil activation and subsequently to reflect disease activity and severity, shedding further light into the pathogenesis of ARDS. Fueled by this prospect, Pugin et al. \[[@B60]\] compared the concentrations of proinflammatory cytokines and collagenases in serum and pulmonary oedema fluids in a small group of patients with ARDS and hydrostatic oedema from congestive heart failure. Authors concluded that elevated pulmonary oedema levels of these mediators could differentiate between these conditions, whereas plasma levels of proinflammatory and metalloproteinase activity proved to be of poor discriminative value. The latter observation mirrors the hypothesis that the inflammatory response characterizing ARDS patients is well compartmentalized, with little spillover into the circulation and that the measurement of circulating proinflammatory cytokines without the appreciation of their inhibitors or receptor antagonists is misleading mainly due to a possible neutralization. To gain a more comprehensive understanding on the role that neutrophils exhibit during the inflammatory cascade resulting to ALI, investigators scrutinized the utility of other chemotactic agents including leukotrienes (LTs). LTs (B4, C4, D4, E4) exert a synergistic role with IL-8 in the neutrophil influx and activation leading to a massive recruitment of neutrophils and to a catastrophic inflammatory response. Their BALF levels have been found elevated in patients with ARDS and their involvement in the alterations of microvascular permeability correlated with the accumulation of pulmonary oedema has been suggested \[[@B61]-[@B63]\]. Moreover, Amat et al \[[@B64]\] utilizing ROC curve analysis demonstrated that LTB4 plasma levels could serve as a valuable predictive marker of ARDS in terms of specificity and sensitivity. In the same study, authors performed serial measurement and reported a strong association of both LTB4 and IL-8 peripheral concentrations with the patients\' survival. However, the small number of patients enrolled, the lack of adjustment with the disease severity and the inability of LTB4 plasma levels to be an independent predictive marker arise major concerns whether they could monitor disease behaviour and predict ARDS development in at risk patients (Table [6](#T6){ref-type="table"}). Ferritin (Table [6](#T6){ref-type="table"}) ------------------------------------------- Ferritin is a 480-kDa iron-storage protein that sequesters iron in the ferric (Fe^3+^) state. It has been speculated that ferritin may serve as a crucial antioxidant mediator because free iron enhances the formation of highly toxic hydroxyl radicals from superoxide anion and hydrogen peroxide. On the other hand, oxidative stress is a condition commonly seen in disorders at risk for ARDS development such as sepsis. Hence, ferritin-derived iron may aggravate oxidative damage in critically ill patients, contributing to the pathologic abnormalities encountered in ARDS. Furthermore, proinflammatory cytokines such as IL-8, IL-6 and TNFa which are increased and presumably participating in the pathogenetic derangements of ARDS have been suggested to promote ferritin synthesis \[[@B39],[@B65]\]. Thereby, it can be concluded that elevated ferritin levels could result from oxidative stress, proinflammatory cytokines and the degree of lung injury, all conditions characterizing the pathogenesis of ARDS and subsequently can be used as prognostic and monitoring tool reflecting the likelihood of ARDS development and the disease severity. The first study attempted to prove such correlation was conducted by Connelly et al \[[@B66]\]. They plotted ROC curves to estimate the utility of ferritin levels as prognostic factors and produced clinically useful cut-off points which could predict the development of ARDS with high sensitivity, specificity, negative and positive predictive value, both in male and female predominantly septic subjects. However, the heterogeneity of the etiologic factors resulting to ARDS development in at risk patients renders major uncertainty to the rigidity and reliability of the results. To ameliorate this hardship, the same group of authors {Sharkey et al. \[[@B67]\]} generalized and extended the latter results in a homogeneous group of at-risk patients with multiple trauma demonstrating a strong correlation of initial ferritin plasma levels with the development of ARDS and multiple organ failure. In addition, an association of serum ferritin levels with the severity of lung injury as well as other markers of endothelial activation was also noted supporting the premise that elevated levels could reflect the inflammatory status encountering in ALI. Nonetheless, authors failed to detect specific predictive cut-off values suggesting that circulating concentrations of this biomarker are unable to predict per se the progression to ARDS. A possible explanation could arise from the hypothesis that elevated levels of this marker must reflect a systemic response to a risk factor, which may prove to reduce its specificity (Table [6](#T6){ref-type="table"}). Lung epithelium-specific proteins (Table [7](#T7){ref-type="table"}) ==================================================================== ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Studies measuring lung-specific proteins in patients with or at risk for ARDS ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Investigator** **Patients Controls** **Biomarker / Summary** **ROC curve analysis Cut-off values** **Specificity -- Sensitivity Diagnostic accuracy** **Limitations** ---------------------- ----------------------- ------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------- ---------------------------------------------------- ------------------------------------------------------------------------------------------------------- Doyle et al. ^81^ 15 ARDS^a^\ SP-A is an acute indicator of lung function and alveolocapillary membrane injury No Not estimated Small number of patients\ 10 at risk\ No ROC curve analysis / cut-off values\ 10 controls No definitive relation with disease severity Doyle et al. ^82^ 22 ARDS^a^\ Superiority of SP-B compared to SP-A plasma levels as a marker of lung function and alveolocapillary membrane injury No Not estimated Only 3 case-control studies\ 10 at risk\ Inadequate sample size\ 33 controls Lack of adjustment with disease behaviour\ No ROC curve analysis / cut-off values Greene et al. ^83^ 41 ARDS^a^\ SP-A, SP-B, SP-D\ Yes\ Poor predictive value\ Limited number of patients\ 22 at risk\ Serum changes found to be neither sensitive nor specific in predicting the onset of ARDS and discriminating survivors from non-survivors. Not reported Low specificity/sensitivity Serial measurements for a short period of time/ Lack of serial measurement for the most severe forms\ 35 controls Heterogeneity of studied population\ Poor predictive value for serum levels Cheng et al. ^84^ 36 ARDS^a^\ SP-A levels were associated with severity of clinical lung injury and with disease outcome No Not estimated Small sample size\ 2 ALI Causal diversity of studied population\ No serial measurement Greene et al. ^85^ 51 at risk\ SP-A levels are predictive for at risk patients who developed ARDS from sepsis and aspiration but not trauma No Not estimated Small sample size\ 26 ARDS^a^\ No ROC curve analysis / cut-off levels 16 controls Bersten et al. ^86^ 54 at risk\ SP-B but not SP-A cut-off plasma levels predict ARDS development, particularly in at-risk patients suffering a direct lung injury Yes\ 78-85-85-78% Small number of patients\ 9 controls SP-B: 4.994 ng/ml Limited follow-up serum data\ Most of patients had already lung injury\ Exclusion of milder at risk patients Eisner et al. ^87^ 565\ SP-A, SP-D\ No Not estimated Only 2 serial measurements\ ALI/ ARDS^a^ Attenuation of SP-D plasma levels by lower volume ventilation strategies Heterogeneity of studied population Potential selection bias\ No ROC curve analysis / cut-off levels Ishizaka et al. ^95^ 35 at risk\ Association of optimal cut-off values of KL-6 serum levels with patients\' mortality Yes\ 100-87% Inadequate sample volume\ 27 ARDS^a^\ KL-6: 253 U/ml Heterogeneity of studied population 21 controls Sato et al. ^96^ 28 ARDS^a^\ Association of KL-6 serum levels with variables of lung injury severity and with mortality rates\ No Not estimated Small sample size\ 10 controls No correlation with ventilation strategies Heterogeneity of studied group\ No serial measurement\ No ROC curve analysis / cut-off levels\ Diversity of ventilatory treatment ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: ALI: Acute Lung Injury, ARDS: Acute Respiratory Distress Syndrome, BAL: Bronchoalveolar Lavage, KL-6: Krebs von den Lungen-6, ROC: Receiver Operating Characteristic, SP: Surfactant Protein, ^a:^Use the American European Consensus Conference definitions ::: Beyond other important functions, the lung epithelium produces complex secretions, including mucus blanket, surfactant proteins, as well as several proteins important for host defense \[[@B68]\]. Sampling the epithelial lining fluid by bronchoalveolar lavage (BAL) represents the common means of studying the proteins secreted by the lung epithelium and investigating their alterations in lung disorders \[[@B69]\]. However, the past fifteen years pioneering studies \[[@B70]\] showed the presence of these proteins in the bloodstream as well, even though in small amounts. Because these proteins are mainly, if not exclusively secreted within the respiratory tract, their occurrence in the vascular compartment can be explained by several hypothetical mechanisms including, leakage from the lung into the bloodstream, increased production by the alveolar type II cells or diminished clearance rates from the circulation \[[@B68]\]. Surfactant-associated Proteins ------------------------------ Pulmonary surfactant is a complex and highly surface active material covering the alveolar space of the lung. Biochemically, surfactant is a molecular mixture composed mainly of structurally heterogeneous phospholipids. A major function of pulmonary surfactant is to reduce the surface tension at the air-liquid interface of the alveolus, thereby preventing alveolar collapse on expiration. It has also been demonstrated that the surfactant contains specific proteins \[[@B71]\]. Four surfactant-specific proteins with different structural and functional properties have so far been identified. They were named surfactant protein-(SP)-A, SP-B, SP-C and SP-D according to the chronologic order of their discovery \[[@B72]\] and have been divided in two distinctive groups, the low-molecular-weight hydrophobic SP-B and SP-C and the high-molecular-weight-hydrophilic SP-A and SP-D. The latter belong to the collectin subgroup of the C-type lectin superfamily and are produced by two types of non-ciliated epithelial cells in the peripheral airway, Clara cells and alveolar type II cells. Studies have demonstrated that SP-B and SP-C seem to play an essential role for the adsorption of phospholipids to the air-water interface resulting to a stable phospholipids film and for the dynamic surface-tension-lowering properties \[[@B73]\]. Additional functions of the alveolar surfactant system include prevention of alveolar edema \[[@B74]\] and a pronounced influence, especially of the collectins SP-A and SP-D in the innate immune system of the lung \[[@B75],[@B76]\] and have been used as useful markers for confirming the diagnosis and evaluation of disease activity of various ILDs since they reflect the epithelial damage and turnover \[[@B77]\]. Thus, it has been speculated that alterations of SPs in biological fluids could serve as valuable markers of the severity of the lung injury or clinical outcome in ARDS patients. Most of our knowledge regarding changes in SP concentrations that occur in patients with or at risk for ARDS and their value in reflecting the disease severity or the likelihood of ARDS development comes from BAL studies. Several reports in the literature have demonstrated the occurrence of low SP-A levels in BALF of patients with ARDS following trauma \[[@B78],[@B79]\] coupled with a strong relation of this biological marker to the severity of endothelial damage \[[@B80]\]. Moreover, a potential value of SP-A plasma levels in discriminating patients with ALI of various etiologic factors has also been shown \[[@B79]\]. On the other hand, little is known about changes in peripheral concentrations of surfactant-associated proteins in patients with ALI and whether these alterations can serve as markers of injury to the epithelial and endothelial barriers in the lungs. Doyle et al. \[[@B81]\] documented elevated circulating concentrations of SP-A in patients with ARDS and in those with acute cardiogenic pulmonary edema possibly resulting from increased alveolocapillary permeability due to excessively high pulmonary capillary pressures. In the same study, blood SP-A levels were inversely associated with blood oxygenation and static respiratory system compliance. These results were fully confirmed by the same group of authors {Doyle et al. \[[@B82]\]} who also illustrated a clear superiority of SP-B compared to SP-A plasma levels as a marker of lung function and alveolocapillary membrane injury. Another study by Greene et al. \[[@B83]\], evaluated the differences that occur in SPs in BALF and serum of a relatively small cohort of patients at risk for ARDS and during the course of the syndrome. Authors demonstrated that only SP-A and SP-D BALF levels were strongly related to outcome and likelihood of disease progression whereas serum changes found to be neither sensitive nor specific in predicting the onset of ARDS and discriminating survivors from non-survivors. These data were further confirmed by a small cohort observational study by Cheng et al \[[@B84]\]. Even though authors reported an association of elevated SP-A plasma levels with a high degree of lung injury, they failed to extend this correlation with the disease mortality. Moreover, serum SP-D levels exhibited weak relation to the disease severity. It should also be noted that the aforementioned results present low statistical power due to the limited number of patients, the causal heterogeneity of the studied group and the absence of serial measurement and therefore no meaningful outcome can be excluded. In harmony with the latter results, Greene et al. \[[@B85]\] found that plasma SP-A was weakly predictive for ARDS development in septic patients and were unable to detect at risk trauma patients that developed the syndrome. Additionally, authors raised the crucial issue whether circulating SPs can reflect pathophysiologic differences between direct and indirect causes of ARDS and subsequently detect biologic changes early after an insult. From a pragmatic clinical perspective the most important question to be answered is which ICU patients requiring ventilatory assistance will develop ARDS. To do so Bersten et al. \[[@B86]\] generated ROC curve analysis and identified practical thresholds for SP-B plasma levels that could be clinically useful in predicting ARDS development, particularly in at-risk patients suffering a direct lung injury. In consistency with earlier studies \[[@B83]-[@B85]\] SP-A blood levels added no significant information on the disease prognosis. Further, an increase of circulating SP-B concentrations was documented on study entry, before changes in commonly applied clinical variables for the assessment of lung injury become apparent. These findings emphasize the usefulness of surfactant-associated proteins for the early detection of ARDS pathophysiologic alterations preceding changes in clinical parameters such as respiratory dysfunction. Arguments that can be made include the small sample size, the limited sequential measurements and the exclusion from the study recruitment of at risk patients with milder pulmonary dysfunction. These caveats coupled with the evidence that a considerable number of patients studied had already lung injury, pose major limitations to the predictive capacity of SP-B plasma levels and raise the necessity for larger prospective studies. The only so far large multicenter randomized controlled trial was performed by Eisner et al. \[[@B87]\] who estimated the prognostic value of SP-A and SP-D levels in an overall of 565 patients with early ALI/ARDS. Authors conducted the first study with adequate statistical power to examine the impact of SPs on mortality and other clinical variables and clearly demonstrated a strong linkage of elevated SP-D levels with worse clinical outcomes such as greater risk of death, fewer ventilator- free and organ failure-free days. One of the most remarkable ascertainments of this study was the attenuation of SP-D plasma levels by the lower volume ventilation strategies which reduces patients\' mortality postulating for the first time a significant association of biological parameters with therapeutic approaches and subsequently emphasizing the role of this mediator as a marker of the disease severity and prognosis. However, this study exhibited substantial weaknesses including the lack of sufficient serial measurements, a potential selection bias of patients recruited and the diversity of predisposing factors for ARDS development. These observations are not to diminish their value as prognostic and monitoring tools but to highlight the need for further confirmation studies using independent and well-defined populations of ALI/ARDS patients. Mucin-associated Antigens ------------------------- Mucins are major components of the mucus layer covering the airway epithelium. They consist of high-molecular-weight glycoproteins belonging to a broad family of mucin peptides \[[@B68]\]. Mucins are either associated with membranes or secreted at the surface of the respiratory tract \[[@B68]\]. Krebs von den Lungen-(KL)-6 is mainly associated with cellular membranes. It was initially described by Kohno et al. \[[@B70]\] as a high-molecular-weight glycoprotein and was classified as human MUC1 mucin. Immunohistochemistry has mainly detected KL-6 in alveolar type II and epithelial cells of the respiratory bronchioles. KL-6 is predominantly expressed by airway cells; however, is not entirely lung specific, since it is also present on other somatic cells, such as pancreatic cells, eosophageal cells and fundic cells of the stomach \[[@B88]\]. Additionally, KL-6 is a sensitive indicator of damage to alveolar type II cells, which strongly express this mucin at their surface \[[@B70]\]. Type II pneumonocytes are regenerated over the alveolar basement membrane after the death of type I pneumonocytes over the first stage of lung injury. Therefore, its raise would theoretically represent the destruction of the normal lung parenchyma and architecture, the increased permeability of the air-blood barrier as long as the regenerating process as expressed by type II pneumonocytes\' activity. Towards this direction, the presence of KL-6 has been extensively used with great promises to monitor the severity of disease in idiopathic pulmonary fibrosis \[[@B89]-[@B91]\] and other interstitial lung diseases \[[@B92]-[@B94]\]. Since damage to, and disruption of, the alveolar epithelial lining coupled with loss of integrity of the air-blood barrier represent key features in the pathophysiology of ARDS, KL-6 serum levels could potentially serve as valuable indicators of the disease severity directly assessing the degree of epithelial damage and predicting the progression to ARDS. Nevertheless, only few studies so far, have evaluated their monitoring and prognostic efficacy in patients with or at risk for ARDS development. One of the first studies to do so was recently carried out by Ishizaka and co-workers \[[@B95]\]. Authors generated ROC curve analysis and documented a highly sensitive and specific association of optimal cut-off values of KL-6 serum levels with patients\' mortality. The latter, further supports the premise that disruption of the alveolar barrier represents a major determinant of prognosis of ALI and that serial measurements of KL-6 plasma levels might be helpful markers of the disease progression. Limitations that should be addressed include the limited number of patients, the retrospective analysis of the results and the diversity of the etiologic factors of ALI generate major concerns about the reproducibility and the reliability of the data. Recently, Sato et al. \[[@B96]\] sought to determine potential correlations of KL-6 circulating concentrations with disease severity, patients\' survival and different predisposing factors of ARDS. The most remarkable ascertainments of this study include strong associations of KL-6 peripheral levels with variables of lung injury severity and with the rates of mortality indicating possible relationship between the degree of epithelial damage and poor outcome in ARDS. Even though, authors attempted to show a modulation of KL-6 serum levels by ventilatory strategies, this relationship failed to reach a statistical significance. Despite substantial weaknesses exhibited such as the lack of serial measurement, the small sample size and the diversity of the applied treatment data derived from this study is highly informative and provides important knowledge regarding the biological impact of mechanical support strategies in this syndrome indicating the monitoring value of the epithelial damage markers. Further and sizeable prospective studies are required to validate the aforementioned hypothesis (Table [7](#T7){ref-type="table"}). Future challenges and limitations ================================= The ARDS represents an overwhelming inflammatory reaction to numerous insults within the pulmonary parenchyma resulting in life-threatening derangements in pulmonary vasomotion, alveolar ventilation and gas exchange. ARDS is a frequent disease with a devastating incidence between 13.5 and 75 per 100,000, thus affecting about 16--18% of all patients ventilated in the ICU \[[@B76],[@B97]\]. Hence, ALI/ARDS is a major public health problem encountered frequently by all physicians who care for critically ill patients. Despite the fact that research efforts over the past several years have provided a more comprehensive knowledge of the potential mechanisms comprising the immunopathogenesis of ALI/ARDS and led to the development of innumerable causative or symptomatic treatment approaches, the mortality rate of these patients remains unacceptably high at 30--40% \[[@B97]\]. Currently, the only therapy that has been proven to be effective at reducing mortality is a protective ventilatory strategy \[[@B98]\]. However, new therapies are still needed. One of the most fruitful applications is monitoring the disease activity and consequently the early identification of at risk patients with increased likelihood of non-response to treatment and progression to ARDS. Nevertheless, there are problems with the sensitivity, effort-dependability and ease of repetition of the current modalities being used for this purpose, including radiological and BAL techniques as well as clinical and physiological indices of pulmonary injury (Murray score), systemic illness (Acute Physiology and Chronic Health Evaluation- APACHE-II score and Simplified Acute Physiological Score -SAPS), ARDS severity (respiratory system compliance and PaO2/FiO2 ratio) and multiorgan system failure (Multi-Organ Dysfunction Score-MODS). Most of these clinical parameters have failed to be independent predictors of mortality in studies of adults with ALI/ARDS \[[@B99],[@B100]\]. Development of a prognostic index that combines clinical and biological determinants may be useful to ameliorate these hardships. On the basis of this conception, a large body of serum markers either cytokines and lung-specific proteins or markers of endothelium and neutrophil activation as well as other serological parameters probing different facets of the immunopathogenesis of ALI/ARDS has been delineated. The applications of these markers in the clinical setting created major expectations in terms of defining categories of patients for different therapies or prognosis for the purpose of counseling families and patients and/or possibly identifying novel therapeutic targets. The determination of a reliable serologic marker reflecting the disease behaviour and adding independent information regarding the development of the syndrome before it becomes obvious in clinical level, easily reproducible and feasible to be measured serially represents a major challenge. The early serial measurement of this biomarker may serve as an independent non-invasive prognosticator of the disease outcome even at the onset of the syndrome and therefore lead to an early detection of at risk patients with increased likelihood of progression to ARDS. The latter, if sufficiently accurate, could prove extremely useful in identifying and counseling families of patients at low or high risk for adverse outcomes and further, will allow ventilatory or other types of treatment to be evaluated or eventually modulated before they have failed in the high risk group. The presented data give credence to the view that multiple biomarkers can be used to measure the lung and systemic response to a protective ventilatory strategy and potentially to discriminate patients who are ineffectively treated and might be candidates for rescue therapies \[[@B49],[@B87]\]. More importantly, use of one or more of these biologic markers to select a group of patients at higher risk of adverse clinical outcome could be used to restrict or stratify enrollment in future clinical trials applying novel ventilator treatments such as high-frequency oscillatory ventilation leading to a better patient care. Thereby, a combination of clinical factors and biologic marker measurements could be crucial for the selection of more homogeneous groups of patients with ALI/ARDS for further studies producing evidence of high scientific rigidity \[[@B101]\]. Finally, understanding the relative roles of markers of systemic and pulmonary endothelial injury and other inflammatory mediators to the pathogenetic process of the syndrome is likely to lead to valuable insights into the final pathway resulting in diffuse alveolar damage and significant lung dysfunction, highlighting therapeutic targets for novel interventions. The aforementioned components can potentially compile a clinician\'s \"wish list\". However, the feeling of excitement arising from the expected clinical utility comes in contrast with important deficiencies exhibited by the new methodologies including non-standardization techniques, lack of knowledge of reproducibility and link to disease behaviour. Furthermore, most of the studies enrolled a limited number of patients, insufficient to extract any meaningful or statistically significant outcome. In addition, the heterogeneity of the studied population resulting from the causal diversity of the syndrome and the use of non-uniform criteria for the definitions of ARDS and at-risk patients (Tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}) render major uncertainty to the reproducibility and the scientific rigidity of these findings and may explain potential discrepancies between various studies investigating different groups of at risk patients. Moreover, many of the caveats arising from this data are generated by the origin disadvantages of the investigated serological parameters to serve as specific markers of the disease activity and severity. In particular, it is well known that assays of circulating cytokine concentration may be misleading, because they do not detect receptor or cellular bound cytokines, or they may fail to detect cytokines when inhibitors or receptor antagonists are present. Thus, measured cytokine concentrations may not reflect the disease activity or the state of inflammation but the increased production or decreased clearance from the circulation. In consistency with these limitations, it should be underlined that cytokines are part of an inflammatory cascade and biological effects are difficult even impossible to interpret without the appreciation of the entire network of the inflammatory response. Hence, data in most of the studies was inconclusive and incomplete since none of them analyzed serum alterations of a considerable number of inflammatory components. Additionally, it is of high importance to note that unfortunately only few studied molecules (VWF, IL-1β, IL-6, ICAM-1, VCAM-1) exhibited independent discriminatory power \[[@B18],[@B44],[@B48],[@B49]\] and associated with the mortality of patients with high sensitivity and specificity \[[@B45]\]. Finally, only the minority of the studies \[[@B18],[@B19],[@B25]-[@B27],[@B43],[@B45],[@B47],[@B51],[@B53],[@B64],[@B66],[@B67],[@B83],[@B86],[@B95]\] clarified the effectiveness and the diagnostic accuracy of the biomarkers by applying ROC curve analysis which is essential to estimate the sensitivity and specificity of a marker and to introduce clinically practical cut-off levels for the prediction of ARDS development in at risk individuals. Collectively, these findings highlight the necessity for further investigations in the context of large prospective studies analyzing homogeneous and well defined group of ARDS or at risk patients and the assessment of novel molecules to serve as diagnostic and prognostic tools, as well as markers of the disease activity and severity. Conclusion ========== Currently, the application status in routine clinical practice for most of these biologic markers is still in its infancy and remains exploratory. Unfortunately, they do not yield independent indications for therapy or mark the end of the inflammatory process and their prognostic value still needs to be established. Although the majority of them have not yet lived up to the \"great hype\" that was generated, markers of endothelium activation and mostly VWF and adhesion molecules (ICAM-1, VCAM-1, E-selectin) show the greatest promise in ARDS and ALI. On the contrary the majority of serum cytokines and ferritin appear to be not ready for routine monitoring since they may reflect an inflammatory response to a risk factor rather than lung injury and disease severity. Additionally, lung specific proteins have proven to be neither specific nor sensitive for the prediction of ARDS development and the disease outcome and moreover they have failed to associate with alterations in the ventilatory strategies in large clinical trials. Further prospective investigations, technical improvements and introduction of novel markers are warranted in order to elevate the association of serum biomarkers with the pathogenesis of ARDS in the same status as for tumour markers with lung cancer. Nevertheless, crossing the boundary from research to clinical application requires validation in multiple settings, experimental evidence supporting a pathophysiologic role, and ideally intervention trials showing that modification improves the outcome. The emergence of pioneering technologies including DNA microarrays which have already been applied with great success in the respiratory research field \[[@B102]\] can help scientists to circumvent this problem and bridge this boundary. In the interim, these markers can be quite useful to supplement the clinical, radiological and physiological monitoring of the disease and identify high-risk patients who would benefit from aggressive management of established risk factors. List of Abbreviations ===================== Acute Lung Injury (ALI) Acute Physiology and Chronic Health Evaluation (APACHE) Acute Respiratory Distress Syndrome (ARDS) Bronchoalveolar lavage fluid (BALF) ICU: Intensive Care Unit ICAM-1: Intercellular Cell Adhesion Molecule-1 ILs: Interleukins ILDs: Interstitial Lung Diseases Krebs von den Lungen-(KL)-6 LTs: Leukotrienes MMPs (Metalloproteinases) MODS: Multi-Organ Dysfunction Score Receiver-operating-characteristic (ROC) Simplified Acute Physiological Score (SAPS) SIRS: Systemic Inflammatory Response Syndrome Soluble-E-selectin: s-E-selectin Soluble IL-2 receptor (sIL-2R) Surfactant protein-(SP) TNFa: Tumor Necrosis Factor-alpha VWF: von Willebrand factor VCAM-1: Vascular Cell Adhesion Molecule-1 Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AT, IP and DB were involved with the study conception. AT and IP performed the data acquisition and interpretation. DB was involved in revising the article for important intellectual content. All authors read and approved the final manuscript. Acknowledgements ================ The authors are grateful to Stavros Anevlavis (M.D) for his valuable assistance in collecting the data and revising the article.

PubMed Central

2024-06-05T03:55:58.272073

2005-6-22

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168906/", "journal": "Respir Res. 2005 Jun 22; 6(1):62", "authors": [ { "first": "Argyris", "last": "Tzouvelekis" }, { "first": "Ioannis", "last": "Pneumatikos" }, { "first": "Demosthenes", "last": "Bouros" } ] }

PMC1168907

Introduction ============ Cystic fibrosis (CF) is the most common serious, monogenic, autosomal recessive disease in Caucasians and results from mutations in the gene encoding the cystic fibrosis transmembrane conductance regulator (CFTR). CF is characterised by variable phenotypic expression, which is not entirely explained by the allelic heterogeneity of the pathogenic CFTR mutations, and there is accumulating evidence that much of this phenotypic diversity is due to the effect of modifier genes \[[@B1]\]. *Pseudomonas aeruginosa*, an environmental organism, is the most important pathogen in patients with CF as chronic infection results in a more rapid decline in lung function and reduced survival \[[@B2]\]. CD14, a key gene of the innate immune system, functions as a receptor for lipopolysaccharide (LPS), a constitutive element of the *P.aeruginosa*cell wall, and is a potential modifier of severity in patients with CF. CD14 is expressed in both a membrane-bound form on macrophages, monocytes and neutrophils (mCD14) and soluble form in serum (sCD14). A polymorphism in the CD14 gene promoter (C-159T) has an allele frequency of approximately 50% in Europeans \[[@B3]\]. The -159C allele is associated with lower circulating levels of sCD14 in healthy children. Higher constitutive levels of mCD14 and sCD14 have been shown to increase the magnitude of airway neutrophil response to LPS \[[@B4]\], whereas CD14 receptor blockade results in a reduction in the deleterious systemic responses that occur in sepsis, due to a reduction in pro-inflammatory cytokines but at the expense of an increased bacterial load \[[@B5]\]. Hence, CD14 may play a pivotal role in determining the balance of infection and inflammation in CF. Increased expression of CD14 may be associated with more inflammation but a reduced bacterial load, whereas reduced expression may result in less inflammation but a greater bacterial load. We hypothesised that *P.aeruginosa*infection would be acquired earlier in children with the -159CC genotype. Methods ======= A prospective, population-based, cohort of children with CF was recruited from the only tertiary paediatric hospital in Western Australia, which cares for all children with CF in the state. Annual bronchoscopy, performed through a laryngeal mask to minimise the risk of upper airway contamination, was done routinely from diagnosis until 7 years of age in all children with CF. Children diagnosed after the neonatal period, whose first BAL was positive for *P.aeruginosa*were excluded from this analysis, as in this situation it could not be determined when they had initially acquired this pathogen. All children with a positive BAL culture for *P.aeruginosa*, which was defined as \>10,000 CFU/ml, were admitted to hospital for further treatment. Blood was collected for genetic studies and subjects were genotyped for the CD14 promoter polymorphism as previously described \[[@B3]\]. Plasma samples taken at the time of the earliest BAL, when the children were not infected with *P.aeruginosa*, was available for measurement of sCD14 levels in 31 children. Plasma sCD14 levels were determined using a commercially available enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems, Minneapolis, USA). The Kaplan-Meier survival method was used to explore the difference in the *P.aeruginosa*free survival rates between CD14 genotype groups. As children in the study were not the same age, data was censored for individual children who were known to have not isolated *P.aeruginosa*at the time of their last BAL and this information was incorporated into the analysis. Breslow test (Breslow Generalized Wilcoxon Test) was employed to compare the difference in the survival curves of acquisition of *P. aeruginosa*in the three CD14 genotype groups. Multivariate Cox regression was used to estimate the relative risks after adjustment for potential confounding factors including age, sex, CFTR mutation and nutritional status. Plasma sCD14 levels were analysed by ANOVA and all the statistical analyses were performed using SPSS for Windows (Version 11). Parents of all participants gave informed consent and the Ethics Committee of King Edward Memorial and Princess Margaret Hospitals, Western Australia, approved the study. Results ======= Forty-five children (22 male), aged 0.6--6.6 years (mean 3.25 years) were studied, of whom 25/45 (55%) were DF508 homozygous and 20/45 DF508 heterozygous. CF was diagnosed at a mean age of 0.23 yrs (95%CI 0.06--0.4 yrs), with 73% diagnosed in the newborn period. There was no significant difference between CFTR mutation (p = 0.74), gender (p = 0.38), nutritional status (weight for height z scores) (p = 0.39) or socio-economic status (p = 0.29) in those who did and did not isolate *P.aeruginosa*. CD14 genotype frequencies at position -159 were: CC 12/45 (27%), CT 20/45 (44%), TT 13/45 (29%). *P.aeruginosa*was isolated from thirteen children (29%) and the mean age of acquisition was 2.26 years (95% CI 1.29--3.25). Subjects with CD14 -159CC appeared to isolate *P.aeruginosa*at a younger age (mean = 1.1 years, 95%CI = 0.2--1.9 years) than -159CT (mean = 2.8 years, 95%CI = 1.3--4.2 years) and TT subjects (mean = 3.3 years), although this difference was not statistically significant (p = 0.19). Figure [1](#F1){ref-type="fig"} shows the *P.aeruginosa*free survival curves of the three CD14 genotype groups. The probability of children remaining free of *P.aeruginosa*with -159CC is consistently below that of children who are CT or TT and the curve for children with CT is below that of children with TT. For example, at 2 years of age, the % children remaining uninfected with *P.aeruginosa*is 55% vs. 82% vs. 100% for CC, CT and TT respectively. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Kaplan-Meier estimates of proportion of children free from *P.aeruginosa*by CD14 C-159T. ^1^Breslow Test (Breslow Generalized Wilcoxon Test) ^2^\"Censored\" represents the censored observations that arise when the duration of a study is limited. ::: ![](1465-9921-6-63-1) ::: Compared with -159TT, children with the CC genotype had a 10-fold (95%CI = 1.09--92.30, p = 0.042) higher relative risk and children with the CT genotype had an intermediate 5.5 fold (95%CI = 0.69--44.63, p = 0.108) higher relative risk of being infected with *P.aeruginosa*. After adjustment for CFTR mutation and nutritional status, the estimated relative risk in children with CC or CT increased further (RR = 13.32, 95%CI = 1.37--129.13, p = 0.025 and RR = 6.0, 95%CI = 0.71--51.03, p = 0.101 respectively). This suggested that independent of CFTR mutation and nutritional status, those with the C allele had a significantly increased risk of being infected with *P.aeruginosa*. In addition, there was a significant linear trend across the three genotype groups, between increasing numbers of C alleles and increasing likelihood of being infected with *P.aeruginosa*(p = 0.015). Compared to children who subsequently isolated *P.aeruginosa*, those who remained free of *P.aeruginosa*had significantly higher plasma sCD14 levels: 1339.43 μg/ml (95%CI 1096.63--1635.98 μg/ml) vs. 492.75 μg/ml (95%CI 55.7--4359.01 μg/ml), p = 0.018. We found no significant association between plasma sCD14 levels and CD14 C-159T (p = 0.38). Discussion ========== This study showed an association between CD14 -159CC and early acquisition of *P. aeruginosa*in children with CF. Many CF centres promote aggressive treatment regimes to eradicate first isolates of *P.aeruginosa*, and, thus, prospectively identifying a \"high risk\" group of infants with CF could have substantial clinical benefit. CD14 receptor activation results in a strong Th1 cytokine response, directed through IL-12, in an attempt to eradicate pathogens. The finding of lower plasma sCD14 levels in children who subsequently isolate *P.aeruginosa*suggests that an inadequate pro-inflammatory response to LPS may place these individuals with CF at greater risk of early *P.aeruginosa*acquisition. The lack of association between CD14 C-159T and plasma sCD14 levels may merely reflect the small number of subjects in this population, but may also indicate that in CF there are factors more critical than CD14 C-159T genotype that determine plasma sCD14 levels. Although children with higher sCD14 levels may be relatively protected from earlier acquisition of *P.aeruginosa*, when they become colonised they may paradoxically have a worse outcome, as a result of an over-aggressive and ineffectual inflammatory response. These latter patients might potentially benefit from aggressive anti-inflammatory treatment. Thus, the timing of anti-inflammatory therapy would need to be carefully considered, as inflammation might be beneficial in early life by delaying colonisation with *P.aeruginosa*, but harmful once infection is established, due to the over-exuberant inflammatory response. Therapy that modulates the inflammatory response might promote the acquisition of pathogens such as *P. aeruginosa*in young children with different genetic susceptibilities. The results of this study are based on a limited number of subjects due to the stringent inclusion criteria and the prospective longitudinal design. In addition, the potential role of other single nucleotide polymorphisms in genes involved in innate immunity such as Toll-like receptor 4 Asp299Gly \[[@B6]\], Toll-like receptor 2 Arg753Gln \[[@B6]\] and mannose binding lectin \[[@B7]\], may influence the acquisition of *P.aeruginosa*in children with CF and more detailed investigation of these pathways in CF is indicated. This study identified the potential association between a common genetic variant and the early isolation of *P.aeruginosa*. The CD14 promoter polymorphism may have a central function in determining the age of first isolation of *P.aeruginosa*, with the CC genotype conferring a higher risk of early isolation and the TT genotype being relatively protective, emphasising the importance of understanding the delicate balance between inflammation and infection that exists in CF. Acknowledgements ================ We would like to thank the children and families who participated in this study and acknowledge the technical assistance of Elizabeth Balding and Samantha Gard. This study was supported by The National Health and Medical Research Council and the Australian Cystic Fibrosis Research Trust.

PubMed Central

2024-06-05T03:55:58.279679

2005-6-23

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168907/", "journal": "Respir Res. 2005 Jun 23; 6(1):63", "authors": [ { "first": "AC", "last": "Martin" }, { "first": "IA", "last": "Laing" }, { "first": "G", "last": "Zhang" }, { "first": "S", "last": "Brennan" }, { "first": "K", "last": "Winfield" }, { "first": "PD", "last": "Sly" }, { "first": "SM", "last": "Stick" }, { "first": "J", "last": "Goldblatt" }, { "first": "PN", "last": "LeSouef" } ] }

PMC1168908

Background ========== Obstructive airway diseases, including asthma, bronchiolitis, obstructive sleep apnea, and chronic obstructive pulmonary disease (COPD), are common in children and adults \[[@B1]-[@B7]\]. Early recognition and accurate assessment of the severity of airway obstruction and the response to therapy are fundamental to the improvement of health for patients with these disorders. However, objective measures of airway obstruction currently available in the Emergency Department (ED) and other acute care settings have significant limitations. Spirometry is frequently not available in acute clinical settings, including the ED. Peak expiratory flow rate (PEFR) has been demonstrated to progressively underestimate airway obstruction with increasing air trapping, making it less reliable as airway obstruction worsens \[[@B8]\]. As well, the ability of a patient with moderate to severe airway obstruction to generate an erroneously normal PEFR and the inability to measure PEFR in young children render this test less useful in the setting of an acute asthma exacerbation \[[@B8]\]. Further, both spirometry and PEFR require patient coordination and cooperation. Validated, objective measures to determine severity of airway obstruction in bronchiolitis are nonexistent \[[@B9]\]. The pulse oximeter plethysmograph waveform reflects dynamic net changes in arteriolar inflow and venous outflow of tissue bed capillaries interrogated by the oximeter light emitting diodes \[[@B10]-[@B12]\]. Indeed, the oxygen saturation output of the device (Sp0~2~) depends upon isolation of the oxygenated, arterialized light signal from those light signals representing tissue, venous blood and other chromophobes \[[@B13]\]. At levels of arterial oxygen saturation (Sa0~2~) approaching 100%, the waveform is derived almost entirely from the infrared (940 nm) signal determined by oxyhemoglobin concentration and arterialized flow. Because oxyhemoglobin concentration is constant, dynamic changes in the waveform are a result of arterialized flow change \[[@B13]\]. Under these conditions the waveform represents a plethysmograph, a device measuring change in volume, in this case change in volume of arterialized blood \[[@B11],[@B12],[@B14]\]. As such, the plethysmograph waveform has been demonstrated to correlate with radial artery Doppler waveforms \[[@B12]\]. Changes in the plethysmograph waveform might be useful clinically to estimate the severity of perturbations in physiologic events influencing arterial flow \[[@B10]\]. Certain pathologic conditions, most notably airway obstruction, influence these physiologic events and result in the phenomenon known as pulsus paradoxus \[[@B15]\]. Although pulsus paradoxus cannot be readily measured directly from the plethysmograph waveform, changes in plethysmograph waveform variables might nonetheless correlate with the physiologic perturbations characteristic of pulsus paradoxus and be useful in assessing the severity of physiologic alterations resulting from airway obstruction. Changes in waveform curve or baseline height, one-dimensional parameters, have been used to estimate pulsus paradoxus \[[@B16]-[@B18]\]. Pulsus paradoxus represents change in left ventricular stroke volume, a three-dimensional parameter. As a two-dimensional parameter, area under the curve may more accurately reflect the physiologic events resulting in pulsus paradoxus. Additionally, the contribution of diastolic blood pressure changes to pulsus paradoxus have been noted, and AUC measurement might more completely and accurately incorporate these events \[[@B17],[@B19]\]. Finally, a general principle of signal analysis maintains that the signal-to-noise ratio improves at a rate proportionate to the square root of the number of data points obtained \[[@B20]\]. Area under the curve data may therefore be less prone to noise artifact than height data, and might provide a more optimal signal to noise ratio. With this in mind, changes in area under the waveform curve might represent a more accurate measure of waveform variability than changes in waveform height. Indeed, Hartert and colleagues have suggested evaluation of area under the waveform baseline during the respiratory cycle, rather than baseline height change, as a more accurate measurement of waveform variation\[[@B18]\]. There are limited data on the levels of intrapleural pressure generated in the presence of most obstructive airway diseases. However, levels of intrapleural pressure generated in adults in severe status asthmaticus have been demonstrated to be (-)24.4 ± 6.5 cmH~2~0 on inspiration and (+)7.6 ± 6.0 cmH~2~0 on expiration \[[@B15]\]. Mouth pressure reflects intrapleural pressure within 4 cmH~2~0 \[[@B21]\]. In this study our primary objective was to determine whether maximal change in area under the pulse oximeter plethysmograph waveform curve correlates with the degree of experimentally applied airway obstruction across a range of mouth pressures up to these levels of obstruction. A secondary objective was to determine whether maximal changes in height of the plethysmograph waveform curve similarly correlate with the degree of airway obstruction. Methods ======= Study Setting and Population ---------------------------- The study was approved by the University of Alabama at Birmingham Institutional Review Board as an expedited study. Informed written consent was obtained from each subject prior to enrollment. This study was conducted in the Pulmonary Function Laboratory of an urban children\'s hospital. A prospective convenience sample of healthy young adult subjects, twenty years of age and above, were recruited. Subjects with doctor-diagnosed asthma, a history consistent with asthma, or either FEV~1~or FEV~1~/FVC less than 80% predicted, were excluded from this study. The subjects underwent spirometry, performed by certified pulmonary function technicians according to American Thoracic Society protocol \[[@B22],[@B23]\]. Study Design and Protocol ------------------------- We utilized a closed airway circuit to generate airway obstruction, consisting of a Hans Rudolph 2600 two-way non-rebreather valve assembly with adjustable spring-loaded inspiratory and expiratory pressure relief valves and a mouthpiece pressure transducer port (Hans Rudolph, Kansas City, MO). Our experimental method was to allow each subject to experience increasing levels of inspiratory and expiratory airway obstruction corresponding to the levels of mouth pressure and to the estimated levels of intrapleural pressure noted previously \[[@B15],[@B21]\]. The pressure relief valves were adjusted accordingly at a minimum of five intervals and a maximum of ten intervals, to provide progressively increasing levels of mouth pressure from approximately (-)15 to (-)26 cmH~2~0 on inspiration and (+)2 to (+)9 cmH~2~0 on expiration. Each subject was allowed to rest for a minimum of one minute before testing at the subsequent, increased level of applied resistance in order to allow the plethysmograph waveform to return to baseline. Pulse oximeter plethysmograph waveform data was acquired for 60 seconds at each level of applied obstruction. Plethysmographic waveforms were acquired with a BioPac MP150 data acquisition system using a TSD123A transducer and an OXY100C pulse oximeter module (BioPac Systems, Santa Barbara, CA). This apparatus utilizes optical transmission at red (660 nm) and infrared (940 nm) wavelengths and employs Novametrix Medical Systems, Inc. artifact rejection and averaging algorithms that use an eight second pulse history signal to output Sp0~2.~The algorithm averages signal only for Sp0~2~calculation \[[@B24]\]. Plethysmograph waveform signal was acquired, processed and analyzed without averaging, smoothing or filtering. Mouth pressure waveforms were acquired with a BioPac TSD160C transducer. Transducers were calibrated according to manufacturer protocol. Waveform data were analyzed with BioPac Acknowledge software (version 3.7.2). The software algorithm calculates area under the curve (AUC) as the area encompassed by a waveform from the point of deflection from baseline to the point of return to baseline, and calculates height (HT) as height from the point of deflection from baseline to the waveform peak. Each subject was studied in the sitting position. A nose clip was applied, and the subject was instructed to exclusively mouth breathe through the airway circuit at a respiratory rate of approximately 10--16/min and at normal to slightly increased inspiratory and expiratory effort. Data were acquired at progressively increasing levels of applied inspiratory and expiratory obstruction for approximately 60 seconds at each level. Data collection and processing ------------------------------ Physiologic perturbations occurring during the respiratory cycle, such as airway obstruction, result in alterations of arterial flow and the phenomenon known as pulsus paradoxus \[[@B15]\]. It is these dynamic changes in arterial flow that we hypothesize might allow estimation of airway obstruction from oximeter plethysmograph waveform changes. Timing the measurement of these changes with the respiratory cycle is difficult in the clinical environment because patients with these disorders often have rapid respiratory rates. For this reason we chose to analyze data during specified time intervals. In order that at least one complete respiratory cycle and the corresponding maximum and minimum mouth pressure be included in each interval, the interval so chosen was 7.5 seconds. Data extracted for each 7.5 second interval consisted of maximum and minimum waveform area under the curve, maximum and minimum waveform height, and maximum and minimum mouth pressure. Maximum change in area under the curve and height were calculated as the difference between the maximum and minimum values of each parameter divided by the maximum value of the respective parameter during the specified 7.5 second interval. Maximum change in mouth pressure was calculated for the corresponding interval as the absolute difference between the maximum and minimum mouth pressure in cmH~2~0. These data were acquired using the Acknowledge software and entered into a spreadsheet program (Excel, Microsoft, Redmond, WA). Using the Excel formula function, mean values for maximal change in area under the curve, height, and mouth pressure for each level of applied obstruction were calculated from the multiple sequential 7.5-second intervals at the corresponding level of obstruction. This data was then entered into a statistical analysis program (SAS^®^v9.0, Cary, NC.) for analysis \[[@B25]\]. Outcome Measures ---------------- The primary outcome measure was the correlation of mean maximum change in area under the plethysmograph waveform curve with mean maximum change in mouth pressure at each successive level of applied obstruction. The secondary outcome measure was the corresponding correlation using mean maximum change in height. Data Analysis ------------- Subjects in this study contributed multiple observations to the dataset. Because of this the fundamental assumption of independence across observations was violated. Performing a separate analysis for each subject would reduce the number of observations in each analysis and increase the potential for Type II errors. On the other hand, if all of the observations were analyzed as independent, ignoring the inherent clustering within subjects, then the potential for Type 1 errors would increase. We utilized a repeated measures model that takes into account the clustering and correlation between subjects. In this analysis, the PROC MIXED procedure in SAS^®^was used to model the relationship between maximum change in area under the curve and maximum change in mouth pressure as well as the relationship between maximum change in height and maximum change in mouth pressure. Each subject contributed a single data point for each level of applied obstruction, representing the average of the 7.5-second intervals for that level of applied obstruction. Akaike\'s Information Criteria was used to compare the fit of the area under the curve vs. height models for mouth pressure \[[@B25]\]. An alpha level of p \< .05 was considered statistically significant. A total sample size of 30 subjects would allow us to construct a 95% CI for correlation and achieve a power of 0.8 and a two-tailed alpha of 0.05. Results ======= Forty-eight subjects were enrolled in the study; no subject experienced any known adverse event during or as a result of this study. Two subjects were found after enrollment to have asthma and were excluded from data analysis. Eight subjects experienced an uncomfortable sensation of dyspnea and could not use the closed airway circuit in accordance with study protocol. Data from these subjects was excluded from analysis. Seven subjects had recurrent electrical interference of the waveform baseline, the source of which could not be determined after consultation with software and hardware engineers (BioPac Systems, Santa Barbara, CA). Data from these seven subjects was excluded from analysis. Overall thirty-one subjects met inclusion criteria and had data included for analysis. Of these thirty-one subjects, eleven were male and twenty were female. The mean age was 29.9 years with a median of 28 years and range of 23 to 48 years. One subject had a prior history of cigarette smoking. No subject had heart or lung disease. One subject performed breathing maneuvers at five levels of applied obstruction, one subject at eight levels, six subjects at nine levels, and twenty-three subjects at ten levels. A total of 297 data points were available for analysis. Plethysmograph waveforms were noted to return to baseline during the period of rest (at least 1 minute) between sequentially increasing levels of applied resistance. Subjects were noted to generate plethysmograph waveforms visually significant for periodic changes with the respiratory cycle, similar to changes characteristic of pulsus paradoxus, when utilizing this apparatus (Figure [1](#F1){ref-type="fig"}). There was a significant relationship between plethysmograph waveform maximum change in area under the curve and maximum change in mouth pressure (P \< 0.0001) (Figure [2](#F2){ref-type="fig"}). The prediction equation for each cmH~2~0 maximum change in mouth pressure was 12.01 + 37.21 × (maximum change in area under the curve), 95% CI for coefficient = 30.56 to 43.87. Similarly, there was a significant relationship between maximum change in height and maximum change in mouth pressure (P \<0.0001). The prediction equation for each cmH~2~0 maximum change in mouth pressure was 16.10 + 35.94 × (maximum change in height), 95% CI for coefficient = 27.57 to 43.30. A comparison of Akaike\'s Information Criteria (AIC) between the models showed that the AIC statistic was smaller for the area under the curve model than the height model, indicating a better model fit for the area under the curve model. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Oximeter plethysmographic waveform (Pleth) generated with inspiratory and expiratory pressure relief valve apparatus. Corresponding mouth pressure indicates pressure at airway circuit mouthpiece. ::: ![](1465-9921-6-65-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Relationship between maximum changes in mouth pressure and area under the plethysmograph waveform curve. ::: ![](1465-9921-6-65-2) ::: Discussion ========== Pulse oximetry is widely available and applied in acute care settings. The device outputs a continuous plethysmographic waveform corresponding to flow of arterialized blood in the tissue bed to which the transducer is applied \[[@B10],[@B12]-[@B14]\]. It is plausible that, in the setting of airway obstruction, such changes in arteriolar flow might reflect alterations in left ventricular stroke volume resulting from the same physiologic perturbations that abnormally increase pulsus paradoxus. It is thus of interest whether the severity of airway obstruction might be estimated from changes in mathematic plethysmograph waveform variables. The study results indicate a correlation between maximum changes in area under the curve or in height of the plethysmograph waveform and the severity of airway obstruction. Analysis of both direct arterial waveform and oximeter plethysmograph waveform data for calculation of arterial flow have previously been explored in the laboratory setting. Cerutti and colleagues provide compelling data from conscious, freely moving Sprague-Dawley rats \[[@B26]\]. These investigators compared different models of central arterial line waveform analysis with simultaneously recorded cardiac output. A model using different waveform parameters identified by multiple linear regression analysis provided a reliable and precise estimation of cardiac output. Although these investigators did not use oximeter plethysmograph waveforms, their findings nonetheless support the principal of waveform analysis. Steele and colleagues performed an unblinded study on one healthy adult, breathing through a valve to which airway resistance was applied to artificially induce pulsus paradoxus. For this subject, the percent decrease in plethysmograph waveform height during the respiratory cycle correlated modestly with pulsus paradoxus calculated similarly from intra-arterial waveform (r = 0.59, 95% CI 0.32 to 0.78). This study was limited by the small subject size (n = 1) and did not measure the degree of airway obstruction generated by the resistance valves in use. The technique relied upon determination of phases of the respiratory cycle and capture of waveform indices in accordance with estimated peak inspiration and expiration \[[@B16]\]. In the clinical setting, variation of the oximeter plethysmograph waveform baseline has been noted to occur during the respiratory cycle and to represent fluctuations in local venous pressure \[[@B14],[@B18]\]. Hartert and colleagues hypothesized that this respiratory waveform variation might occur in response to pleural pressure changes and thus reflect changes in left ventricular stroke volume and pulsus paradoxus. This was studied in adult patients admitted to an ICU with obstructive airway disease, 46% of whom were receiving mechanical ventilation. Respiratory waveform variation was significantly correlated with manually measured pulsus paradoxus (R^2^= 0.88) as well as with auto-PEEP (R^2^= 0.96) \[[@B18]\]. Frey and Butt compared simultaneous 1 minute paper recordings of intra-arterial pressure and plethysmograph waveforms in 62 non-intubated children with and without respiratory disease. Correlation was noted (r = 0.85) between changes in plethysmograph waveform height and pulsus paradoxus determined from intra-arterial waveform height change \[[@B17]\]. Our study demonstrates that maximal change in height and in area under the plethysmograph waveform curve might provide a non-invasive, clinically relevant estimate of the severity of airway obstruction. A possible limitation to our study was the method of artificially inducing airway obstruction. The dynamic biomechanical changes occurring during an asthma exacerbation are not ideally simulated by externally applied resistance \[[@B27]\]. Also, in lieu of invasive, intra-arterial waveform analysis as the dependent variable and reference standard, the study protocol utilized change in mouth pressure as a surrogate measure of obstruction induced. The levels of progressive obstruction were not standardized, except insofar as the mouth pressure generated reflects intrapleural pressure \[[@B21]\]. As well, subjects were exposed to both inspiratory and expiratory obstruction during the test period. It is of interest whether correlations of waveform parameters may differ during isolated inspiratory or expiratory obstruction. Other variables that may influence the plethysmograph waveform, including hydration status, hyperinflation, and tidal volume, were likewise not controlled for in this study. Our method of using time intervals to measure changes in plethysmograph waveform AUC, HT and mouth pressure is unique. Pulsus paradoxus has traditionally been determined by noting the difference between the systolic pressure at which heart sounds are heard only during expiration and the point at which they are heard continuously \[[@B28]-[@B30]\]. However, in the tachypneic patient it is often difficult to correlate auscultation of heart sounds with the corresponding phase of the respiratory cycle. With this in mind, we chose to analyze data during specified time intervals that would encompass at least one respiratory cycle. The chosen interval, 7.5 seconds, was based upon the expected duration of the respiratory cycle in our subjects. We additionally chose to utilize the average values of data extracted from sequential intervals at each level of applied obstruction. Frey and Freezer demonstrated significant intrasubject variation of breath-to-breath measurement of pulsus paradoxus utilizing arterial waveform tracings, and averaging of pulsus paradoxus determined from multiple consecutive respiratory cycles was reported to be more accurate \[[@B19]\]. Pulse oximeters have incorporated an analogous technology for calculation of Sp0~2~, running weighted signal averaging, to minimize the effect of signal artifact and to thus enhance the reliability and validity of the calculated Sp0~2~\[[@B13]\]. Oxygen saturation is calculated 30 times per second with values averaged over a minimum of several seconds. Each instantaneous value is first compared with this moving average and assigned a weighted value based upon variation from the moving average. This weighted value then contributes to the moving average that in turn is displayed as the Sp0~2~value \[[@B13]\]. Our analysis may minimize the influence of individual waveform and respiratory cycle artifact and thus enhance the internal validity of the estimated airway obstruction. With these elements of waveform analysis in mind, our method of measuring waveform parameters may represent a strength of study design rather than a limitation. Conclusion ========== There is accumulating evidence that the plethysmograph waveform might provide clinically useful information. Our results suggest that analysis of oximeter plethysmograph waveform data may be feasible for real-time estimation of airway obstruction. To our knowledge this is the first investigation of area under the curve as a waveform parameter of potential value, and our results indicate that this parameter may achieve better correlation with airway obstruction than analyses based on waveform height. A non-invasive, real-time method to estimate the severity of airway obstruction, as well as other disorders involving pulsus paradoxus physiology, might enhance the ability of clinicians to identify and quantify the severity of such disorders \[[@B31]\]. An essential step in the development of such technology is to validate the physiologic relevance of estimating the severity of these pathophysiologic events from the oximeter plethysmograph waveform. Future study of patients with obstructive airway disease in the clinical environment, using a quantifiable, objective criterion standard such as FEV~1~will enable further assessment of oximeter plethysmograph waveform parameters to predict severity of airway obstruction. Should the accuracy and feasibility of such a tool be demonstrated in the clinical environment, development of this technology for routine clinical practice may be justified. Competing interests =================== Don Arnold has applied for patent protection for methods of waveform analysis discussed in this manuscript. Authors\' contributions ======================= DA was the principal investigator and participated in study concept and design, acquisition of the data, drafting of the manuscript and obtained institutional funding for this study to be conducted. DS was a co-investigator and participated in study concept and design, acquisition of the data, drafting of the manuscript and critical revision of the manuscript for important intellectual content. RD assisted in the statistical design and analysis and interpretation of the data, and provided critical revision of the manuscript for important intellectual content. JH participated in study concept and design, acquisition of the data, drafting of the manuscript, critical revision of the manuscript for important intellectual content, and supervised the study. Grants ====== This study was funded by a grant from The Research Institute at The Children\'s Hospital of Alabama. Acknowledgements ================ The authors are gratified for the assistance of Sheila S. Gibson, R.R.T., R.P.F.T. Johanna Kimbrough, R.P.T., and Bettye Mitchell, R.P.T. in the conduct of this study.

PubMed Central

2024-06-05T03:55:58.280752

2005-6-28

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168908/", "journal": "Respir Res. 2005 Jun 28; 6(1):65", "authors": [ { "first": "Donald H", "last": "Arnold" }, { "first": "David M", "last": "Spiro" }, { "first": "Renee' A", "last": "Desmond" }, { "first": "James S", "last": "Hagood" } ] }

PMC1168909

Background ========== Despite the standardization of the operative technique, improvement of preoperative risk assessment, and postoperative intensive care management, surgical therapy for esophageal carcinoma continues to be associated with a high incidence of operative complications, and a high mortality rate. Risk stratification in the selection of patients for surgery, and the choice of the surgical procedure are therefore important considerations. The present study was aimed to identify preoperative and intraoperative factors that could potentially influence morbidity and mortality after esophagectomy for esophageal carcinoma. Patients and methods ==================== Between September 1985 and March 2004, a total of 424 patients underwent esophagectomy for esophageal carcinoma in the Department of General and Abdominal Surgery of the Johannes Gutenberg-University Hospital Mainz. Of these, 186 (43.9%) patients had transhiatal esophagectomy, 231 (54.5%) patients underwent abdominothoracic esophagectomy with two-field lymphadenectomy (abdominal and mediastinal) and 7 patients had a proximal esophageal resection with a free jejunal graft. Reconstruction was accomplished by gastric tube pull-up in 384 (93.0%) patients, by colon interposition in 21 (5.1%), and small intestine interposition in 7 (1.7%) patients; the anatomic prevertebral esophageal bed was used for the majority of these procedures. Extra-anatomic reconstruction by the retrosternal route with cervical anastomosis after pull-up was carried out in patients with a high risk for loco-regional recurrence only (n = 42; 10.2%). Abdominothoracic esophagectomy was routinely performed for squamous cell carcinoma (n = 229). A transhiatal procedure was selected for tumors with a distal location, for malignancies without esophageal wall penetration, or in the presence of a high general risk. Transhiatal esophagectomy with abdominal and posterior mediastinal lymphadenectomy was carried out for the majority of adenocarcinomas (n = 150), the two-field procedure was performed in the presence of advanced tumor growth or extended lymph node involvement. Long-term results of this choice of the operative procedure adjusted to the histological tumor type had shown a significant prognostic advantage in patients undergoing transthoracic compared to transhiatal resection in squamous cell carcinoma whereas there was no survival benefit in patients with adenocarcinoma of the esophagus \[Junginger T, et al.; Gockel I et al.; unpublished data\]. Data were collected prospectively in a specially established database. Preoperative and intraoperative variables as well as postoperative morbidity and mortality were documented, in addition to routine demographic data. The following **preoperative risk factors**were recorded: ASA-classification (I-IV) according to the preoperative anesthesiology evaluation, BMI (Body Mass Index) based on body weight and height in kg/m^2^, and the nutritional status on a scale from 0 (= no alcohol or tobacco consumption), 1 (= tobacco alone), 2 (alcohol alone) to 3 (= combined nutritional risk with tobacco and alcohol use). Among the preoperative diseases, cardiovascular risk factors were defined as a history of coronary heart disease, or myocardial infarction, arterial hypertension, valvular disease (\>II°), arrhythmia requiring therapy (\>III° according to the Lown-classification), congestive heart failure NYHA (New York Heart Association) \> Grade II, and peripheral occlusive arterial disease (\>IIb according to Fontaine). A history of chronic obstructive pulmonary disease (COPD), regular tobacco consumption and/or the use of bronchodilators were subsumed under pulmonary diseases. The preoperative assessment of the vital capacity (VC) and forced expiratory volume (FEV~1~= Tiffeneau-test) served to ensure a more accurate assessment. Prior cirrhosis of the liver (\>/= CHILD-Pugh A) was defined as hepatic disease, and determined on the basis of the assessment of serum albumin (g/dl), serum bilirubin (mg/dl), Quick-value (%), and the presence of ascites or encephalopathy. The evaluation of additional risk factors included the prevalence of diabetes mellitus (insulin-dependent or requiring oral antidiabetic therapy), and the history of a secondary carcinoma. Intraoperative variables included in the risk analysis were tumor location, the operative procedure, and the transfusion requirement of packed erythrocytes. The group of tumor characteristics comprised tumor size, R-classification, TNM-stage, and the number of dissected lymph nodes. Postoperative variables were not part of this risk analysis. Morbidity, surgical (anastomotic leakage, graft necrosis, mediastinitis, recurrent laryngeal nerve paralysis, chylothorax, tracheal fistula, bleeding requiring reoperation) and general complications (pneumonia, atelectasis, ARDS i.e. adult respiratory distress syndrome, myocardial infarction, cardiac failure, pulmonary embolism, renal insufficiency, pancreatitis, deep vein thrombosis), length of intensive care unit stay, 30-day mortality, and the mortality rate served as factors for assessment. Statistical analysis -------------------- The SSPS 10.0 software package was used for statistical data analysis (SSPS, Chicago, IL, USA: 1999). Data are expressed as median with ranges (minimum -- maximum), or as percentages (%). Factors with a possible influence on perioperative morbidity and mortality were calculated using the logistic regression model (univariate and multivariate). The χ^2^test with Pearson\'s correction and Fisher\'s exact test were used for comparison of the parameters for squamous cell carcinoma and adenocarcinoma. The Mann-Whitney U test served as the non-parametric method. A *p*-value of \<0.05 was considered statistically significant for all procedures. Results ======= Preoperative and intraoperative parameters, tumor characteristics ----------------------------------------------------------------- Median patient age at the time of surgery was 58 (28--84) years; the proportion of males was 83.0%. The median BMI (Body Mass Index) was 24.4 (13.8--39.3) kg/m^2^. A total of 36.9% patients were classified as ASA grade II, 58.2% as ASA grade III, and 4.9% as ASA grade IV. A combined nutritional risk (alcohol and tobacco use) was determined in 41.5% of patients. Tobacco use alone was found in 59.1%, and habitual alcohol consumption in 58.2% of patients. Preexisting cardiovascular diseases were noted in 28.2%, pulmonary disease in 13.5%, hepatopathy in 4.9%, and diabetes mellitus in 5.9% of patients undergoing surgery for esophageal carcinoma. The history of 9.8% of all patients showed the presence of secondary carcinoma. Median preoperative vital capacity (VC) ranged at 3.8 (1--7.2) l, FEV~1~(Tiffeneau-test) was 2.9 (0.5--9.0) l/sec. Out of all tumors, 56.3% were located in the lower, 34.6% in the middle, and 9.1% in the upper third of the esophagus. Squamous cell carcinoma was identified in 55.3%, adenocarcinoma in 35.9%, and an undifferentiated carcinoma in 7.6% of patients (1.2% with other malignant esophageal tumors, e.g. melanoma) who underwent esophagectomy. Operative time ranged at 300 (160--560) minutes. The median number of packed erythrocyte units used was 1 (0--38) (54.6% of all patients did not require packed erythrocyte transfusion). Median length of intensive care therapy was 10 (0--176) days, at a total postoperative hospital stay of 22 (0--189) days. Median tumor size was 4 (0.3--20) cm. A R0 resection was accomplished in 81.0% of all patients (R1 resection: 16.0%, R2 resection: 3.0%). Based on results of the pathological examination, the majority of tumors were assigned to the T3 category (57.5%), while 34.6% of all tumors were allocated to the T1 or T2, and 7.9% to the T4-category. Distribution for the N category was as follows: N0: 34.0%, N1: 66.0%. A M1-situation (M1-lymph node or M1-organ) was identified in 26.5% of all patients, which was treated with curative intent in the course of the same surgical procedure. Morbidity and mortality: Prevalence and influence factors --------------------------------------------------------- There was a 35.5% prevalence of surgical complications in the total population of 424 patients undergoing esophagectomy for esophageal carcinoma. General complications were observed in 36.0% of all patients. The 30-day mortality rate was 6.7%, at a mortality rate for the entire observation period of 11.5%. Anastomotic leakage was the most common surgical complication (18.2%), followed by recurrent laryngeal nerve paresis (15.7%). There was a similar incidence of graft necrosis (3.2%) and postoperative hemorrhage (2.9%) requiring surgical revision, while 1.3% of all patients developed a chylothorax. Univariate analysis showed tumor location (p = 0.0194) and the surgical procedure (p = 0.0116) to be significantly related to the incidence of surgical complications; none of the other pre- and intraoperative factors, including tumor characteristics, were of relevance in this model (Table [1](#T1){ref-type="table"}). Multivariate analysis identified the transthoracic surgical procedure (p = 0.0004), tumor location (upper third) (p = 0.02), and the transfusion of packed erythrocytes (p = 0.04) as factors significantly related to the incidence of surgical complications (Table [2](#T2){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Factors with a potential influence on surgical and general complications, as well as on the perioperative mortality rate after esophagectomy for esophageal carcinoma: univariate analysis (p values). ::: **Surgical complications** **General Complications** **Perioperative Mortality** --------------------------------------- ---------------------------- --------------------------- ----------------------------- [Preoperative factors:]{.underline} \- Age 0.8996 0.0073\* 0.1349 \- Sex 0.7367 0.3396 0.5658 \- BMI 0.3710 0.1200 0.0719 \- ASA 0.3992 0.0076\* 0.0274\* \- Nutritional status 0.5093 0.1897 0.2223 \- Cardiovascular PD 0.7803 0.0171\* 0.0172\* \- Pulmonary PD 0.2910 0.0405\* 0.0059\* \- Hepatopathy 0.1696 0.0282\* 0.0165\* \- Diabetes mellitus 0.8218 0.0635 0.9430 [Intraoperative factors:]{.underline} \- Tumor location 0.0194\* 0.0156\* 0.0391\* \- Surgical procedure 0.0116\* 0.0001\* 0.0057\* \- Packed erythrocyte transfusion 0.1045 0.0006\* 0.0022\* [Tumor characteristics:]{.underline} \- Tumor size 0.5041 0.1595 0.1502 \- R-classification 0.5573 0.3372 0.0439\* \- T-category 0.2017 0.0072\* 0.0036\* \- N-category 0.6805 0.9928 0.0584 \- M-category 0.0621 0.8128 0.4359 \*statistically significant ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Significant factors with a potential influence on surgical and general complications, as well as on the perioperative mortality rate after esophagectomy for esophageal carcinoma: multivariate analysis (p value). ::: **Surgical complications** **General complications** **Perioperative Mortality** --------------------------------------------- --------------------------------------------- --------------------------------------------- Surgical procedure (p = 0.0004) Surgical procedure (p = 0.0001) Surgical procedure (p = 0.0068) Tumor location (p = 0.0204) Packed erythrocyte transfusion (p = 0.0009) Pulmonary PD (p = 0.0096) Packed erythrocyte transfusion (p = 0.0493) Patient age (p = 0.0039) Packed erythrocyte transfusion (p = 0.0099) Nutritional status (p = 0.0284) ASA-classification (p = 0.0486) ::: Pulmonary complications were most common among the general complications (32.9%). Adult respiratory distress syndrome (ARDS) occurred in 6.5% of patients. Postoperatively, 3.7% of patients developed renal failure requiring dialysis, and myocardial infarction or pulmonary embolism occurred in 1.2%, respectively, of all patients. Postoperative pancreatitis was found in 0.5% of patients. Univariate analysis determined patient age (p = 0.0073), ASA-classification (p = 0.0076), preexisting cardiovascular (p = 0.0171), pulmonary (p = 0.0405), and hepatic (p = 0.0282) disease as significant preoperative variables in relation to postoperative general complications. Similarly significant among the intraoperative factors were tumor location (p = 0.0156), the surgical procedure (p = 0.0001), transfusion of packed erythrocytes, as well as the T-category in the group of tumor characteristics (Table [1](#T1){ref-type="table"}). On multivariate analysis, independent parameters with a significant influence on postoperative general complications were in decreasing order: the surgical procedure (p = 0.0001), transfusion of packed erythrocytes (p = 0.0009), patient age (p = 0.0039), nutritional status (p = 0.0284), and ASA-classification (p = 0.0486) (Table [2](#T2){ref-type="table"}). The cause of postoperative mortality (11.5%; n = 49 patients) in 19 patients was sepsis (accounted for by anastomotic insufficiency in 8, and in 3 patients by graft necrosis). Twelve patients died from pulmonary failure, and 5 patients died after myocardial infarction. Fulminant pulmonary embolism was the cause of death in 3 patients (one out of these occurred in a patient with suture dehiscence). Two patients developed hemorrhagic shock, and another 2 patients died in hospital from rapid progression of the malignant disease. Further causes of hospital death in 6 additional patients after esophagectomy were; hepatorenal syndrome, syncopal attack with cardiovascular failure, right ventricular heard failure, medial cerebral infarction, intraoperative cardiac arrest, and tracheogastric fistula. Preoperative variables as predictors of a fatal postoperative course found on univariate analysis comprised ASA-classification (p = 0.0274), preexisting cardiovascular (p = 0.0172) and pulmonary disease (p = 0.0059), and hepatopathy (p = 0.0165). Further factors exerting an influence on mortality were: tumor location (p = 0.0391), the surgical procedure (p = 0.0057), packed erythrocyte transfusion (p = 0.0022), R-classification (p = 0.0439), and T-category (p = 0.0036) (Table [1](#T1){ref-type="table"}). Multivariate analysis identified the surgical procedure (p = 0.0068), preexisting pulmonary disease (p = 0.0096), and the transfusion of packed erythrocytes (p = 0.0099) as the most significant predictors of mortality (Table [2](#T2){ref-type="table"}). Risk profile: comparison between squamous cell carcinoma and adenocarcinoma --------------------------------------------------------------------------- From among the group of preoperative factors, patients with adenocarcinoma were characterized by significantly more advanced age and a higher BMI (p \< 0.0001) compared to those with squamous cell carcinoma. Differences in gender distribution and ASA-classification were not significant. There was a significantly higher nutritional risk (alcohol and tobacco use) in patients with squamous cell carcinoma than in the comparison group of patients with adenocarcinoma undergoing esophagectomy (p \< 0.001). The incidence of cardiovascular and pulmonary disease, as well as preexisting diabetes mellitus, or secondary carcinoma in the patient history was similar in both groups. However, patients with squamous cell carcinoma had significantly poorer preoperative lung function (p = 0.0078), and a higher prevalence of hepatopathy (p \< 0.0001) (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of the preoperative risk profile between patients with squamous cell carcinoma and adenocarcinoma of the esophagus. ::: **Parameter** **Squamous cell carcinoma (n = 229)** **Adenocarcinoma (n = 150)** **p-value** ---------------------------- --------------------------------------- ------------------------------ ------------- \- Age (years) 56.0 (29--84) 61.0 (28--78) 0.0001\* \- Sex (% males) 79.8 87.4 0.071 \- BMI (kg/m^2^) 23.2 (13.9--34.9) 25.8 (15.8--39.3) 0.0001\* \- ASA-classification (%) -II 33.2 43.2 -III 62.4 52.1 0.072 -IV 4.4 4.8 \- Nutritional risk (%) -No nutritional risk (%) 16.2 43.2 0.0001\* -Alcohol use (%) 69.5 42.9 0.0001\* -Tobacco use (%) 73.5 38.1 0.0001\* -Combined risk (%) 83.8 56.9 0.0001\* \- Cardiovascular PD (%) 30.4 24.7 0.240 \- Pulmonary PD (%) 14.7 14.4 0.593 -VC (l) 3.7 (1.0--6.6) 3.8 (1.9--6.1) 0.3063 -FEV 1 (l/sec) 2.8 (0.5--5.2) 3.1 (1.3--5.9) 0.0078\* \- Hepatopathy (%) 8.1 1.4 0.0001\* \- Diabetes mellitus (%) 4.9 7.5 0.369 \- Secondary carcinoma (%) 11.2 8.2 0.382 \*statistically significant ::: The surgical procedure used was different with respect to the significantly more frequent performance of transhiatal esophagectomy in patients with adenocarcinoma (p \< 0.0001). In the presence of this histological subtype, the gastric tube (p = 0.027) brought up in the existing esophageal bed (p = 0.001) was used more often as the interposed organ. The number of dissected abdominal lymph nodes was significantly higher (p = 0.0146) for adenocarcinoma, while a significantly higher number of thoracic lymph nodes (p = 0.0011) was removed in patients with squamous cell carcinoma. The operative time was significantly shorter (p \< 0.0001) at a lower packed erythrocyte transfusion requirement for adenocarcinoma (p = 0.0495) compared to squamous cell carcinoma (Table [4](#T4){ref-type="table"}). There was no significant difference between the two groups with regard to UICC stages and tumor characteristics (p \> 0.05). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Intraoperative Factors: Comparison between squamous cell carcinoma and adenocarcinoma of the esophagus. ::: **parameter** **Squamous cell carcinoma (n = 229)** **Adenocarcinoma (n = 150)** **p-value** ------------------------------------------------ --------------------------------------- ------------------------------ ------------- \- Surgical procedure (% transhiatal) 30.6 68.7 0.0001\* \- Esophageal substitute (% gastric tube) 90.8 97.3 0.027\* \- Repositioning (% esophageal bed) 83.7 95.3 0.001\* \- Number of removed abdominal LN (n) 11 (0--55) 13 (0--51) 0.0146\* \- Number of removed thoracic LN (n) 11 (0--47) 6 (0--83) 0.0011\* \- Operative time (min) 305 (115--560) 270 (160--540) 0.0001\* \- Intraoperative blood loss (ml) 1000 (200--5000) 800 (0--7500) 0.2224 \- Units of transfused packed erythrocytes (n) 1.5 (0--38) 0 (0--14) 0.0495\* \*statistically significant ::: Although there was no significant difference in surgical complications between the two groups, the rate of general complications (p = 0.012), and thus the total complication rate, (p = 0.039) was significantly higher in patients with squamous cell carcinoma. Comparable to the morbidity rate, the duration of postoperative intensive care therapy was significantly longer (p \< 0.0001) for this type of tumor than for adenocarcinoma. The 30-day mortality rate was identical for both groups, while the mortality rate in patients with squamous cell carcinoma was significantly higher (p \< 0.0001) (Table [5](#T5){ref-type="table"}). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Postoperative morbidity and mortality for squamous cell carcinoma compared with adenocarcinoma of the esophagus. ::: **parameter** **Squamous cell carcinoma (n = 229)** **Adenocarcinoma (n = 150)** **p-value** ------------------------------- --------------------------------------- ------------------------------ ------------- \- Surgical complications (%) \- General complications (%) 37.9 33.7 0.447 -Total complications (%) 40.7 28.0 0.012\* \- Length of ICU -- stay (d) 68.1 57.6 0.039\* \- 30-day-moratlity (%) \- Mortality (%) 11 (2--176) 8 (1--107) 0.0001\* 6.1 6.1 0.8569 15.1 6.6 0.0001\* \*statistically significant ::: Discussion ========== Although the morbidity rate after esophagectomy for esophageal carcinoma has been markedly reduced in recent years as a result of improvements in patient selection, surgical technique, and advances in perioperative management, the morbidity rate remains high \[[@B1]-[@B7]\]. Pulmonary function disturbances are the major contributing factor here \[[@B8]-[@B10]\]. Risk factors for the development of impaired pulmonary function include smoking, patient age (\>70 years), obesity, and preexisting COPD \[[@B11]\]. Parameters with an influence on perioperative mortality defined by Bartels *et al*, are in decreasing order: reduced general status of the patient, impaired cardiac and hepatic function, and respiratory function \[[@B12]\]. The evaluation of the preoperative function of the described organ systems using a scoring system developed by the authors enables the classification of risk groups, and leads to a decrease in the operative risk as a result of adequate patient selection \[[@B12]\]. The aim of our analysis of data collected prospectively in 424 patients undergoing surgery for esophageal carcinoma was to answer the question as to what extent the selection of the surgical procedure, i.e. the transthoracic or the transhiatal approach, and the operative course assessed on the basis of intraoperative blood loss, in addition to patient and tumor related factors, exert an influence on the postoperative course. The data were collected in 424 consecutive patients undergoing surgery for esophageal carcinoma over the period from 1985 to 2004. Surgical complications occurred in 35.5% and general complications in 36.0% of patients. The 30-day mortality rate ranged at 6.7%, at a mortality rate of 11.5%. Univariate analysis identified the selection of the surgical procedure as the main risk factor affecting the mortality rate. Transthoracic esophagectomy was associated with a higher complication and mortality rate than transhiatal dissection. Various authors have investigated the question as to the effectiveness of an extended radical procedure with an associated increased operative risk after the transthoracic technique. Results of two randomized studies did not demonstrate a difference between these procedures, although the meaningfulness of these findings is limited due to the small number of patients enrolled and the lack of information on the oncological radicality, especially the extent of lymph node dissection \[[@B13],[@B14]\]. In addition, only patients with early tumor stages \[[@B14]\] or with a distal location of the carcinoma \[[@B13]\] were taken into consideration, respectively. A higher pulmonary complication rate after the transthoracic compared to the transhiatal procedure for adenocarcinoma was found by a prospective randomized study \[[@B15]\]. Postoperative ventilation time, intensive care unit and postoperative in-hospital stay, reflecting perioperative morbidity, were significantly longer after transthoracic in contrast to transhiatal esophagectomy in this trial. However, the authors did not determine any differences in the long-term course \[[@B15]\]. This prospective study confirmed the outcome of a meta analysis revealing significantly higher early (pulmonary) morbidity and mortality after the transthoracic procedure published previously by the same author with 5-year survival rates of approximately 20% after both kinds of resection \[[@B16]\]. In our patient population, the long-term prognosis for patients with squamous cell carcinoma undergoing transthoracic surgery -- though exhibiting a higher perioperative morbidity and mortality -- was significantly better than that for patients after transhiatal resection \[Junginger T, et al.; unpublished data\]. Patients with adenocarcinoma did not differ in survival undergoing transhiatal or transthoracic esophagectomy \[Gockel I, et al; unpublished data\]. In concert with other authors, we therefore favor the transhiatal technique with posterior mediastinal and upper abdominal lymph node dissection for adenocarcinoma of the esophagus, and the transthoracic procedure with abdominal and mediastinal lymphadenectomy for squamous cell carcinoma. Thus -- long term results regarding the surgical technique of our own patient population have to be viewed critically, especially for adenocarcinoma, as the two groups differed significantly in UICC-stage and R-classification in contrast to patients with squamous cell carcinoma with a rather equal distribution \[[@B1],[@B2]\]. Independent of the operative procedure, surgical blood loss had a significant influence on the postoperative morbidity and mortality rates. This is in accordance with experiences reported by Whooley *et al*\[[@B8]\], and indicates that the selection of a limited resection technique can be crucial for the development of the postoperative course. Patient-associated parameters (age, nutritional status, ASA-classification) were of relevance only with respect to the occurrence of general complications. Preexisting pulmonary disease was an independent predictor of postoperative mortality. This confirms the findings of Chan *et al*, who identified impaired pulmonary function as a preoperative variable predictive of postoperative mortality \[[@B17]\]. The implementation of an appropriate preoperative therapy, discontinuation of smoking, more frequent use of epidural analgesia, and early bronchoscopy in the presence of the suspicion of postoperative pulmonary secretion impairment are essential factors for risk reduction \[[@B8]\]. In contrast to results of risk analyses by Law *et al*\[[@B18]\] and Lund *et al*\[[@B19]\], tumor characteristics as, e.g. TNM classification, were of no influence on the postoperative course in the patient population of this study. A different operative risk was determined for the two histological tumor types of the esophagus: while there were similar surgical complications in both groups, overall morbidity and mortality rates were significantly higher in patients with squamous cell carcinoma than in the group with adenocarcinoma. In accordance with reports in the literature \[[@B20]-[@B22]\], this reflects, on the one hand, the different preoperative risk profile of both entities, consisting of an increased nutritional risk, higher prevalence of hepatopathy, and poorer lung function in patients with squamous cell carcinoma. Additionally, in this study there was a significantly higher incidence of transthoracic esophagectomy with a higher complication rate in patients with squamous cell carcinoma than in those with adenocarcinoma (69.4 vs. 31.3%). Conclusion ========== The present analysis shows that the selection and the type of the surgical procedure are crucial factors for both the incidence of postoperative complications and the mortality rate. The transhiatal procedure is associated with a significantly lower morbidity and mortality rate and thus represents -- as long term survival does not favor the transthoracic approach -- the surgical technique of choice for adenocarcinoma of the esophagus. In contrast, our previous long-term experience and results obtained by this study advocate the performance of the transthoracic procedure for squamous cell carcinoma. The higher operative risk is justified with a view to a better long-term prognosis. Independent of the choice of the operative approach, a less-invasive surgical procedure and the implementation of measures designed to minimize the risk of pulmonary complications are essential to achieve a reduction in the morbidity rate of esophageal carcinoma. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= **IG:**study design, collection of data, statistical analysis, sequence alignment, draft of manuscript **ChE:**collection of data, statistical analysis **ThJ:**conceived of the study, design and coordination of the study, draft and revision of the manuscript

PubMed Central

2024-06-05T03:55:58.283343

2005-6-21

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1168909/", "journal": "World J Surg Oncol. 2005 Jun 21; 3:37", "authors": [ { "first": "Ines", "last": "Gockel" }, { "first": "Christoph", "last": "Exner" }, { "first": "Theodor", "last": "Junginger" } ] }

PMC1171969

Introduction ============ There is an impressively large body of physiological, imaging, and psychopharmacological data regarding the phasic activity of dopaminergic (DA) cells in the midbrains of monkeys, rats and humans in classical and instrumental conditioning tasks involving predictions of future rewards \[[@B1]-[@B5]\]. These data have been taken to suggest \[[@B6],[@B7]\] that the activity of DA neurons represents temporal difference (TD) errors in the predictions of future reward \[[@B8],[@B9]\]. This TD theory of dopamine provides a precise computational foundation for understanding a host of behavioural and neural data. Furthermore, it suggests that DA provides a signal that is theoretically appropriate for controlling learning of both predictions and reward-optimising actions. Some of the most compelling evidence in favour of the TD theory comes from studies investigating the phasic activation of dopamine cells in response to arbitrary stimuli (such as fractal patterns on a monitor) that predict the proximate availability of rewards (such as drops of juice). In many variants, these have shown that with training, phasic DA signals transfer from the time of the initially unpredictable reward, to the time of the earliest cue predicting a reward. This is exactly the expected outcome for a temporal-difference based prediction error (*eg.*\[[@B1],[@B2],[@B10]-[@B13]\]). The basic finding \[[@B7]\] is that when a reward is unexpected (which is inevitable in early trials), dopamine cells respond strongly to it. When a reward is predicted, however, the cells respond to the predictor, and not to the now-expected reward. If a predicted reward is unexpectedly omitted, then the cells are phasically inhibited at the normal time of the reward, an inhibition which reveals the precise timing of the reward prediction \[[@B10]\], and whose temporal metrics are currently under a forensic spotlight \[[@B14]\]. The shift in activity from the time of reward to the time of the predictor resembles the shift of the animal\'s appetitive behavioural reaction from the time of the reward (the unconditioned stimulus) to that of the conditioned stimulus in classical conditioning experiments \[[@B7],[@B10]\]. In a most interesting recent study, Fiorillo *et al.*\[[@B15]\] examined the case of partial reinforcement, in which there is persistent, ineluctable, prediction error on every single trial. A straightforward interpretation of the TD prediction error hypothesis would suggest that in this case (a) dopamine activity at the time of the predictive stimuli would scale with the probability of reward, and (b) on average over trials, the dopaminergic response after the stimulus and all the way to the time of the reward, should be zero. Although the first hypothesis was confirmed in the experiments, the second was not. The between-trial averaged responses showed a clear ramping of activity during the delay between stimulus onset and reward that seemed inconsistent with the TD account. Fiorillo *et al.*hypothesised that this activity represents the uncertainty in reward delivery, rather than a prediction error. In this paper, we visit the issue of persistent prediction error. We show that a crucial asymmetry in the coding of positive and negative prediction errors leads one to expect the ramping in the between-trial average dopamine signal, and also accounts well for two further features of the DA signal -- apparent persistent activity at the time of the (potential) reward, and disappearance (or at least weakening) of the ramping signal, but not the signal at the time of reward, in the face of trace rather than delay conditioning. Both of these phenomena have also been observed in the related instrumental conditioning experiments of Morris *et al.*\[[@B16]\]. Finally, we interpret the ramping signal as the best evidence available at present for the nature of the learning mechanism by which the shift in dopamine activity to the time of the predictive stimuli occurs. Uncertainty in reward occurrence: DA ramping ============================================ Fiorillo *et al.*\[[@B15]\] associated the presentation of five different visual stimuli to macaques with the delayed, *probabilistic*(*p*~*r*~= 0, 0.25, 0.5, 0.75, 1) delivery of juice rewards. They used a delay conditioning paradigm, in which the stimulus persists for a fixed interval of 2s, with reward being delivered when the stimulus disappears. After training, the monkeys\' anticipatory licking behavior indicated that they were aware of the different reward probabilities associated with each stimulus. Figure [1a](#F1){ref-type="fig"} shows population histograms of extracellularly-recorded DA cell activity, for each *p*~*r*~. TD theory predicts that the phasic activation of the DA cells at the time of the visual stimuli should correspond to the *average*expected reward, and so should increase with *p*~*r*~. Figure [1a](#F1){ref-type="fig"} shows exactly this -- indeed, across the population, the increase is quite linear. Morris *et al.*\[[@B16]\] report a similar result in an instrumental (trace) conditioning task also involving probabilistic reinforcement. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Averaged prediction errors in a probabilistic reward task**(a) DA response in trials with different reward probabilities. Population peri-stimulus time histograms (PSTHs) show the summed spiking activity of several DA neurons over many trials, for each *p*~*r*~, pooled over rewarded and unrewarded trials at intermediate probabilities. (b) TD prediction error with asymmetric scaling. In the simulated task, in each trial one of five stimuli was randomly chosen and displayed at time *t*= 5. The stimulus was turned off at *t*= 25, at which time a reward was given with a probability of *p*~*r*~specified by the stimulus. We used a tapped delay-line representation of the stimuli (see text), with each stimulus represented by a different set of units (\'neurons\'). The TD error was *δ*(*t*) = *r*(*t*) + **w**(*t*- 1)·**x**(*t*) - **w**(*t*- 1)·**x**(*t*- 1), with *r*(*t*) the reward at time *t*, and **x**(*t*) and **w**(*t*) the state and weight vectors for the unit. A standard online TD learning rule was used with a fixed learning rate *α*, **w**(*t*) = **w**(*t*- 1) + *αδ*(*t*)**x**(*t*- 1), so each weight represented an expected future reward value. Similar to Fiorillo *et al*., we depict the prediction error *δ*(*t*) averaged over many trials, after the task has been learned. The representational asymmetry arises as negative values of *δ*(*t*) have been scaled by *d*= 1/6 prior to summation of the simulated PSTH, although learning proceeds according to unscaled errors. Finally, to account for the small positive responses at the time of the stimulus for *p*~*r*~= 0 and at the time of the (predicted) reward for *p*~*r*~= 1 seen in (a), we assumed a small (8%) chance that a predictive stimulus is misidentified. (c) DA response in *p*~*r*~= 0.5 trials, separated into rewarded (left) and unrewarded (right) trials. (d) TD Model of (c). (a,c) Reprinted with permission from \[15\]©2003 AAAS. Permission from AAAS is required for all other uses. ::: ![](1744-9081-1-6-1) ::: By contrast, at the time of potential reward delivery, TD theory predicts that *on average*there should be no activity, as, on average, there is no prediction error at that time. Of course, in the probabilistic reinforcement design (at least for *p*~*r*~≠ 0, 1) there is in fact a prediction error at the time of delivery or non-delivery of reward on every single trial. On trials in which a reward is delivered, the prediction error should be positive (as the reward obtained is larger than the average reward expected). Conversely, on trials with no reward it should be negative (see Figure [1c](#F1){ref-type="fig"}). Crucially, under TD, the average of these differences, weighted by their probabilities of occurring, should be zero. If it is not zero, then this prediction error should act as a plasticity signal, changing the predictions until there is no prediction error. At variance with this expectation, the data in Figure [1a](#F1){ref-type="fig"} which is averaged over both rewarded and unrewarded trials, show that there is in fact positive mean activity at this time. This is also evident in the data of Morris *et al.*\[[@B16]\] (see Figure [3c](#F3){ref-type="fig"}). The positive DA responses show no signs of disappearing even with substantial training (over the course of months). Worse than this for the TD model, and indeed the focus of Fiorillo *et al.*\[[@B15]\], is the apparent *ramping*of DA activity towards the expected time of the reward. As the magnitude of the ramp is greatest for *p*~*r*~= 0.5, Fiorillo *et al.*suggested that it reports the *uncertainty*in reward delivery, *rather than*a prediction error, and speculated that this signal could explain the apparently appetitive properties of uncertainty (as seen in gambling). Both the ramping activity and the activity at the expected time of reward pose critical challenges to the TD theory. TD learning operates by arranging for DA activity at one time in a trial to be *predicted away*by cues available earlier in that trial. Thus, it is not clear how any seemingly predictable activity, be it that at the time of the reward or in the ramp before, can persist without being predicted away by the onset of the visual stimulus. After all, the *p*~*r*~-dependent activity in response to the stimulus confirms its status as a valid predictor. Furthermore, a key aspect of TD \[[@B17]\], is that it couples prediction to action choice by using the *value*of a state as an indication of the future rewards available from that state, and therefore its attractiveness as a target for action. From this perspective, since the ramping activity is explicitly not predicted by the earlier cue, it cannot influence early actions, such as the decision to gamble. For instance, consider a competition between two actions: one eventually leading to a state with a deterministic reward and therefore no ramp, and the other leading to a state followed by a probabilistic reward with the same mean, and a ramp. Since the ramp does not affect the activity at the time of the conditioned stimulus, it cannot be used to evaluate or favour the second action (gambling) over the first, despite the extra uncertainty. We suggest the alternative hypothesis that both these anomalous firing patterns result directly from the constraints implied by the low baseline rate of activity of DA neurons (2--4 Hz) on the coding of the *signed*prediction error. As noted by Fiorillo *et al.*\[[@B15]\], positive prediction errors are represented by firing rates of \~270% *above*baseline, while negative errors are represented by a decrease of only \~55% *below*baseline (see also \[[@B14],[@B18]\]). This asymmetry is a straightforward consequence of the coding of a signed quantity by firing which has a low baseline, though, obviously, can only be positive. Firing rates above baseline can encode positive prediction errors by using a large dynamic range, however, below baseline firing rates can only go down to zero, imposing a restriction on coding of negative prediction errors. Consequently, one has to be careful interpreting the sums (or averages) of peri-stimulus-time-histograms (PSTHs) of activity over different trials, as was done in Figure [1a](#F1){ref-type="fig"}. The asymmetrically coded positive and negative error signals at the time of the receipt or non-receipt of reward should indeed *not*sum up to zero, even if they represent correct TD prediction errors. When summed, the low firing representing the negative errors in the unrewarded trials will not \"cancel out\" the rapid firing encoding positive errors in the rewarded trials, and, overall, the average will show a positive response. In the brain, of course, as responses are not averaged over (rewarded and unrewarded) trials, but over neurons within a trial, this need not pose a problem. This explains the persistent positive activity (on average) at the time of delivery or non-delivery of the reward. But what about the ramp prior to this time? At least in certain neural representations of the time between stimulus and reward, when trials are averaged, this same asymmetry leads TD to result exactly in a ramping of activity toward the time of the reward. The TD learning mechanism has the effect of propagating, on a trial-by-trial basis, prediction errors that arise at one time in a trial (such as at the time of the reward) towards potential predictors (such as the CS) that arise at earlier times within each trial. Under the asymmetric representation of positive and negative prediction errors that we have just discussed, averaging these propagating errors over multiple trials (as in Figure [1a](#F1){ref-type="fig"}) will lead to positive means for epochs within a trial before a reward. The precise shape of the resulting ramp of activity depends on the way stimuli are represented over time, as well as on the speed of learning, as will be discussed below. Figure [2](#F2){ref-type="fig"} illustrates this view of the provenance of the ramping activity. Here, a tapped delay-line representation of time since the stimulus is used. For this, each unit (\'neuron\') becomes active (i.e., assumes the value 1) at a certain lag after the stimulus has been presented, so that every timestep after the stimulus onset is consistently represented by the firing of one unit. Learning is based on the (dopaminergically-reported) TD error, formalized as *δ*(*t*) = *r*(*t*) + *V*(*t*) - *V*(*t*- 1), with *V*(*t*) the weighted input from the active unit at time *t*, and *r*(*t*) the reward obtained at time *t*. Updating the weights of the units according to the standard TD update rule with a fixed learning rate, allows *V*(*t*) to, on average, represent the expected future rewards (see Figure [1](#F1){ref-type="fig"} caption). As each subsequent timestep is separately represented, TD prediction errors can arise at any time within the trial. Figure [2a](#F2){ref-type="fig"} shows these errors in six consecutive simulated trials in which *p*~*r*~= 0.5. In every trial, a new positive or negative error arises at the time of the reward, consequent on receipt or non-receipt of the reward, and step-by-step the errors from previous trials propagate back to the time of the stimulus, through the constant updating of the weights (*eg.*the error highlighted in red). When averaging (or, as in PSTHs, summing) over trials, these errors cancel each other on average, resulting in an overall flat histogram in the interval after the stimulus onset, and leading up to the time of the reward (black line in Figure [2b](#F2){ref-type="fig"}, summed over the 10 trials shown in thin blue). However, when summed after *asymmetric scaling*of the negative errors by a factor of *d*= 1/6 (which simulates the asymmetric coding of positive and negative prediction errors by DA neurons), a positive ramp of activity ensues, as illustrated by the black line in Figure [2c](#F2){ref-type="fig"}. Note that this rescaling is only a *representational*issue, resulting from the constraints of encoding a negative value about a low baseline firing rate, and should not affect the learning of the weights, so as not to learn wrong values (see discussion). However, as PSTHs are directly sums of neuronal spikes, this representational issue bears on the resulting histogram. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Backpropagation of prediction errors explains ramping activity**. (a) The TD prediction error across each of six consecutive trials (top to bottom) from the simulation in Figure 1b, with *p*~*r*~= 0.5. Highlighted in red is the error at the time of the reward in the first of the trials, and its gradual back-propagation towards the time of the stimulus in subsequent trials. Block letters indicate the outcome of each specific trial (R = rewarded; N = not rewarded). The sequence of rewards preceding these trials is given on the top right. (b) The TD error from these six trials, and four more following them, superimposed. The red and green lines illustrate the envelope of the errors in these trials. Summing over these trials results in no above-baseline activity on average (black line), as positive and negative errors occur at random 50% of the time, and so cancel each other. (c) However, when the prediction errors are asymmetrically represented above and below the baseline firing rate (here negative errors were asymmetrically scaled by *d*= 1/6 to simulate the asymmetric encoding of prediction errors by DA neurons), an average ramping of activity emerges when averaging over trials, as is illustrated by the black line. All simulation parameters are the same as in Figure 1b,d. ::: ![](1744-9081-1-6-2) ::: Figures [1b,d](#F1){ref-type="fig"} show the ramp arising from this combination of asymmetric coding and inter-trial averaging, for comparison with the experimental data. Figure [1b](#F1){ref-type="fig"} shows the PSTH computed from our simulated data by averaging over the asymmetrically-represented *δ*(*t*) signal in \~50 trials for each stimulus type. Figure [1d](#F1){ref-type="fig"} shows the results for the *p*~*r*~= 0.5 case, divided into rewarded and unrewarded trials for comparison with Figure [1c](#F1){ref-type="fig"}. The simulated results resemble the experimental data closely in that they replicate the net positive response to the uncertain rewards, as well as the ramping effect, which is highest in the *p*~*r*~= 0.5 case. It is simple to derive the average response at the time of the reward (*t = N*) in trial *T*, *i.e.,*the average TD error ![](1744-9081-1-6-i3.gif)*δ*~*T*~(*N*)![](1744-9081-1-6-i4.gif), from the TD learning rule with the simplified tapped delay-line time representation and a fixed learning rate *α*. The value at the next to last timestep in a trial, as a function of trial number (with initial values taken to be zero), is ![](1744-9081-1-6-i1.gif) where *r*(*t*) is the reward at the end of trial *t*. The error signal at the last timestep of trial *T*is simply the difference between the obtained reward *r*(*T*), and the value predicting that reward *V*~*T*-\ 1~(*N*- 1). This error is positive with probability *p*~*r*~, and negative with probability (1 - *p*~*r*~). Scaling the negative errors by a factor of *d*∈ (0, 1\], we thus get ![](1744-9081-1-6-i2.gif) For symmetric coding of positive and negative errors (*d*= 1), the average response is 0. For asymmetric coding (0 \<*d*\< 1), the average response is indeed proportional to the variance of the rewards, and thus maximal at *p*~*r*~= 0.5. However, *δ*~*T*~is positive, and concomitantly, the ramps are positive, and in this particular setting, are related to uncertainty, *because of*, rather than *instead of*, the coding of *δ*(*t*). Indeed, there is a key difference between the uncertainty and TD accounts of the ramping activity. According to the former, the ramping is a within-trial phenomena, coding uncertainty in reward; by contrast, the latter suggests that ramps arise only through averaging across multiple trials. Within a trial, when averaging over simultaneously recorded neurons rather than trials, the traces should not show a smooth ramp, but intermittent positive and negative activity corresponding to back-propagating prediction errors from the immediately previous trials (as in Figure [2a](#F2){ref-type="fig"}). Trace conditioning: a test case =============================== An important test case for our interpretation arises in a variant of Fiorillo *et al.\'s*\[[@B15]\] task, as well as in the analogous instrumental task of Morris *et al.*\[[@B16]\], both involving trace conditioning. In contrast to delay conditioning (Figure [3a](#F3){ref-type="fig"}) in which the reward coincides with the offset of the predictive stimulus, here there is a substantial gap between the offset of the predictive stimulus and the delivery of the reward (Figure [3b](#F3){ref-type="fig"}). Clearly, in this case, uncertainty about the reward could only get larger, owing to noise in timing the interval between stimulus and reward \[[@B19]\], so under the uncertainty account, there should be comparable or even larger ramps. However, the experimental results show the ramping activity to be *smaller*, or even negligible (Figure [3c;d](#F3){ref-type="fig"}). Note, though, that the magnitude of the trial-average activity at the expected time of reward is maintained, pointing to a dissociation between the height of the ramp and the amount of positive activity at the expected time of reward. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Trace conditioning with probabilistic rewards**. (a) An illustration of one trial of the delay conditioning task of Fiorillo *et al.*\[15\]. A trial consists of a 2-second visual stimulus, the offset of which coincides with the delivery of the juice reward, if such a reward is programmed according to the probability associated with the visual cue. In unrewarded trials the stimulus terminated without a reward. In both cases an inter-trial interval of 9 seconds on average separates trials. (b) An illustration of one trial of the trace conditioning task of Morris *et al.*\[16\]. The crucial difference is that there is now a substantial temporal delay between the offset of the stimulus and the onset of the reward (the \"trace\" period), and no external stimulus indicates the expected time of reward. This confers additional uncertainty as precise timing of the predicted reward must be internally resolved, especially in unrewarded trials. In this task, as in \[15\], one of several visual stimuli (not shown) was presented in each trial, and each stimulus was associated with a probability of reward. Here, also, the monkey was requested to perform an instrumental response (pressing the key corresponding to the side in which the stimulus was presented), the failure of which terminated the trial without a reward. Trials were separated by variable inter-trial intervals. (c,d) DA firing rate (smoothed) relative to baseline, around the expected time of the reward, in rewarded trials (c) and in unrewarded trials (d). (c,d) Reprinted from \[16\] ©2004 with permission from Elsevier. The traces imply an overall positive response at the expected time of the reward, but with a very small, or no ramp preceding this. Similar results were obtained in a classical conditioning task briefly described in \[15\], which employed a trace conditioning procedure, confirming that the trace period, and not the instrumental nature of the task depicted in (b) was the crucial difference from (a). ::: ![](1744-9081-1-6-3) ::: The TD model of DA readily explains these puzzling data. As shown in Figure [4](#F4){ref-type="fig"}, the shape of the ramp, though not the height of its peak, is affected by the learning rate. The size of the back-propagating prediction errors is determined, in part, by the learning rate, as these errors arise as part of the online learning of new predictions. Indeed, there is a continuous updating of predictions such that after a rewarded trial, there is a higher expectation of reward (and thus the next reward incurs a smaller prediction error), and conversely after a non-rewarded trial \[[@B18]\] (see Figure [2a](#F2){ref-type="fig"}). This updating of predictions is directly related to the learning rate -- the higher the learning rate, the larger the update of predictions according to the current prediction error, and the larger the fraction of the prediction error which is propagated back. In this way, with higher learning rates, the difference in expectations after a rewarded versus an unrewarded trial will be larger, and thus the prediction errors when the next reward is or is not available will be larger -- hence the larger and more gradual ramp. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Dependence of the ramp on learning rate.**The shape of the ramp, but not the height of its peak, is dependent on the learning rate. The graph shows simulated activity for the case of *p*~*r*~= 0.5 near the time of the expected reward, for different learning rates, averaged over both rewarded and unrewarded trials. According to TD learning with persistent asymmetrically coded prediction errors, averaging over activity in rewarded and unrewarded trials results in a ramp up to the time of reward. The height of the peak of the ramp is determined by the ratio of rewarded and unrewarded trials, however, the breadth of the ramp is determined by the rate of back-propagation of these error signals from the time of the (expected) reward to the time of the predictive stimulus. A higher learning rate results in a larger fraction of the error propagating back, and thus a higher ramp. With lower learning rates, the ramp becomes negligible, although the positive activity (on average) at the time of reward is still maintained. Note that although the learning rate used in the simulations depicted in Figure 1b,d was 0.8, this should not be taken as the literal synaptic learning rate of the neural substrate, given our schematic representation of the stimulus. In a more realistic representation in which a population of neurons is active at every timestep, a much lower learning rate would produce similar results. ::: ![](1744-9081-1-6-4) ::: Indeed, compared to delay conditioning, trace conditioning is notoriously slow, suggesting that the learning rate is low, and thus that there should be a lower ramp, in accord with the experimental results. A direct examination of the learning rate in the data of Morris *et al.*\[[@B16]\], whose task required excessive training as it was not only a trace conditioning one but also involved an instrumental action, confirmed it indeed to be very low (Genela Morris -- personal communication, 2004). Discussion ========== The differential coding of positive and negative values by DA neurons is evident in all the studies of the phasic DA signal, and can be regarded as an inevitable consequence of the low baseline activity of these neurons. Indeed, the latter has directly inspired suggestions that an opponent neurotransmitter, putatively serotonin, be involved in representing and therefore learning the negative prediction errors \[[@B20]\], so that they also have full quarter. Here, however, we have confined ourselves to consideration of the effects of asymmetry on the trial-average analysis of the dopamine activity, and have shown that ramping DA activity, as well as an average positive response at the time of reward, result directly from the asymmetric coding of prediction errors. Apart from a clearer view of the error signal, the most important consequence of the new interpretation is that the ramps can be seen as a signature of a TD phenomenon that has hitherto been extremely elusive. This is the progressive back-propagation of the error signal represented by DA activity, from the time of reward to the time of the predictor (Figure [2a](#F2){ref-type="fig"}). Most previous studies of dopaminergic activity have used *p*~*r*~= 1, so making this back-propagation at best a transitory phenomenon apparent only at the beginning of training (when, typically, recordings have not yet begun), and potentially hard to discern in slow-firing DA neurons. Further, as mentioned, the back-propagation depends on the way that the time between the predictive stimulus and the reward is represented -- it is present for a tapped delay-line representation as in \[[@B6]\], but not for representations which span the entire delay, such as in \[[@B21]\]. Note that the shape of the ramp also depends on the use of eligibility traces and the so-called TD(*λ*) learning rule (simulation not shown), which provide an additional mechanism for bridging time between events during learning. Unfortunately, as the forms of the ramps in the data are rather variable (figure [1](#F1){ref-type="fig"}) and noisy, they can not provide strong constraints on the precise TD mechanism used by the brain. More recent studies involving persistent prediction errors also show activity suggestive of back-propagation, notably Figure 4 of \[[@B13]\]. In this study, prediction errors resulted from periodic changes in the task, and DA recordings were made from the onset of training, thus back-propagation-like activity is directly apparent, although this activity was not quantified. We expect the ramps to persist throughout training only if the learning rate does not decrease to zero as learning progresses. Pearce & Hall\'s \[[@B22]\] theory of the control of learning by uncertainty suggests exactly this persistence of learning -- and there is evidence from partial reinforcement schedules that the learning rate may be higher when there is more uncertainty associated with the reward. Indeed, from a \'rational\' statistical point of view, learning should persist when there is substantial uncertainty about the relationship between predictors and outcomes, as can arise from the ever-present possibility of a change in the predictive relationships. This form of persistent uncertainty, together with uncertainty due to initial ignorance regarding the task, have been used to formalize Pearce & Hall\'s theory of the way that uncertainty drives learning \[[@B23]\]. Thus, our claim that uncertainty may not be directly represented by the ramps, should certainly not be taken to mean that its representation and manipulation is not important. To the contrary, we have suggested that uncertainty influences cortical inference and learning through other neuromodulatory systems \[[@B24]\], and that it also may determine aspects of the selection of actions \[[@B25]\]. Various other features of the asymmetry should be noted. Most critical is the effect of the asymmetry on DA-dependent learning \[[@B26]\], if the below baseline DA activity is responsible by itself for decreasing predictions which are too high. In order to ensure that the learned predictions remain correct, we would have to assume that the asymmetric representation does not affect learning, *i.e.,*that a mechanism such as different scaling for potentiation and depression of the synaptic strengths compensates for the asymmetric error signal. Of course, this would be rendered moot if an opponent neurotransmitter is involved in learning from negative prediction errors. This issue is complicated by the suggestion of Bayer \[[@B14]\] that DA firing rates are actually similar for all prediction errors below some negative threshold, perhaps due to the floor effect of the low firing rate. Such lossy encoding does not affect the qualitative picture of the effects of inter-trial averaging on the emergence of ramps, but does reinforce the need for an opponent signal for the necessarily symmetric learning. Finally, the most direct test of our interpretation would be a comparison of intra- and inter-trial averaging of the DA signal. It would be important to do this in a temporally sophisticated manner, to avoid problems of averaging non-stationary signals. In order to overcome the noise in the neural firing, and determine whether indeed there was a gradual ramp within a trial, or, as we would predict -- intermittent positive and negative prediction errors, it would be necessary to average over many neurons recorded simultaneously within one trial, and furthermore neurons associated with similar learning rates. Alternatively, single neuron traces could be regressed against the backpropagation response predicted by their preceding trials and TD learning. A comparison of the amount of variability explained by such a model, compared to that from a regression against a monotonic ramp of activity, could point to the most fitting model. A less straightforward, but more testable prediction is that the shape of the ramp should depend on the learning rate. Learning rates can be assessed from the response to the probabilistic rewards, independent of the shape of the ramp (Nakahara *et al.*\[[@B18]\] showed in such a way, that in their partial reinforcement trace conditioning task, the learning rate was 0.3), and potentially manipulated by varying the amount of training or the frequency with which task contingencies are changed and relearned. Indeed, quantifying the existence and shape of a ramp in Nakahara *et al.\'s*recorded DA activity, could well shed light on the current proposal. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= YN, MD and PD jointly conceived and executed this study, and helped draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We are very grateful to H. Bergman, C. Fiorillo, N. Daw, D. Joel, P. Tobler, P. Shizgal and W. Schultz for discussions and comment, in some cases despite varying interpretation of the data. We are particularly grateful to Genela Morris for analyzing her own published and unpublished data in relation to ramping. This work was funded by the EC Thematic Network (YN), the Gatsby Charitable Foundation and the EU BIBA project.

PubMed Central

2024-06-05T03:55:58.286424

2005-5-4

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1171969/", "journal": "Behav Brain Funct. 2005 May 4; 1:6", "authors": [ { "first": "Yael", "last": "Niv" }, { "first": "Michael O", "last": "Duff" }, { "first": "Peter", "last": "Dayan" } ] }

PMC1173078

Introduction ============ In 1996, Salkeld et al. \[[@B1]\] found that screening the average-risk Australian population for colorectal cancer using a faecal occult blood test (FOBT), compared to existing practice, would cost \$24,660 per life-year gained (LYG). Due to uncertainty regarding the effectiveness of FOBT screening, they report a range of values, between \$12,695 and \$67,848 per LYG. Randomised controlled trials of population screening with FOBT conducted in the UK \[[@B2]\] and Denmark \[[@B3]\], but published after Salkeld et al.\'s study, have reduced this uncertainty. The cost-effectiveness analyses based on the UK trial data \[[@B4]\] suggest a cost per life-year gained between £1,371--£5,685 (approximately \$AU3,370--13,974) and the analysis of the Danish trial data \[[@B5]\] suggest a cost per life-year gained between 17,000--42,000DKK (approximately \$AU3,916--9,672). Three years before publication of Salkeld\'s study Bolin \[[@B6]\] discussed the advantages of using colonoscopy for population screening, and in 1996, he suggested colonoscopy was cost-effective \[[@B7]\]. In 1997, he asked whether the time had come to use colonoscopy for population screening in Australia \[[@B8]\]. Kermond \[[@B9]\] responded suggesting double contrast barium enema (DCBE) should not be overlooked arguing colonoscopy is 10 times more expensive, false negatives still occur and complication rates are higher. Bolin argued, in the same issue of the MJA, that the sensitivity of colonoscopy exceeds DCBE, the complication rate is only 0.1% and cost differentials are actually less than those suggested by Kermond \[[@B9]\]. Bolin also claimed that FOBT at one and three years and colonoscopy at 10 years, assuming a 10-year period during which time the cancer is detectable and curable (known as the dwell time), are cost-effective modes of CRC screening, probably referring to data subsequently published in 1999 \[[@B10]\]. For this research, the authors substituted Australian values for cost parameters into a US model of the cost-effectiveness of CRC screening \[[@B11]\] and generalised the results to the Australian population. They reported change in cost and change in life-years gained, as compared to existing practice, for competing screening strategies that encompass FOBT, colonoscopy, flexible sigmoidoscopy and DCBE \[[@B10]\]. By assuming that society is willing to pay up to \$US40,000 (approximately \$AU65,449 in 2002 prices) per LYG, Bolin proposed that annual FOBT, triennial FOBT, triennial DCBE, five-yearly DCBE, five-yearly colonoscopy and ten-yearly colonoscopy are all cost-effective and concluded that physicians have the option of offering individuals a range of screening alternatives, including colonoscopy \[[@B10]\]. Since publishing the research Bolin has argued, on four separate occasions that population screening with colonoscopy is cost-effective \[[@B12]-[@B15]\]. The last of these, in 2002 \[[@B14]\], provoked Macrae and Hebbard \[[@B16]\] to criticise Bolin\'s interpretation of epidemiological data. In 2004, O\'Leary et al. \[[@B17]\] also addressed the economic questions around population screening in Australia. They estimated the cost-effectiveness, compared to existing practice, of FOBT, flexible sigmoidoscopy and colonoscopy, and found flexible sigmoidoscopy and colonoscopy were cost-effective but FOBT was not. There are important differences in the way that Salkeld et al. \[[@B1]\], Bolin et al. \[[@B10]\] and O\'Leary et al. \[[@B17]\] report the Australian cost and effectiveness data. Salkeld et al. reports an incremental cost-effectiveness ratio, O\'Leary et al. reports both average and incremental cost-effectiveness ratios, however they draw their conclusions from an average analysis. Bolin only calculates average cost-effectiveness ratios. The correct ratio for decision-making is an incremental cost-effectiveness ratio: McMahon \[[@B18]\] argues the use of average ratios is not meaningful; Drummond \[[@B19]\] and Gold \[[@B20]\] both discuss why incremental rather than average cost-effectiveness ratios are relevant for decision making; and, both Torgerson \[[@B21]\] and Neuhauser & Lewicki \[[@B22]\] provide examples of how average analyses muddy the waters. In their much-cited 1975 paper, Neuhauser & Lewicki \[[@B22]\] reviewed data on screening for CRC. They illustrated that repeatedly testing a stool sample up to six times, when a previous test result was negative, would capture all cases of CRC, at an average cost per case of \$2451. They also did an incremental analysis, with the same data, and showed the incremental cost per case detected, from the fifth to sixth round of testing was \$47 million. This illustrates that average analyses can be grossly misleading. We have four objectives in this paper: first, to demonstrate why incremental, not average, cost-effectiveness ratios should be used for decision-making; second, to update the cost data reported by Salkeld et al. \[[@B1]\], Bolin et al. \[[@B10]\] and O\'Leary et al. \[[@B17]\] to 2002 Australian dollar prices; third, to calculate incremental cost-effectiveness ratios from the Bolin data; and fourth, to discuss the results of our re-analysis, comparing the outcomes from the three previous studies. This will provide readers with an up-to-date and appropriate assessment of the existing cost-effectiveness data for population-based CRC screening programmes in Australia. Defining average & incremental cost-effectiveness ratios ======================================================== Average cost-effectiveness ratios for health care interventions are the amounts by which costs change from a baseline comparator (Δ*C*) divided by the amount by which health benefits change from a baseline comparator (Δ*E*). The baseline comparator is often existing practice. We illustrate this method with hypothetical data in Table [1](#T1){ref-type="table"} by presenting the change in costs and health effects that arise from four competing health care alternatives. If we remain with Existing Practice, there is no change in cost or health effect. However, if we are committed to generating health effects and we wish to be efficient, then we should choose the cheapest option that improves health outcomes. The data in Table [1](#T1){ref-type="table"} show that Intervention 4 generates the best ratio of (Δ*C*) and (Δ*E*) when compared to Existing Practice. Cost changes by \$145,000 and health effects change by 150 and the cost per LYG is \$967. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### An illustration of average cost-effectiveness ratios for four competing hypothetical health care interventions ::: Change in cost (\$) Change in health effect (Life-years Gained) Average cost-effectiveness (\$) ------------------- --------------------- --------------------------------------------- --------------------------------- (Δ*C*) (Δ*E*) (Δ*C*) divided by (Δ*E*) Existing Practice 0 0 Intervention 1 200,000 12 16,667 Intervention 2 75,000 15 5,000 Intervention 3 300,000 250 1,200 Intervention 4 145,000 150 967 ::: We reject Interventions 1 and 2. The reasons lie in Figure [1](#F1){ref-type="fig"} (which is a graph of the data in Table [1](#T1){ref-type="table"}). Intervention 1 generates less health benefit and higher costs than Intervention 4, a situation described as \'simple dominance\'. For this comparison, Intervention 4 is preferred on both costs and outcomes. Intervention 2 also generates less health effect but differs from Intervention 1 in that it\'s cheaper than Intervention 4; however, the cost per LYG from Intervention 2 is greater than the cost per LYG from Intervention 4. This situation is known as \'extended dominance\' and is only relevant if the cost of Intervention 4 (\$145,000) exceeds the total amount of money available to the decision-maker. Rather than choosing Intervention 2 over Intervention 4, it would be better (more productively efficient) to choose some blend of existing practice and Intervention 4. This implies that some proportion of the population would receive Intervention 4 and the remainder would receive existing practice. This raises questions of equity of access and so poses another set of problems for decision-makers. If the available budget exceeds \$145,000, there is a further question to consider. Do we invest in the more costly but more effective Intervention 3? Some care is required when making this decision. The average cost-effectiveness ratio for Intervention 3 (\$1,200 per LYG), represented by the dashed line on Figure [1](#F1){ref-type="fig"}, is misleading. It\'s calculated by comparing Intervention 3 to existing practice; yet, the relevant decision is whether we should invest in Intervention 3 given that we have established Intervention 4 as the most cost-effective option. We must consider the incremental changes in cost and health effects compared to the next best alternative, Intervention 4. The incremental cost-effectiveness ratios for Interventions 4 and 3 are marked with solid lines on Figure [1](#F1){ref-type="fig"}. When a more effective alternative also costs more, then the decision-maker must compare the increased cost with the increased effects \[[@B19]\]. The only way to achieve this is to conduct an incremental analysis, which we illustrate in Table [2](#T2){ref-type="table"}. Investing in Intervention 3, as compared to 4, changes total costs by \$155,000 and LYG by 100, yielding an incremental cost-effectiveness ratio of \$1,550 per LYG not the \$1,200 per LYG as previously estimated using average cost-effectiveness ratios and illustrated in Table [1](#T1){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### An illustration of average and incremental cost-effectiveness ratios for the two remaining hypothetical health care interventions ::: Intervention Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ----------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- **(a)** **(b)** **(c)** **(d)** **(a)/(c)** **(b)/(d)** Existing Practice 0 0 Intervention 4 145,000 145,000 150 150 967 967 Intervention 3 300,000 155,000 250 100 1,200 1,550 Note: Interventions 1 and 2 have been rejected on the grounds of \'simple\' and \'extended\' dominance, see text for a discussion ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Change in cost and change in effect from four hypothetical health care interventions. ::: ![](1743-8462-2-10-1) ::: Imagine you have decided to take a holiday in a beach resort. You face a decision between a standard apartment for \$1,000 and a Penthouse apartment for \$1,300. Because you have decided to take the holiday (and so one of the apartments), it is the difference in cost (\$300) that you compare to the difference in benefit (penthouse vs. standard apartment). If you don\'t perceive the additional benefit to be worth the extra \$300, then you reject the penthouse. This simple example illustrates the importance of thinking about decisions in terms of incremental changes. An average analysis with both options compared to \'no holiday\' may lead to a bad decision. Methods for re-analysis and the results ======================================= We converted Bolin\'s \$US dollar estimates to Australian dollars with the exchange rate reported in the original article \[[@B10]\] and adjusted the estimates reported by Bolin, Salkeld and O\'Leary to 2002 prices using a health price index \[[@B23]\]. As Salkeld and O\'Leary reported incremental cost-effectiveness ratios, we only need to calculate incremental cost-effectiveness ratios from the Bolin data. We achieved this by inputting the reported estimates of (Δ*C*) and (Δ*E*) into decision analysis software \[[@B24]\]. Assuming a 5-year dwell time (the period during which cancer can be detected and cured), we include 13 screening strategies and the existing practice comparator. For a 10-year dwell time, estimates of (Δ*C*) and (Δ*E*)were reported for different, additional, frequencies of FOBT, flexible sigmoidoscopy and DCBE screening, resulting in 27 strategies and the existing practice comparator. We ranked all strategies by increasing cost, estimated incremental cost and effectiveness, and excluded all strategies for which other options prevailed on the basis of \'simple\' or \'extended\' dominance. Finally, we reported the strategies not excluded due to either \'simple\' or \'extended\' dominance and present the relevant incremental cost-effectiveness ratios. In Table [3](#T3){ref-type="table"}, we describe all strategies evaluated by the authors of the three studies. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Descriptions of the screening strategies included in the re-analysis ::: Screening strategy Description of screening strategy Salkeld \[1\] Bolin \[10\] (5-year dwell time) Bolin \[10\] (10-year dwell time) O\'Leary \[17\] -------------------- --------------------------------------------------------------------------- --------------- ---------------------------------- ----------------------------------- ----------------- Existing practice Existing screening practices X X X COL10 10-yearly colonoscopy X X X COL5 5-yearly colonoscopy X X COL one off screening colonoscopy at age 50 X X DCBE one off double contrast barium enema X DCBE10 10-yearly double contrast barium enema X DCBE15 15-yearly double contrast barium enema X DCBE20 20-yearly double contrast barium enema X DCBE3 3-yearly double contrast barium enema X X DCBE5 5-yearly double contrast barium enema X X FOBT10 10-yearly faecal occult blood test X FOBT15 15-yearly faecal occult blood test X FOBT2 2-yearly faecal occult blood test X X FOBT20 20-yearly faecal occult blood test X FOBT5 5-yearly faecal occult blood test X FOBT one off faecal occult blood test X FOBT1 annual faecal occult blood test X X X X FOBT1+DCBE3 annual faecal occult blood test and 3-yearly double contrast barium enema X X FOBT1+DCBE5 annual faecal occult blood test and 5-yearly double contrast barium enema X X FOBT1+FSIG3 annual faecal occult blood test and 3-yearly flexible sigmoidoscopy X X FOBT1+FSIG5 annual faecal occult blood test and 5-yearly flexible sigmoidoscopy X X FOBT3 3-yearly faecal occult blood test X X FSIG flexible sigmoidoscopy once only X FSIG10 10-yearly flexible sigmoidoscopy X X FSIG15 15-yearly flexible sigmoidoscopy X FSIG20 20-yearly flexible sigmoidoscopy X FSIG3 3-yearly flexible sigmoidoscopy X X FSIG5 5-yearly flexible sigmoidoscopy X X ::: In Figure [2](#F2){ref-type="fig"}, we illustrate the costs and effects of all strategies reported by Bolin, assuming a 5-year dwell time. The strategies to the left and above the cost-effective frontier, defined by the solid line, are excluded by either \'simple\' or \'extended\' dominance. The three remaining (un-dominated) strategies, that define the cost-effective frontier, are EXISTING PRACTICE, DCBE3 and FOBT1+DBCE3. In Table [4](#T4){ref-type="table"}, we report all corresponding costs, health benefits and cost-effectiveness ratios, and indicate the options that are \'simply\' dominated. In Table [5](#T5){ref-type="table"}, we report the incremental cost-effectiveness ratios for the options that survive the tests of \'simple\' or \'extended\' dominance; these options define the cost-effective frontier. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Updated costs and effects for Bolin et al\'s 13 tested strategies, assuming a five-year dwell time. ::: ![](1743-8462-2-10-2) ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Estimates of costs in 2002 prices, health benefits and cost-effectiveness from Bolin *et al*. \[10\], assuming a five-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- **(a)** **(b)** **(c)** **(d)** **(a)/(c)** **(b)/(d)** EXISTING PRACTICE 0 0 (Undefined) FOBT3 109,167,314 109,167,314 2,010 2,010 54,312 54,312 COL 195,987,062 86,819,748 1,166 -844 168,085 (Simply Dominated) FSIG5 205,570,434 96,403,120 3,365 1,355 61,091 71,146 FOBT1 230,156,355 24,585,921 4,447 1,082 51,755 22,723 DCBE5 253,881,621 23,725,266 5,050 603 50,274 39,345 FSIG3 271,052,169 17,170,548 3,909 -1,141 69,341 (Simply Dominated) COL10 278,701,522 24,819,901 3,718 -1,332 74,960 (Simply Dominated) DCBE3 307,911,416 54,029,795 6,184 1,134 49,792 47,645 COL5 360,264,079 52,352,663 6,181 -3 58,286 (Simply Dominated) FOBT1+FSIG5 364,595,167 56,683,751 5,849 -335 62,335 (Simply Dominated) FOBT1+DCBE5 373,803,843 65,892,427 6,573 389 56,870 169,389 FOBT1+FSIG3 420,786,416 46,982,573 6,032 -541 69,759 (Simply Dominated) FOBT1+DBCE3 424,911,339 51,107,496 7,020 447 60,529 114,334 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Incremental cost-effectiveness ratios in 2002 prices for the preferred (not dominated) strategies from Bolin et al. \[10\], assuming a five-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- **(a)** **(b)** **(c)** **(d)** **(a)/(c)** **(b)/(d)** EXISTING PRACTICE 0 0 (Undefined) DCBE3 307,911,416 307,911,416 6,184 6,184 49,792 49,792 FOBT1+DBCE3 424,911,339 116,999,923 7,020 836 60,529 139,952 Note: Other options rejected on the grounds of \'simple\' and \'extended\' dominance, see text for a discussion. ::: In Figure [3](#F3){ref-type="fig"}, we illustrate the costs and effects for all strategies reported by Bolin, assuming a 10-year dwell time. In this case EXISTING PRACTICE, FOBT2, DCBE5, DCBE3, FOBT1+DCBE5 and FOBT1+DCBE3 define the cost-effective frontier. In Table [6](#T6){ref-type="table"}, we report all corresponding costs, health benefits and cost-effectiveness ratios, and indicate the options that are \'simply\' dominated. In Table [7](#T7){ref-type="table"}, we report the incremental cost-effectiveness ratios for the options that survive the tests of \'simple\' or \'extended\' dominance; again, these options define the cost-effective frontier. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Estimates of costs in 2002 prices, health benefits and cost-effectiveness from Bolin *et al*. \[10\], assuming a 10-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- **(a)** **(b)** **(c)** **(d)** **(a)/(c)** **(b)/(d)** EXISTING PRACTICE 0 0 (Undefined) FOBT 24,429,089 24,429,089 352 352 69,401 69,401 FOBT20 31,633,953 7,204,864 558 206 56,692 34,975 FOBT15 37,620,230 5,986,277 681 123 55,243 48,669 FOBT10 47,472,852 9,852,622 951 270 49,919 36,491 FOBT5 75,561,379 28,088,527 1,674 723 45,138 38,850 FSIG 102,900,905 27,339,526 1,222 -452 84,207 (Simply Dominated) FOBT3 105,521,804 29,960,425 2,605 931 40,507 32,181 FSIG20 120,362,301 14,840,497 1,892 -713 63,616 (Simply Dominated) FSIG15 131,780,307 26,258,503 2,294 -311 57,446 (Simply Dominated) DCBE 134,697,657 29,175,853 1,896 -709 71,043 (Simply Dominated) FOBT2 140,117,791 34,595,987 3,551 946 39,459 36,571 FSIG10 148,850,170 8,732,379 3,127 -424 47,602 (Simply Dominated) DCBE20 158,390,199 18,272,408 2,939 -612 53,893 (Simply Dominated) DCBE15 173,448,383 33,330,592 3,566 15 48,639 2,222,039 COL 190,862,405 17,414,022 2,368 -1,198 80,601 (Simply Dominated) DCBE10 194,089,298 20,640,915 4,778 1,212 40,621 17,030 FSIG5 204,328,539 10,239,241 3,583 -1,195 57,027 (Simply Dominated) FOBT1 224,367,390 30,278,092 5,271 493 42,566 61,416 DCBE5 248,112,291 23,744,901 6,023 752 41,194 31,576 COL10 265,297,565 17,185,274 5,970 -53 44,438 (Simply Dominated) FSIG3 270,484,399 22,372,108 3,993 -2,030 67,740 (Simply Dominated) DCBE3 304,301,903 56,189,612 6,720 697 45,283 80,616 COL5 357,822,831 53,520,928 6,583 -137 54,356 (Simply Dominated) FOBT1+FSIG5 361,026,560 56,724,657 6,344 -376 56,908 (Simply Dominated) FOBT1+DCBE5 370,181,240 65,879,337 7,076 356 52,315 185,054 FOBT1+FSIG3 417,677,588 47,496,348 6,457 -619 64,686 (Simply Dominated) FOBT1+DCBE3 422,777,702 52,596,462 7,299 223 57,923 235,859 ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Incremental cost-effectiveness ratios in 2002 prices for the preferred (not dominated) strategies from Bolin et al. \[10\], assuming a 10-year dwell time ::: Strategy Cost (\$) Incremental changes in cost (\$) Effectiveness (LYG) Incremental changes in effectiveness (LYG) Average cost-effectiveness ratio (\$) Incremental cost-effectiveness ratio (\$) ------------------- ------------- ---------------------------------- --------------------- -------------------------------------------- --------------------------------------- ------------------------------------------- **(a)** **(b)** **(c)** **(d)** **(a)/(c)** **(b)/(d)** EXISTING PRACTICE 0 0 (Undefined) FOBT2 140,117,791 140,117,791 3,551 3,551 39,459 39,459 DCBE5 248,112,291 107,994,500 6,023 2,472 41,194 43,687 DCBE3 304,301,903 56,189,612 6,720 697 45,283 80,616 FOBT1+DCBE5 370,181,240 65,879,337 7,076 356 52,315 185,054 FOBT1+DCBE3 422,777,702 52,596,462 7,299 223 57,923 235,859 Note: Other options rejected on the grounds of \'simple\' and \'extended\' dominance, see text for a discussion ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Updated costs and effects for Bolin et al\'s 27 tested strategies, assuming a ten-year dwell time. ::: ![](1743-8462-2-10-3) ::: The incremental cost-effectiveness ratio derived from the Salkeld data, in 2002 prices, for FOBT, is \$30,556 per LYG, and the 2002 incremental cost-effectiveness ratios from the O\'Leary data are \$17,356 per LYG for FSIG and \$26,587 per LYG for COL. These results define the screening options that are preferred (ie, not dominated), by the measure of cost-effectiveness, for population-based CRC screening in Australia. However, the decision over which to choose depends on additional factors that we discuss next. Discussion ========== We defined average and incremental cost-effectiveness ratios and emphasise the latter are relevant for decision-making. We calculated incremental cost-effectiveness ratios, in 2002 prices, for a number of population screening strategies, for which Bolin had previously reported average cost-effectiveness ratios. For Bolin\'s estimates of (Δ*C*) and (Δ*E*), for a 5-year dwell time, we found only DCBE3 (\$49,792 per LYG) and FOBT1+DCBE3 (\$139,952 per LYG) were preferred (not dominated). For a 10-year dwell time, FOBT2 (\$39,459 per LYG), DCBE5 (\$43,687 per LYG), DCBE3 (\$80,616 per LYG), FOBT1+DCBE5 (\$185,054 per LYG) and FOBT1+DCBE3 (\$235,859 per LYG) were preferred (not dominated). Incremental cost-effectiveness ratios are useful for decision-making when a ceiling value for a LYG is specified. Bolin argued in 1996 \[[@B7]\] that \$US40,000 per LYG was the relevant cut-off (approximately \$AU65,449 in 2002 prices). We prefer a decision rule described by Garber & Phelps \[[@B25]\] that states a LYG is worth approximately twice the median annual per capita income. They derived this value from a model of optimal lifetime spending for medical care and explored its relationship to the cost-effectiveness criterion. They evaluated the model in terms of maximizing utility for individuals, with utility a function of income and health. Their rule implies, for Australia, a rational cut-off for one LYG is approximately \$AU39,000 \[[@B26]\]. If we apply this rule to our interpretation of Bolin\'s data, we wouldn\'t recommend any additional population screening activities for the 5-year dwell time, and for a 10-year dwell time, we would only recommend FOBT2. Based on the O\'Leary and Salkeld data we recommend COL and FOBT1, respectively. The results in Tables [4](#T4){ref-type="table"} and [6](#T6){ref-type="table"} illustrate the colonoscopy strategies, championed by Bolin, would cause between 33 and 1,322 years of life to be lost and between \$M17 and \$M87 to be wasted. We showed that all colonoscopy options were dominated by more cost-effective alternatives. Despite our re-analysis, the decision over which model of CRC screening is optimal for the Australian, average-risk population remains ambiguous. While the incremental cost-effectiveness ratios from two studies support annual FOBT \[[@B1]\] or biennial FOBT \[[@B10]\], the most recent study supports colonoscopy \[[@B17]\]. At least we now have incremental cost-effectiveness ratios in 2002 prices. While two of the studies agree that FOBT screening is preferred, we have not investigated why the O\'Leary \[[@B17]\] analysis leads to a different conclusion. The answer may be sought in a careful assessment of model structures, the particular perspectives adopted for each analysis and the values used for the parameters, which is beyond the scope of this commentary. In addition, we haven\'t attempted to model the effect of uncertainty on the conclusions. This would require access to the models, data and software used in each of the previous three studies. Policy-makers should review the three papers and make a judgement over which they believe produces the best estimates of change in cost and health benefit, identify their willingness to pay for the proposed health benefits and make their decision in the context of other logistic, social and political issues. Competing interests =================== The author(s) declare that they have no competing interests.

PubMed Central

2024-06-05T03:55:58.289772

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173078/", "journal": "Aust New Zealand Health Policy. 2005 May 18; 2:10", "authors": [ { "first": "Nicholas", "last": "Graves" }, { "first": "Loretta", "last": "McKinnon" }, { "first": "Barbara", "last": "Leggett" }, { "first": "Beth", "last": "Newman" } ] }

PMC1173079

Background ========== Coexistence of public and private health insurance, such as in Australia, has been the subject of intense debate among health economists and policy makers \[[@B1]\]. The main issue surrounding this debate has been how the mix of public and private health care financing influences the demand for private health insurance (PHI) and whether PHI takes pressure off the public system \[[@B2]\]. Falling PHI membership, observed since the introduction of Medicare in 1984, was thought to have increased the demand on the public system \[[@B3]\], prompting the federal government to implement policies aimed at encouraging possession of PHI to take the pressure off public hospitals and restore balance to the health care system \[[@B3]-[@B5]\]. Since 1995 three major policy reforms have been introduced in Australia \[[@B6]\]. Firstly, in 1995 selective contracting was introduced. The then federal Labor government passed legislation allowing private health plans to contract selectively with hospitals and doctors so as to improve competition on price and quality. Secondly, government subsidisation of PHI was introduced in 1997 by the Conservative coalition federal government as a means-tested rebate, capped at a flat amount irrespective of the cost of PHI, combined with an income tax surcharge for high income earners without PHI. The rebate part of the policy was subsequently replaced in 1999 by a non-means-tested 30% rebate for PHI available to everyone. Finally, in 2000, lifetime community rating was introduced. This policy relaxed the previous stringent community rating system by allowing the price of PHI to be varied according to the age at which a member joined \[[@B6]\]. The Australian Healthcare Agreement 1998--2003 committed the Commonwealth and states to review the relationship between PHI cover and the use of hospital services by private patients. The investigation of this relationship is becoming a priority as there is disagreement among commentators as to the financial efficiency of the 30% rebate \[[@B5],[@B7]-[@B10]\]. To date, analyses of the effects of policies aimed at supporting PHI in Australia have primarily centred on changes in the proportion of the population covered by PHI \[[@B3],[@B6],[@B11],[@B12]\]. However, changing the proportion of the population covered may not directly translate to increased utilisation and, therefore, reduce pressure on the public system. The relationship between PHI cover and type of hospital use is complex because the universality of Medicare in Australia means that anyone can be treated in a public hospital at no charge regardless of their insurance status. The right of individuals to choose between public and private insurance, regardless of the status (public or privately insured) of previous hospitalisations or the possession of PHI is protected by the principles of Medicare as set out in the Health Insurance Act of 1973. Thus switches between use of the public and PHI system are initiated solely by patients based on choice rather than being mandated by government, notwithstanding that this \'choice\' may be constrained in some instances by socioeconomic and locational access factors. We propose that since possession and utilisation of PHI are not equivalent, analysing the effectiveness of the recent government strategies in relieving the pressure on the public system cannot be accomplished by evaluations of changes in possession of PHI alone. Rather, changes in choice, as reflected by patient-initiated switching between the public and private insurance systems must be analysed. The aim of this study was to identify and measure changes in the behavioural patterns of switching between public and privately insured status for hospitalisation by the population of Western Australia using our novel couplet methodology for analysing longitudinal data. Method ====== Hospital morbidity data extraction and case selection ----------------------------------------------------- The WA Data Linkage System \[[@B13]\] was used to extract all hospital morbidity data system (HMDS) records from 1^st^January 1980 to 31^st^December 2001, containing encrypted patient identification and episode numbers, age, gender, date of admission, date of separation, payment classification (public, insured private, or \"other\"), and hospital type. The \"other\" payment category, which included the private uninsured (2.2% of the total episodes), workers compensation (1.8%), motor vehicle (0.7%), defence force personnel (0.3%) and Veteran Affairs (1.7%) classifications was removed from the data set, leaving only the categories of public and private insured. This was done because the study was concerned with elective shifts between PHI and public categories; not enforced payment classifications due to mandatory funding arrangements, or private episodes for which the patients paid the full cost. Data coverage ------------- This research has made use of records from the HMDS which is the inpatient information system for WA acute care hospitals. The data collected by the HMDS are patient identification, socio-demographic, services, administration and clinical diagnosis information. Every WA hospital defined as an acute care facility has to provide data via the Health Act or for private hospitals as part of their license and every free-standing day surgery unit must also provide data. The HMDS includes patient information for all acute A class hospitals, day surgery units, geriatric and psychiatric inpatient facilities in general hospitals and, as of July 1993, healthy newborn infants born in hospital. It does not include patients of stand-alone geriatric and psychiatric institutions, geriatric hostels or rehabilitation units; however, most metropolitan hospitals have geriatric and psychiatric units attached for which data are provided for the HMDS. Over the lifetime of this study there were no significant changes in the HMDS coverage. Formation of episodes of care and assignment of payment classification ---------------------------------------------------------------------- For each individual in the data set all eligible hospital records were grouped into episodes of care, using the separation and admission dates to define temporally contiguous periods of health care service utilisation. Thus one episode of care could have contained one or more inter hospital transfers. Each episode of care was assigned to one of the two eligible payment classifications (public or privately insured) on the basis of the initial payment classification at admission, where inter-hospital transfers were involved. Allocation of public or PHI status was based solely on payment classification and not on hospital type. This was done because the focus of this paper (consistent with the focus of public policy making in Australia) was on use of PHI rather than hospital type. Stratification by age group --------------------------- Stratification was based on age at admission of the first episode in each couplet. Each episode of care was assigned to one of four broad age categories (0--16 years, 17--39 years, 40--69 years and 70+ years) chosen to represent PHI market segmentation (children, young adults, middle age and old age) following consultation with a local private health insurer. Formation and classification of hospital couplets ------------------------------------------------- For each individual in the data set, eligible episodes of care were grouped incrementally, starting with the index (first episode of care) to form hospital couplets such that episodes 1 and 2 formed hospital couplet 1, episodes 2 and 3 formed hospital couplet 2 and so on (see figure [1](#F1){ref-type="fig"}). Hospital couplets were categorised depending upon the sequential combination of payment classifications of their two contributing episodes of care as follows: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The formation of hospital couplets. ::: ![](1743-8462-2-12-1) ::: \(i) Concordant public couplets containing only public episodes of care. \(ii) Concordant private couplets containing only privately insured episodes of care. \(iii) Discordant (mixed) public to private couplets had the first episode of care as public and the second as privately insured. \(iv) Discordant (mixed) private to public couplets had the first episode of care as privately insured and the second as public. Analysis of the mean intra-couplet interval ------------------------------------------- The intra-couplet interval was defined as the time in days between the final separation from the first episode of care and admission to the second episode of care. The mean intra-couplet interval was calculated for each couplet classification by the year of admission of the second episode of care (a term we subsequently refer to as the couplet year) and expressed as a ratio of the sum of the mean intra-couplet interval of all couplets in that year, regardless of classification. We subsequently refer to this measure as the grand mean. Thus a ratio greater than one for a particular couplet classification was indicative of an intra-couplet interval longer than the grand mean for that couplet year. The couplet year indicates the year the switch (choice) was made. Analysis of behavioural patterns in switching of payment classification ----------------------------------------------------------------------- The proportion of each type of discordant couplet relative to the total number of couplets having a first episode of care in the baseline payment classification was determined independently for all intra-couplet intervals (in whole years) in the data set. This analysis was performed separately for each age group for the whole observation period and two predefined time periods (1981 to 1989 and 1990 to 2001) chosen to represent the two main eras of health care policy in Australia. The first related to the removal and re-introduction of free public hospital care, while the second related to changes in federal health policies aimed at supporting PHI \[[@B5],[@B12],[@B14],[@B15]\]. Hospital couplets were partitioned into the two time periods using the year of admission of second episodes of care. The values obtained were plotted as segmented trend lines and the average rates of loss from each payment classification per year of intra-couplet interval were calculated using least squares fit. Results ======= Characteristics of the hospital couplets in the HMDS data file -------------------------------------------------------------- The HMDS data file contained data pertaining to 1,979,946 individuals of which 1,185,014 (60%) had at least one valid couplet. The 7,561,486 episodes of care recorded for these individuals formed 6,376,472 distinct hospital couplets (see table [1](#T1){ref-type="table"}). Of these, approximately one half (51%) had an intra-couplet interval of one year or greater. Details of the distribution of the hospital couplets by intra-couplet interval are shown in table [2](#T2){ref-type="table"}. Significant differences were observed in the characteristics of couplets having intra-couplet intervals of less than 1 year compared with those of one year or over. The largest differences were observed in relation to hospital type and couplet category. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of individuals, hospital episodes and hospital couplets ::: **Characteristic** **Individuals** **Episodes** **Couplets**(2^nd^Episode) -------------------- ---------------------- -------------- ---------------------------- --------- -------- --------- -------- **Sex** Male 522167 44.1 3156438 41.7 2634271 41.3 Female 662841 55.9 4405035 58.3 3742194 58.7 Indeterminate 4 0.0 13 0.0 7 0.0 TOTAL 1185014 100 7561486 100 6376472 100 **Age Group** 0--16 Years 313244 26.4 1045092 13.9 731848 11.5 17--39 Years 443602 37.4 2495039 33.1 2051437 32.3 40--69 Years 326974 27.6 2708572 35.9 2381598 37.5 70+ Years 100870 8.5 1292462 17.1 1191592 18.7 TOTAL 1184690 100^1^ 7541165 100^2^ 6356475 100^3^ **Hospital** Teaching 367964 31.1 2852257 37.7 2484293 39.0 **Type** Public Metropolitan 198456 16.7 902286 11.9 703830 11.0 Private Metropolitan 313563 26.5 1838354 24.3 1524791 23.9 Public Country 279315 23.6 1745527 23.1 1466212 23.0 Private Country 25097 2.1 208883 2.8 183786 2.9 TOTAL 1184395 100^4^ 7547307 100^5^ 6362912 100^6^ ^1^324 missing ^2^20321 missing ^3^19997 missing ^4^619 missing ^5^14179 missing ^6^13560 missing Missing records were caused by missing data in the relevant fields of the individual records and do not reflect deficiencies in the linkage process. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distribution of hospital couplets by characteristics and intra-couplet interval ::: **Characteristic** **Intra-Couplet Interval (I-CI)** ---------------------- ----------------------------------- --------- -------- -------- --------- -------- ------ **Sex** Male 1397957 44.6 53.1\* 1236314 38.2 46.9 Female 1739643 55.4 46.5\* 2002551 61.8 53.5 Indeterminate 5 0.0 71.4 2 0.0 28.6 TOTAL 3137605 100 49.2\* 3238867 100 50.8 **Age Group** 0--16 Years 318255 10.2 43.5\* 413593 12.8 56.5 17--39 Years 871234 27.9 42.5\* 1180203 36.5 57.5 40--69 Years 1271495 40.7 53.4\* 1110103 34.4 46.6 70+ Years 664439 21.3 55.8\* 527153 16.3 44.2 TOTAL 3125423 100^1^ 49.2\* 3231052 100^2^ 50.8 **Hospital Type** Teaching 1554708 49.6 62.6\* 929585 28.7 37.4 Public Metropolitan 249979 8.0 35.5\* 453851 14.0 64.5 Private Metropolitan 573729 18.3 37.6\* 951062 29.4 62.4 Public Country 672844 21.4 45.9\* 793368 24.5 54.1 Private Country 77205 2.5 42.0\* 106581 3.3 58.0 Other 9140 0.3 67.4\* 4420 0.1 32.6 TOTAL 3137605 100 49.2\* 3238867 100 50.8 **Couplet Category** Concordant Public 2104333 67.1 56.7\* 1605368 49.6 43.3 Private 887622 28.3 43.0\* 1176629 36.3 57.0 Discordant Public -- Private 74132 2.4 31.5\* 161140 5.0 68.5 Private -- Public 71518 2.3 19.4\* 296730 9.1 80.6 TOTAL 3137605 100 49.2\* 3238867 100 50.8 \* Statistically significantly different (at the 0.05% level) to the proportion of couplets with an intra-couplet interval 1 year or greater. ^1^12182 missing ^2^7815 missing ::: The distribution of first couplet episodes in each year of observation was more uniform in couplets with less than a one year intra-couplet interval compared with couplets longer intra-couplet interval or all couplets in the data file as shown in figure [2(A)](#F2){ref-type="fig"}. The reduction of first couplet episodes in couplets with intervals = one year was directly proportional to the number of years remaining in which a second episode (thus completing a couplet) could be observed. The lack of first episodes observed in 1980 was the result of a reduced volume of data in the original HMDS file for that year, most likely caused by an extraction error. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Distribution of the proportion of first and second episodes in couplets by year. ::: ![](1743-8462-2-12-2) ::: Figure [2(B)](#F2){ref-type="fig"} shows the distribution of second episodes in couplets over the observation period. The proportion of second episodes increased over the observation period regardless of the duration of the intra-couplet interval. This was a function of the increased number of individuals eligible to complete a couplet with a second episode as time progressed. Mean intra-couplet interval --------------------------- The ratio of the mean intra-couplet interval observed for each couplet category relative to the grand mean by age group and couplet year is shown in figure [3](#F3){ref-type="fig"}. Discordant couplets had the longest intra-couplet intervals, having on average a ratio relative to the annual grand mean intra couplet interval of 1.35, while concordant couplets types had the shortest intra-couplet intervals, their ratio being 0.65. The overall pattern indicated that the longer the time between the first and second episode of a couplet, the more likelihood there was of a change in payment classification, especially where the first payment classification was private. The trends also indicated that, within each age group, individuals with public concordant couplets, on average, had shorter intervals between episodes (ratio 0.5) than individuals with private concordant couplets (ratio 0.8). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Ratio of the mean intra-couplet interval to the grand mean for each couplet category by age group and couplet year. ::: ![](1743-8462-2-12-3) ::: The removal / re-introduction of free public hospital care (1980 -- 1984) has previously been shown to have had a significant effect on use of in-patient insurance classifications \[[@B16]\]. In this study no significant difference was observed with regard to when the first episode of a couplet occurred (pre or post Medicare); however, the point in time of the second episode did correspond with behavioural change as can be seen in figure [3](#F3){ref-type="fig"}. Differences in trend were observed across the four age groups. The 0--16 years age group trended towards a reduction in the intra-couplet interval associated with discordant private to public couplets after 1985. This was not observed during the late 1980s and early 1990s in the other three age groups. The 40--69 years age group showed a trend pattern similar to the 17--39 years age group until 2000, when the average intra-couplet interval associated with discordant private to public couplets reduced sharply. In addition there was a sharp increase in the average intra-couplet interval associated with discordant public to private couplets (also observed in the 0 to 16 years age group) observed at that time. In the oldest age group (70+ years) the average difference in intra-couplet interval between public and private concordant couplets was much smaller than observed in any of the other three age groups. Over all age groups the trend in concordant couplets was that of a slowly reducing average intra-couplet interval, excluding the 1983 -- 1984 period, where the average intra-couplet interval for private concordant couplets increased in all age groups. Behavioural patterns in switching of payment classification ----------------------------------------------------------- Differences were observed in the rate of loss from the public and privately insured payment classifications across age groups as shown in table [3](#T3){ref-type="table"}. Across all age groups the largest losses, at all intra-couplet intervals, occurred from the privately insured payment classification. The largest differences in the rates of loss were observed in the 70+ years and the 17--39 years age groups, where an additional 2.39 and 1.78 percent of privately insured episodes, respectively, were lost for every year of intra-couplet interval. The rates of loss from the privately insured payment classification over shorter intra-couplet intervals (defined as 0 to 13 years) were greater than the rates of loss over longer intra-couplet intervals (defined as greater than 14 years) in all age groups with the largest difference being observed in the 40--69 years age group and the smallest in the 70+ years age group. The definition of short and long intra-couplet interval was based on an observed substantial change of slope (inflection) in the segmented trend lines. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Loss from each payment classification as a function of age and intra-couplet interval ::: **Rate of Loss (percent per intra-couplet year)** **Difference in Rate (percent per intra-couplet year)** --------------------- --------------------------------------------------- --------------------------------------------------------- ------ ------ ------ ------ **0--16 Years**^\$^ 0.63 1.77 2.37 0.40 1.14 1.97 **17--39 Years** 0.49 2.27 2.39 0.55 1.78 1.84 **40--69 Years** 0.81 1.94 2.56 0.46 1.13 2.10 **70+ Years** 0.25 2.64 2.67 1.91 2.39 0.76 \* Averaged over all intra-couplet intervals in years. ^\$^Maximum intra-couplet interval 16 years as age determined at second episode. ::: Figure [4](#F4){ref-type="fig"} shows the degree of switching from private to public episodes over the two designated eras in health care policy. In all age groups the average rate of switching away from the private sector in 1991--2001 (era 2) was lower than that observed in 1981--1990 (era 1). The decrease in rate was small (average across all age groups -0.35% per intra-couplet year) but statistically significant. In addition, significance testing of the difference between each pair of proportions (era 1 versus era 2) indicated a significant difference for the majority as indicated in figure [4](#F4){ref-type="fig"}. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### The proportionate discordance among hospital couplets with a private first episode by decade of the second couplet episode. \# Significant difference (p \< 0.01) between the percentage of discordant second episodes occurring in 1981--1990 versus 1991--2000. ::: ![](1743-8462-2-12-4) ::: We also found that as the intra-couplet interval increased, the difference between the proportions of discordant couplets having a private first episode versus a public first episode increased from 4.2% at one year to 34.3% at 18 years (data not shown). This indicated that overall, regardless of age, there was greater switching away from PHI than away from the public classification. Discussion ========== As expected, due to their greater capacity to move between respective payment classifications, private patients were found to be more likely to switch payment classifications in their next admission than public patients, irrespective of the length of time between the two episodes. There are a number of possible explanations for this including putative structural and cognitive reasons for the observed behaviour. Structurally, the average patient who begins with a public classification is likely to be of lesser socioeconomic means than the average patient who begins with a private insured classification. The option of the former patient to use PHI at the next hospitalisation will be dependant upon them taking out, or at the very least maintaining (if they had private cover at the initial episode but did not use it) private cover in the meantime. The same pre-requisite does not exist for an initially privately insured patient accessing a public classification on the next occasion. Also patients with private insurance experiencing trauma or an acute disease event may, in some circumstances, be admitted in an emergency as a public patient, thus to some extent disallowing the patient from exerting their preference. Cognitive explanations include the possibility that PHI may not be as entrenched culturally as Medicare (the public system) in this population and, as a consequence, use of PHI may be more dependant upon marketed value propositions than use of Medicare. In other words, PHI may be perceived by the populace as a market good, whereas the public system (Medicare) may be perceived as a fundamental right. We also found that the degree of switching from PHI towards the public system was inversly proportional to the length of time between episodes. Possibly, healthier individuals were more likely to switch to the public system than sicker individuals, assuming that a relatively short duration of intra-couplet interval can be taken as an indicator of increased morbidity. This phenomenon is consistent with reports that the decline in the proportion of the eligible population holding PHI since the introduction of Medicare in 1984 has been largely attributed to younger and healthier individuals dropping out \[[@B11],[@B12]\]. This finding indicated the presence of a substantial cross-subsidisation between low risk and high risk individuals further exacerbating the adverse selection price \'death spiral\' of PHI \[[@B11]\]. Improving the risk profile of PHI holders has been a major focus of recent federal government policy, with the introduction of a lifetime community rating in 2000 in an effort to encourage younger, healthier individuals to take out and remain in PHI funds. Our finding that in all age groups the overall rate of switching away from PHI was slightly higher in the period 1981 -- 1990 compared with the rate in the period 1991 -- 2001, suggests that these policies have had an effect on behaviour consistent with the government\'s intention. An alternative, but less likely explanation for our findings, may be systematic differences in the cognitive decision making process between long intra-couplet intervals compared with between shorter intra-couplet intervals. For example, individual historical preferences may play a role in short interval switching but may not be important over longer intervals, where decisions may be made in isolation. While it is important to consider this alternative, we feel that if the changes observed in switches away from PHI were largely due to differences in cognitive decision making, one would expect to see a similar phenomenon in switching away from a public classification. Such a phenomena was not observed in our data. In all age groups, our analysis indicated that the overall rate of switching from the private payment classification was slightly higher in the period 1981 -- 1990 compared with the rate in the period 1991 -- 2001, suggesting that the policies had an effect on behaviour consistent with the government\'s intention. Assumptions and limitations of the approach ------------------------------------------- This study made use of the WA Data Linkage Project which is unique in Australia and is one of only a small number of population-based record linkage systems in the world. The use of administrative data has strengths and weaknesses. For example data can be inaccurate due to recording or coding errors \[[@B17]\] or linkage errors. For this study individual patient records were linked by probabilistic matching, using an automated computer algorithm based on the probability of two records being from different people having the same identifier and two records from the same person having different identifiers. The probabilities were then aggregated into a score and checked against a threshold to determine if a match was made. This technique typically has been found to have a true positive predictive value of 95--99% and a negative predictive value of 98--99% \[[@B18]\]. Extensive validation of the quality of the performance of matching has been undertaken on the WA Record Linkage Project using sampling techniques and the proportions of mismatches and missed matches found were in the order of approximately 0.11% \[[@B18]\]. Linked data have the advantage of supporting a large and diverse research programme at relatively low cost, once the infrastructure is in place. They have the capacity to provide a population-based view of events experienced longitudinally by individuals across all institutions \[[@B17]\]. Given the objectives of this study, the latter point makes the use of linked data particularly appropriate. The approach we have taken in order to analyse the use of PHI and Medicare by the population of Western Australia is unique in at least two respects. Firstly the study was conducted at the population level, due to the use of hospital morbidity data. Therefore, the \"reference population\" was not merely an abstract concept as in conventional quantitative research, but an operationalised descriptor of the study sample. Secondly, our \'couplet methodology\' has not been reported previously in the literature and represents a new method of analysing longitudinal data on use of health insurance. The couplet methodology has enabled patterns to be measured based the behaviour of individuals rather than average shifts in private -- public mix generated from unlinked episodes of care. However, we recognise that since the analysis was based on episodes of care, patient initiated switching as part of a hospital transfer could not be analysed. This limitation may have affected our conclusions about intra-couplet intervals. This paper is dedicated to explaining the couplet technique for the first time and applying it to address an initial set of relatively descriptive questions (ie teasing out what is happening). The next stage of investigation is a more analytic analysis which recognises that a wide range of potential explanatory variables might predict the couplet-based phenomenon of switching (the why). For example hospitalisation rates in set periods before and after a couplet, stratified by different lengths of stay and/or different DRG weights. In addition, the admission types (emergency/elective) of the members of the couplet may also be important predictors. We feel that the couplet methodology described in this paper will enable significant inroads to be made into the investigation of such explanatory variables. Conclusion ========== Our study found that the population of Western Australia exhibited distinct behavioural patterns in the switching of payment classifications for inpatient hospitalisation between 1980 and 2001. Private patients were more likely to switch payment classification than public patients with shorter intervals between episodes corresponding to a greater probability of private-to-public switching. However, the average rate of switching from a privately insured classification was greater between 1981 and 1990 than between 1991 and 2001, indicating that recent health care policy reforms implemented by the federal government to promote uptake of PHI have had an impact on behaviour. Competing interests =================== Professor D\'Arcy Holman is an independent director of HBF Health Funds inc which is the largest provider of private health insurance in Western Australia. Authors\' contributions ======================= The manuscript has been read and approved by all authors and the requirements for authorship have been met as outlined below. REM was responsible for the conception and design of the study; analysis and interpretation of the data; and drafting and revising the paper. CDJH was responsible for conception and design of the study; interpretation of the data; and revising the paper. Acknowledgements ================ The initial construction of the Data Linkage System was funded by the Western Australian Lotteries Commission. We would like to thank the WA Department of Health for on-going support of the Data Linkage Unit.

PubMed Central

2024-06-05T03:55:58.294160

2005-6-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173079/", "journal": "Aust New Zealand Health Policy. 2005 Jun 27; 2:12", "authors": [ { "first": "Rachael E", "last": "Moorin" }, { "first": "C D'Arcy J", "last": "Holman" } ] }

PMC1173080

Background ========== In their recent review of protein sequence analysis *in silico*, Michalovich et al. \[[@B1]\] described the methodology for transferring functional annotation of known proteins to a novel protein. Computer-assisted technology is used to search for and assign the similarity from databases of well-maintained and previously annotated sources. Sequence-based and profile-based searches are conducted using BLAST and PSI-BLAST, respectively. Meanwhile, the Hidden Markov model is more efficient in searching for a distant family. Furthermore, structure-based annotation conducted by using a combination of PSI-BLAST and GenThreader (matching of substitution energy in evolution) may facilitate rapid functional annotation from structure \[[@B1]\]. However, proteins sharing the same fold can have different functions, and structure determination and analysis will not always mean that function can be derived \[[@B2]\]. There are examples of protein families, such as the four-helical cytokine and cytochrome super families, whose sequence similarities are either very low or not detectable \[[@B3]\]. Instead, their topography is more conserved than their sequences. This is rational, since protein functions are classified based on function per se, regardless of whether their sequences or 3D structures are similar or different. An example is the classification of a protein as possessing the function of lysozyme activity, as long as the protein possesses the ability of hydrolyzing peptidoglycans. Another direct approach for peptide QSAR has been simultaneously investigated in peptide sequence analysis \[[@B4]\]. A critical difference between those two approaches, namely bioinformatics and QSAR, is the prerequisite of 3D structure information on the basis of evolutionary conservation in the case of former; on the other hand, the 3D information is helpful but not always indispensable in the case of latter, by substituting with simpler steric parameters to account for the functional mechanism \[[@B5]\]. For example, Hellberg et al. \[[@B4]\] used altogether 29 properties of side chains of bioactive peptides. After dimension reduction using principal components analysis (PCA) the resultant three main PC scores, i.e., z~1,~z~2~and z~3,~representing hydrophobicity, molecular size and electronic parameter, respectively, were used as independent variables in regression analysis on the dependent variable of functionality \[[@B4]\]. Meanwhile, by using the homology similarity analysis (HSA), we have found the importance of functional segments within 15-residue sequences of lactoferricin derivatives to correlate with the minimum inhibitory concentration (MIC) \[[@B6]\]. Pattern similarity constant (a correlation coefficient) of the pattern of segments within a test derivative, in comparison to the reference pattern of the corresponding segment and the average of property values of the amino acid side-chains in the most potent derivative, was computed and correlated with MIC of the derivatives. In order to obtain the best (lowest) MIC, the pattern similarity should be close to 1.0 and the average property value should be close to that of the reference potent peptide (template). In the case of the above lactoferricin derivatives, higher correlation coefficients were obtained for log MIC predicted by HSA vs. measured log MIC computed as the output variables of regression ANN (artificial neural networks) than by sequence analysis based on the Hellberg approach \[[@B4]\]. More recently, a different approach, namely \"additive QSAR\" obtained by substituting with other amino acid residues at different positions in the same sequences, was reported to correlate well with peptide functions \[[@B7]\]. Lejon et al. \[[@B8]\] reported that in PCA analysis of peptide sequences, information of the positions of side chains in the sequence should be included for improving *R*^2^*X*value compared to the results obtained by computing without side chain position data (0.99 vs. 0.60, respectively). Our HSA approach, by segregating segments with and without α-helix propensity, was in good agreement with theirs (*R*^2^*X*of 0.90--0.94 compared to corresponding value of 0.60 but with a much larger number of derivatives). We have further extended this approach to infer the mechanism of emulsifying capacity of peptides with 10--32 residues as a function of hydrophobic periodicity \[[@B9]\]. For the study of emulsification function, a new homology similarity search (HSS) was introduced to plot similarity constants and average property values of segments (3--7 residues) by shifting the segment stepwise from N-terminus towards C-terminus of the sequences; the reference segment used was ELE, i.e., alternate cycle of charged (E), hydrophobic (L) and charged (E) residues. However, emulsification ability is a rather general function of peptides that is not dependent on specific active sites within the sequences; overall, the emulsification ability of peptides was highly correlated with hydrophobic periodicity of their entire sequences. There are cases of \"peptides\" which do not have definitive functional sites but requiring specific segments, or \"functions\" which do require neither specific sites nor segments. The lactoferricin derivatives described in the above study \[[@B4]\] are an example of the former since all of the mutants were prepared as derivatives of the corresponding wild-type lactoferricin 15-residue sequence, which has distinct helical and cationic segments. In contrast, the peptide emulsions \[[@B9]\] are an example of the latter. Typical examples of proteins with definitive functional sites are enzymes, for which the positions of active sites are critical to elucidate the functional mechanisms. Defective protein folding leading to amyloid fibril formation has been associated with various human diseases, such as Alzheimer\'s and Creutzfelds-Jacob diseases. In 1993, hereditary non-neuropathic systemic amyloidosis was reported to be caused by naturally occurring variants of human lysozyme that aggregated in the liver \[[@B10]\]. Similarly, cystatin C mutation in an elderly man was reported to be the cause of amyloid angiopathy and intracerebral hemorrhage \[[@B11]\]. The recent discovery of a new-fold enzyme named Cellosyl \[[@B12]\] led us to select the lysozyme family in this study as an important one to use for validating the HSS approach to search for functional sites \[[@B13]\]. Meanwhile, loss of papain inhibitory activity in recombinant human cystatin C was reported to be due to insolubilization \[[@B14]\]. Assuming that this loss was induced by an amyloidosis, changes of helix-to-strand in the inhibitory sites as well as the binding sites with papain could also be used for a rational example of application of the HSS approach in this study. The objective of this paper was to extend application of this new HSS approach to search for functional sites, such as active and substrate binding sites in lysozyme and amyloidosis of cystatin families, to verify the reliability of our new method. The intention was to validate the hypothesis that the evaluation of pattern similarity of short segments with 3--7 residues or even slightly longer in protein sequences is useful in predicting functionality, assuming that they are within allowable topographical units. Accordingly, it is not our intention to replace the 3D approach by the new peptide QSAR proposed in this study; rather, it is anticipated to be supplemental. Results ======= Lysozyme families ----------------- ### PCS classification Figure [1](#F1){ref-type="fig"} shows a scattergram derived from principal components similarity (PCS) analysis of 25 lysozyme sequences using the hydrophobicity index of side chains and hen lysozyme as a reference. A scattergram similar to this figure was also obtained when a charge index was used (instead of the hydrophobicity index) for classification. CH-type lysozymes used herein were from a fungus and *Streptomyces globisporus*. The v-type lysozymes were T4 and PA2 phage lysozymes, while g-type lysozymes were from goose, black swan, cassowary, ostrich and chicken G. Sixteen lysozymes (chicken C, human, horse, dog, rat, mouse, red deer, rainbow trout, pigeon, turkey, duck, California quail, Japanese quail, common bobwhite, fruit fly Drosophila, and tobacco hornworm) belong to the c-type family. The large deviations in slope (about 2) of CH-type lysozymes as seen in Figure [1](#F1){ref-type="fig"} are suggestive of explicit difference in their molecular structures from those of other lysozyme types. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **PCS scattergram of lysozyme families (C-, G-, V-, and CH-types) when hydrophobicity was used as property index.**Hen lysozyme was used as the reference with \[coefficient of determination\] = 1.0 and \[slope\] = 1.0. ::: ![](1471-2091-6-9-1) ::: ### Segment pattern similarity search for active sites The HSS computer program was applied to the sequence alignment patterns shown in Figure [2](#F2){ref-type="fig"} using the active sites of hen lysozyme at positions 54--57(F^34^ESN) and 80--83(T^51^DYG) as the references. The position numbers in parenthesis are the positions in un-gapped sequences of individual lysozymes, while the immediately preceding numbers outside of parenthesis are the position numbers of the multiple sequence alignment (gapped) in Figure [2](#F2){ref-type="fig"}. Since the peptidoglycan-lysing activity is associated with glutamic 55(E^35^) and aspartic 81(D^52^) side-chains in the hen lysozyme \[[@B13]\], the segments flanking these residues in the sequences were the focus of pattern similarity search using the HSS. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Multiple sequence alignment of five lysozymes belonging to four families. ::: ![](1471-2091-6-9-2) ::: The search patterns illustrated in Figure [3A](#F3){ref-type="fig"} (charge) and 3B (turn propensity) are a trial run to validate the HSS approach, applied to human lysozyme vs. hen lysozyme, employing the five residues flanking E^35^(K^33^FESN of hen) as a search unit. The rules herein for selecting active segments are that the greater the similarity approaching to 1.0 and the nearer the average value to that of the reference, the more likely to be active sites in the test sequence. In Figure [3A](#F3){ref-type="fig"}, in addition to E^35^and D^53^, three residues of E^7^, S^80^and D^120^show about the same similarity constants (upper pattern) as well as similar average charge values (shown by arrows on the lower pattern). However, none of the above latter three positions are acceptable as the active site, as N^44^, I^59^, C^65^, Q^86^, and Q^126^in Figure [3B](#F3){ref-type="fig"} for β-turn search do not have matching peaks in Figure [2A](#F2){ref-type="fig"} for charge search. Therefore, only E^35^and D^53^are qualified as the active positions of human lysozyme. The same result was obtained when five residues flanking D^52^(S^50^TDYG) of hen lysozyme were used as an alternative reference segment. The two regions around E^35^and D^53^determined for human lysozyme are in good agreement with those being reported for the active site in the literature \[[@B13]\]. Despite the fact that the charge is of prime importance in defining active sites of lysozymes, the turn propensity values of the segment rather than its pattern similarity appear to play an important role in the enzyme, as exemplified in a low similarity value of 0.30 for D^53^compared to 0.98 for E^35^, whereas similar average turn values of 1.2 and 1.1, respectively, were computed (Fig. [3B](#F3){ref-type="fig"}). All these results may imply that the exposure of active sites to react with the substrate binding sites is essential. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **HSS search patterns for active sites of human lysozyme.**Segment 53--57 of hen lysozyme was used as the reference. A: HSS search pattern based on charge. B: HSS search pattern based on turn propensity. ::: ![](1471-2091-6-9-3) ::: The same search was conducted for goose and T4 lysozymes as well as the new fold CH-lysozyme \[[@B12]\], with the assumption that it is an un-annotated sequence, to confirm validation of the HSS approach (Table [1](#T1){ref-type="table"}). Two positions of 79(E^73^) and 103(D^97^) were determined as the active sites of the goose lysozyme sequence. D^97^may not be essential for the catalytic activity of the goose lysozyme \[[@B15]\]. In comparison, 11(E^11^) and 20(D^20^) were noted for T4 lysozyme, which are in good agreement with those reported in the literature \[[@B13]\]. As shown in Figure [4](#F4){ref-type="fig"}, in the case of the sequence of Cellosyl (CH-type), 14(D^9^), 104(D^98^) and 106(E^100^) were identified as candidates to be the active positions of catalysis, which are in good agreement with Rau et al. \[[@B12]\]. These results would support the reliability of the HSS approach. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Determination of active sites in sequences of lysozymes in different families ::: Family Active sites Position Charge 1 Charge 2 Turn 1 Turn 2 -------- -------------- ----------- ---------- ---------- ---------- ----------- Hen 1 55(35)E 1.0/5.0 .98/5.0 1.0/1.12 2 81(52)D .98/5.2 1.0/5.2 1.0/1.21 Hman 1 55(35)E 1.0/5.9 .98/5.0 .98/1.17 2 81(53)D .98/5.2 1.0/5.2 1.01/1.22 Goose 1 79(73)E .67/6.9 .78/6.9 .06/1.03 2 103(97)D .83/7.3 .83/7.3 .67/,92 T4 1 11(11)E .61/4.5 .55/4.5 .76/1.16 2 20(20)D .55/5.4 .55/5.4 .53/1.07 CH 1 14(9)D .98/5.2 1.0/5.2 .95/.95 2 104(98)D .74/4.5 .65/4.5 .81/.92 3 106(100)E .95/5.6 .86/5.6 .87/.97 Position: the first number is the multiple sequence alignment position shown in Figure 2. Number in bracket is that in the sequence of each lysozyme. Charge 1 and 2 and Turn 1 and 2: Sites 54--57 and 80--83 (alignment position numbers in Fig. 2) were used as references. These position numbers correspond to 34--37 and 50--54, respectively in the sequence of hen lysozyme. The first and second digits n each cell of Charge and Turn are similarity constant and average of property value. ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **HSS search patterns for active sites in *Strep. coelicolor*lysozyme.**Segment 79--83 (STDYG) of hen lysozyme was used as the reference based on charge. ::: ![](1471-2091-6-9-4) ::: ### HSS search for substrate binding sites Compared to the active sites that are explicitly negative in charge at single positions, the substrate binding sites of lysozymes are rather loosely defined, mainly due to the greater complexity of determining the 3D structure of enzyme-substrate complexes than that of enzyme alone. It is generally agreed that the substrates locate inside the cleft formed between the helix lobe and the strand lobe of most lysozyme molecules, except for Cellosyl, the bacterial muramidase from *Streptomyces coelicolor*, which can be attributed to structural difference in the catalytic crevice \[[@B12]\]. The six-residue segment at alignment positions 84--89 (I^55^LQINS) of hen lysozyme was employed as the reference segment \[[@B16]\] using the hydrogen bonding scale (Table [2](#T2){ref-type="table"}). Based on the high pattern similarity constants and the average hydrogen bonding index values, eight potential sites were identified for hen and human lysozymes. The same rule as that for selecting active site was used herein for selecting of binding sites. Position 77(D^48^) and 188(W^111^) shown in \"Hen 1\" of Table [2](#T2){ref-type="table"} with lower similarity constant and hydrogen bonding value, respectively, may not be the potent substrate binding sites in the hen lysozyme. In the 3D structure \[[@B17]\], those two positions are far away from the catalytic cleft where the substrates snugly fit in. In the human lysozyme, 188(W^112^) is more likely to be the binding site than 102(A^73^), with higher pattern similarity and strength of hydrogen bonding of 0.89/0.45 than 0.80/0.32, respectively. Not only strong hydrogen bonding, but also high pattern similarity of a segment may be required to be qualified for substrate binding sites. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Determination of substrate binding sites in amino acid sequences of lysozyme families ::: **Potential sites** --------- --------------------- ------------- --------------- --------------- ------------- --------------- --------------- --------------- --------------- Hen 1 **55(35)E** **70(42)A** 77(48)D **84(55)I** **90(61)R** **102(72)S** **117(83)L** 118(111)W **.81/.71** **.79/.53** .34/.65 **1.0/.41** **.56/.58** **.78/.56** **.61/.42** .66/.35 Hen 2 **36S** **45R** **49G** **56L** **62W** **73R** **90A** **100S** 112R **.82/.71** **.86/.44** **.79/.47** **1.0/.39** **.59/.58** **.91/.47** **.98/.41** **.55/.46** .76/.29 Human 1 **54(34)W** **70(42)A** 77(49)D **84(56)I** **90(62)R** 102(73)A **116(83)A** **118(112)W** **.48/.52** **.42/.51** .34/.65 **.91/.47** **.50/.61** .80/.32 **.61/.46** **.89/.45** Human 2 27N **43T** **50R** **57F** **63Y** 74V 84L 101R **117Q** .56/.39 **.56/.51** **.61/.47** **1.0/.45** **.55/.61** .72/.33 .48/.47 .58/.29 **.79/.45** Goose 43(73)I 99(93)L **119(113)I** **143(133)S** 181(143)G **201(163)G** .85/.31 .76/.34 **.67/.46** **.81/.55** .82/.36 **.81/.58** T4 **6(6)M** **16(16)K** **53(49)A** **66(57)V** **75(66)L** 94(85)K **123(112)A** **186(136)S** 207(153)F **.74/.34** **.99/.38** .78/.30 **1.0/.42** **.52/.54** .67/.38 **.73/.45** **.58/.51** .78/.36 CH **40(35)T** **47(42)D** **55(50)T** 80(74)A **92(86)W** **124(118)T** 153(147)C 186(180)T **1.0/.58** **.63/.52** **.80/.51** .51/.44 **.73/.49** **.83/.59** .51/.56 .56/.54 The first/second digits are similarity constants and average hydrogen bond index values of binding site with segments of sox residues beginning with the positions shown. Bold digits show more likely binding sites than non-bold digits. Hen 1 and Human 1: from the alignment shown in Figure 2; Hen 2 and Human 2; from the alignment of the lysozyme C family exclusively. ::: For the goose lysozyme, six potential binding positions were detected (Table [2](#T2){ref-type="table"}); I^113^and G^163^appear more likely to be the binding sites than other positions considering the location of the cleft in the molecule. Similarly, nine sites were found to be potential sites of T4 lysozyme, especially three positions, i.e. M^6^, L^66^and S^136^(Table [2](#T2){ref-type="table"}). In the case of Cellosyl, eight positions were identified as the potential binding sites (Table [2](#T2){ref-type="table"}). Probably due to the considerable 3D-structure difference of this lysozyme from those of other lysozyme families \[[@B12]\], the alignment positions 40--45 (T^34^EGTNY) instead of hen\'s 84--89 (I^55^LQINS) were used as a reference segment for obtaining more rational search results. Instead of hydrogen-bonding motivated interactions, less polar van der Waals interaction with the aromatic side chains in CH-lysozyme may be regarded as the second important stereochemical forces in the substrate binding \[[@B16]\]. In the literature, the most frequently cited substrate-binding sites in c-type lysozymes family have been W^62^and D^101^of hen lysozyme \[[@B13]\]. Since Figure [2](#F2){ref-type="fig"} includes the distant family of CH-type, the segment similarity search was repeated within the c-type lysozymes alone to restrict the search within similar fold. The results are shown as \"hen 2\" and \"human 2\" in Table [2](#T2){ref-type="table"}. Those results almost perfectly match to the substrate binding mechanism based on X-ray crystallographic analysis, e.g. D^101^, N^103^, N^104^, A^107^, V^109^, E^35^, N^46^, V^110^, E^52^, N^59^, and W^63^\[[@B15]\]. Almost all of these side-chains are very close or adjacent to the segments listed in \"Hen 1\" and \"Hen 2\" of Table [2](#T2){ref-type="table"}. ### Substrate binding sites reported by site-directed mutagenesis Among three mutants obtained by replacing W^62^with Y, F or H, the W62H mutant, and especially the double mutant W62H/D101G, reduced substrate binding drastically \[[@B15]\]. This change can be explained by a decrease in the hydrogen bond average value from 0.58 to 0.54 and from 0.46 to 0.29 in V62H and D101G, respectively, when the index values employed in this study were used in computation. The double mutant changed substrate-binding mode while maintaining the overall protein structure almost identical to that of the wild type \[[@B18]\]. An extensive cluster of hydrophobic structure is involved in distinct regions of the sequence, but is all disrupted by a single point mutation of W62G located at the interface of the two major structural domains in the native lysozyme \[[@B19]\]. Similar effects were observed in mutants Y63L and D102E of human lysozyme \[[@B20]\]. The double mutants R41N/R101S and V74R/Q126R of human lysozyme were better catalysts for lysis of *Micrococcus lysodeikticus*\[[@B18]\]. The average hydrogen bond value of both R41N and R102S was shown to increase in our HSS search, but similar effects could not be observed for V74R/Q126R. An interesting finding is that these two mutations have both resulted in the side chains being identical to those of hen lysozyme. R^41^and V^74^are near A^42^and A^73^, respectively. Importance of R^115^in substrate binding of human lysozyme was reported \[[@B21]\], which is in good agreement of W^112^within the same subsite F (Table [2](#T2){ref-type="table"}). Cystatins --------- ### PCS classification The PCS scattergram of the 17 cystatins using human cystatin C (HCC) as a reference and hydrophobicity as side-chain property index is shown in Figure [5](#F5){ref-type="fig"}; alteration of the index to α-helix and β-strand propensities did not appreciably change the grouping results. The three groups include human cystatins C, D, S, SA, SN and hen cystatin (Group I), human cystatins E, F and M (Group II), and human cystatins A and B (Group III) which are the stefin group cystatins that are smaller in molecular size and have slightly lower papain inhibitory activity than HCC \[[@B22]\]. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **PCS scattergram of cystatins when hydrophobicity was used as property index.**Human cystatin C (HuC) was used as the reference. Human cystatins M, E, S, SA, SN, D, F, A and B are labelled as 1, 2, 7, 8, 9, 10, 15, 16 and 17. Labels 3, 4, 6, and 11--14 are for cystatins from mouse C, rat C, bovine, hen, rainbow trout, chum salmon and carp, respectively. ::: ![](1471-2091-6-9-5) ::: ### Active sites The HSS patterns using hydrophobicity to search for active sites of the four cystatins (HCC, EWC, HCA and HCB), when L^9^VGG of HCC has been employed as the reference, are shown in Figure [6](#F6){ref-type="fig"}. The active sites shown with arrows are at about the same location in all cystatins sequences used in this study. Similar results were obtained when bulkiness was used as side-chain index. In these cases, the similarity peaks are the major clue for identifying active sites. However, when similarity peaks appear in the neighbourhood, the average index values would become more reliable for identifying active sites as shown in Figure [6B](#F6){ref-type="fig"}; hen cystatin has two probable active positions side by side with the same similarity constants. The active sites, therefore, should be near the N-terminus with Prosite-type patterns of \[L,I,M\]-x(4)-G- \[G,A\]; the active sites are L^9^VGG and L^7^LGA in human cystatin C and hen cystatin, respectively. The similarity constants and the average hydrophobicity of the active sites are shown in Table [3](#T3){ref-type="table"}. Stefin B (HCB) shows much lower similarity constant and average hydrophobicity than those of other cystatins. This result is in good agreement with *k*~*i*~difference reported by Abrahamson \[[@B23]\]. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Active and substrate binding sites of cystatins ::: **HCC** **EWC** **HCA** **HCB** ------------------------ ------------ ------------ ----------- ------------ **Active site** Segment L^9^VGG L^7^LGA I^2^PGG M^2^SGA SimConst/Av.Hydroph\* 1.00/1.35 0.87/1.80 0.97/1.00 0.55/-0.01 **Binding site 1** Segment Q^55^IVAG Q^53^LVSG Q^46^VVAG Q^46^VVAG SimConst/Av.Hydroph 1.00/0.90 0.70/0.92 0.90/0.95 0.90/0.95 **Binding site 2** Segment V^104^PWQG I^102^PWLN L^73^PGQN L^73^PHEN SimConst/Av.Hydroph 1.00/1.01 0.35/1.95 0.30/0.98 -0.16/0.42 SimConst/Av.Turn\*\* 1.00/1.05 0.90/0.96 0.56/1.22 0.99/1.06 \* Pattern similarity constant / average hydrophobicity \*\* Pattern similarity constant / average turn propensity Similar to the corresponding tables for lysozymes (Tables 1 and 2), Sim/Const is 1.0 for reference HCC, and the greater the Av.Property the greater the property strength. ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **HSS search patterns for active sites of cystatins against human cystatin C based on hydrophobicity.**A: HCC (reference), B: EWC (egg white cystatin), C: HCA and D: HCB. ::: ![](1471-2091-6-9-6) ::: ### Substrate binding sites A HSA study similar to our previous paper \[[@B6]\] was conducted at the active and two binding sites of cystatins, yielding results (Table [3](#T3){ref-type="table"}) which are in good agreement with Turk et al. \[[@B22]\]. Substrate-binding site 1 had the pattern Q-x(3)-V- \[S,A\]-G, while substrate-binding site 2 had the pattern \[L,I,V\]-P-x(3)-x(3)- \[N,G\]. Similarity constants of binding loop 2 of egg white cystatin (EWC) and HCA are lower than that of HCC, whereas not only similarity but also average hydrophobicity are lower in HCB. Similarity constants at the active site (against 1.0 for HCC) using hydrophobicity index were \>0.8 for cystatins A, D, F and hen, \~0.5 for E and M, and 0.1--0.2 for cystatins B, S, SA and AN. Similarity constants at binding loop 2, when strand propensity was used for PCS computation, were lower for stefins A and B with values of 0.8 and 0.6, respectively, than \>0.9 for other cystatins. On the other hand, similarity constants for strand at binding site 1 were not much different among different cystatins, with values \>0.9 (not included in Table [3](#T3){ref-type="table"}). It is interesting to note that stefins A and B do not have the PW pair which is in the binding site 2 of HCC and EWC; instead they have PG and PH pairs, respectively (Table [3](#T3){ref-type="table"}). The W → G replacement increased strand propensity, while W → H replacement did so moderately. The values shown in Table [3](#T3){ref-type="table"} were almost inversely proportional to the equilibrium inhibition constant *k*~*i*~except for human cystatin S that was weak in the inhibitory activity, which might have been due to the difference in phosphorylation of serine at N-terminal region \[[@B23]\]. Although stefins A and B are classified differently from other groups on PCS scattergram (Fig. [5](#F5){ref-type="fig"}), the weak binding at the binding site 2 may not have considerable effects on the *k*~*i*~values. Turk et al. \[[@B22]\] have stated that the differences in the binding constants between cystatins and various cysteine proteases arise primarily from differences in the structure of enzyme active site clefts. The inhibition of endopeptidases, i.e. papain and cathepsins S and L, by cystatins is extremely tight and rapid, whereas the inhibition of exopeptidases, i.e. cathepsins B and H, is considerably weaker. The active site cleft of known endopeptidases is free to accommodate inhibitors, while in the case of exopeptidases, the active site cleft contains extra residues in it. In the N-terminal region of cystatins, it was observed that the affinity for target proteases decreased with both size and charge of substituting residues \[[@B22]\]. These observations are in good agreement with the results when bulkiness of side chains was used for the HSS computation for the binding site 1, \"SimConst /Av.bulk\" values were 1.00/12.3, 0.92/14.2, 0.98/11.4 and 0.91/11.34 for HCC, EWC, HCA and HCB, respectively. As expected, stefins A and B were less bulky. Furthermore, HCC and EWC include longer chains at the N-terminal sides with bulkier residues than those of stefins. These findings are in good agreement with the effect of bulkiness of G^4^in the stefin A sequence, implying that the bulkier the residue at position 4, the weaker the papain inhibitory activity \[[@B24]\]. Amyloidosis ----------- ### Lysozymes Similarity constants and average propensities of α-helix and β-strand were computed for G^54^IL and G^54^ILQIN of hen lysozyme (G^55^IF and G^55^IFQIN in the case of human lysozyme) as shown in Table [4](#T4){ref-type="table"}. The amyloidogenic mutant I55T showed increased strand propensity from 0.82 to 0.85, without a substantial change in the similarity constants. With regard to the helix structure, the similarity constant decreased from 1.00 to 0.93 while the average value increased from 1.08 to 1.14, thereby implicating a decrease in helix, since the helix index used herein is inversely related to the content of helix structure. These changes are favourable for amyloidosis. The fact that position 56 in human lysozyme is near its active position at D^53^may explain its dramatic effect on the enzymatic activity, more effective than other positions in the sequence. The six-residue computation did not show as clear a difference as was found in the three-residue computation. Nine double-site mutations in addition to I55T at other positions selected at random in the sequences of the amyloidogenic mutant I55T by using the RCG program did not restore the activity of wild-type lysozyme (unpublished). These results infer that the amyloid, once formed by detrimental mutation at the active site, cannot be restored by mutation at other locations in the sequence. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### HSA computation for I55T lysozyme ::: **Hen** **Human** **I55T**(Hen) ----------------- --------- ----------- --------------- **Helix** G^54^IL(F)QINS SimConst\* 1.000 0.917 0.952 Average 0.960 0.993 0.985 G^54^IL(F) SimConst 1.000 1.000 0.927 Average 1.082 1.082 1.141 **Strand** G^54^IL(F)QINS SimConst 1.000 0.978 0.997 Average 0.788 0.762 0.797 G^54^IL(F) SimConst 1.000 1.000 0.998 Average 0.822 0.822 0.845 L^56^for hen and F^57^for human. \* Similarity constant. ::: ### Cystatins Heat treatment of HCC induced its dimer formation at an early stage of separation, resulting in a complete loss of its activity \[[@B25]\]. Based on a dramatic decrease in the monomer form as shown by its CD spectrum, polymerization such as amyloidosis could be a cause of the loss of papain inhibitory activity of mutated HCC \[[@B14]\]. Of 35 residues (positions 1--35) of the helix domain of HCC, 17 residues were mutated in the 22 single-site mutants using the RCG program \[14 (Table [1](#T1){ref-type="table"})\]. When 33 mutants obtained by adding one extra residue each in both side of the original 17 residues after eliminating duplication were used for PCS computation, the resultant PCS demonstrated that helix propensity and bulkiness were playing important roles in thermostability (data not shown). Employment of three residues flanking the mutated residue was important in pattern similarity computation. This conclusion is in good agreement with Hall et al. \[[@B26]\] who did an exhaustive study showing that mutations at positions 8--10 enhanced thermostability of cystatin. With regard to the papain inhibitory activity, the importance of hydrophobicity and bulkiness was demonstrated (the PCS scattergrams, similar to Fig. [7](#F7){ref-type="fig"}, are not shown here). ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **Effects of mutating the strand domain of human cystatin C (mutated 21 residues).**A: Strand index, B: Helix index. The numbers show multiples of activity increases from wild-type. 1 shows no increase of inhibitory activity. 5\* is the reference mutant 12W86V with the highest inhibitory activity. ::: ![](1471-2091-6-9-7) ::: Of 86 residues in the strand domain (positions 36--121) of HCC, 21 residues (positions 36--120) were mutated in the 23 double-site mutants using the RCG program \[[@B14]\]. Thirty-seven residues were used for the PCS computation of single-site mutations as described above. Hydrophobicity appeared to be playing an important role in thermostability, while strand propensity was important for inhibitory activity (data not shown). Strand and helix propensities in the strand domain were influential to the papain inhibitory activity of HCC (Fig. [7](#F7){ref-type="fig"}). The figures show 12 data points only by eliminating data from single site mutation in the helix domain, which did not show distinct trends with broader scatter in these figures. The second mutations in addition to the above single mutations were conducted at the strand domain of the enzyme \[[@B14]\]. Coefficient of determination of 1.0 and slope of 1 indicate perfect match with the reference sample (5\*) that is mutant G12W/H86V with the lowest strand propensity along with highest helix propensity in the strand domain among 23 double mutants. It is worth noting that the PCS is a classification program comparing pattern similarity without demonstrating quantitative relationships with functions but providing the information of the extent of involvement of side chain properties in the functions of interest. For mutant G12W/H86V that gained the greatest activity increase of 4.98 ± 0.09 times (mean ± SD at n = 3) that of recombinant wild-type \[[@B14]\], the strand propensity decreased from 0.78 to 0.69 (H86V) with a slight increase in the helix propensity (corresponding to decrease in the index values). The same was true for mutant D15P/H86I with 2.65 ± 0.30 times activity increase. A similar result was observed in mutant G4L/D40I with 2.11 ± 0.29 times activity increase, due to strand decrease along with almost no change in helix (D40I). However, mutant V10S/R93G with 4.50 ± 0.07 times activity increase behaved differently with increased strand and simultaneous decrease in helix. It is worth noting that the single mutation of V10S alone increased the activity 2.96 ± 0.06 times, therefore, changes in the helix domain may have a more predominant effect on the inhibitory activity than mutations in the strand domain. The activity change due to mutation helix → strand in the strand domain in the sequence may be slight in this case. It is well known that a single-point mutation of human lysozyme, namely I56T, has been identified as the origin of hereditary systemic amyloidosis \[[@B27]\]. The amyloidogenic nature of the lysozyme variants arises from a decrease in the stability of the native fold relative to partially folded intermediates. Accordingly, in a low population of soluble, partially folded species, the protein can aggregate in a slow and controlled manner to form amyloid fibrils. Similarly, sporadic amyloid angiopathy and intracerebral hemorrhage was reported in an elderly man due to cystatin C mutation \[[@B11]\]. In the case of human cystatin C, the decrease in strand along with an increase in helix might have prevented amyloidosis, despite the fact that helix change was not always as evident as in the case of lysozyme. Some inconsistency in the amyloidosis as a cause of inhibitory activity of human cystatin C in our mutation optimization \[[@B14]\] may be due to lack of the data of single-site mutation in the strand domain of the cystatin sequence. Unfortunately, the objective of that study \[[@B14]\] was for mutation optimization and not for investigation of the mechanism of amyloidosis. It has been reported that stefin B (HCB) readily formed amyloid \[[@B28]\], which may imply declined importance of the role being played by the binding site 2 in amyloidosis of HCC. Discussion ========== In a review on the quest to deduce protein function from sequences \[[@B29]\], the author stated that the searching of pattern databases would be more sensitive and selective than searching of sequence database. It was predicted that the sequence pattern databases, especially by comparing the pattern similarity, would play an increasingly important role, as the post-genome quest to assign functional information to raw sequence data gains pace \[[@B29]\]. Pattern similarity computation requires at least three residues in segments to represent a nonlinear curve, which is unlikely to be due to the effect of a single point mutation per se. With regard to an apparent effect of the single residue mutations of hen lysozyme on substrate binding, the structural analysis by NMR of the position-62 mutant of hen lysozyme \[[@B18],[@B30]\] found major changes in the chemical shift of back bone protons, especially in a loop region (positions 61--78), which contains W62 influencing the local folding. Similarly, Muraki et al. \[[@B20]\] reported that compared to the wild-type human lysozyme, the N-acetylglucosamine residue at subsite B of the L63 mutant markedly moved away from the 63^rd^residue, with substantial loss of hydrogen-bonding interaction. In Figure [5](#F5){ref-type="fig"} of Ref 17, involvement of not only Y63 but also W64 is evident. These results are supportive of the importance of pattern similarity of ≥ 3 residues, which are affected by single-residue mutation. The predictability of the active and binding sites solely on the basis of protein sequences \[[@B31]\] may be useful for investigating the underlying mechanisms of unknown functions of human genes after translation to protein sequences. Usually, two or three essential residues are directly involved in the bond making and breaking steps leading to formation of enzyme catalysis; however, the removal of an essential group often does not abolish activity, but can significantly alter the catalytic mechanism \[[@B32]\]. T4 lysozyme was cited by Peracchi \[[@B32]\] as an example of the alteration of catalytic mechanisms; the lytic activity of lysozyme changed to that of a transglucosidase. An approach utilizing the property of side chains in a sequence for identifying functional motifs has already been utilized in the computer-assisted selection of antigenic peptide sequences \[[@B33]\]. The authors stated that an antibody produced in response to a simple linear peptide with 7--9 residues in a protein would most likely recognize a linear epitope. Furthermore, this epitope must be solvent-exposed to be accessible to the antibody. In a large scale data mining study, Binkowski et al. \[[@B34]\] described the importance of local sequence and spatial surface patterns in inferring functional relationships of proteins. The general feature of protein structure that would correspond to these criteria could be turns or loop structures, which are generally found on the molecular surface connecting to other elements of secondary structure, and the area of high hydrophobicity, especially for those containing charged residues. Successful identification of active sites of new-fold of CH-lysozyme using the HSS approach in this study suggests that this approach could be applied to query proteins translated from unknown RNA segments of the human genes against templates with known functions, when their 3D structure information is still unavailable. It has been shown that the inhibition of the papain family by cystatin is due to a tripartite wedge-shaped structure with a good supplement to the active site clefts of the enzyme \[[@B35]\]. Todd \[[@B36]\] stated that despite highly homologous relationship as seen in Figure [8](#F8){ref-type="fig"}, lysozyme functions as an O-glycosyl hydrolase, while α-lactalbumin lacks this activity and instead regulates the substrate specificity of galactosyltransferase. The active site of peptidoglycan lysis is disrupted in α-lactalbumin. The ESS computation showed that although a pattern equivalent to hen\'s D^53^exists in the form of T**E**YG/Y**D**YG, there is no residue equivalent to hen\'s E^35^around corresponding positions as in the form of HTSG/W**E**SG. Two side-chain carboxyl radicals are required for the lysozyme activity within the crevice between the helix and strand domains of the molecule belonging to C-type family \[[@B16]\]. According to Alvarez-Fernandez et al. \[[@B35]\], the three parts of the cystatin polypeptide chain included in the enzyme-binding domain are the N-terminal segment, a central loop-forming segment with motif QXVXG and second C-terminal loop typically containing a PW pair \[[@B31],[@B37],[@B38]\]. ::: {#F8 .fig} Figure 8 ::: {.caption} ###### **An example of enzyme vs. non-enzyme**. Adopted from Figure 10.3b \[36\]. Fig. 8A: 1IWT human α-lactalbumin, Fig. 8B: 1B9O human lysozyme C using Swiss-PdbViewer (spdbv). Blast 2 sequences \[1\] showed that the identities and the positives between the two proteins were 35% and 55% respectively. ::: ![](1471-2091-6-9-8) ::: For multiple sequence alignment of an uncharacterized protein or peptide, many Web alignment servers are available for use \[[@B1]\], such as Blast and NPSA, as was done in this study. For classification of uncharacterized sequences, the PCS scatterplots are also useful as shown in Figures [1](#F1){ref-type="fig"}, [5](#F5){ref-type="fig"} and [7](#F7){ref-type="fig"}. The PCA demonstrated the classifying capacity superior to that of distance-based cluster analysis \[[@B39]\]. The PCS is more flexible than cluster analysis as different pattern similarity patterns can be drawn by rotating the reference segment for searching. It implies that similarity is not always \[1 -- dissimilarity\]. This difference resulted in the possibility of selecting outliers, which is critical in deriving true classes or ranking \[[@B40]\]. Most of the currently available peptide QSAR, such as the method of Hellberg et al. \[[@B4]\], intends to be based on whole sequence data. The new HSS approach reported in this study could be just the beginning of more detailed, reliable peptide QSAR to be developed in the future. Analysis of a variety of bioactive proteins contributing to human health is a potential future application of the HSS software package as well as multifunctional PCS. Considering the multifunctional nature of human diseases, the functionality of food proteins also can be manipulated based on combinations of bioactive segments in different or even single natural protein sequences. Therefore, for an uncharacterized protein or peptide, a new plan is proposed: (1) A reference sequence is chosen from multiple sequence alignment (MSA) as discussed above; PCS scattergrams would assist this selection in addition to BLAST search. (2) Based on segments with high similarity in MSA, segments to be used for search are selected within the reference sequence. Then, (3) HSS search is conducted to identify functional segments in the uncharacterized sequence. (4) From the above PCS computation, important PC scores are screened (PCA is a subroutine subprogram of PCS). (5) Regression neural networks are conducted using selected PC scores as input variables as exemplified in our lactoferricin derivative study \[[@B6]\]. (6) RCG would be useful for confirming the HSS data and also to find the best segment or sequence as exemplified in our HCC mutation \[[@B14]\]. One of the original purposes of our new approach in unsupervised data mining was to verify the hypothesis that there might be adaptability of different human cystatins to better inhibit different human cathepsins \[[@B41]\]. This hypothesis has not been fully pursued in the past, probably because of costly separation of pure cystatins and cathepsins. An advantage of our approach is to derive potential hypothesis for enzyme/substrate interactions exclusively from their sequence data. Although the verification of those hypotheses may need to await future 3D-structure study, it is important that most of the useful QSAR data could become available, which would promote the functional mechanism study based on 3D structure. However, we admit that more examples of application should be performed in the future to more thoroughly verify and establish this method for predicting functions based on sequences. This work is underway in our laboratory. Conclusion ========== Although the importance of pattern similarities of motifs with 20--30 residues as a whole has been reported for peptide QSAR in the past, the importance of a search for segments with three or more residues as functional sites of protein sequences has not been investigated. Lysozymes and cystatins were used as examples of proteins to demonstrate the capacity of segment pattern similarity analysis to predict functions, such as active and binding sites, amyloidosis and thermostability as a tool for quantitative functional sequence analysis. Methods ======= Amino acid sequences of proteins -------------------------------- Multiple sequence alignments of lysozymes were conducted using the Network Protein Sequence Analysis of Pôle Bio-Informatique Lyonnais \[[@B42]\] based on Clustal W. Similarly, multiple sequence alignments were obtained for human cystatins A (HCA), B (HCB) and C (HCC) and hen egg white cystatin (EWC) as well as for papain as host proteases of cysteine protease inhibitors, i.e. cystatins. For PCS analysis, a total of 17 cystatins were used: human A, B, C, D, E, F, M, S, SA, SN, hen (EWC), bovine, ratC, mouseC, Chum salmon, Rainbow trout and carp. Principal components similarity analysis of protein sequences ------------------------------------------------------------- The method described in the previous papers \[[@B9],[@B39]\] was followed. Principal components analysis (PCA) was modified to principal components similarity (PCS) by incorporating linear regression of PC scores to be able to account for more than three PC scores on a 2D scatter plot. The PCS was then modified to apply to peptide sequences. Homology similarity search -------------------------- Homology similarity search (HSS) was conducted as reported previously \[[@B9]\]. The similarity constant used in this study is eventually a correlation coefficient \[[@B43]\]. A preliminary study was carried out by changing the size of segment (normally 3--7) flanking the potential functional position to determine the most appropriate size of segment in differentiating the functional site from other segments within the sequence of lysozymes and cystains. The property indices used for amino acid side chains were hydrophobicity, charge, propensities of α-helix, β-strand and β-turn, hydrogen bonding, and bulkiness as reported previously \[[@B14],[@B44]\]. Segments with pattern similarity close to 1.0 and average values similar to that of the reference segment were sought within each gapped sequence. All software used in this study along with the instructions on how to use the computer programs are available in the form of ftp files on the Web \[[@B45]\] to download to PC computers. List of abbreviations ===================== EWC Egg white cystatin or hen cystatin. HCC Human cystatin C. HCA Human cystatin A or stefin A. HCB Human cystatin B or stefin B. HSA Homology similarity analysis: the PCS software was modified to compute pattern similarity constants and average side-chain property index values of segments in sequences \[[@B6]\]. HSS Homology similarity search: A step-wise search program initiated from N-terminus of query sequences by shifting the search unit (reference segment) towards C-terminus based on similar segments in terms of pattern similarity constant and average property values compared to those of template sequences \[[@B9]\]. MIC Minimum inhibitory concentration. PCA Principal components analysis. PCS Principal components similarity: PCA modified for multi-functional variables using linear regression of deviation of PC scores on the reference PC scores. Scatter plot is drawn as slope vs. coefficient of determination (r^2^) \[[@B44]\]. RCG Random-centroid optimization of site directed mutagenesis. QSAR Quantitative structure-activity relationships. Authors\' contributions ======================= EL participated in laboratory investigation to verify the hypothesis set in this study, while JD was taking care of the computer programming. SN was responsible mainly for the creation and application of software used in this study. All authors read and approved the final manuscript. Acknowledgements ================ This work was financially supported by a Multidisciplinary Network Grant entitled \"Structure-function of food biopolymers\" (Dr. Rickey Y. Yada of University of Guelph as the principal investigator) from the Natural Sciences and Engineering Research Council of Canada. The authors acknowledge the collaboration of all co-authors listed in our past publications as shown in the following references. Especially, the drawing of 3D structures to compare lysozyme and α-lactalbumin by Dr. Yasumi Horimoto is highly appreciated.

PubMed Central

2024-06-05T03:55:58.298066

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173080/", "journal": "BMC Biochem. 2005 May 18; 6:9", "authors": [ { "first": "Shuryo", "last": "Nakai" }, { "first": "Eunice CY", "last": "Li-Chan" }, { "first": "Jinglie", "last": "Dou" } ] }

PMC1173081

Background ========== Discovery of regulatory elements in human gene promoters is one of current bioinformatics challenges. Although transcriptional control mechanisms have been investigated in various organisms for at least three decades, it is still almost impossible to predict tissue-specific or developmental-stage-specific expression of a given gene by simply analyzing its promoter sequence \[[@B1]\]. The 5\' segment immediately adjacent to the TSS includes the core promoter and the proximal promoter, which usually extends about 200--300 nucleotides. This region is involved in the modulation of transcription. The distal part of a promoter is variable with respect to composition and length, which may encompass from 100 nucleotides to over 2 kb. There is no clear-cut defined 5\'-boundary for promoters \[[@B2]\]. Regulatory elements binding the same transcription factor can be found in different promoters as short DNA sequences, differing among them to some extent; they are, in general, from 5 to 25 nucleotides long \[[@B3],[@B4]\], often separated by un-conserved sequences. Control regions are modular in nature and expression of a given gene depends on specific combination of its regulatory elements and sometimes from their order and orientation \[[@B5]\]. Searching by computational methods for promoters and for regulatory elements in DNA sequences spanning several Kb, produces a large number of false-positive results. A possible solution to this problem is to identify a \"sheltered environment\" in which specificity of pattern discovery might be enhanced. Unknown binding sites for transcription factors might be detected by searching for common elements in upstream regulatory regions of genes with common biological function and/or expression. In fact, genes with similar expression are frequently co-regulated and genes with related function are often similarly expressed \[[@B6]\]. In this study, we attempted to detect putative regulatory elements in promoters of genes expressed in an adult human tissue (retina), by a multi-step approach involving computational analysis of large-scale expression data, selection of a subset of putatively co-expressed genes, retrieval of the upstream portion of their complete genomic sequence and application of pattern discovery on promoter regions. Results ======= Analysis of known regulatory sequence elements binding transcription factors ---------------------------------------------------------------------------- Before applying COOP software on a selected group of promoters, we attempted to exploit information on known regulatory sequences available in TRANSFAC \[[@B10]\], to establish some \"rules\" which could facilitate the discovery of novel regulatory elements. In particular, TRANSFAC matrix data were analysed in order to describe number, percent and localization of fixed and variable positions in consensus sequences. We first considered 385 matrices including information on mammalian regulatory elements. Average length of consensus sequences was 13.0 and mode 12; motifs of even length were more represented (even lengths two times more represented than odd lengths among consensus sequences of length ranging from 8 to 17). Less than 5% of the motifs showed only invariant positions (average and mode of length of completely invariant motifs were 10.3 and 9, respectively). About 33% of motifs showed more than 75% fixed positions (average length 11.5, mode 10), whereas about 73% showed more than 50% fixed positions (average length 12.3, mode 8). In general, the shortest the motif, the less variable appeared its consensus sequence. By separately considering three regions of consensus sequences (left, center and right), we observed that lateral positions are variable in 37% of sequences, whereas central positions are variable only in 20% of them. Most regulatory elements included in TRANSFAC seem to be symmetrical, being equally variable in their left and right sides. We obtained very similar conclusions from the analysis of the group of 610 eukaryotic matrices. Results of this analysis suggested that pattern discovery on mammalian promoter sequences might focus on patterns 10, 12 or 14 nucleotides long, showing from 0% to 25% variable positions, and possibly, less variable in the central region. COOP : Clustering Overlapping Occurrences of approximate Patterns ----------------------------------------------------------------- Since sequence signals with biological significance are frequently subtle, stringency of pattern discovery analyses in biological sequences cannot be set too high. This implies that results are often too numerous. A novel tool for Clustering Overlapping Occurrences of approximate Patterns (COOP) was implemented in Python (Figure [1](#F1){ref-type="fig"}). This software allows identification of tractable numbers of possibly interesting motifs, starting from large numbers of exact or approximate patterns. Selection of genes and retrieval of putative promoter regions ------------------------------------------------------------- Among 1,814 genes expressed in retina, statistical analysis of differential expression, by Audic and Claverie test \[[@B11]\], picked out 80 genes significantly more expressed in retina than in all other tissues. We selected as well a group of 59 known genes whose mutation is known to cause retinal diseases, recorded in OMIM and/or in RetNet databases, and/or encoding proteins for which a specific function in retina has been described. In total, 129 were selected. For each of these genes, the Reference Sequence or the longest sequence of the mRNA with complete CDS was compared by BLAT \[[@B12]\] to human genome sequence, for annotation of the intron/exon structure and for prediction of the most probable TSS (Transcription Start Site). We predicted with good confidence TSS of 90 genes (45 overexpressed UniGene clusters, 28 retinal disease genes, 7 both overexpressed and retinal disease-genes and 10 genes encoding proteins with a retinal function)\[[@B13]\]. Sequences from 90 selected genes, each corresponding to 1 Kb upstream the predicted TSS, were retrieved for further analyses. For one gene, USH3A (LL 7401) two alternative promoters (USH3A-A and USH3A-B) controlling transcription of messenger RNA encoding retinal products were found. Therefore, 91 gene promoters pertaining to 90 different retinal genes were considered for the study. Search for approximate patterns ------------------------------- ### Retinal datasets We considered datasets including sequences corresponding to the 500 bp upstream the predicted TSS of the selected genes (unmasked sequences, 500U91; partially masked, 500PM91; masked, 500M91) and the group of 91 fully masked sequences corresponding to 1 Kb upstream the predicted TSS (1000M91 dataset). Fully masking of sequences in datasets 500M91 and 1000M91 produced, on average, 435 and 788 unmasked nucleotides, respectively. In each group of sequences, approximate patterns of length ranging from 10 to 14 nucleotides, with at most two variable positions (10-2, 12-2, 14-2 patterns), were searched by SPEXS (Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}). For each dataset, patterns were ranked in different classes, according to the number of sequences in which they were represented (Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}, Figure [2](#F2){ref-type="fig"}). In the 500 bp datasets the number of sequences in which most represented patterns were found as relatively low, reaching a maximum of 44 (48%) for 10-2 patterns in unmasked sequences (500U91) and only 5 (5.5%) for 14-2 patterns in masked sequences (500M91). When considering sequences of length 1,000, the number of sequences with occurrences of most represented patterns slightly increased. For instance, most represented 10-2 patterns were found in 32 sequences in 500M91 and in 40 sequences in 1000M91. ### Negative control datasets Random groups of human promoter sequences were established as negative control. One thousand groups of 91 promoter sequences randomly selected among 17,156 (1 Kb long) human gene promoters were generated (RAN1000M91i, with *i*ranging from 1 to 1,000). Each of these 1,000 groups included the same number of sequences of the 1000M91 set of retinal gene promoters and sequences fully masked and of the same length of retinal gene promoters. Moreover, TSS predictions were done by Promoser \[[@B14],[@B15]\] according to identical criteria adopted for TSS prediction of retina genes. Results of pattern discovery in the 91 retina gene promoters group (1000M91) and in the dataset of 1,000 groups of 91 human gene promoters randomly selected among 17,156 (RAN1000M91i) are shown in Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"} and in Figure [2](#F2){ref-type="fig"}. The number of patterns with different *quorum*in the retinal datasets and in the negative control groups of promoter sequences are reported. In the last two rows of Table [4](#T4){ref-type="table"} the comparison of pattern discovery results in the 1000M52 retinal gene promoters group and in the dataset of 1,000 groups of 52 sequences (RAN1000M52i) randomly selected among 17,156 human gene promoters is shown. The number of patterns found in at least 10 and in at least 20 out of 91 retinal promoter sequences is higher than the average of number of patterns found in at least 10 and in at least 20 out of 91 randomly selected promoter sequences, calculated over 1,000 different samples. Statistical significance (P-value) of number of patterns found in each number of sequences of the retina datasets was calculated as the proportion of negative control random datasets in which the number of patterns found was higher or equal to those found in retinal promoters. When this number was below 0.05, the difference was considered statistically significant. Over 1,000 random samples, 351.3 10-2 patterns were found in average in at least 20 promoter sequences, while 719 patterns were found in at least 20 retinal promoters (P-value = 0.017). Similarly, 714 12-2 patterns were found in 7 or more retinal promoter sequences, whereas 410.2 were found in average in random samples (P-value = 0.060). Statistical significance of this observed difference resulted just above 0.05. This effect could be due to possible heterogeneity of the class of 91 gene promoters. When considering the group of 52 promoters corresponding to genes significantly more expressed in retina than in all other considered tissues (1000M52) and the corresponding negative control dataset (RAN1000M52i), the difference between retinal and random datasets is more evident. For instance, 114 12-2 patterns were found in 7 or more retinal promoter sequences, whereas 33.7 in average were found in 1,000 random samples (P-value = 0.027). In Figure [2A](#F2){ref-type="fig"}, the cumulative number of patterns is plotted against the number of sequences in which they were found, among sequences of dataset 1000M52 and RAN1000M52i. In Figure [2B](#F2){ref-type="fig"} the comparison between the 1000M91 and the RAN1000M91i is shown. For negative control (random groups of promoters), the average value of 1,000 sets of sequences is given, with an interval of two standard deviations centred to the mean (e.g. 4,895 patterns were found in at least five sequences out of 91 in the retina dataset, whereas the average number of patterns found in at least five sequences out of 91 was 3,057 in the negative control dataset). It should be noticed that in both comparisons the number of patterns found in the retinal dataset is always considerably higher than average in the negative control dataset. Putative novel regulatory elements in retina gene promoters identified by COOP ------------------------------------------------------------------------------ We considered the group of 716, 12-2 patterns detected by SPEXS in at least 7 out of 91 promoter sequences of genes expressed in human retina (dataset 1000M91). We clustered the 6,611 occurrences of 716 selected patterns by using different sets of parameters, in order to identify combinations maximizing the biological meaning of resulting motifs. Distance parameters ranging from 2 to 5 nucleotides were used, each associated with *o*values of 0.6 or 0.7. The number of clusters decreases when *d*increases and increases with *o*. For instance, when *o*is set to 0.6, the increase of *d*from 2 to 5 changes the total number of clusters from 211 to 183, changing the number of clusters containing only the occurrences of one SPEXS pattern (\"single clusters\") from 136 to 116, with maximum number of patterns per cluster increasing from 119 to 180. We kept *o*high, in order to cluster patterns according to their \"very frequent\" overlapping occurrences. The change of *o*from 0.6 to 0.7 changes the number of clusters obtained with d = 3 from 195 to 279, while the number of \"single clusters\" changes from 123 to 183. The number of clusters containing the occurrences of different SPEXS patterns is quite stable in respect to variation of *o*(72 and 97, respectively), whereas the number of \"single clusters\" increases considerably with *o*. In all cases, the number of extracted motifs, ranging in these experiments from 169 (*d*= 5, *o*= 0.6) to 300 (*d*= 2, *o*= 0.7), is considerably lower than the number of patterns considered. Starting from 6,611 occurrences of 716 selected patterns found in at least 7 sequences with *d*set to 3 and *o*to 0.7 COOP produced 279 clusters with an average number of occurrences per cluster of 13.9 (median 9 and mode 7, maximum 150). Clusters are associated to 279 sequence alignments, 89 of which (31.9%) are longer than 12 nucleotides, and to position-specific scoring matrices. In average, 13.9 occurrences per motif were observed. These 279 motifs occurred in 87 sequences out of 91. The mean number of motifs per sequence is 26.7, whereas the mean number of motif occurrences per sequence is 42.6. Subgroups of these motifs could be similar or overlapping. Motifs occurrences were ranked into four classes according to their position in promoter sequences (bp distance from the predicted TSS). The observed distribution significantly deviated from expectation under assumption of randomness (1,558 motif occurrences from -1 to -250; 1,002 in -251/-500; 626 in -501/-750 and 690 in -751/-1000; chi squared test: P = 3.03\*10--121). There is a positive correlation between density of motif occurrences and proximity to TSS. In fact, over 40% of total motifs occurrences are concentrated in 250 nucleotides close to the predicted TSS and the two thirds of the total number of occurrences fall within 500 nucleotides adjacent to the first exon. Regions of promoters sequences corresponding to the 500 bp proximal to TSS were in average less masked than the upstream regions. When normalizing the number of motifs occurrences to the percentage of unmasked nucleotides in the different regions, a strong difference remains, with significant deviation from a random distribution. P value of chi squared test for the comparison among two groups (from -1 to -500 and from -501 to -1000) is 3.3\*10--48. From each sequence alignment pertaining to a cluster, a consensus sequence representing the motif was built. Different sets of parameters were used to this purpose. The choice of different stringency of parameters for building consensus sequences influences the length of obtained motifs and the fraction of variable positions included in them. Threshold *i*ranged from 0.1 to 0.5 and *f~l~* and *f~c~* ranged from 0.6 and 0.8 (data not shown). If *f~l~* and *f~c~*are sufficiently stringent, a low value for *i* could be used, in order to maximize information extracted. When a *i*= 0.1 is applied for construction of consensus sequences, the average length of motifs remains almost unchanged (12.4), with 89 motifs of length over 12. Part of the information about each motif is lost when a matrix is built from the alignment of pattern occurrences and when frequency data are converted into a consensus sequence. We used moderately stringent thresholds for the minimum frequency of a single nucleotide determining \"fixed\" positions. In particular, *f~l~* was set to 0.5, whereas *f~c~*, referring to the core regions of consensus sequences, was set to 0.7. Out of 279 motifs, only 62 \"most informative\" motifs were selected, which showed a completely conserved consensus or a consensus showing variable positions only in the side regions (Table [5](#T5){ref-type="table"}). The average length of this group of sequence motifs was 13.0, with about 55% of them longer than 12 nucleotides. Two pairs of motifs were represented by the same consensus sequence. The resulting group of 60 motif consensus sequences, representing putative functional elements in retinal promoters, were compared with TRANSFAC data, by TESS program \[[@B16]\]. In particular, we used the \"Filtered String-based Search Query\" tool \[[@B17]\] for comparison only with known regulatory sites of mammals, with no mismatches allowed and by using the entire length of known sites instead of their core positions. Out of 60 motifs, 53 (88%) exactly matched at least one sequence known to bind a mammalian transcription factor. Sequences corresponding to common or general transcription factors (Sp1, Sp3, MAZ, GCF, CUP or Yi) were matched by 47 out of 60 consensus sequences (78%), 22 of which matches also additional factors (AP-1, AP-2, WT1, Krox-20, GR, PPUR or ER). In total, 26 consensus sequences resulted similar to sequences recognized by AP-1, AP-2, WT1, Krox-20, GR, PPUR or ER. Analysis of positive control datasets ------------------------------------- To the purpose of both analysing COOP efficiency with selected conditions and of indirectly comparing the performances of the method with those of different software, we analysed by COOP a group of 26 human positive control datasets prepared by Tompa and colleagues for the assessment of computational tools for the discovery of transcription factor binding sites \[[@B18],[@B19]\]. This benchmark included 26 groups of promoters, for which it is known which regulatory signals should be detected. The number and the length of promoter sequences and of known signals per dataset are reported in Table 6, along with the adopted quorum \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. The number of signals per group of promoters which are shorter than the length of approximated patterns taken as input by COOP (12 nucleotides) and which were, in principle, very difficult to find by adopted settings, is also reported. For each experiment, among different motifs predicted we selected only the one corresponding to the motif represented in the highest number of sequences. For each dataset the nucleotide-level and site-level \[[@B18]\] overlap between pattern occurrences belonging to the cluster (i.e. the selected motif) and known signals, evaluated by different measures (see Methods), are included in Table 6 \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. Moreover, the \"combined\" statistics summarizing COOP performance over the collection of human datasets, was compared with the same statistics calculated for the 14 different programs tested by Tompa and colleagues on the same datasets \[[@B18]\]. Results are presented in Figure [3](#F3){ref-type="fig"} and in Table 7 \[see [Additional file 2](#S2){ref-type="supplementary-material"}\]. The comparison showed that the performances of COOP on the \"very difficult\" human dataset are in line with those of the top rated tools. In comparison with other software, COOP resulted to be in the best 20% of evaluated tools according to four measures (nSn, nPC, sSn, sASP) and in the best one third according to six measures (nSn, nPC, nCC, sSn, sPPV, sASP). Discussion ========== Pattern discovery in sequences of putative and sometimes incomplete promoters is a considerably complex problem \[[@B20]-[@B25]\]. It may be reasonably assumed that some regulatory regions of a group of co-regulated genes share similar sequence elements. In yeast, pairs of genes showing over 0.84 Pearson correlation between their expression profiles, have over 50% probability of sharing at least one common transcription factor binder \[[@B26]\]. Since patterns with biological significance could be subtle \[[@B27]\], a main difficulty in pattern discovery approaches is a priori establishing a \"quorum\" and defining search parameters (e.g. pattern length, number of allowed wildcards or distance of occurrences from the model). By increasing the distance and/or decreasing the quorum, the number of false positives becomes excessively large. A possible solution to the problem of output explosion is to use biological knowledge both before and after application of automated pattern discovery. We analysed TRANSFAC data to obtain information about frequent properties of regulatory elements. It should be noticed that although not all TRANSFAC matrices are based on high-quality data and on large samples of sequences, this database represents the largest existing collection of known regulatory elements in different organisms. In this study we developed a novel tool, called COOP, for analysing promoter sequences of putatively co-regulated genes, aiming at extraction of sequence motifs with possible regulatory function. The motif extraction method is based on Clustering of Overlapping Occurrences of approximate Patterns, which allows identification of tractable numbers of possibly interesting motifs, starting from large numbers of exact or approximate patterns. Our method is somehow related to two approaches proposed by van Helden \[[@B28]\] and by Caselle \[[@B29]\], although these studies considered yeast promoter sequences and dealt with exact patterns. The originality of our approach, mainly resides principally in adopting a new similarity measure between patterns, based on the frequency of pattern co-occurrences, and in designing a flexible procedure, with seven parameters which could be varied in order to modulate stringency of different analysis steps. Motifs reconstruction was designed to maximize information included in each extracted sequence motif avoiding generation of spurious elements, given that clustering parameters (*d*and *o*) are appropriately set. Each obtained motif is represented by a consensus sequence, derived from the alignment of strings grouped in a specific cluster by adjustable criteria. In particular, the *i*threshold affects the length of the consensus sequence, whereas the *l*, f~c~and f~l~determine the number and the distribution of variable positions in the consensus sequence. In order to evaluate the performance of the method, we analysed positive control datasets, such as all the human benchmark groups of promoter sequences, containing known signals at known positions, proposed by Tompa for a systematic assessment of motif discovery tools. COOP analyses were carried out with the same settings used for analysing retina gene promoter. The quorum was established by using an unique criterion for different datasets, based on the total number of sequences in the sample. In the first analysis phase, approximated patterns of 12 nucleotides, with at most 2 variable positions, were searched. Benchmark datasets were 26 groups of different number of promoters sequences. Each set of sequences contained a group of known signals. Lengths of signals ranged from 4 nucleotides to 71 nucleotides. Thus, several datasets contained a number of very short signals, which were very hard to find by a motif discovery approach designed to find motifs of length equal or higher than 12 nucleotides. We predicted a motif for all but two human datasets, which included only two sequences. COOP performed comparably well than the tools which were top rated in the Tompa assessment. It should be noticed that some of the other tools gave no predictions for a number of datasets, thus being advantaged from the averaging nPPV, sPPV and nCC scores when calculating combined statistics over all the human datasets \[[@B18]\]. The method we developed was applied, in a case study, to a collection of human promoter sequences pertaining to a group of 91 putatively co-regulated genes expressed in the retina. One Kb long promoter sequences were identified by predicting the most probable TSS according to the consensus of information about cDNA and ESTs alignments with the genome sequence. Even neglecting the possible presence of alternative promoters, definition of exact(s) TSS it is a still open problem, because of low sensitivity of promoter prediction programs and of incomplete cDNA coverage of 5\' exons. However, cDNA coverage of the majority of genes selected for this study is almost complete, since genes considered are well known and/or highly expressed. The adequacy of our method for selecting gene promoter sequences is supported by results obtained by Trinklein and colleagues \[[@B30]\]: over 90% 152 human 600 bp promoter sequences, randomly selected among 10,276 TSS predictions (based only on alignment with full-length cDNA clones from Mammalian Gene Collection) resulted active in at least two cell lines. Although the size of 1 Kb upstream the TSS might be insufficient to cover all possible regulatory regions for all genes, it could reasonably include both core and proximal promoters and at least part of the distal promoter. A very high number of approximate patterns of length from 10 to 14 with at most 20% variable positions was found in retina promoter sequences. This is partially due to the fact that they include similar sequences or sequences shifted only by few positions. By using COOP for clustering approximate patterns, on the basis of their frequent overlapping occurrences, we identified a number of interesting sequence motifs, often longer than the original patterns. In order to test the significance of the group of sequence patterns frequently found in retinal promoters and used as input for COOP, we generated and studied different negative control datasets corresponding to a thousand groups of sequences, randomly selected among a very large group of human promoters. The average pairwise level of pattern sharing in these groups of promoters was expected to reflect the general level of pattern sharing between human gene promoters. We observed that the selected group of retinal promoters (pertaining to a sample of genes putatively co-expressed, co-regulated and/or with similar function) is enriched in common patterns as compared to random groups. Sequence motifs produced by COOP resulted more frequently in the regions close to the TSS. Moreover, the group motifs consensus sequences selected according to very low variability in their central region was compared with sequences which are known to bind mammalian transcription factors under stringent criteria. Signals similar to those for general and widespread transcription factors, such as Sp1 or MAZ, are the most represented. Moreover, a number of selected motifs resulted to be similar to signals recognised by transcription factors expressed in tissues of ectodermal origin and relevant to development and function of retina (AP-1, AP-2, WT1, Krox-20, GR, PPUR or ER). For instance, AP-1 elements were found in a number of retinal gene promoters including cGMP-phosphodiesterase beta subunit \[[@B31]\] and hydroxyindole-O-methyltransferase \[[@B32]\] whereas the WT1 zinc finger factor is essential for normal development of retina and specifically involved in regulation of retinal genes \[[@B33]\]. Conclusion ========== We developed a method to detect sequence motifs corresponding to putative regulatory elements in gene promoters, starting from lists of approximate patterns with occurrences in promoter sequences. This method could be profitably applied to different datasets, including promoter sequences of different groups of genes in humans or in other Eukaryotes, for which co-regulation could be demonstrated or inferred. The method could be used to investigate on different kinds of regulatory sequences, such as intronic enhancers, or other sequence motifs with non-regulatory function. Availability and requirements ============================= COOP can be downloaded free-of-charge from the web page <http://telethon.bio.unipd.it/bioinfo/Retina/suppl_material.html>. COOP was developed in Python. The software works under Linux and requires Python 2.3 or higher, BioPhyton 1.40b and ClustalW. COOP is provided \'as is\' with no guarantee or warranty of any kind and it is freely available for all. Methods ======= COOP : Clustering Overlapping Occurrences of approximate Patterns ----------------------------------------------------------------- COOP takes as input a FASTA file of nucleotide sequences and a list of patterns with their number of occurrences in sequences or in their reverse complement. We used the SPEXS program \[[@B7]\] for producing the list of approximate patterns frequently represented in selected groups of promoter sequences. SPEXS code is freely available and it provides a number of advantages in terms of execution time and flexibility of parameters determining search conditions and output appearance. Seven COOP parameters can be varied in order to select stringency at different stages of the analysis (Table [1](#T1){ref-type="table"}). In the first step, patterns represented in more than *q*promoters are searched by COOP in promoter sequences. In particular, both direct and reverse complement sequences of each pattern are compared against promoter sequences in order to collect pattern occurrences. Then, pattern occurrences (strings) are clustered according to a similarity measure based on frequency of their co-occurrences and by a joining algorithm derived from the so called \"quick-find\" algorithm \[[@B8]\]. In order to be included in the same cluster, two different strings must occur in promoter sequences much more frequently together than separately. Given the physical distance between pattern occurrences, measured as nucleotide distance between the 5\'-ends of two corresponding sequences, the threshold *d*defines the maximum value for the distance between two pattern occurrences to be considered overlapping. Threshold *o*indicates the minimum ratio between observed overlapping occurrences of two strings and their average number of occurrences, allowing their inclusion in a unique cluster. The total number of clusters obtained in this way is influenced by the number of pattern occurrences to be clustered, depending on the *q*parameter, and depends as well on values selected for the minimum distance between patterns and for threshold *o*. Once clusters are obtained, all sequence elements corresponding to pattern occurrences belonging to each given cluster are multi-aligned by ClustalW \[[@B9]\], with Gap Opening Penalty set to 100. Each alignment is then analysed in order to build up a matrix describing nucleotide counts in alignment positions. The cell *x*~*A*1~of the matrix contains the number of times the A nucleotide has been observed in the first alignment position. Later, consensus sequences are built from matrices (Table [2](#T2){ref-type="table"}). The maximal number of adjacent matrix positions fulfilling the established threshold for *i*(minimum ratio between number of nucleotides per alignment position and total number of lines in the alignment) is further analysed to determine the *m*nucleotides long motif consensus sequence. Once the extension of lateral regions (*l*, ranging from 0 to m/2) is fixed, motif positions are considered invariable if the frequency of a single nucleotide exceed selected thresholds (*f~l~* for lateral regions and *f~c~* for the core region of the motif). Once the length of motif and of lateral regions is known, the extension of the core region is fixed. Moreover, an additional procedure is available, which uses IUB/IUPAC nucleic acid single letter, double-degenerate codes (M = \[A\|C\], K = \[G\|T\], S = \[G\|C\], W = \[A\|T\], R = \[G\|A\], Y = \[T\|C\]) and four-degenerate code (N = \[A\|G\|;C\|T\]) and follows IUPAC rules for string consensus reconstruction: i. A single nucleotide is shown if its frequency is greater than 50% and at least twice as high as the second most frequent nucleotide; ii. A double-degenerate code indicates that the corresponding two nucleotides occur in more than 75% of the underlying sequences (but the criteria for a single nucleotide assignment are not met); iii. All other frequency distributions are represented by the letter \"N\". The output of the program is a collection of sequence clusters, each one representing a sequence motifs. Each cluster is associated to an alignment, to a matrix describing nucleotide counts in alignment positions and, ultimately, to a consensus sequence. Moreover, information about promoter sequences and nucleotide positions in which each cluster string occurs is given. Selection of genes and of putative promoter regions --------------------------------------------------- ### Analysis of genomic expression data For the study, genes significantly more expressed in retina than in all other tissues were identified by analysis of genomic expression profiles of several human tissues. Genomic expression profiles were reconstructed *in silico*by using 41 unbiased (un-subtracted and/or un-normalized) UniGene cDNA libraries pertaining to 11 adult human normal tissues (retina, bone, hyppocampus, liver, lung, marrow, melanocyte, muscle, pancreas, prostate and testis) for which at least 6,000 ESTs per tissue were available \[[@B34]\]. The whole dataset included 270,871 ESTs, corresponding to 27,924 UniGene \"clusters\". The expression profiles were merged in an expression data matrix, which was then analysed by the Audic and Claverie test of differential expression in order to identify genes significantly more expressed in retina than in all other tissues considered. Significance threshold was set to a = 0.05. ### Disease genes and genes encoding proteins with a specific function in the retina By searching in OMIM \[[@B35]\], Retinal Information Network \[[@B36]\], LocusLink \[[@B37]\] and GeneCards \[[@B38],[@B39]\] we selected a group of known genes whose mutation causes retinal diseases and genes encoding proteins which play specific functions in retina. ### Retrieval of putative promoter regions Reference Sequences of selected genes (RefSeq) \[[@B37]\] were extracted from corresponding LocusLink entries. When RefSeq for a gene was unavailable, the longest mRNA sequence with complete CDS was used. Sequences were then searched by BLAT \[[@B40]\] against release 15 of human genome sequence, for prediction of Transcription Start Site (TSS), obtained by analysis of mRNA/genomic DNA alignment, 5\' ESTs placement and Acembly gene boundaries \[[@B41]\] annotation. Genomic sequences of 1 Kb upstream the predicted TSS were retrieved. These sequences were masked by RepeatMasker \[[@B42]\], in order to remove repetitive DNA. Negative controls ----------------- Negative control groups of promoter sequences were established as 1,000 sets of promoter sequences, sampled at random among 17,156 human gene promoters. Reference sequences of 27,427 human mRNA were obtained from GenBank. A group of 20,315 reference sequences was obtained, after exclusion of all sequences referring to unknown genes (chromosome open reading frames, hypothetical or predicted proteins) or to genes including in their sequence record words referring to vision, eye or retina. Retrieval of promoter sequences corresponding to 1 Kb upstream the predicted TSS of genes was done by PromoSer \[[@B14],[@B15]\]. TSS prediction options were set in order to retrieve for each gene 1 Kb upstream the most 5\' TSS, with the same criteria used for retrieval of retinal gene promoters. In addition, exclusion options of PromoSer were set to extract at most one promoter per gene and to avoid retrieval of overlapping sequences. By this way, 17,156 promoters were localized and retrieved. A Python script was developed for iterative random extraction of groups of m sequences from a list of N sequences, with N \> m, without repetition. Analysis of known regulatory sequence elements binding transcription factors ---------------------------------------------------------------------------- We used the TRANSFAC version available through BIOBASE. TRANSFAC \[[@B43]\] is a database of eukaryotic cis-acting regulatory DNA elements and trans-acting factors containing information on transcription factors, regulated genes, regulatory sites and nucleotide distribution matrices for binding sites of transcription factors. By using Perl scripts developed to the purpose, flat files pertaining to matrices data were parsed in order to extract information about consensus sequences length and about number, percent and localization of fixed and variable positions. Analysis of positive control datasets ------------------------------------- We analysed by COOP groups of promoters for which it is known which regulatory signals should be detected. Analyses were done of the complete set of groups of human sequences included in public benchmarks prepared by Tompa and colleagues \[[@B18]\], including three different types of sequences: 9 groups of real genomic promoter sequences containing known transcription factors binding sites, 9 groups of randomly chosen human genomic promoter sequences in which the binding sites were planted and 8 groups of sequences randomly generated according to a Markov chain of order 3 (that was constructed from human promoter sequences) in which the binding sites were planted. We analysed such datasets by using the same methodology applied for the identification of putative novel regulatory elements in retinal gene promoters, including searching for 12-2 patterns by SPEXS and motifs reconstruction by using COOP (clustering parameters: *d*= 3 and *o*= 0.7). The quorum *(q*) was set to the highest integer equal or less than one third of the total number of sequences in the dataset. When with the selected quorum no results were obtained, *q*was set to the highest integer giving results. When analysing datasets composed of from 3 to 6 sequences, *q*was set to *2*. We made no motif predictions for the two datasets of two sequences each. For each group of promoters, among different clusters obtained, we selected only the one corresponding to the motif represented in the highest number of sequences. For each dataset out of the 26 considered, we checked the overlap between pattern occurrences belonging to the cluster (i.e. the motif) and known signals. The efficiency of COOP was evaluated according to different measures, defined as follows. At nucleotide-level nTP (number of nucleotide positions in both known sites and predicted sites), nFN (number of nucleotide positions in known sites but not in predicted sites), nFP (number of nucleotide positions not in known sites but in predicted sites) and nTN (number of nucleotide positions in neither known sites nor predicted sites). A predicted site overlaps a known site if they overlap by at least one-quarter the length of the known site. Thus, at site-level we calculated: sTP (number of known sites overlapped by predicted sites), sFN (number of known sites not overlapped by predicted sites) and sFP (number of predicted sites not overlapped by known sites). We then calculated the following measures of accuracy. At either the nucleotide (x = n) or site (x = s) level: Sensitivity, xSn = xTP/(xTP + xFN); Positive Predictive Value, xPPV = xTP/(xTP + xFP). At the nucleotide-level: Specificity, nSP = nTN/(nTN + nFP); Performance Coefficient, nPC = nTP/(nTP + nFN+ nFP); Correlation Coefficient, ![](1471-2105-6-121-i1.gif); Average Site Performance, sASP = (sSn + sPPV)/2 \[[@B18]\]. In addition, the statistics (nSn, nPPV, nSp, nPC, nCC, sSn, sPPV, sASP) summarizing COOP performance (with selected settings) over the collection of human datasets, were computed with the \"combined\" method \[[@B18]\] and compared with the same statistics calculated for the 14 different programs tested by Tompa and colleagues on human datasets. Authors\' contributions ======================= SB and GAD conceived the study. SB and AC carried out pattern discovery analyses, developed and tested COOP software and drafted the manuscript. AB and CP participated to the early phases of the work, contributing to gene promoter selection and to the development of the COOP algorithm, respectively. GAD revised the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table 6. Results of the analysis of 26 human positive control datasets with COOP. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Table 7. Comparative evaluation of COOP performance on 26 human positive control datasets. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ Financial support of the Italian Ministry for Technological and Scientific Research and of Padova University to G.A.D and of Italian Association for Cancer Research (AIRC) to S.B is gratefully acknowledged. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Flow-chart of COOP program**. Input, output and main steps are shown. ::: ![](1471-2105-6-121-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Comparison of patterns discovery results in retinal gene promoter sequences and in 1,000 negative control datasets**. Plots of number of patterns (12 bp long, with at most two variable positions) *vs*number of sequences in which they were found, in retinal gene promoter sequences (open squares) and in 1,000 negative control datasets (filled diamonds). For negative control datasets, the average value of 1,000 sets of sequences is given, with a two standard deviations interval. Statistically significant differences (0.05 threshold) are marked by stars. (A) Comparison between the 1000M52 dataset (52 promoter sequences of genes overexpressed in the retina) and the RAN1000M52i dataset (1,000 groups of 52 randomly chosen human promoters); (B) Comparison between the 1000M91 dataset (91 retinal gene promoter sequences) and the RAN1000M91i (1,000 groups of 91 randomly chosen human promoters). ::: ![](1471-2105-6-121-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Statistics comparing the accuracy of COOP and of 14 different motif discovery tools on 26 human positive control datasets**. Combined measures of correctness over all 26 human datasets, as defined in Methods. The number of datasets (out of 26) for which no motif was predicted by each tool is reported in brackets, following the name of the tool. ::: ![](1471-2105-6-121-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Description of COOP parameters. ::: **Step** **Parameter** **Description** **Range** **Example** -------------------------------- --------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ----------- --------------------------- Search for pattern occurrences *q* *Quorum*(minimum number out of N input sequences in which a given pattern must be represented) 1 - N ≥ 20 out of 100 sequences Clustering *d* Physical distance between 5\'-ends of occurrences of patterns of length p 0 - \|p\| ≤ 2 nucleotides *o* Ratio between observed overlapping occurrences of two patterns and their average number of occurrences 0 -- 1 ≥ 0.8 Consensus building *i* Ratio between the number of nucleotides per alignment position and the total number of lines in the alignment. The maximal number of adjacent positions exceeding the threshold *i*(*m*) is further analysed for determining the consensus sequence of the motif 0 -- 1 ≥ 0.5 *l* Nucleotide length of the lateral region of the motif 0 - m/2 3 bp (out of 10) *f*~*l*~ Frequency of a single nucleotide in each position of the lateral region to be considered specified 0 -- 1 ≥ 0.6 *f*~*c*~ Frequency of a single nucleotide in each position of the core region to be considered specified 0 -- 1 ≥ 0.8 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Procedure for building a consensus sequence starting from a matrix of nucleotide counts, according to selected parameters. Rows from two to five represent the matrix of nucleotide counts in different positions of an alignment associated to a cluster of pattern occurrences. The sixth row contains, for each alignment position, the ratio between number of sequences in the position and the total number of lines in the alignment. Out of 11 positions of the matrix, positions from one to ten (shaded in grey) fulfil the minimum *i*(0.5) and are considered for building the consensus. If the lateral region length is set to 3 nucleotides, a 3-4-3 motif is obtained. The *f*~*l*~(0.6) threshold is applied to the positions in the lateral regions, whereas the *f*~*c*~(0.8) is applied to positions in the core region. Cells containing values fulfilling the condition reported on the left are in bold. In the last row, the derived consensus sequence is shown. ::: **1** **2** **3** **4** **5** **6** **7** **8** **9** **10** **11** -------------------- --------- --------- ------- --------- ------- ------- ------- ------- ------- --------- -------- **A** 0 0 0 4 0 0 0 0 0 0 0 **C** 0 0 5 0 5 2 0 0 0 0 2 **G** 0 4 0 0 0 3 5 5 0 4 0 **T** 3 0 0 1 0 0 0 0 5 0 0 *i*(0.5) **0.6** **0.8** **1** **1** **1** **1** **1** **1** **1** **0.8** 0.4 *f*~*l*~(0.6) **1** **1** **1** **1** **1** **1** *f*~*c*~(0.8) **0.8** **1** 0.6 **1** Consensus sequence **T** **G** **C** **A** **C** **N** **G** **G** **T** **G** **-** ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Number of sequences in which most represented patterns were found in different retinal datasets. ::: **Pattern** **500U91** **500PM91** **500M91** **1000M91** ------------- ------------ ------------- ------------ ------------- **10-2** 44 42 32 40 **12-2** 19 19 10 12 **14-2** 11 11 5 7 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Statistics about patterns found in different groups of retina gene promoter sequences and in the corresponding negative control random datasets. ::: ***Quorum*** **Pattern** **Obs.** **Exp.** **P-value** ---------------------------------------------------------------- -------------- ------------- ---------- ---------- ------------- **10-2** 719 351.3 0.017 **20** **12-2** 0 \- \- **14-2** 0 \- \- **1000M91**(91 promoters of retinal genes *vs*RAN1000M91~i~) **10-2** 18683 12846.4 0.016 **10** **12-2** 41 35.4 0.324 **14-2** 0 \- \- **7** **12-2** 714 410.2 0.060 **1000M52**(52 promoters of retinal genes *vs*RAN1000M52~*i*~) **5** **12-2** 1537 429.4 0.001 ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### List of 60 consensus sequences corresponding to selected motifs showing most conserved central regions. For each motif, consensus sequence, length and total number of occurrences in the 1000M dataset are reported, along with LocusLink symbols of corresponding genes. In the last column, for each consensus, the list of mammalian transcription factors recognising similar DNA sequences is reported. ::: **Consensus sequence** **Length** **Total occ.** **Genes in which promoter the motif was found** **Transcription factors** ------------------------ ------------ ---------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------ AAAAAAAAAAAAAA 14 151 EFEMP1, CCNI, CNGB3, KCNV2, IMPDH2, SLC24A1, DHRS3, G2AN, RTP801, MGC15WIF11, USH3A, CRX, 18, HMGA1, SLC24A2, RDS, TULP1, DC-TM4F2, OPN1SW, RP1, MGAT4B, GAPD, ELOVL4, RRAD, ARR3 NGGCCCCGCCCCCN 14 114 EEF1G, HMGA1, EFEMP1, CYBA, KRT18, OPA1, DPYSL4, RAX, FLJ1415, MGC15WIF11, FLJ1415, ALMS1, EIF3S8, G2AN, ALMS1, DC-TM4F2, MSH6, RCV1, KRT19, DHRS3, PITPNC1, RRAD, HPCAL1, MGAT4B, SLC38A3, IMPDH2, CNGB1, RDH5, EFEMP1, CRABP1, C7orf20, CCNI, GNB1, CRX, GAPD, ARF4L, AIPL1, DKFZP564K0822 AP-1, GCF, Sp1, Sp3, TFIID GCACCCCCAGCCCCN 15 101 RHO, G2AN, EFEMP1, SLCO4A1, CYBA, HPCAL1, KIFC3, RCV1, NK4, KRT18, CRX, ARR3, PPP1R3F, MGAT4B, NRL, RRAD, CCNI, SAG, ALMS1, MGC15WIF11, DKFZP564K0822, VMD2, DPYSL4, GNAT1, GAPD, OPN1SW, RAX, DHRS3, COPEB, SLC38A3, TMEM16B, SLC24A1 Sp1 NGAGGGCAGGGGCNN 15 94 GNB1, KRT19, ELOVL4, VMD2, MSH6, HMGA1, RHO, NK4, SLC38A3, LRRCGUCA1B, CYBA, RCV1, RRAD, GUCY2D, MGC15WIF11, AIPL1, MGAT4B, KIFC3, CRX, CRABP1, G2AN, ALMS1, RTP801, EEF1G, COPEB, OPA1, EFEMP1, KCNV2, PDE6A, AOC2, RLBP1, FLJ1415, RAX, DPYSL4, WIF1, DC-TM4F2 Sp1 CCTCCCTCCCTCCC 14 76 ARF4L, COPEB, RHO, SLC38A3, FLJ1415, WDR17, ELOVL4, DHRS3, KCNV2, OPA1, CCNI, GUCA1B, RDH5, RAX, ALMS1, DKFZP564K0822, NK4, RGS19IP1, RRAD, KIFC3, KRT19, SLCO4A1, HPCAL1, DPYSL4, TNFRSF6, CNGB1, DC-TM4F2 MAZ NCTCCCCCTCCCCC 14 43 CNGB1, GAPD, RPE65, ALMS1, COPEB, MSH6, RRAD, CRABP1, TNFRSF6, CRX, WIF1, FLJ1415, DKFZP564K0822, PDE6A, RDH5, SLC38A3, CYBA, GNB1, MERTK, WDR17 Sp1, AP-2, MAZ GNNTGGGGGAGGGGN 15 41 CYBA, RLBP1, KCNV2, CNGB1, COPEB, KIFC3, RDH5, CCNI, FLJ1415, MGC15WIF11, AIPL1, NK4, HPCAL1, CNGB1, GUCA1A, ALMS1 MAZ, Sp1 CNCCCCCACCCCCACC 16 40 RCV1, SLC38A3, HPCAL1, KIFC3, RLBP1, RPE65, DHRS3, RTP801, CYBA, DPYSL4, RDH5, RRAD, COPEB AP-2alphaB, Sp1, WT1 CTCCCCCTCCCCNNC 15 26 CNGB1, CRX, GAPD, RHO, CNGB1, COPEB, CYBA, AIPL1, RAX AP-2, MAZ, Sp1 CCCCAGCCCCNCA 13 23 CCNI, EFEMP1, SLCO4A1, MGC15WIF11, ARR3, CYBA, HPCAL1, KIFC3, RAX, RLBP1, MGAT4B, AIPL1, RGS19IP1, ALMS1 Sp1 NNGGCCCCTGCCCN 14 23 HMGA1, NK4, LRRCGUCA1B, FLJ1415, GNB1, KRT19, AIPL1, GUCA1A, DHRS3 Sp1 NCCCCCTCCACCN 13 22 ARR3, HMGA1, KRT19, VMD2, DHRS3, ARF4L, RAX, CCNI, SIRT3, GUCA1B, DC-TM4F2 Sp1 NCNGGGCTGGGGN 13 22 CYBA, HPCAL1, RRAD, GAPD, GUCA1A, RHO, G2AN, EFEMP1 Sp1 NNTCCCCCTCCCNN 14 22 TNFRSF6, CNGB1, CRX, EEF1G, GAPD, RPE65, ALMS1, DKFZP564K0822, COPEB, AIPL1 AP-2alphaB, MAZ, Sp1, WT1 -KTS NNCCCAGCCCCCAN 14 20 RDH5, SLC38A3, EFEMP1, ARR3, CYBA, GAPD, HPCAL1, NK4, PPP1R3F Sp1 NTGGGGGAGGGGNA 14 20 COPEB, CYBA, RLBP1, PITPNC1, CNGB1, CRX, GAPD, MERTK, CCNI MAZ, Sp1, Sp3 CCNGCCCTGGCCT 13 18 GUCA1A, GUCY2D, RCV1, VMD2, EFEMP1, LRRCGUCA1B, C7orf20, 4, RRAD, UNC119, MERTK Sp1 GCNGCCCCTGCCN 13 18 CRX, CYBA, GNB1, HMGA1, RHO, SLC38A3, MGAT4B, FLJ1415, KRT18 NCNGGGGGCGGGG 13 18 CYBA, RRAD, FLJ1415, HMGA1, RDH5, RGS19IP1, G2AN, RTP801, DC-TM4F2 AP-1, ER, Sp1 CTNCCCCTCCCC 12 17 RLBP1, AIPL1, PITPNC1, CNGB1, GAPD, RHO, CNGB1, EFEMP1, COPEB, CYBA, GNB1, PDE6A AP-2alphaB, MAZ, Sp1 GGGGTGGGGNTG 12 17 GUCY2D, FLJ1415, AIPL1, RDH5, CRABP1, HPCAL1, KIFC3, DHRS3, RTP801, CYBA, RLBP1 AP-2alphaB, Sp1, Sp3 CCCGCCCCTGNCC 13 16 GNB1, HPCAL1, KRT19, MGAT4B, G2AN, Sp1 NGGGGGTGGGGGN 13 16 HPCAL1, RRAD, DHRS3, FLJ1415, CYBA, GNB1, DPYSL4 Sp1 NNCCCCCGCCCCNN 14 16 GNB1, RGS19IP1, LRRCGUCA1B, ALMS1, DC-TM4F2, KRT18, SAG AP-1, AP-2alphaB, ER, Krox-20, Sp1, WT1, WT1 I, WT1 I -KTS AGNGGGAGGGGCN 13 14 CYBA, EFEMP1, RAX, MGC15WIF11, ARF4L, CRX, SLCO4A1 MAZ, Sp1, Sp3 CCCTGTCCCTGGAN 14 14 ARR3, HPCAL1, FLJ1415, DC-TM4F2, KRT19, LRRCGUCA1B, TMEM16B GR CGGGGCCGCCNCN 13 14 FLJ1415, DC-TM4F2, MGC15WIF11, COPEB, MGAT4B, SLCO4A1, RAX CUP, Sp1 CTCTCTCTCCNTN 13 14 GAPD, GUCA1A, NRL, RRAD, FLJ1415, GNAT2, KCNV2 NANCTCTGCACCC 13 14 LRAT, TNFRSF6, CYBA, KIFC3, DPYSL4, G2AN, RTP801 NCCGCCCCCGCCN 13 14 GNB1, IMPDH2, SLC38A3, COPEB, CYBA, KRT18, SLCO4A1 AP-1, ER, Kxox-20, Sp1, WT1 I -KTS, WT1-del2 NGGCCTCTGGNCN 13 14 CYBA, GAPD, KRT19, RDH5, DPYSL4, HPCAL1, MGAT4B NGGGAGGGGGAAG 13 14 GAPD, AIPL1, FLJ1415, EEF1G, RPE65, ALMS1, WDR17 AP-2alphaB, MAZ, Sp1, WT1 I -KTS NGNCCCCAGCCCC 13 14 GAPD, GUCA1A, RHO, ARR3, CYBA, NK4, PPP1R3F AP-2, Sp1 NNCCCAGCCCAGNN 14 14 GAPD, RHO, ARR3, CRABP1, CYBA, RRAD, MGAT4B Sp1 TGGGGGTGGGGGN 13 14 HPCAL1, RLBP1, DHRS3, CYBA, HMGA1, RRAD, DPYSL4 Sp1 NGGCGGGGGCGGGG 14 13 EFEMP1, KRT18, RRAD, SLCO4A1, IMPDH2, EFEMP1, COPEB AP-1, Krox-20, Sp1, WT1 I -KTS, WT1-del2 GGNAGGGGCGGG 12 11 ELOVL4, REA, G2AN, GNB1, MSH6, GUCY2D, RGS19IP1, LRRC21, SLCO4A1, PITPNC1 MAZ, Sp1 CCCGCCCGCCCC 12 9 GNB1, RGS19IP1, WIF1, PITPNC1, DC-TM4F2, HMGA1, DPYSL4, KRT18, RAX Sp1 GGGCGGGGCNGG 12 9 CYBA, DPYSL4, MGAT4B, MSH6, RCV1, ALMS1, FLJ1415 ER, GCF, Sp1 GGGCTGGGGGTG 12 9 CYBA, HPCAL1, KIFC3, RCV1, RHO, G2AN, DKFZP564K0822 Sp1 GGGGAAGGGNGG 12 9 TULP1, CRX, MSH6, KRT19, CNGB1, SLC38A3, AIPL1, HMGA1, FLJ1415 GGGGCGGGCNNG 12 9 EEF1G, KRT19, DC-TM4F2, GUCY2D, RGS19IP1, PITPNC1, C7orf20, RTP801 ER, Sp1 GGNGCGGGCGGG 12 9 HMGA1, KRT19, DPYSL4, DC-TM4F2, RGS19IP1, WIF1, PITPNC1, FLJ1415 AP-2, ETF, Krox-20, Sp1, WT1 I -KTS GNNGGGGCTGGG 12 9 GAPD, HPCAL1, KIFC3, RCV1, RAX, COPEB, RDH5 WT1 -KTS CAGGGGGCGGGG 12 8 CYBA, EFEMP1, HPCAL1, FLJ1415, GAPD, HMGA1, G2AN, DC-TM4F2 AP-1, ER, Sp1, Yi CNCCCCCACCCC 12 8 CYBA, HMGA1, RCV1, SLC38A3, HPCAL1, RLBP1, DHRS3 AP-2alphaB, CACCC-binding, factor, Sp1, WT1 GAGTGGGGGAGG 12 8 DHRS3, KCNV2, COPEB, CYBA, HMGA1, WIF1, FLJ1415, MGC15WIF11 GCCTGGGGGAGG 12 8 CYBA, SIRT3, KIFC3, CCNI, DKFZP564K0822, DC-TM4F2, MGC15WIF11 AP-2 GGGCAGGGGCNG 12 8 CYBA, GNB1, HPCAL1, HMGA1, RHO, SLC38A3, MGAT4B, G2AN Sp1 GGGCGGGGCTGG 12 8 CYBA, HPCAL1, RAX, MSH6, RCV1, ALMS1, DC-TM4F2 ER, GCF, Sp1 CCCTGTCCCTGG 12 7 CNGB1, GNB1, FLJ1415, KRT19, ELOVL4, TMEM16B, FLJ1415 GR CCTTCCCCCNGC 12 7 GNB1, SLC38A3, AIPL1, SLCO4A1, RDH5, TULP1, NK4 MAZ CNCCTCCTGCNC 12 7 CRABP1, GUCA1A, PDE6A, RGR, DPYSL4, WIF1, HPCAL1 PPUR, Sp1 CNGCCCCCAGNC 12 7 RHO, EFEMP1, DC-TM4F2, CNGB1, CYBA, NK4, MERTK Sp1 GCNCCCCTCCCC 12 7 COPEB, CRX, HPCAL1, RGR, CNGB1, MERTK, RAX MAZ, Sp1 GGGCAGGGGCGG 12 7 ELOVL4, HMGA1, HPCAL1, RHO, SLC38A3, MGAT4B, G2AN Sp1 GGGGCTGGGGNC 12 7 ARR3, CYBA, HPCAL1, NK4, RAX, PPP1R3F, RLBP1 AP-2alphaB, Sp1 GNAGGGGGCAGG 12 7 GAPD, NK4, GUCA1B, SLC38A3, WIF1, G2AN, EFEMP1 Sp1 TGGGGGAGGNNA 12 7 KCNV2, COPEB, HMGA1, KIFC3, RDH5, CCNI, FLJ1415 MAZ, Sp1 TTTTTTTTTNTA 12 7 IMPDH2, G2AN, SLC24A2, RTP801, KCNV2, USH3A-PROMB, CCNI TBP :::

PubMed Central

2024-06-05T03:55:58.302732

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173081/", "journal": "BMC Bioinformatics. 2005 May 18; 6:121", "authors": [ { "first": "Stefania", "last": "Bortoluzzi" }, { "first": "Alessandro", "last": "Coppe" }, { "first": "Andrea", "last": "Bisognin" }, { "first": "Cinzia", "last": "Pizzi" }, { "first": "Gian Antonio", "last": "Danieli" } ] }

PMC1173082

Background ========== Protein modelling and structure prediction as well as binding and interaction prediction have become very valuable instruments for researchers in biology and medicine. In order to build reasonable and useful models, as much information as possible has to be incorporated into the protein modelling process. To refine protein models, chemical as well as spatial information about ligand structures can be considered, specifically to optimise side-chain conformations around binding-sites \[[@B1]\]. Several databases delivering structures and different additional information about ligand molecules from the Protein Data Bank (PDB) \[[@B2]\], \[[@B21]\] have been provided on the Internet. Ligand Depot \[[@B3]\], \[[@B22]\] comprises chemical and structural information for small molecules found in the PDB and also provides a graphical interface for performing chemical substructure searches. Idealized three-dimensional structures and additional information about PDB ligands can be retrieved via the search interface of the E-MSD macromolecular structure relational database \[[@B4]\], \[[@B23]\]. Besides many other features, Relibase \[[@B5]\], \[[@B24]\] allows for two-dimensional similarity and substructure search among the ligands as well as for sequence similarity search among the corresponding proteins. LigBase \[[@B6]\], \[[@B25]\] is a database of ligand binding sites aligned with related protein structures and sequences. Various information about ligands bound to macromolecules deposited in the PDB can be retrieved from many further sources like HIC-Up \[[@B7]\], \[[@B26]\], PDBsum \[[@B8]\], \[[@B27]\] and the IMB Jena Image Library of Biological Macromolecules \[[@B9]\], \[[@B28]\]. The latter can be searched after geometrical properties of the ligand binding sites. Information contained in these databases can help identifying ligands which are likely to bind to a given protein structure. The opposite question, namely to find target proteins for a certain ligand, was addressed in \[[@B10]\], where a collection of protein active sites was extracted from the PDB and scanned with aid of a docking algorithm. Further data collections emphasize the link between binding affinities and structures of the protein-ligand complexes and, inter alia, provide experimentally measured binding data, e.g. PLD \[[@B11]\], \[[@B29]\] LPDB \[[@B12]\], \[[@B30]\], PDBbind \[[@B13]\], \[[@B31]\]. For modelling and simulation purposes, chemical and spatial information about protein ligands is vitally important. Addressing this fact, *SuperLigands*is a collection of small molecule structures contained in the PDB, facilitating comparison of the molecules regarding their two-dimensional similarity. As spatial comparison of compounds can deliver valuable information in addition to this, *SuperLigands*also allows for three-dimensional superpositions. Spatial coordinates of the compounds can be retrieved as MDL Mol files, which include information about multiple bonds. Construction and content ======================== Native conformations of small molecules contained in the PDB and additional information were collected from the PDB \[[@B2]\], \[[@B21]\], Ligand Depot \[[@B3]\], \[[@B22]\] and MSD \[[@B4]\], \[[@B23]\] and deposited in the database *SuperLigands*. The database has been designed as a MySQL relational database and supplemented with a user-friendly web interface. Database queries are performed and HTML pages are generated via PHP scripts. The freely available MDL^®^Chime plug-in is used to display molecules and allows the user some manipulations of the view and to store the displayed molecule in the MDL Mol file format. In order to enable fast two-dimensional searches, 960 bit binary fingerprints (MDL MACCS Keys \[[@B14]\]) were calculated and stored in the database for all ligands. Tanimoto coefficients are calculated via a PHP script. Here, all 960 MDL keys are included and equally weighted. The Tanimoto coefficient for two structures a and b is defined as follows: T(a,b)=Nab/(Na+Nb-Nab), where Na and Nb are the numbers of bits set in the fingerprint of structure a and b, respectively, and Nab is the number of bits which are common to both fingerprints. Three-dimensional superposition of two different PDB ligands is performed in the following way: each conformation of one molecule occurring in the PDB is superposed with each conformation of the second molecule. Those two instances matching best are displayed. The best match is defined by maximizing the *score*defined by ![](1471-2105-6-122-i1.gif) where RMSD is the Root Mean Square Deviation of the superposed atoms. For completion, PDB codes, chain identifiers and positions in the PDB files of the matched conformations as well as the atom numbers of both ligands, the number of superposed atoms, the number of superposed atoms of the same type and the RMSD of the superposition are returned. For detailed information regarding the superposition algorithm see \[[@B15]\]. Utility ======= *SuperLigands*can be searched by hetero-ID (i.e. the three-letter PDB code for hetero-compounds), name, molecular formula or PDB identifier. In the results table, hetero-ID and names of the compounds are given. Moreover, the molecular structure is displayed in one cell of the table where it can be rotated by the user and different displaying options can be chosen. More information like molecular formula, atom numbers and occurrence in the PDB can be retrieved additionally. The database SuperLigands contains compounds defined by \'HETATM\' records in PDB files. Some of these molecules may be bound to pseudopeptides but can also be separate ligands. Coming across such a molecule, the user is given a hint and is provided with a list of pseudopeptides in which this molecule is bound. The user can search the database for molecules that have a significant two-dimensional similarity to a given ligand or assess the three-dimensional similarity of two compounds by superposing them with each other. Such similarity queries can be performed starting from the search results tables or directly using separate forms. Using such a form, the Tanimoto coefficient of two given structures can also be retrieved. A typical example for a query to *SuperLigands*is a search for tobramycin, known as antiinfective and antibacterial drug, starting in the main form. Searching the database for similar compounds in the next step supplies the drug kanamycin as PDB ligand with the highest Tanimoto similarity (98.6%). Now, a three-dimensional superposition of all instances of tobramycin (32 atoms, two instances) and kanamycin (33 atoms, four instances) occurring in the PDB can be performed. The best fitting structures are superposed and then displayed. In this case, best fitting are the instances of tobramycin in PDB entry 1m4d and kanamycin in 1m4i with an RMSD of 0.14Å (32 atoms superposed). Navigating through the website, the topological and spatial similarities of PDB ligands can be obtained easily. For example, tobramycin and ribostamycin (31 atoms, four instances) have a Tanimoto coefficient of 96.5%, the RMSD of their best superposition is only 0.95Å (25 atoms superposed). In turn, geneticin (34 atoms, five instances) delivers a Tanimoto coefficient of only 81.1% and a much better RMSD (0.16Å, 30 atoms superposed). As an additional feature of *SuperLigands*, similarity of PDB ligands to known drugs can be assessed in a comfortable manner. Starting with a ligand, a two-dimensional similarity search as described above can be initiated, not only among the PDB ligands but also in a database containing the structures of known drugs (*SuperDrug Database*\[[@B16]\], \[[@B32]\]). The drug structures found can be superposed spatially (for an example, see Figure [1](#F1){ref-type="fig"}). Discussion ========== Statistics: comparison of PDB ligands with drugs ------------------------------------------------ Recently, a database containing 2396 drug molecules and having the same design as *SuperLigands*has been created (*SuperDrug Database*\[[@B16]\], \[[@B32]\]). To answer the question, how many drugs or drug-like molecules are bound to PDB structures, Tanimoto coefficients have been calculated for all pairwise combinations of molecules from *SuperLigands*and the *SuperDrug Database*. A set of 5,040 PDB ligands has been incorporated into these calculations. Considering two molecules having a Tanimoto coefficient of 100% (or greater than 95% ; 90%) identical or very similar, this analysis reveals that 413 (771 ; 1,457) of 5,040 PDB ligands are drugs or drug-like compounds. Furthermore, some chemical properties of PDB ligands and drugs have been compared (see Figure [2](#F2){ref-type="fig"}). The distributions of numbers of hydrogen bond donors for PDB ligands and drugs differ most significantly. A bigger percentage of the drugs (26%) have no hydrogen bond donor, the largest fraction of the PDB ligands (19%) have two of them. About half of the drugs have no or only one hydrogen bond donor, which applies for only a quarter of the PDB ligands. About one third of the drugs have three or four hydrogen bond acceptors, the fractions of drugs with nine or more hydrogen bond acceptors drop below 3%. For the PDB ligands, the distribution is more flat: only 22% of them have three or four hydrogen bond acceptors, still over 3% of them have 11 hydrogen bond acceptors. Most drugs have a logP value around 3, and the logP values of the PDB ligands accumulate around the negative value -1. Approximately the same fraction of PDB ligands and drugs are \"drug-like\" according to the Lipinski \"Rule of five\" \[[@B17]\]: 92 and 91%, respectively, have a logP value less than 5, although altogether the logP values of the drugs are closer to this critical value. A majority of the PDB ligands have very low molecular weights in comparison to the drugs, which supposedly is be caused by the fact that in proteins often very small solvent molecules are bound. Nevertheless, slightly more (5%) drugs than PDB ligands fulfil the Lipinski \"Rule of five\" regarding the molecular weight. The same applies for the numbers of hydrogen bond donors (and acceptors): 7% (5%) more drugs fulfil the Lipinski \"Rule of five\". Compounds violating more than one of the Lipinski Rules are assumed to have problems with bioavailability and are therefore presumably not suitable as drugs. Table [1](#T1){ref-type="table"} shows the percentages of PDB ligands and drugs violating the Lipinski Rules. From this table can be seen that a total of approximately 19% of the PDB ligands and 10% of the drugs, respectively, violate more than one of the Lipinski Rules. This analysis reveals that there are only marginal differences between PDB ligands and drugs regarding single chemical properties. But, not surprisingly, from a general point of view, PDB ligands are significantly less drug-like than drugs. Discussion ---------- *SuperLigands*is a collection of PDB ligands freely accessible via a user-friendly web site. Molecular coordinates can be retrieved as MDL Mol files, supplementing the connectivity records contained in PDB files with bond types, which are necessary for modelling and simulation purposes. The database can be searched for compounds similar to a given ligand by comparison of Tanimoto coefficients. As stated in \[[@B15]\] and shown in the example in the section *Utility*, spatial comparison of small molecules can reveal more similarities, and thus similar kinds of interaction, than a pure two-dimensional topology comparison. With aid of *SuperLigands*, such three-dimensional comparisons can be performed easily. Moreover, the topological similarity of PDB ligand structures to known drugs can be assessed by calculation of Tanimoto coefficients. Conclusion ========== The database presented here supplements the set of existing resources of information about small molecules bound to PDB structures. As novel features, three-dimensional comparison of molecules as well as topology comparison of PDB ligands with known drugs are made possible. Thus, *SuperLigands*represents a valuable means of analysis and prediction in the field of biological and medical research. Availability and requirements ============================= The database is publicly accessible at <http://bioinf.charite.de/superligands/>. For visualisation, the free browser plug-in MDL^®^Chime is required. Chime runs on Windows systems with Microsoft Internet Explorer (6.0 or 5.5 SP2) or Netscape 4.75, 4.79 or on Mac OS 9.0 or 8.6 with Netscape 4.75 (please see <http://www.mdlchime.com/products/framework/chime/sys_req.jsp> for detailed information). Authors\' contributions ======================= EM designed the database and the web site and finished its functionality, was responsible for data acquisition and processing and drafted the manuscript. MD delivered the basic part of the website functionality and contributed to database conception and data processing. AG provided the tool for three-dimensional superposition and helped to draft the manuscript. RP conceived of the project, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by the BMBF (German Federal Ministry of Education and Research). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Usage of the web interface of *SuperLigands***. From the main menu, the form *Compound search*can be reached. Here, a PDB ligand can be searched after hetero-ID, name, molecular formula or PDB identifier. In the first column of the results table, two buttons can be found to retrieve more information. The *FULL info*button delivers detailed information about the selected PDB ligand like molecular formula, atom numbers and occurrence in the PDB. After clicking the *DRUGS*button, a two-dimensional similarity search among the drugs in the *SuperDrug*database \[16\] is performed. The best hits are displayed in a new window. From here, they can be spatially superposed. In the figure, this procedure was carried out for celecoxib, a COX-2 inhibitor which was recently categorised as problematic (see the \"Pfizer Statement on New Information Regarding Cardiovascular Safety of Celebrex\" \[18\]). The two-dimensional similarity search in the *SuperDrug*database delivers only hits below 72% Tanimoto similarity. A following spatial superposition of the best hits reveals a further COX-2 inhibitor, namely valdecoxib, (RMSD 0.26Å and 21 of 22 atoms superposed) as very similar to celecoxib. The Tanimoto similarity of celecoxib and valdecoxib is only 65% and there are two drugs more similar to celecoxib: Sulfaphenazole (71% Tanimoto similarity, spatial superposition with 0.65Å RMSD and 15 of 22 atoms superposed) and Sulfamazone (67% Tanimoto similarity, spatial superposition with 0.32Å RMSD and 17 of 26 atoms superposed). Nevertheless, the three-dimensional comparison here proves to be very important to reveal molecular similarities in addition to topological comparison (as also shown in the example in the section *Utility*), which is confirmed by the fact that valdecoxib was categorised as toxic \[19\] and sales of this drug were suspended recently \[20\]. ::: ![](1471-2105-6-122-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Statistics: comparison of PDB ligands with drugs**. Chemical properties of 5040 PDB ligands and 2396 drug molecules are compared: histograms for numbers of hydrogen bond donors and acceptors, logP value and molecular weight are shown. Molecular weight within \[100,200) means that the molecular weight is greater than or equal to 100 and less than 200. Those areas for which the Lipinski \"Rule of 5\" is fulfilled, are highlighted in grey. ::: ![](1471-2105-6-122-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Percentage of PDB ligands and drugs violating certain numbers of Lipinski Rules ::: Number of violated Lipinski Rules PDB ligands Drugs ----------------------------------- ------------- ------- 0 64.4 75.7 1 16.9 14.0 2 10.3 5.7 3 8.3 4.7 4 0.1 0.0 :::

PubMed Central

2024-06-05T03:55:58.309870

2005-5-19

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173082/", "journal": "BMC Bioinformatics. 2005 May 19; 6:122", "authors": [ { "first": "Elke", "last": "Michalsky" }, { "first": "Mathias", "last": "Dunkel" }, { "first": "Andrean", "last": "Goede" }, { "first": "Robert", "last": "Preissner" } ] }

PMC1173083

Background ========== The introduction of a microarray based comparative genomic hybridization method (arrayCGH) in 1997 paved the way for higher resolution detection of DNA copy number aberrations \[[@B1]\]. ArrayCGH is founded on the same principles as metaphase CGH, but uses mapped reporters instead of whole chromosomes. One of the major challenges in arrayCGH studies remains the accessibility, management, and interpretation of the vast amount of data generated in single experiments, and parallel comparison of multiple experiments. Typically, these arrays contain 3,000 to 30,000 reporters, each of which has multiple biological annotations (chromosomal position, sequence information, gene name, biological and molecular function,\...) as well as physical (grid layout) and quality control (sequence verification, FISH mapping information,\...) annotations. In addition, the description of the DNA samples under investigation and the applied lab protocols should be easily accessible. For classical CGH, several commercial software packages are available to analyze and interpret the data of a CGH experiment. Also for arrayCGH there are a number of separate software systems that individually address some of the needs, such as databases for data storage (BASE \[[@B2]\]), applications for clustering and visualization of microarray data (seeGH \[[@B3]\], M-CGH \[[@B4]\], CGHAnalyzer \[[@B5]\], aCGH-smooth \[[@B6]\] and CGH-Miner \[[@B7]\]), public genome databases that contain reporter information, commercially available Laboratory Information Management Systems (LIMS), and various storage methods for recording biomaterial annotations. However, none of these software packages or databases combine all these features (see Supplemental Table). In this paper, we present the development of a web based open source arrayCGH analysis platform, *arrayCGHbase*, that combines all these features and on top provides additional unique aspects making the analysis and sharing of arrayCGH data easily implementable for both research and routine purposes. Implementation ============== MIAME compliant database ------------------------ *arrayCGHbase*runs in Windows, Linux, Macintosh, and Unix environments. Particular attention was paid to the use of open source software for the development of *arrayCGHbase*. The software was developed in the PHP scripting language, with all data being stored in a relational, MIAME \[[@B8]\] (Minimal Information About a Microarray Experiment) supportive, MySQL database and communicated to the user through an Apache Web server (Figure [1](#F1){ref-type="fig"}). After installation on a private server, experiments can be shared by different users over the internet or a local intranet. *ArrayCGHbase*integrates DNA sample information, lab protocols, extracted data, and contains a plug-in architecture for data transformation, analysis, and graphical display, allowing users to develop their own modules. Reporters can be directly linked to the Ensembl \[[@B9]\] or UCSC \[[@B10]\] genome browsers, providing additional up-to-date information on each reporter. Reporters can also be manually imported into the MySQL database with the ability to update all linked experiments. The structure of *arrayCGHbase*was designed to follow the laboratory workflow and is compatible with all types of arrayCGH experiments and data formats (dual colour genomic clone, cDNA \[[@B11]\], or oligonucleotide \[[@B12]\] arrays spotted on any substrate, physical layout, type of array, as well as single channel hybridizations such as the Affymetrix SNP chips \[[@B13]\]). With a personal account and administrated access levels, a user can enter new DNA samples, annotate these, and append all relevant sample information such as quantity, quality, and applied lab protocols at each step. Each user can group experiments together into projects and, in a uniform and streamlined fashion, apply filters and transformations and run analyses. Data is exportable in several formats for offline analysis using other (dedicated) software tools, for publication or for sharing data with the research community. For advanced users, an SQL query window allows interrogation of the underlying MySQL database. Data processing and visualization routines ------------------------------------------ A first and important step in data analysis of arrayCGH experiments is the processing of large, possibly noisy data sets to identify the specific reporters that are differentially hybridized and hence show an aberrant copy number. Data processing is performed in a streamlined four-step manner: (1) the local noise or background associated with the experiments is removed, (2) the quality of the experiment is assessed and poor quality features are removed, (3) ratios are calculated, transformed to log~2~scaled ratios, and normalized, and finally (4) reporters that show altered ratios are identified and hence, reporters with aberrant copy number are identified. In the past, this normally required the sequential processing of data by different, often incompatible programs. Using established and widely used microarray (CGH) data processing procedures, *arrayCGHbase*will automatically correct the signal intensities, filter out unwanted poor quality features (based on signal to noise ratio, image processing software related flags, or other user defined filters), normalize the fluorescence intensity ratios, score levels of differential hybridization, combine the results of replicate experiments and assess the quality of individual and replicate experiments. All these steps are user adjustable. Input data and local background correction ------------------------------------------ The experimental input data for *arrayCGHbase*consists of export files generated by image analysis software. Currently, the program recognizes files from GenePix Pro versions 2.0--4.0, Scanalyze version 2.0, UCSF SPOT version 2.0, Imagene versions 4.0 -- 5.5 and the Affymetrix Chromosome Copy Number Tool. The program can easily be updated for the recognition of other data input formats upon request. Moreover, *arrayCGHbase*has an interactive import wizard, which makes it possible to import data at your own desire. The processing steps may be changed by altering the parameters at the input stage. By default, the results for each feature are defined as the median foreground minus background intensities for each dye (as determined by the image processing software). The ratio of each feature is determined as the relative background corrected signal between the two dyes or in the case of single color experiments as the corrected signal intensity. Poor quality flagging --------------------- Nearly every experiment contains features of poor quality, comprising features that have unusual morphology (e.g. doughnut patterns), exhibit uneven hybridization, or have saturated signal intensity. After background corrections, *arrayCGHbase*can automatically flag features of inferior quality using different criterions (e.g., the standard deviation between replicates), by a manually set signal or signal-to-noise threshold, or using image processing generated flag annotations. Normalization ------------- Following calculation of the corrected signal intensities and filtering for good quality features, the relative contributions of the fluorescence intensities are compared. To go from a multiplicative space to an additive space, ratios are log~2~transformed. Ideally, the signals of the two dyes should be equal for nucleic acid reporters that have equal amounts in the test and reference samples (i.e., the log~2~transformed ratios of the two corrected signals should approach zero for reporters hybridizing to an equal degree in both fluorescence channels). However, in practice the ratio of the corrected signal intensities deviates from the expected ratio due to the different molecular and physical characteristics of the dyes, the different amounts of DNA used for labeling with the different dyes, the spatial heterogeneity in the hybridization conditions across the slide, and many other factors. Normalization compensates these effects by applying a data transformation such that ratios of reporters with unchanged copy-number are close to zero. In the normalization step, an appropriate term is added or subtracted from the log~2~transformed ratio for each feature. The program allows normalization in several ways, either by global normalization or subgrid (or pin) normalization, or by a combination of different normalization procedures. A major issue in microarray normalization is the definition of the set of constant probes to which the data are normalized. The most widely accepted method employs the \'constant majority\' method, which assumes that the majority of reporters do not change in ratio. This method, which is implemented in *arrayCGHbase*, is generally applicable to most experiments as it is valid even in cases where up to 50% of reporters have altered ratios, it does not require prior knowledge of which features remain constant, and allows for intensity and spatial variation. Hence, this method calculates a scaling term from the median of all ratios, excluding all outliers. In this way the distribution of all ratios is transformed so that it centers around zero. Quality control --------------- ### Percentage of good quality spots This first quality assessment is a basic calculation of the number of reporters (or features) that are not flagged based on quality measures (user defined parameters and thresholds, see above). ### Intra- and inter-array hybridization quality Three other major quality parameters can be determined with *arrayCGHbase*for each experiment. The first assesses the variation between reporters present in replicates on the array (typically duplicates or triplicates). An increased variation typically reflects lower quality hybridizations resulting in less reliable ratios. A second quality parameter is the standard variation between the different reporters on the array that show a normal (unaltered) copy number. This quality measure is only applicable in experiments with few reporters with aberrant copy number. The third quality measure is the average ratio for reporters with aberrant copy number. This ratio should significantly differ from zero to allow identification of differentially hybridized reporters. This last quality measure is only applicable in experiments where DNA copy number aberrations are known or validated. These parameters provide an objective quality measure and can also be helpful to compare different experiments. In addition to these parameters, different graphical displays, such as ratio-intensity plots (usually referred to as MA plots), dual channel intensity scatter plots, and ratio histograms give an idea of the quality of an individual experiment or series of experiments (Figure [2](#F2){ref-type="fig"}). In all these visualizations, thresholds for gains and losses are displayed and can be adjusted. The slide viewer generates a virtual spatial view of all features on the array using the ratio, or signal and background intensities; this viewer allows the identification of problematic regions or artifacts on the slide surface. Clicking on an individual feature shows specific data associated with this feature (e.g., reporter name, signal intensities, and data quality flags). Scoring chromosomal regions with aberrant copy number ----------------------------------------------------- The final step in arrayCGH data processing is the identification of reporters that exhibit differential hybridization, corresponding to chromosomal regions that have altered copy number. The major issue is to identify those reporters whose relative ratios stand out from the experimental noise with sufficient statistical significance. *arrayCGHbase*currently incorporates two scoring methods. The most widely used approach is to define a ratio threshold and identify the probes that exhibit ratios greater or smaller than this threshold. Another, statistically more sound approach, is to use a floating threshold based on the standard deviation of all reporters in a given experiment. Reporters that exhibit ratios greater than this threshold will be defined as differential \[[@B14]\]. Both methods are implemented in *arrayCGHbase*and can be applied on each individual feature, or on the mean value of replicates. Besides the aberrant feature scoring methods, two other algorithms are available: a universal data smoothing algorithm, as well as a breakpoint-identification algorithm, which both consist of a moving window along the chromosomes and hence make use of the spatial \"along the chromosome\" distribution of the reporters. With these algorithms, chromosomal breakpoints can be easily identified in more noisy datasets. By writing custom plug-ins (in PHP or R), sophisticated algorithms that use segmentation methods (e.g. Cluster Along Chromosomes, CLAC \[[@B7]\]) or others, can be implemented by any user in a straightforward way. Chromosome visualization ------------------------ A wide variety of result viewers are available. The results can be mapped upon standard ISCN (International Standard on Cytogenetic Nomenclature) ideograms in an electronic karyotype, or visualized per chromosome or zoomed in on a region of interest (Figure [3](#F3){ref-type="fig"}). Moreover, various CGH profile views provide the user with a tool to compare different experiments and to identify regions with relevant copy number alterations. Views are returned to the user either as PNG (Portable Network Graphic) or as SVG (Scalable Vector Graphic) files, with the ability to scale images according to screen width. Data export ----------- Processed data can be exported as MIAME compliant text files and figures; these include the original feature signal and background intensities, the normalized ratio value, a list of reporters that are differentially hybridized, and the data quality parameters. Additionally, a file can be generated for submission of arrayCGH results directly into Progenetix \[[@B15]\], a comprehensive collection of published cytogenetic abnormalities in human neoplasms. Lastly, BED files can be created to map results and visualize the experiment from within the Ensembl or UCSC genome browser. ArrayCGHbase at work ==================== In several publications from our research group, *arrayCGHbase*has been successfully used to analyse arrayCGH data to identify and delineate copy number aberrations \[[@B16]-[@B19]\]). At the demo site, users can explore the data published in Hellemans et al. \[[@B16]\], a small \~5 Mb deletion in chromosome 12q identified using SNP chips), the results of a case report of the identification of an unbalanced X-autosome translocation by arrayCGH in a boy with a syndromic form of chondrodysplasia punctata brachytelephalangic type \[[@B17]\], a distal 9p trisomy and distal Xp nullisomy caused by an unbalanced X;9 translocation: 46, Y, der(X)t(X;9)(p22.32;p23) detected with a 1 Mb BAC array), and the copy number profile of a cancer cell line NGP.1A.TR \[[@B18]\]). It is possible to look at the raw data of these hybridizations and more importantly, test the performance of the program using different settings. Conclusion ========== We present *arrayCGHbase*, a versatile web based, platform independent data storage and analysis tool for processing microarray CGH data. Routines were implemented for feature flagging, data normalization, data quality assessment and the identification of chromosomal regions with aberrant copy number. A zoomable graphical interface allows immediate identification of altered genomic regions and the underlying gene content by several database links. A multitude of export functions allow the user to further process the results. The easy plug-in architecture makes it possible for each user to add custom algorithms for data analysis and visualization and share these with the user community. This webtool and database will enable investigators to interpret single experiments and compare large data sets efficiently throughout different array platforms and provides all of the essential features and links for further investigation of the genomic regions of interest. Future developments =================== *arrayCGHbase*will continually be updated to incorporate new processing methods that will be developed both within and outside our laboratory. Immediate plans include the addition of export and import functions to R \[[@B20]\] or Bioconductor \[[@B21]\] to be able to apply several available mathematical algorithms such as two-dimensional LOWESS normalization \[[@B22]\]. Immediate export functions to the DECIPHER web site \[[@B23]\] to link phenotypical data to actual experiments will also be included. The *arrayCGHbase*source code is freely available under a Creative Commons License, to encourage others to develop new analysis methods and utilities that will further improve its capabilities. Availability and requirements ============================= An *arrayCGHbase*demo site is available at <http://medgen.ugent.be/arrayCGHbase/>. At this site, all quality control features and other features can be tested for several experiments with BAC arrays as well as SNP chips (see \'*arrayCGHbase*at work\'). At the same site, the complete package can be freely downloaded for local installation on a private hosted web server. For local use, additional software is required such as the MySQL database \[[@B24]\], a web server (e.g. Apache \[[@B25]\]), and PHP hypertext preprocessor \[[@B26]\]. These software packages are freely available and are key parts of LAMP (Linux, Apache, MySQL, PHP), an open source web platform. Enquiries for *arrayCGHbase*should be made to <arrayCGHbase@medgen.ugent.be>. Glossary ======== Reporter: any DNA fragment (BAC, PAC, cosmid, fosmid, cDNA clone, oligonucleotide, genomic PCR product) used for hybridization Feature: physical reporter spotted, printed, or otherwise linked to a substrate at a specific location PHP: Hypertext PreProcessor (server-side scripting language) MIAME: Minimal Information About a Microarray Experiment MySQL: My Structured Query Language ISCN: International System for human Cytogenetic Nomenclature BED: Browser Extendable Data Authors\' contributions ======================= BM was the principle programmer of arrayCGHbase. FP, KDP, PR and SVV contributed ideas for different features and display requirements. JV oversaw the project; all other authors have reviewed the manuscript and FS and JV were the final editors of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Comparison between different already available arrayCGH software programs and arrayCGHbase for the analysis and visualization of arrayCGH data. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ Jo Vandesompele and Katleen De Preter are supported by a grant from the Flemish Institute for the Promotion of Innovation by Science and Technology in Flanders (IWT). Filip Pattyn is a Research Assistant of the Research Foundation -- Flanders (FWO -- Vlaanderen). This study is supported by GOA-grant 12051203, FWO-grant G.0185.04, G.0200.03 and G.0106.05 and VEO project 011V1302, research grant of Kinderkankerfonds vzw (a non-profit childhood cancer foundation under Belgian law). This text presents research results of the Belgian program of Interuniversity Poles of attraction initiated by the Belgian State, Prime Minister\'s Office, Science Policy Programming (IUAP). Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### ***arrayCGHbase*scheme**. The data is stored in a MIAME supportive MySQL database (red) and reporter info is updated using the NCBI, Ensembl and UCSC genome database. Data and results are presented to the user through a web browser via the PHP scripting language. Data-normalization and other analysis or result visualization methods can be integrated using the plug-in architecture. Further data processing using the R statistical scripting language will be implemented in the near future. Results can be exported to a Progenetix or MIAME compatible format, or visualized on the genome using the Ensemble or UCSC genome browser. ::: ![](1471-2105-6-124-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Quality control graphs**. Graphical displays to assess the quality of an experiment, such as a dual channel intensity scatter plot, ratio histogram, ratio-intensity plot and a virtual slide view. ::: ![](1471-2105-6-124-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Selected result viewers**. Graphical displays of *arrayCGH*results of neuroblastoma cell line NGP.1A.TR1: line view (all reporters ordered by chromosome and chromosomal position on one line), karyo view (al reporters mapped on their chromosomal position on a standard ISCN ideogram), chromosome view (zoom on one chromosome or chromosomal region) with breakpoint identification algorithm, and genome browser view (neuroblastoma cell line IMR32), with all reporters and their copy number status displayed in the UCSC genome browser. ::: ![](1471-2105-6-124-3) :::

PubMed Central

2024-06-05T03:55:58.310909

2005-5-23

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173083/", "journal": "BMC Bioinformatics. 2005 May 23; 6:124", "authors": [ { "first": "Björn", "last": "Menten" }, { "first": "Filip", "last": "Pattyn" }, { "first": "Katleen", "last": "De Preter" }, { "first": "Piet", "last": "Robbrecht" }, { "first": "Evi", "last": "Michels" }, { "first": "Karen", "last": "Buysse" }, { "first": "Geert", "last": "Mortier" }, { "first": "Anne", "last": "De Paepe" }, { "first": "Steven", "last": "van Vooren" }, { "first": "Joris", "last": "Vermeesch" }, { "first": "Yves", "last": "Moreau" }, { "first": "Bart", "last": "De Moor" }, { "first": "Stefan", "last": "Vermeulen" }, { "first": "Frank", "last": "Speleman" }, { "first": "Jo", "last": "Vandesompele" } ] }

PMC1173084

Background ========== Eukaryotic gene transcription is largely known to be orchestrated by protein factors like activators, co-activators and co-repressors \[[@B1]\]. However, nucleosomal organisation, non-passive structural scaffolds and global structure of chromatin are increasingly being recognised as major players in the regulation of gene expression. The ability of sequences to position nucleosomes and to be anchored to the nuclear matrix to provide a spatial context for regulation of expression are measurable parameters that may influence the interactions with transcription machinery \[[@B2],[@B3]\]. This level of regulation may be distinctly different for genes whose expression is constitutive in comparison to genes that exhibit tissue specific expression. The latter would demand an open chromatin configuration in certain tissues and repressive organisation in others. In this study, we examined whether the potential to utilise global regulatory mechanisms to control gene expression through chromatin organisation varies between housekeeping and tissue specific genes (Hkg and Tsg respectively) by virtue of their organisation. An *in-silico*comparison of chromatin related organisational differences in the 5\' and 3\' regulatory regions of housekeeping and tissue specific genes was carried out to shed light in this direction. Results and discussion ====================== Chromatin landscape of a region plays a major role in determining and modulating the expression status of its neighbouring genes \[[@B4]\]. The role played by chromatin in the 5\' regulatory regions of genes in transcriptional regulation has been extensively studied \[[@B5],[@B6]\]. In the present study, we have taken 2 distinct sets of genes differing predominantly in their spatial expression aspect, namely, housekeeping and tissue specific, to understand the various attributes of the regulatory role played by chromatin organisation in the 5\' region. Analysis of scaffold/matrix associated sequences ------------------------------------------------ Scaffold/matrix attachment regions (S/MARs) are defined as sequences, which can attach themselves to the nuclear matrix and hence help in the formation of independent chromatin loops \[[@B7]\]. Transcriptional regulation of gene expression is known to involve formation of dynamic chromatin loops mediated by S/MAR attachment to the nuclear matrix \[[@B3]\]. The attachment of a DNA sequence to the matrix will place the neighbouring genes in proximity of the transcription factors. The abundance of S/MARs in the 5\' *cis*-regulatory regions of genes further demonstrates their role in transcriptional regulation \[[@B8]\]. We have analysed the predicted S/MAR sites in the 5\' and 3\' flanking regions of human Hkg and Tsg (Table [1](#T1){ref-type="table"}). We used MAR Finder (new version) and ChrClass programs for predicting S/MAR binding sites in the sequences (Table [1](#T1){ref-type="table"}). Glazko *et al*have classified 5\' flanking regions up to 1500 bp of human tissue specific genes as an out-group, assuming that these regions have no significant association with S/MAR binding \[[@B7]\]. On the contrary, our study reveals that S/MAR binding sequences are enriched in 5\' regulatory regions of Tsg in comparison to the Hkg. The common predictions of both the programs were taken for the analysis. This data indicates a significant enrichment of S/MAR binding sequences in the 5\' flanking regions of Tsg and depletion of S/MARs in the 3\' Hkg regions as compared to Tsg. Chi-square test was applied for both 5\' and 3\' region S/MAR predictions of Hkg and Tsg, to ascertain whether the distributions are significantly different. The chi-square value of 11.37 (df = 1) and *P*-value ≤ 0.001 obtained for the distribution of S/MARs in 5\' regions of Hkg and Tsg indicate a significant difference in the distribution of S/MAR elements between the two sets. Similarly, for the distribution of S/MARs in 3\' regions of Hkg and Tsg the chi-square value of 5.033 (df = 1) and *P*-value of ≤ 0.025 show that the Hkg 3\' regions are significantly depleted of S/MARs as compared to Tsg. The observation that the 5\' regulatory regions of Hkg are less enriched in S/MARs in comparison with Tsg might be related to the distribution of housekeeping genes in the genome. Housekeeping genes cluster in chromosomes and therefore, they often would be present in distinct chromatin domains along with housekeeping genes that have a co-ordinated expression \[[@B9],[@B10]\]. The data showing preferential absence of S/MARs in the 3\' regions in Hkg further lend support to this hypothesis. On the other hand, tissue specific genes are known to be dispersed in gene dense as well as heterochromatic regions \[[@B9],[@B11]\]. It may be necessary for them to shield themselves against the effects of positive and negative *cis*-acting elements of adjacent regions in order to maintain tissue specific expression profile. In this context, the boundary elements or the insulator model has been proposed earlier \[[@B11]\]. S/MARs function as boundary elements and their co-localisation with insulators such as the Drosophila gypsy element is also reported \[[@B12],[@B13]\]. They also function as boundary elements in *in vitro*systems by shielding away the position effect \[[@B14]\]. Some earlier reports have suggested a role for S/MARs in maintaining tissue specific gene expression \[[@B15]\]. More recently, the 5\'-HS4 chicken-globin insulator is known to have a CTCF protein binding dependent matrix association \[[@B16]\]. Hence, the over representation of S/MARs seen in Tsg set might possibly be associated with a boundary element function. Our results on the prediction performance of the programs have been quite different from the previous reports \[[@B7]\]. We find that MAR Finder (an under predictor) predicts more number of S/MAR regions in our dataset in comparison to ChrClass program (an over predictor) \[[@B7]\]. This may be attributed to the use of the advanced version of MAR Finder in our study wherein, new parameters/features have been added in the form of the \"New MAR Rules\" option. Analysis of nucleosomal organisation ------------------------------------ The primary template for local and global changes in the chromatin structure of a chromosome is the nucleosomal unit \[[@B4]\]. Chromatin structure and nucleosomal organisation over the promoter regions play a major role in regulation of expression of downstream gene(s) \[[@B6],[@B17]\]. The nucleosome distribution would depend upon the occurrence of nucleosome destabilising elements as well as nucleosome forming sequences. We have analysed both these parameters in our study. Nucleosome destabilising elements --------------------------------- Nucleosome destabilising/excluding elements such as poly (dA.dT) and (CCGNN)~n~in promoter regions have been implicated in maintaining constitutive gene expression \[[@B18]-[@B21]\]. At the functional level, it is known that poly (dA.dT) elements increase the accessibility of promoters of HIS3, URA3 and Ilv1 in yeast to the cognate transcription factor \[[@B18]\]. With the increasing length of poly (dA.dT) repeat, the availability of the sequences to transcription factors improves and similarly, with increasing lengths, the propensity to exclude nucleosomes increases for (CCGNN)~n~sequence motif as demonstrated in yeast and mammalian systems \[[@B19]-[@B21]\]. It has been demonstrated that (CCGNN)~n~sequences promote meiotic recombination and activated HIS4 expression by generating open chromatin \[[@B22]\]. We hypothesised that the differential distribution of nucleosome exclusion elements might be one of the mechanisms involved in maintaining distinct nucleosomal organisation of the housekeeping and tissue specific genes. The frequency of pure poly (dA.dT) stretches \>10 bp and (CCGNN)~2--5~in the 2000 bp 5\' *cis*-regulatory regions of human Hkg and Tsg(s) were analysed. A significant enrichment of poly (dA.dT) elements in the upstream regions of Hkg is seen in comparison to Tsg (Table [2](#T2){ref-type="table"}). The *t*-test for the difference in distribution of poly (dA.dT) stretches (\>10 bp) between Hkg and Tsg show significant *P*-values in the different lengths of the stretches examined. In Hkg, 670 repeats of (CCGNN)~2--5~were detected as against 430 in Tsg. (CCGNN)~2~was the most prevalent repeat unit and uninterrupted repeat units (\>5 mers) were not found in the sequence sets. Although shorter repeat units (2--5 mers) have not been studied for nucleosome exclusion, they might play a role in destabilising the histone octamer \[[@B20]\]. Further, many of them form a part of longer interrupted stretches. The *t*-test for difference in distribution of (CCGNN)~2--5~between Hkg and Tsg shows a significant *P*-value of 1.71E-06. Nucleosome formation potential scores and expression level of genes ------------------------------------------------------------------- Using Recon, Levitsky *et al*(2001) have examined the nucleosome formation potential of 3 classes of human genes namely, Hkg, Tsg and widely expressed genes that differ in their spatial expression status \[[@B2]\]. Their report, based on a small sample size of around 200 genes shows the difference in the nucleosome formation potential between these 3 classes of genes in the upstream 50 bp from the transcription start site. In this study, we examined the nucleosome formation potential values in upstream 2000 bp of 5\' regions of Hkg and Tsg and their correlation with gene expression levels with the complete set of 1083 genes. The Tsg and Hkg sequences show a considerable difference in their nucleosome formation potential scores over an extended upstream region of 2000 bp (Figures [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}). The Tsg region is enriched in nucleosome formation potential scores (peak at 1) in all upstream positions analysed (till 2000 bp). For Hkg, the distribution seems to be shifted towards the negative scores at 400 bp region and this shift diminishes gradually as we move further upstream to finally peak at 1 in 2000 bp upstream region (Figure [1](#F1){ref-type="fig"}). *t*-test was applied to ascertain the difference in distribution of Recon scores between Hkg and Tsg (Table [3](#T3){ref-type="table"}). The resultant *P*-values in various intervals of relevance (0.8 to 1, 1 to 1.2, -0.8 to -1 and -1 to -1.2) reflect that the scores in the upstream 400 bp from the gene start site show the maximum difference in all the intervals and at 2000 bp, the difference gradually fades away in intervals 0.8 to 1 and 1 to 1.2 (Table [3](#T3){ref-type="table"}). A correlation analysis between nucleosome formation potential and expression levels was carried out considering the Recon scores at upstream 400bp region, where the *P*-values reflect the largest difference and the log~10~values of expression levels were taken as inputs \[\"see [Additional file 1](#S1){ref-type="supplementary-material"}\"\]. Initially, we analysed the gross dependence of total expression levels on nucleosome potential in the upstream regions of the two sets of genes (Table [4](#T4){ref-type="table"}). In all the four intervals, no correlation is seen, indicating that chromatin plays an insignificant role in global modulation of levels of expression in these two sets of genes. These results are similar to that observed in case of *Saccharomyces cerevisae*whole genome analysis (unpublished results). Further, we refined the analysis to examine the correlation, if any, between nucleosome formation potential in upstream regions and extreme expression levels of genes. The Hkg and Tsg groups were further categorised separately into high and low expression level groups as described under \"Methods\" section and their correlation with the nucleosome formation potential was analysed (Table [5](#T5){ref-type="table"}). The high and low expression genes of Hkg show a low negative correlation with scores in intervals 0.8 to 1.0 and 1 to 1.2 and a low positive correlation with scores in intervals -1.2 to -1 and -1.0 to -0.8. In Tsg, except in one interval, there was no valid correlation seen. This solitary value was not considered since the correlation coefficients in other intervals didn\'t reflect this trend. Our data restates that chromatin in 5\' region plays a major role in determining the ubiquitous or restricted tissue expression of a gene as shown by Levitsky *et al*(2001) \[[@B2]\]. The abundance of nucleosome exclusion elements in Hkg 5\' regions and the low Recon scores reflect their poor preference for nucleosome assembly. The expression analysis suggests that although chromatin plays a role in bringing about extreme variations of gene expression levels in certain classes of genes such as the housekeeping genes, the relation is not linearly correlated with the total, wider range of expression levels. It is possible that nucleosomes might be involved in fine-tuning of expression levels that may escape our attention, since the difference in the range of expression considered is fairly large. The difference detected in nucleosome formation potential between the two sets might reflect the accessibility to basal transcription factors for Hkg and gene/tissue specific transcription factors for Tsg, considering the difference in spatial and temporal expression patterns of the two groups. Analysis of repetitive sequences -------------------------------- Repetitive sequences are implicated in chromatin organisation and heterochromatinisation \[[@B23]-[@B25]\]. They are differentially enriched in various functional categories of genes and are predicted to play an important role in gene regulation \[[@B24],[@B26]\]. We analysed the distribution of various repeat classes in the 5\' regions of Hkg and Tsg using RepeatMasker software. The total repeat content in Hkg regions is seen to be more than in Tsg regions. As reported earlier, our data shows enrichment of SINES (Alu) in comparison to other classes of repetitive sequences in both the sets \[[@B24]\]. Further, the 5\' sequences of Hkg are more enriched in Alu sequences in comparison to those of Tsg regions (Table [6](#T6){ref-type="table"}). The difference in the distribution of Alu repeats in the two classes of sequences was determined by applying *t*-test for the number of repeats and the repeat content in terms of length in base pairs in each sequence set (Table [7](#T7){ref-type="table"}). The low total repeat content seen in Tsg upstream regions lends support to the hypothesis that condensed chromatin disfavours transposable element insertions in comparison to open chromatin (Hkg promoters)\[[@B27]\]. Genes with high expression levels are clustered in genomic regions known as ridges. These gene rich regions also have high (G+C) content, SINES and genes with short introns \[[@B9]\]. Eisenberg and Levanon \[[@B28]\] have reported the presence of significantly shorter introns and an overall compact gene structure in Hkg as compared to non-Hkg \[[@B28]\]. We have used the gene list provided by Eisenberg and Levanon \[[@B28]\] for our analysis. The enrichment of SINES in the 5\' regions of Hkg suggests that Hkg might be localised in the ridge regions of the genome. More recently, it has been suggested that the contrasting attributes of gene compactness, GC content and the length of the intronic and intergenic sequences in Hkg and Tsg might be involved in chromatin mediated regulation for maintaining distinct expression patterns in the gene sets \[[@B29]\]. Recently, Alu elements have been shown to house transcription factor binding sites and the presence of such regulatory elements might influence the chromatin structure and gene expression \[[@B30]\]. The paradigm for regulation of gene expression in human tissues has shifted the focus from involvement of a battery of transcription regulators to global regulatory mechanisms \[[@B31]\]. These mechanisms have also gained significance in the context of the low estimates of gene numbers in the human genome \[[@B32]\]. It is in this framework that we have analysed the chromatin characteristics of two groups of genes, one that needs almost a continuous and ubiquitous expression and another demanding tissue specific regulation. It had been predicted that the nucleosomal density in a chromatin domain and the buffering of supercoiling waves by repetitive DNA will play a major role in establishing coordinated gene regulation in a domain in the context of the relevance of maintenance of repetitive sequences during evolution \[\[[@B25],[@B33]\], and \[[@B34]\]\]. A recent report also infers the role of chromatin-mediated mechanisms in the differential gene expression patterns seen in housekeeping and tissue specific genes \[[@B29]\]. Our data and analyses lend support to these hypotheses (Figure [3](#F3){ref-type="fig"}). Another recent report, which addresses the chromatin architecture of the human genome, provides experimental evidence that open chromatin correlates with high gene density regions but not with gene expression \[[@B35]\]. This data further supports our *in-silico*observations and strengthens the domain concept for concerted expression of clustered genes. The role of nucleosome formation potential is apparent from the present analysis in both the housekeeping genes as well as tissue specific genes but with an opposing correlation. Housekeeping genes apparently discourage nucleosome formation to match their expression profile in space and time by ensuring accessibility to transcription machinery. In addition, they also show a significant enrichment in poly (dA.dT) stretches, which are known to destabilise nucleosomes. On the other hand, the tissue specific genes show higher scores for nucleosome formation potential through which they perhaps provide selective accessibility to the transcriptional machinery. Further, our analysis suggests that tissue specific genes resort to additional global regulatory features such as matrix association, which would facilitate maintenance of functionally distinct domains to insulate themselves from both silencing and activating regulatory influence of adjacent domains. The differential distribution of repetitive sequences in housekeeping and tissue specific genes might also play an important role in maintaining distinct chromatin landscape over these regions. Conclusion ========== We have demonstrated that the regulatory regions of housekeeping and tissue specific genes have differential chromatin architecture with respect to S/MAR binding, nucleosome positioning potential and repetitive sequences. This has potential implications for regulation of gene expression in eukaryotic genomes. Methods ======= In this study, the 5\' and 3\' flanking regions of genes were analysed for various attributes of chromatin organisation. The list of human housekeeping genes (Hkg) was retrieved from <http://www.compugen.co.il/supp_info/Housekeeping_genes.html>\[[@B28],[@B36]\]. 532 genes have been categorised as housekeeping because of their ubiquitous and high expression levels in 47 tissues. The list and expression levels of the human tissue specific genes were obtained from Eli Eisenberg (personal communication). 566 genes expressed in only a single tissue were taken as tissue specific genes (Tsg) and analysed. We could unambiguously retrieve sequences of 525 Hkg and 558 Tsg from human genome build 33 (NCBI). Approximately, 2000 bp of the 5\' and 3\' regions from each of these genes were taken for analysis. Scaffold/matrix associated regions (S/MAR) analysis --------------------------------------------------- MAR Finder was used for prediction of S/MAR regions \[[@B37],[@B38]\]. All the default options and the \"New MAR Rules\" were selected for predicting S/MARs. ChrClass program was used for S/MAR prediction \[[@B39],[@B40]\]. Nucleosome organisation and gene expression correlation analysis ---------------------------------------------------------------- The upstream regions (2000 bp) were scanned for nucleosome exclusion elements \[[@B18],[@B20]\] -- poly (dA.dT) pure stretches of \>10 bp length and \[5\' (CCGNN) 3\'\]~2--5~using in-house programs. Recon was used for evaluating nucleosome formation potential in the sequences \[[@B2],[@B41]\]. The score outputs of the 5\' regions were categorised in frequency intervals of 0.2 with a range from -3.2 to +3.2. The Recon scores around +1 and -1 imply strong nucleosome formation and exclusion potentials respectively. The scores in the four intervals of relevance (0.8 to 1, 1 to 1.2, -0.8 to -1 and -1 to -1.2) were taken for all the analyses. Since the promoter region information was not retrieved for these genes, the 2000 bp upstream region from the gene start site was split into 400, 800, 1200 & 1600 bp and analysed. The Recon scores at 400 bp were used to draw correlation between the nucleosome formation potential and expression levels in the two sets of genes. In each sequence set, genes with expression levels \<500 and \>5000 affymetrix expression units were classified as low and high expression genes respectively. We considered a minimum ten fold difference in the expression levels of genes as a relevant criterion for classifying them as high and low expression genes. In Hkg, this criterion yielded 33 low expression and 35 high expression genes. In Tsg, we categorised 416 low expression genes and 24 high expression genes. Repetitive sequence analysis ---------------------------- RepeatMasker version: 20040306-web was used to calculate the repeat content in 2000 bp upstream sequences of the two groups of genes \[[@B42]\]. List of abbreviations ===================== S/MAR: scaffold/matrix attachment regions Hkg: housekeeping genes Tsg: tissue specific genes df: degree of freedom Authors\' contributions ======================= MG, VB and SKB contributed to the study design, analyses and in drafting the manuscript. MG, PS, SKDS, DD, KK and GP were involved in the data retrieval and analyses. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 gene list and expression levels of housekeeping genes, gene list and expression levels of tissue specific genes. \'supplementaryfile1.xls\' contains the list of housekeeping and tissue specific genes and their expression levels used for the analysis. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ MG acknowledges the financial support provided by Council for Scientific and Industrial Research (CSIR), India. SKB thanks Council for Scientific and Industrial Research, India and VB thanks Indian Council for Medical Research (ICMR), for financial assistance through a grant. The authors wish to acknowledge Dr. Beena Pillai, Dr. Rakesh Sharma, Dr. Neeraj Pandey and Dr. Mitali Mukerji for their valuable discussions and suggestions. We would also like to acknowledge Samira for careful checking and helping with the manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Nucleosome formation potential score distributions for 5\' regions of housekeeping and tissue specific genes**. The 5\' sequences of human housekeeping and tissue specific genes were analysed by Recon for distribution of nucleosome formation potential scores. Frequency distribution histograms were plotted for scores in various intervals (range -3.2 to +3.2). (A) and (B) show the distribution of nucleosome formation potential scores at 400 and 2000 bp upstream from the gene start site respectively. Nucleosomal density is significantly lower for housekeeping genes as compared to tissue specific ones, in regions close to the gene start site. ::: ![](1471-2105-6-126-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Nucleosome formation potential score distributions for 5\' regions at different positions from the gene start site in housekeeping and tissue specific genes**. The 5\' regions of 800, 1200 and 1600 bp from the gene start site of housekeeping and tissue specific genes were taken for the analysis. Frequency distribution histograms were plotted for Recon scores in various intervals (range -3.2 to +3.2). (A), (B) and (C) show the distribution of nucleosome formation potential scores at 800, 1200 and 1600 bp upstream from the gene start site respectively. As we move upstream from the gene start site, the difference in the nucleosome formation potentials between housekeeping and tissue specific genes gradually fades away. ::: ![](1471-2105-6-126-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **A model for chromatin landscape in 5\' regions of tissue specific and housekeeping genes**. (A) depicts the repressive role of chromatin in maintaining tissue specific gene expression profiles in a chromosome. The chromatin organisation in the 5\' regions of Tsg1 and Tsg2, two different tissue specific genes dispersed in the chromosome is shown. Nucleosome formation potentials and S/MARs -- the boundary elements, are enriched in their upstream regions and might play a major role in facilitating tissue specific expression. This is likely to be a local effect since neighbouring genes might have a different expression pattern. (B) depicts the chromatin organisation in the 5\' regions of Hkg1, Hkg2 and Hkg3, three housekeeping genes clustered in the chromosome. The presence of low nucleosome formation potential regions and enrichment of nucleosome destabilising elements ensure an open chromatin configuration in this domain. As Hkg generally cluster together, they are depleted in S/MARs relative to tissue specific genes as shown in the present analysis by the significant absence of predicted S/MARs in both 5\' and 3\' regions of housekeeping genes as compared to tissue specific genes. ::: ![](1471-2105-6-126-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Distribution of putative S/MARs in housekeeping and tissue specific genes. ::: **Putative S/MARs in 5\' regions (%)\*** **Putative S/MARs in 3\' regions (%)\*** ------------------- ------------------------------------------ ------------------------------------------ ------ ------ presence of S/MAR 26.1 34.1 19.1 20.6 absence of S/MAR 26.1 19.2 34.5 25.1 ^\#^Housekeeping genes, ^§^Tissue specific genes, \* ChrClass and MAR Finder programs were used for prediction of S/MARs in housekeeping and tissue specific genes regulatory regions (5\' & 3\' regions -- 2000 bp each). The common predictions of both the programs were used for the analysis. The data is represented as percentage of genes with predicted S/MARs in 5\' and 3\' regions of 525 housekeeping and 532 tissue specific genes. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Distribution of poly (dA.dT) repeats of various lengths in the 5\' upstream regions of housekeeping and tissue specific genes. ::: **Poly (dA.dT) stretch (bp)** **No. of repeat stretches in the two classes** **No. of genes with repeats in 5\' region (%)** ------------------------------- ------------------------------------------------ ------------------------------------------------- ------------ ------------ ---------- \>10 443 345 268 (51.0) 240 (43.0) 1.31E-04 \>11 381 297 243 (46.3) 214 (38.4) 3.25E-04 \>12 339 248 226 (43.1) 184 (33.0) 6.10E-05 \>13 295 207 209 (39.8) 156 (28.0) 4.29E-05 \>14 251 168 188 (35.8) 128 (22.9) 2.77E-05 \>15 209 140 164 (31.2) 111 (19.9) 8.83E-05 \>16 180 116 146 (27.8) 99 (17.7) 7.58E-05 \>17 155 103 134 (25.5) 88 (15.8) 2.42E-04 \>18 138 79 120 (22.9) 71 (12.7) 2.23E-05 \>19 112 66 101 (19.2) 59 (10.6) 2.61E-04 \>20 100 58 92 (17.5) 53 (9.5) 5.32E-04 ^\#^Housekeeping genes, \*Tissue specific genes. A total of 525 housekeeping and 558 tissue specific genes were analysed. The numbers in parentheses (4^th^& 5^th^columns) represent the percentage of genes containing the repeat stretch. ^§^Difference in the distribution of poly (dA.dT) stretches in Hkg and Tsg analysed by applying *t*-test (for normalizing the difference in sample size). The repeat lengths from \>12 to \>18 bp are showing very significantly different distributions between Hkg and Tsg. The distributions were examined in 2000 bp upstream region from the gene start site. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### *t*-test *P*-values for the difference in the distribution of nucleosome formation potential scores between housekeeping and tissue specific genes. ::: **Length (bp)\*** ***P*-value in intervals of scores** ------------------- -------------------------------------- ---------- ---------- ---------- 400 3.53E-13 8.73E-17 1.28E-17 4.45E-23 800 1.16E-24 6.27E-24 6.64E-12 6.65E-22 1200 6.91E-26 1.44E-24 2.10E-09 1.64E-18 1600 1.72E-24 6.99E-24 2.72E-07 5.63E-15 2000 2.55E-25 1.84E-25 3.22E-05 6.71E-13 \*denotes the length of 5\' upstream region from the gene start site taken for the analysis. The scores were compared in the four Recon score intervals of relevance -1.2 to -1, -1 to -0.8, 0.8 to 1 and 1 to 1.2. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Correlation coefficients of total expression levels (log~10~) with nucleosome formation potential scores in housekeeping (Hkg) and tissue specific genes (Tsg). ::: **Category** **Correlation coefficient** -------------- ----------------------------- ------- ------ ------- Hkg^\#^ 0.10 0.14 0.04 -0.01 Tsg\* -0.10 -0.12 0.15 0.17 ^\#^Housekeeping genes, \*Tissue specific genes. The correlation was drawn in the four Recon score intervals of relevance -1.2 to -1, -1 to -0.8, 0.8 to 1 and 1 to 1.2. ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Comparison of the level of correlation between nucleosome formation potential scores and contrasting expression levels of genes. ::: **\*Category** **Correlation coefficient** ---------------- ----------------------------- ------- ------- ------- Hkg ↑↑ 0.17 0.30 -0.10 -0.26 Hkg ↓↓ 0.36 0.35 -0.28 -0.39 Tsg ↑↑ 0.03 0.02 0.26 0.11 Tsg ↓↓ -0.12 -0.14 0.15 0.16 \*High and low expression level genes were categorised in Hkg (housekeeping genes) and Tsg (tissue specific genes) groups separately. ^\#^Recon score intervals. The genes classified as high and low expression genes in both Hkg and Tsg had atleast a 10-fold difference in their expression levels. The up (↑↑) and down (↓↓) arrows denote high expression and low expression respectively. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### The distribution of Alu repeats in 5\' upstream regions of housekeeping (Hkg) and tissue specific genes (Tsg) is represented in terms of the number of copies and basepairs covered by Alu repeats. ::: **Repeat category** **No. of copies** **% of the total sequences covered by the repeat** --------------------- ------------------- ---------------------------------------------------- ------ ------ Alu 866 575 20.1 12.3 ^\#^Housekeeping genes, \*Tissue specific genes. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### *t*-test *P*-values for the difference in the distribution of Alu repeats in 5\' upstream regions of housekeeping and tissue specific genes. ::: **Repeat Category** ***P*-value** --------------------- --------------- ---------- Alu 3.02E-08 7.63E-11 :::

PubMed Central

2024-06-05T03:55:58.312808

2005-5-26

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173084/", "journal": "BMC Bioinformatics. 2005 May 26; 6:126", "authors": [ { "first": "Mythily", "last": "Ganapathi" }, { "first": "Pragya", "last": "Srivastava" }, { "first": "Sushanta Kumar Das", "last": "Sutar" }, { "first": "Kaushal", "last": "Kumar" }, { "first": "Dipayan", "last": "Dasgupta" }, { "first": "Gajinder", "last": "Pal Singh" }, { "first": "Vani", "last": "Brahmachari" }, { "first": "Samir K", "last": "Brahmachari" } ] }

PMC1173085

Background ========== The introduction of DNA microarray technology has enabled investigators to screen thousands of genes simultaneously. One of the main goals of these studies is to identify differentially expressed genes between two biological conditions. For example, many studies \[[@B1]-[@B4]\] have been performed in prostate cancer research to find candidate markers. Since laboratory protocols, microarray platforms and analysis techniques used in these studies were not identical, it is difficult to make a comparison among the results obtained from them. However, systematic integration of gene expression data from different sources increases statistical power of detecting differentially expressed genes and allows assessment of heterogeneity. Meta-analysis is a classical statistical methodology for combining results from different studies addressing the same scientific questions, and it is becoming particularly popular in the area of medical and epidemiological research \[[@B5]\]. Meta-analysis methods have recently been applied to the analysis of microarray data \[[@B6]-[@B11]\]. Rhodes et al. \[[@B6]\] focused on combining p-values for each gene from the individual studies to estimate an overall p-value for each gene across all studies. Their method has been applied to four prostate cancer microarray datasets, two of which are cDNA microarray data and the remainder Affymetrix microarray data. Samples in each data set were taken from prostate cancer cases, but were analyzed with different platforms. Differential expression was assessed independently for each gene in each dataset. Since the method chosen to combine results across studies was based on the statistical confidence measure, the p-value, not on the expression level, this strategy avoids direct comparisons of data sets and related cross-platform normalization issues. Choi et al. \[[@B7]\] focused on integrating effect size estimates to obtain an overall estimate of the average effect size. Effect size is used to measure the magnitude of treatment effect in a given study. Using the same datasets as those used by Rhodes et al. \[[@B6]\], they demonstrated that their method can lead to the discovery of small but consistent expression changes with increased sensitivity and reliability. Parmigiani et al. \[[@B9]\] developed a correlation-based method for assessing reproducibility of gene expression studies with application to lung cancer. They demonstrated that their method can improve correlation across the various studies. Jiang et al. \[[@B10]\] used a distribution transformation method to integrate two lung cancer studies and proposed a gene shaving-based classification approach to identify a small list of differentially expressed genes between lung cancer and normal patients. They noted that many of the selected genes have been experimentally validated. Although some of the above studies (for example Rhodes et al. \[[@B6]\], Choi et al. \[[@B7]\], Parmigiani et al. \[[@B9]\]) demonstrated the utility of integrating cDNA and Affymetrix microarray data, other investigators argued against this approach. Kuo et al. \[[@B12]\] compared Affymetrix and spotted cDNA gene expression measurements based on 60 cell lines from the National Cancer Institute. They found low correlation between the actual gene measurements from the two technologies, and concluded that \"data from spotted cDNA microarrays could not be directly combined with data from synthesized oligonucleotide arrays.\" Moreover, they concluded that it was unlikely that the two types of data could be transformed or normalized into a common standardized index. Jarvinen et al. \[[@B13]\] determined the level of concordance between microarray platforms by analyzing breast cancer cell lines with in situ synthesized oligonucleotide arrays, commercial cDNA microarrays and custom-made cDNA microarrays. Their results demonstrated that data from different microarray platforms are variable to the extent that direct integration of data from different platforms may be complicated and unreliable. In classical meta-analysis, quality measures have often been used when combining results across studies. It has been argued that studies of a higher quality will give more accurate estimates of the true parameter of interest, and therefore studies of high quality should receive a higher weight in the analysis summarizing across studies \[[@B14]\]. The Affymetrix microarray technology has been used worldwide. Its success has been demonstrated by numerous publications in scientific journals. However, it is well-known that only part (approximately 40--50%) of the whole genome is expressed in any specific tissue type, so there are many genes showing low expression and random variability across samples. Furthermore, some genes will be measured less precisely by some technologies, or studies, than others. Therefore, our ability to develop powerful statistical methods for efficiently integrating and weighting information from related genomic experiments will be critical in the success of the massive investment made on genomic studies. The focus of this paper is to design and implement a quality measure appropriate for Affymetrix microarray data. Using our quality measure, we weighted the importance of each gene in each experiment and incorporated our quality measure into the effect size model proposed by Choi et al. \[[@B7]\] to model inter-study variation of gene expression profiles. We believe that applying this approach can lead to a more accurate description of expression patterns than Choi et al\'s method \[[@B7]\]. Results ======= We used two data sets consisting of gene expression profiles in lung cancer and normal subjects. These datasets were collected using different chip types of the Affymetrix oligonucleotide microarrays and were conducted by two research groups, one from Harvard and the other from Michigan (see Methods section for details). A list of 6124 common probe sets found in the two datasets was used for data analysis in this study \[[@B10]\]. We developed a quality weight for each gene in each study by modeling the log of the detection p-values with an exponential distribution, and then summarizing across arrays and groups within each study (see Methods). In order to visualize the effect of the proposed quality weighting, we calculated the mean expression value of each probe set across all normal samples, where the expression variation presumably is less heterogeneous than among the cancer samples. The mean expression value of a probe set in a study is the estimated average of the probe set\'s intensity values across all normal samples in the study. Figure [1](#F1){ref-type="fig"} shows the scatter plot of the average expression values of the probe sets in the Harvard dataset plotted against that of the Michigan dataset: (a) weighted by the quality score (see Methods section for our definition of the quality score), and (b) un-weighted. This plot is intended to be illustrative only -- our algorithm weights the test statistics, rather than the gene expression measures. In Figure [1(a)](#F1){ref-type="fig"}, it can be seen, as expected, that many of the genes with low levels of expression are associated with low quality weights. We show examples of quality scores for selected probe sets in Table [1](#T1){ref-type="table"}. The two datasets may have very different detection p-value distributions, which are reflected in the quality scores. Figure [2](#F2){ref-type="fig"} shows a box plot of the detection p-values for one probe set from Table [1](#T1){ref-type="table"}. When the two datasets give small p-values (e.g., last line in Table [1](#T1){ref-type="table"}), the minimum p-value may be much smaller in one dataset than another. Both, however, will give high quality scores with an appropriate choice of the sensitivity parameter *s*that adjusts how the quality measure interprets the detection p-values. Figure [3](#F3){ref-type="fig"} shows the adjusted and unadjusted quantile -- quantile (Q-Q) plots of the observed vs. expected Q values. Q is the test statistic we used for assessing heterogeneity, and is described in detail later in the Methods section. In the adjusted Q-Q plot, the quality score was used as a weight in the computation of Q while it was not considered in the unadjusted Q-Q plot. From these graphs, we can see that the quantiles of the observed Q values are far from the expected quantiles of a ![](1471-2105-6-128-i2.gif) distribution, suggesting that these two datasets generated heterogeneous results beyond random sampling errors. Therefore, we applied the random effect model in this study. The quantiles of the Q statistic were closer to the quantiles of the expected chi-square distribution when quality-adjustment was considered (Figure [3(a)](#F3){ref-type="fig"}) than when it is was not (Figure [3(b)](#F3){ref-type="fig"}). The variance for the unadjusted *Q*values was 9.45, but it was reduced to 3.31 when quality adjustment was used. This result suggests that the incorporation of the adjusted quality measure into effect size estimation can increase consistency and decrease heterogeneity between these two datasets. To identify a list of potentially \"significant\" genes, we adapted the false discovery rate (FDR) algorithm implemented in \[[@B30]\]. We first calculated the adjusted z statistics for all genes based on random-effects model (REM). Genes were then ranked by the magnitude of their z statistic values. A permutation-based approach was used to obtain the corresponding expected ordered z statistic. The potentially \"significant\" genes are genes with a distance between the ordered z statistic from the observed data and that of the permuted data exceeding a given threshold (delta). Figure [4](#F4){ref-type="fig"} shows the relationship between the number of significantly differentially expressed genes and different delta levels. As we see in this figure, the quality-adjusted REM can identify many more significant genes than the quality-unadjusted REM model at any fixed level of delta. We calculated the estimated FDR for each given delta. As expected, the number of genes called significant increased as the value of delta decreased, but at the cost of an increasing FDR. The estimated FDR was zero down to a delta of 0.6, where 228 genes were called significant in the quality-adjusted model and 153 genes in the quality-unadjusted model. In order to get a manageable gene list, we arbitrarily set delta at 1.1. At this delta level, we selected 29 differentially expressed genes (representing 32 probe sets) at a FDR of 0% when the quality weight was incorporated into the estimation of the effect size. However, when the quality measure was not used (Choi et al\'s method \[[@B7]\]), we only selected 20 differentially expressed genes (representing 21 probe sets) at a FDR of 0%. All the 20 genes were also in the top of the list of the 29 genes identified with the quality-weights. Tables 2 and 3 show the selected differentially expressed genes between normal and lung adenocarcinoma patient samples using the quality-adjusted and quality-unadjusted models, respectively, with genes ordered based on their z statistic values \[see [Additional files 1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}\]. As can be seen in Tables 2 and 3, 4 of the 9 genes that were selected by our method, but not by Choi et al\'s method, have also been identified by several other groups including Jiang et al. \[[@B10]\], Beer et al. \[[@B15]\] and Bhattacharjee et al. \[[@B16]\]. In particular, some of these 4 genes, such as TEK and TGFBR2, have been experimentally validated (shown in Table 3 of Jiang et al.\[[@B10]\]). For a cutoff of an adjusted z value of 1.96 (corresponding to a 5% level of significance), the quality-adjusted model identified 9 significantly expressed genes while the quality-unadjusted model identified only 2 of the 9 significantly differentially expressed genes. All these results suggest that our proposed method may have increased sensitivity to detect more differentially expressed and biologically relevant genes than Choi et al\'s method \[[@B7]\]. We compared genes identified with our method with genes identified by Jiang et al. \[[@B10]\], Beer et al. \[[@B15]\] and Bhattacharjee et al. \[[@B16]\], sixteen of the 29 genes identified by our proposed model were also detected in at least one of these studies. In particular, we observed that 6 of the 29 genes were consistently identified in the other three studies. There were 13 of the 29 genes that were uniquely identified in our study. Some of these are plausible candidates for lung adenocarcinoma. For example, Walker et al. \[[@B17]\] found that G protein-coupled receptor kinase 5 (GRK5) is a key gene regulating airway response, that may have implications in obstructive airway diseases. Discussion ========== In this study we proposed a measure to quantify Affymetrix gene chip data quality for each gene in each study. The quality index measures the performance of each probe set to detect its intended target. Furthermore, we extended a traditional effect size model by using the quality index as a weight for combining information from different chip types of Affymetrix microarrays, and incorporating this weight into a random-effects meta-analysis model. We illustrated the advantages of our proposed methods using the Harvard and Michigan gene expression datasets used in \[[@B18]\]. This approach of using the detection p-values to weight the gene expression estimates can be applied in a more general context and to other microarray meta-analysis methods, such as that of Rhodes et al \[[@B6]\]. The assumption of a group-specific exponential distribution for the negative log detection p-values is a very rough approximation. The true distribution is probably closer to the two-parameter log-beta distribution. However, due to the fact that there are only 16--20 probe pairs per probe set used in the Wilcoxon test statistic for the detection p-values, the p-values follow a somewhat discrete distribution. In particular for highly-expressed genes, the p-values for all samples in a group may have the same near-zero value. Estimation of two-parameters is therefore impossible. It would be worth investigating the sensitivity of the performance of this quality-weighting approach to the distributional assumption. The performance of the weighting also depends on the sensitivity parameter s. In this study, a high *s*was needed since a lower cutoff gave almost all genes extremely low quality scores. This is a consequence of the observed detection p-value distributions. These data were created in 2001, and since that time the quality of the oligonucleotide array technology has improved with more specific probe selections and optimized experimental conditions. For expression data obtained today, smaller values of s would be more appropriate. For example, in an analysis of a dataset generated more recently using improved oligonucleotide technology (not shown), using *s*= 0.05 gave good performance. The discrepancy between the calculated and expected values for the Q test of heterogeneity was smaller when our quality measure was incorporated than when it was not, as suggested by the quantile-quantile plots, but overall there still remains some variability that has not been fully captured by the weight function, emphasizing the need for a more elaborate weighting strategy and sensitivity analyses. From a biological point of view, lung adenocarcinoma may be heterogeneous originating from different causes \[[@B19]\] and methods like cluster analysis in which subtypes of relatively homogeneous groups of disorders can be identified might be useful. The focus of our paper was to introduce a methodology that can be used to integrate data, and as such the grouping of the samples as \"diseased\" versus \"normal\" was primarily used to identify genes that discriminate the two groups in a broader sense than at a higher resolution in which sub types could be identified. Choi et al. \[[@B7]\] have also provided a brief Bayesian interpretation of their effect size model to integrate information from multiple microarray studies. They argued that a Bayesian approach could offer a more flexible and robust modeling strategy. We did not consider this issue in this study. It will be interesting to investigate how our proposed quality measure could be incorporated into their Bayesian framework in the future. Methods ======= Data source and preprocessing ----------------------------- We selected two Affymetrix microarray data sets from the 4 lung cancer data sets provided by the organizers of the Critical Assessment of Microarray Data Analysis (CAMDA) conference \[[@B18]\]. These datasets were collected using different versions of the Affymetrix oligonucleotide microarrays and were conducted by two research groups, one from Harvard and the other from Michigan. The Michigan study \[[@B15]\] used the HuGeneFL Affymetrix chip, containing 7,129 probe sets, each with 20 probe pairs. This study included 86 lung adenocarcinoma patient samples and 10 normal samples. The Harvard study \[[@B16]\] used the HG\_U95Av2 chip with 12,625 probe sets, each with 16 probe pairs. This study included 17 normal and 127 lung adenocarcinoma patient samples. We use the word \"sample\" interchangeably with \"array\". The main objectives addressed in these two studies were to identify differentially expressed genes related to lung adenocarcinoma, and genes whose expression was related to patient survival. Our interest is in the former. We developed a new method for identifying genes that are differentially expressed between the cancer and normal samples, by modelling the effect size and integrating information from the Harvard and Michigan studies. We converted the probe level data to a single expression measure for each probe set using the robust multi-array average (RMA) algorithm \[[@B20]\], which provides higher specificity and sensitivity in detecting differentially expressed genes. We used 6124 common probe sets in these two studies for the data analysis in our study. These probe sets were selected based on a sequence-based probe matching method, which is believed to produce more consistent results when comparing similar biological data sets obtained by different microarray platforms \[[@B10],[@B21]\]. The detailed procedure used to select these probe sets can be found in the Data processing Section of \[[@B10]\]. Quality measure for Affymetrix Genechip data -------------------------------------------- Affymetrix genechip microarrays are used to monitor gene expression for thousands of transcripts. Each transcript is represented as a probe set, and a probe set is made up of probe pairs comprised of Perfect Match (PM) and Mismatch (MM) probe cells. The level of expression of a gene product is estimated using the intensities of each probe pair in a probe set. Therefore, the probe-specific variability in a probe set can be used as a measure of the performance of that probe set. The detection algorithm proposed by \[[@B22]\] generates a detection p-value, which represents the probability that the probe set (gene) expression is above zero (i.e., turned on), and measured reliably and consistently. A lower p-value is considered as a useful indicator that the measured gene expression is valid and reliable \[[@B22]\]. Specifically, the p-value is based on testing whether the probe-specific differences (PM-MM) are almost always positive. We used the detection p-values to define quality measures for probe sets, summarizing across the arrays and experiments. We realize that some genes may be truly \"off\" under some experimental conditions and hence a large detection p-value may be providing useful information. However, if a gene is \"off\" under all experimental conditions, we argue that analysis of this gene contributes little to understanding of the experiment. For any gene, let *pvalue*denote its detection p-value and *x*~lg~denote - log(*pvalue*~lg~) for sample *l*in group *g*. We assume that each study compares G groups, where there are *n*~*g*~samples in group g, and g = 1, 2, \..., G. For example, in the lung cancer data, G = 2, since adenocarcinoma samples are compared to controls. It is well known that p-values follow a uniform distribution when there is no signal, and therefore, we expect *x*~lg~to follow an exponential distribution with mean λ = 1 if the gene is not expressed. In order to develop a single quality measure for each gene across all samples in one study, we use this relationship with the exponential distribution to motivate a quality measure. We assumed that the detection p-value for a single gene of sample *l*in group *g*follows the distribution *x*~lg~= - log (*pvalue*~lg~) \~ *Exponential*(λ~*g*~), where different distributions of expression can be expected in each group g. It should be noted that we are modeling the p-value of one gene, across the samples in a group. This is different from the approaches of Allison et al. \[[@B23]\] and Pounds et al. \[[@B24]\] who modeled the distribution of p-values across genes. Although the true distribution of the *x*~lg~may not be exponential, this assumption leads to a simple model where the one parameter can be estimated by a closed-form expression. Hence, the parameter λ~*g*~for each gene, study and group *g*can be estimated by: ![](1471-2105-6-128-i3.gif) where ![](1471-2105-6-128-i4.gif) is the usual sample mean. This is a maximum likelihood estimator (MLE) with well-known asymptotic optimality properties \[[@B25]\]. To combine across the groups, we assumed a sensitivity parameter *s*. It is defined as the probability that a representative probe set in a particular treatment group shows a detectable signal, assuming that the relevant distribution (exponential or beta) holds. It can be thought of as the equivalent to a detection p-value defined for a whole group rather than for one array, and an appropriate value for *s*should be chosen with this interpretation in mind. For example, the default settings by Affymetrix software for the detection calls are p = 0.06 for a \"Marginal\" call, and p = 0.04 for a \"Present\" call, although these can be altered by the user. In practice, the appropriate choice of *s*may depend on the signal-detection capability of particular technologies. We recommend plotting the distribution of quality scores for different choices of *s*, and choosing a value that clearly distinguishes genes of low quality (scores near zero) from high quality genes (scores near 1). The sensitivity parameter *s*is a chosen cutoff, so that genes that are \"off\" or poorly measured across all experimental conditions will have *pvalue*≥ *s*, or in other words, ![](1471-2105-6-128-i5.gif). Therefore, we can define a quality measure across the groups, for each gene and each study as: ![](1471-2105-6-128-i6.gif) The choice of the maximum gives more weight to genes measured with high quality in at least one group, thereby allowing a gene that is \"off\" in one condition and \"on\" under another condition to provide useful information in the analysis. We treat this quality score as a weight for each gene in the subsequent analysis. Modelling effect size with quality-adjusted weights --------------------------------------------------- In order to simplify the discussion, we consider two groups, treatment (*t*) and control (*c*) groups, in study *i*= 1,2,\...,*k*. Let *n*~*it*~and *n*~*ic*~denote the number of treatment and control samples in study *i*, respectively. For each gene, let μ denote its overall mean effect size, a measure of the average differential expression for that gene, and let *y*~*i*~denote the observed effect size for study *i*. We modeled effect size using the hierarchical model: ![](1471-2105-6-128-i7.gif) Where τ^2^is the between-study variability and ![](1471-2105-6-128-i8.gif) is the within-study variance, measuring the sampling error for the *i*^*th*^study. Choi et al. \[[@B7]\] used the standardized mean difference as a measure of the observed treatment difference *y*~*i*~. This well-known estimator of treatment difference found in Hedges and Olkin\'s \[[@B26]\] work is ![](1471-2105-6-128-i9.gif) where ![](1471-2105-6-128-i10.gif) and ![](1471-2105-6-128-i11.gif) are the average gene expression values in the treatment and control groups of study *i*, respectively, and ![](1471-2105-6-128-i12.gif) is the pooled standard deviation. For a study with *n*samples, an approximately unbiased estimator of ![](1471-2105-6-128-i13.gif) is given by ![](1471-2105-6-128-i14.gif)\[[@B26]\]. The estimated variance ![](1471-2105-6-128-i8.gif) of the unbiased effect size is given by \[[@B27]\] ![](1471-2105-6-128-i15.gif) In a fixed-effects model (FEM), the error of the observed effect sizes is fully assigned to sampling error only, ignoring the between study variance, so τ^2^= 0 and *y*~*i*~\~ *N*(μ, ![](1471-2105-6-128-i8.gif)). On the other hand, a random-effects model (REM) considers that each study estimates a different treatment effect θ~*i*~. These parameters are drawn from a normal distributionθ~*i*~\~ *N*(μ, τ^2^). To assess whether FEM or REM is most appropriate, we tested the hypothesis τ = 0 using the following test statistic, which is a modification of Cochran\'s test statistic \[[@B28]\] that incorporates our quality measure for each study ![](1471-2105-6-128-i16.gif) where ![](1471-2105-6-128-i17.gif) and ![](1471-2105-6-128-i18.gif) ![](1471-2105-6-128-i19.gif) is the weighted least squares estimator that ignores between study variations. Under the null hypothesis of τ = 0, this statistic follows a ![](1471-2105-6-128-i20.gif) distribution. We followed Choi\'s approach \[[@B7]\] to draw quantile-quantile plots of *Q*to assess whether a FEM or REM model is appropriate. If the null hypothesis of τ = 0 is rejected, we estimate τ based on the method developed by DerSimonian and Laird \[[@B29]\] ![](1471-2105-6-128-i21.gif) Therefore, we can estimate μ that corresponds to a random effects model by ![](1471-2105-6-128-i22.gif) where ![](1471-2105-6-128-i23.gif). Under the REM, ![](1471-2105-6-128-i24.gif) The *z*statistics to test for treatment effect under REM is ![](1471-2105-6-128-i25.gif) However, when there are only a small number of arrays in each group, the estimates of standard error *s*for each gene can be very variable. Some genes might by chance have very small standard errors, and therefore appear highly significant. To address this problem, different approaches have been developed for \"smoothing\" the variance estimates by borrowing information from the ensemble of genes. This can assist in inference about each gene individually. For example, Tusher et al. \[[@B30]\], Efron et al. \[[@B31]\] and Broberg \[[@B32]\] used t-statistics where an offset was added to standard deviation while Smyth \[[@B33]\] proposed a t-statistic with a Bayesian adjustment to the denominator. We took the offset *s*~0~as the quantile of the gene-wise standard errors that minimizes the coefficient of variation of the z statistics \[[@B30]\]. Therefore, we can calculate the adjusted *z*statistics (used in this study) to test for treatment effect under REM as ![](1471-2105-6-128-i26.gif) The adjusted *z*statistics for FEM is the same as that for REM except that τ^2^= 0. Note that all these expressions refer to a single gene. The adjustment for computing z statistic was also used by Garrett-Mayer et al. \[[@B34]\]. Assessment of differentially expressed genes -------------------------------------------- We performed a multiple testing procedure, as described by Dudoit et al. \[[@B35]\], to evaluate statistical significance for differentially expressed genes in the combined studies. The false discovery rate (FDR) \[[@B36]\] has become a popular error measure for this purpose. Tusher et al. \[[@B30]\] developed a permutation-based method to calculate FDR for evaluating differentially expressed genes. We adapt their approach to our meta-analytic framework as follows: 1\. For each gene *j, j*= 1,2,\..., *J*, in the original data, compute the adjusted z statistic *Z*~1~,\...,*Z*~*J*~based on the meta-analysis procedure described in the previous section 2\. Order these z statistic values to obtain *Z*~(1)~≤,\..., ≤ *Z*~(*J*)~ 3\. Create B random permutations within both studies. For each permutation *b*= 1,2,\... *B*, produce the adjusted z statistic *Z*~1,*b*~,\..., *Z*~*J*,*b*~for gene *j*= 1,2,\..., *J* 4\. Compute expected order z statistics: ![](1471-2105-6-128-i27.gif) 5\. The potentially \"significant\" genes are those that have a distance between *Z*~(*j*)~and ![](1471-2105-6-128-i28.gif), greater than a given threshold delta (Δ). Therefore, we can find the smallest positive *Z*~(*j*)~such that ![](1471-2105-6-128-i29.gif), say *t*~1~. Similarly, we can find the largest negative *Z*~(*j*)~, say *t*~2~ 6\. The estimated FDR for the selected significant genes at the given delta (Δ) is given by: ![](1471-2105-6-128-i30.gif) Authors\' contributions ======================= JB initiated, designed and managed the study. CG proposed the quality measure method and participated in designing and managing the study. PH conducted data analysis and drafted the manuscript. JB and CG revised the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Selected differentially expressed genes between normal and lung adenocarcinoma patient samples (Genes selected based on *quality-adjusted model*with delta = 1.1) ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Selected differentially expressed genes between normal and lung adenocarcinoma patient samples (Genes selected based on *quality-unadjusted model*with delta = 1.1) ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We would like to thank Dr. Hongying Jiang who provided us with the list of common probe set IDs for the two data sets used in our study. We also acknowledge helpful suggestions from two anonymous reviewers that greatly improved the quality of the manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Scatter plots of (a) quality-weighted and (b) quality-unweighted mean expression values for the 6124 common probe sets in the Harvard and Michigan datasets based on normal (healthy) samples. ::: ![](1471-2105-6-128-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Detection p-values for a sample probe set (38249\_at). H1 and H2 denote the detection p-values in normal and lung cancer groups, respectively, for the Harvard study; whereas M1 and M2 denote the detection p-values in normal and lung cancer groups, respectively, for the Michigan study. ::: ![](1471-2105-6-128-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Quantile -- Quantile plots of the observed versus the expected Q statistic: (a) with quality adjustment, and (b) without quality adjustment. ::: ![](1471-2105-6-128-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Relationship between number of significantly expressed genes and different delta levels, obtained from fitting the random effects model. ::: ![](1471-2105-6-128-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Quality scores for selected probe sets at a sensitivity parameter s = 0.2 ::: Quality Score -------------- --------------- ------- ------- ------- 38249\_at 0.663 0.001 0.255 4.050 32180\_s\_at 0.263 0.732 0.829 0.194 37174\_at 0.001 0.495 4.050 0.437 32318\_s\_at 0.795 0.795 0.142 0.142 :::

PubMed Central

2024-06-05T03:55:58.316211

2005-5-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173085/", "journal": "BMC Bioinformatics. 2005 May 27; 6:128", "authors": [ { "first": "Pingzhao", "last": "Hu" }, { "first": "Celia MT", "last": "Greenwood" }, { "first": "Joseph", "last": "Beyene" } ] }

PMC1173086

Background ========== Response thresholds and exclusion criteria are applied when presenting a summary of a microarray study; otherwise the data set can be unmanageable. The precise effect of such criteria is, however, rarely discussed or investigated. We believe that many researchers are typically not aware of the effect that their chosen thresholds or \'qualification\' criteria have had on their final set of \"significant genes\". A common example would be data exclusion. In studies using the Affymetrix platform the selection criteria typically uses the absent, moderate or present \'call\' system, calculated with algorithms implemented in MAS5 (microarray suite 5 from Affymetrix) or D-chip \[[@B1]\]. The decision of whether the detection threshold is set to 10% or 20% (for example) will change the number of significant genes when the analysis is followed by the use of a parametric test with multiple testing corrections (for example). How predictable the imposition of such thresholds, on the composition of the output, is not clear. A recent method to identify significant genes is \"Significance Analysis of Microarrays\" (SAM)\[[@B2]\]. An ISI search indicates that SAM is the most popular method employed for microarray analysis (635 citations of the original publication as of October 2004). In SAM, the relative difference (d(i)) is compared to the distribution of d(i) following random permutation of the sample categories. For each d(i), a certain proportion of all genes in the permutation set (control set) will be found to be \'significant\' by chance and this parameter is then used to calculate a \"False Discovery Rate\" (FDR). This is presented as a q value for each gene in the final list of significant genes. The q-values are influenced by the variability in the data set. This implies that changes in the entire data set composition will affect the d(i) distribution in the permuted control set and thus the q-value assigned to a given gene. To what extent this alters the final output when using SAM or subsequent post-hoc analysis has so far not been discussed. As SAM is arguably the most widely utilized method in the microarray analysis field, we felt it was critical to evaluate these considerations. The number of reported significant genes is influenced by the FC setting within SAM and the use of \"Present/Absent\" calls for data inclusion ---------------------------------------------------------------------------------------------------------------------------------------------- The first point to emphasise is that we are examining the effect of a variable FC setting, implemented in the Excel SAM, on a final gene list with a fixed 1.5 FC criteria. Our principle findings can be extended to any relevant final FC criteria but in the example we provide we have focused on a 1.5 FC criteria, for simplicity. When using the FC setting, implemented in the Excel addin, the researcher selects a proportion of the all genes on the chip, and this selection is also utilised for the permutated control data set, however the settings from the full data set SAM analysis are maintained (s\_0, and pi\_0). The FC hurdle setting can therefore change the resulting q-values as genes that pass a certain FC when the original sample categories are used, are less likely to generate high d(i) when the categories are permutated. To test the extent of this effect, we performed SAM at several different FC hurdles within Excel and scored the number of genes reported at the final 1.5 FC criteria. A difference in the number of genes reported as significant above the final FC would indicate that by using the Excel SAM FC restriction setting, during SAM, we change the outcome by altering the d(i) distribution of the permutation data set. We used three different biological data sets to assess how wide-spread any effects were. The first data set was a paired data set from a human skeletal muscle study which examined subjects before and after endurance training using the U95A-E platform (Affymetrix) \[[@B3]\]. The data was RMA normalized \[[@B4]-[@B6]\] and SAM was performed using different biological sample subgroups (groups were formed on the basis of a variety of physiological parameters) and chipset identity (A, B or D). Changing the FC setting in SAM (Excel) altered the final list of significant genes at the 1.5 FC criteria (Figure [1A--D](#F1){ref-type="fig"}). Importantly, the effect was not uniform across all conditions. For some samples a sequential increase in the FC hurdle in SAM correlated with an increased number of reported genes (Figure [1A--B](#F1){ref-type="fig"}) while under other conditions a very small change in FC had an apparently random impact on the composition of the significant gene list (e.g. Figure [1C--D](#F1){ref-type="fig"}). (The number of genes that passed each fold change criteria can be found in \[[Additional file 1](#S1){ref-type="supplementary-material"}\]). The second data set was derived from an *in-vitro*mouse senescence study performed on U74Av2 chips \[[@B7]\] and normalized using RMA \[[@B4]-[@B6]\]. When comparing two of the time points during the induction of senescence we were unable to observe any effect of FC selection within SAM on the yield of significant genes (Figure [2](#F2){ref-type="fig"}). We believe this reflected the very low q-values obtained when originally using SAM, which in turn most likely reflects the low experimental variation that one can achieve using *in-vitro*models. The third data set is derived from a study of the aging human brain, containing 21 samples split into two categories, young and old (unpaired data) \[[@B8]\], and normalised using D-chip \[[@B1]\]. When varying the FC hurdle within SAM we again saw a large impact on the number of genes reported as being significant, above the [1.5 FC criteria]{.underline} (a q-value threshold of 1% was considered statistically significant similarly to the original report) (Figure [3A--B](#F3){ref-type="fig"}). The \'significant\' gene list increased from 283 genes when using no FC hurdle (during SAM in Excel) to a maximum of 465 significant genes (67% more) when using a ***1.51***FC hurdle within Excel SAM (Figure [3A](#F3){ref-type="fig"}). A similar effect was observed when we examined the methods utilised by the original authors (20% \"Present\" call by D-chip). No FC hurdle generates 314 significant genes and a 1.51 FC generates 538 genes (Figure [3B](#F3){ref-type="fig"}). This is a 71% increase in \"significant\" genes associated with aging in human brain, despite the fact that the FC hurdle utilised in SAM was actually greater than the final FC of 1.5! Similar effects arise when other final FC criterions are used during SAM. For example if we only consider the significant genes that pass a final FC \>2 criteria in the aging brain study, then 5% more \>2 FC genes can be found if the FC hurdle within SAM is set to 2.0 compared to 1.5. Clearly, genes that are modulated to a large extent may be of high biological interest and thus our observation is important. Also, there are 73 more significant genes at the FC hurdle of 1.51 if the reduced data set is used compared the full data set (less genes in the data set gave more significant genes). This indicates that Present/Absent filtering also influences the outcome of the analysis, in this case increasing the number of genes defined as being modulated. Changing the FC setting in SAM changes the reported q-value ----------------------------------------------------------- The analysis presented above demonstrates that the q-value obtained for a specific gene depends on the FC hurdle applied during SAM in Excel. To monitor the q-values generated, for individual genes, we obtained the q-values during all SAM calculation using various FC hurdles for all genes that were reported as significant at the 1.51 FC setting (538 genes) in the brain aging study (Figure [4A](#F4){ref-type="fig"}). The highest q-value for a subset of the genes that passed the final fold change criteria was 3.6% when SAM was performed with a 1.0 FC hurdle. The same genes appear as being significant when the \"optimal\" (in the sense that these setting produced the largest significant gene list) Excel SAM FC hurdle setting 1.51 is used while the highest q-value reported was now only 0.97%. Running SAM using different FC settings can change data interpretation ---------------------------------------------------------------------- Often the main reason for carrying out a microarray experiment is to gain greater insight into the molecular processes that contribute to a complex biological phenotype. One standard method for carrying out such analysis is the use of gene-grouping, such as Gene ontology classification \[[@B9]\]. To address whether the biological interpretation may be influenced, we compared overrepresentation of classifications using EASE \[[@B10]\]. We selected the 314 genes and the 538 genes obtained in the brain aging study using the 1.0 and a 1.51 FC hurdles (described above) and calculated the overrepresented classifications. There was a considerable difference in the number of significantly over-represented classifications identified by EASE (figure [4B](#F4){ref-type="fig"}). 35 classifications were overrepresented in both gene lists, 8 were found in the GO analysis of the 1.0 FC list and 29 were unique to the 1.51 FC list. We feel it is unlikely that greater inclusion of randomly determined genes (i.e. false positives) would give rise to a significant increase in statistically enriched functional groups. This indicates that both the number of genes and the interpretation may be substantially altered by arbitrary data filtering, as exemplified by the use of the FC hurdle during the operation of Excel SAM or through the application of present-absent calls using e.g. D-chip. Conclusion ========== The average researcher is highly dependent on the use of \'standard procedures\' for their microarray analysis. An arbitrary filtering option in the Microsoft Excel Addin (we have called this the FC hurdle to distinguish from FC criteria -- which is the *final*FC value selected by the researcher, to define a functionally significant change in gene expression) or data exclusion (e.g. present or absent call thresholds) can impact on a study, in a less than predictable manner. A clear problem also arises when SAM is utilised on different software platforms. In the R package *Siggenes*, no FC hurdle criteria can be made unless an additional programming function is implemented by the individual researcher. This is in contrast to the more widely used Microsoft Excel *SAM Addin*where the researcher can introduce a FC hurdle prior to the q-value calculations (while the SAM algorithm maintains the initial parameters from the analysis of the full data set). It is often unclear in the literature if SAM was performed in Excel or R; and if an FC hurdle was applied within SAM or if the fold change criteria was introduced after completing the operation of SAM, in EXCEL. Indeed, one might question how robust the SAM methodology is, if it is heavily dependent on both pre SAM data selection and within SAM (Excel) data filtering. However, one of the appreciated strengths of SAM is that the real data set is used to estimate experimental variation. One could also question whether it is valid to reduce the data set prior to using SAM. It would seem intuitive that much of the data being removed using a FC hurdle during SAM operation would be below an acceptable response level to be considered as being biologically relevant. However, the FC hurdle may also remove data that is essential for an accurate estimation of the experimental variation. The challenge would then be to remove \'genuine\' noise from non-expressing genes without removing genes that are genuinely expressed and necessary to approximate the data set variation. The effects will be pronounced in data sets demonstrating a large range of gene expression FC\'s or where significant inter experimental variation exists. It is clear, that investigators must be made aware that the impact of \'qualification criteria for inclusion\' and \'out-put filtering\' is less than predictable, when using SAM. Methods ======= Data sets --------- Three data sets were used. The first is a human skeletal muscle study, comparing eight subjects across the U95A-E Affymetrix chips before and after endurance training \[[@B3]\]. A subset of this data set can be derived by creating a low responder group based on their improvement upon training. This subset is referred to as the low responder group in figure [1](#F1){ref-type="fig"}. The data can be obtained from the authors (JAT). The second data set is a part of an mouse in-vitro senescence study performed on U74Av2 (Affymetrix) \[[@B7]\]. Two conditions were selected for comparison: EpiA1-ts58 cells at the basal condition and EpiA1-ts58 cells after 72 hours of senescence induction. This data set can be downloaded from the journal web site in a MAS5 (Affymetrix microarray suite 5) normalised format or will be distributed in RMA normalised format by the authors (OL). The third data set is a brain aging study performed on the U95A chip (Affymetrix) and normalised using D-chip \[[@B8]\]. The authors separated their data set into two groups: young (\<43 years old) and old (\>72 years old) and we used the same grouping in our analysis. The authors further used a data discrimination based on \"present\" calls given by D-chip (20% present in all samples). We used both the full data set and the 20% data set which is referred to as the reduced data set. The data set can be access at GEO, accession number GSE1572. SAM analysis ------------ SAM analysis was performed using the Microsoft Excel addin v1.21. The analysis was performed using different settings of the \"Fold change\". We also performed SAM analysis using the R (R-project.org) package \"siggenes\" as a comparison to the Excel addin without using the fold change setting with consistent results. Classification analysis ----------------------- To look for overrepresented classifications we used EASE \[[@B10]\]. We used all possible classifications and considered a classification as positively overrepresented if the EASE score was lower then 0.05. Abbreviations ============= SAM -- significance analysis of microarrays FDR -- false discovery rate FC -- fold change EF -- extraction factor MAS5 -- microarray suite 5 RMA -- robust multi array average GEO -- gene expression omnibus Authors\' contributions ======================= OL -- Hypothesis generation, practical work with all data sets and drafting of the manuscript CW -- Study design and revising the manuscript. JAT -- Hypothesis generation, practical work on the human muscle data section and drafting of the manuscript Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Genes that pass various FCs The additional file describes how many genes that passed various FCs from all three data sets ::: ::: {.caption} ###### Click here for file ::: Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **FC effects on the endurance training data set:**SAM analysis was used at various fold changes studying the exercise data set while scoring genes with a q-value of \<0.05 and FC\>1.5. This was done to asses the effect of the fold change option in the SAM Excel addin on genes reported as significant at a higher fold change. The figure shows the number of scored genes using 4 different chip and sample combinations: **(A)**All eight subjects before and after training (totally 16 arrays) were compared in a paired analysis using the U95A chips. **(B)**All eight subjects before and after training (totally 16 arrays) were compared in a paired analysis using the U95B chips. **(C)**The reduced group consisting of low four low responders (totally 8 arrays) were compared in a paired analysis using the U95B chips. **(D)**The reduced group consisting of low four low responders (totally 8 arrays) were compared in a paired analysis using the U95D chips. ::: ![](1471-2105-6-129-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **FC effects on the senescence data set:**SAM analysis was used at various fold changes using the senescence data set while scoring genes with a q-value of \<0.01 and FC\>1.5. A comparison between non-senescent cells and senescent cells was used (two replicates of the senescent cells and four replicates of the non-senescent cells). ::: ![](1471-2105-6-129-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **FC effects on the brain aging data set:**SAM analysis at various fold changes using the brain aging data set (11 old samples and 10 young samples) while scoring genes with a q-value of \<0.01 and FC\>1.5. We used both the full data set and the reduced data set suggested by the authors. **(A)**The full data set. **(B)**The reduced data set (20% \"present\" calls by D-chip\"). ::: ![](1471-2105-6-129-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **FC effects individual q-values:**q-values of all genes scored as significant in the brain aging study (reduced data set) at FC 1.51 at other FC settings. Genes only acquires discrete q-values and all 538 genes are shown, but overlap. **(B) Running SAM with different FC settings changes the biological interpretation:**Venn diagram comparing the number of significantly overrepresented classifications (EASE score \<0.05) using the reduced brain aging data set analysed either with a 1.0 FC setting (314 genes) or a 1.5 FC setting (538 genes). ::: ![](1471-2105-6-129-4) :::

PubMed Central

2024-06-05T03:55:58.318758

2005-5-29

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173086/", "journal": "BMC Bioinformatics. 2005 May 29; 6:129", "authors": [ { "first": "Ola", "last": "Larsson" }, { "first": "Claes", "last": "Wahlestedt" }, { "first": "James A", "last": "Timmons" } ] }

PMC1173087

Background ========== Whenever experimental data is gathered to examine a process, researchers will try -- implicitly or explicitly -- to define a model describing the process. The purpose of building a model is to generalize the experimental data, allowing the prediction of new experimental outcomes, or to gain insights into the experimental process. In the biological sciences, models are often implicitly built and verbally formulated, such as \"the proteasome preferably cleaves after hydrophobic amino acids\". Such a model is easy to understand and often reflects all the knowledge that can be gathered from a few, difficult experiments. However, with the advent of high-throughput experiments to the biosciences, it has become feasible to generate more quantitative, mathematical models, based on large volumes of data. For the purpose of building a model, conducting experiments can be formally described as collecting pairs of experimental parameters x and experimental outcomes (measurements: y~meas~). Building a model is then equivalent to finding a function f(x) = y~pred~≈ y~meas~. Herein, we call the model function f a prediction tool, and the experimental observations used to generate it its training set T = (x, y~meas~). Developing a prediction tool is a two step process. First, a general prediction method capable of generating specific tools has to be chosen, such as a certain type of neural network, classification tree, hidden markov model or regression function. This choice is to some degree determined by the experimental data available. However, only a few prediction methods are clearly unsuited for a certain type or amount of data, leaving several potentially appropriate methods to choose from. In practice, the personal experience of a scientist often determines which prediction method he applies, as learning to apply a new method is often more costly than the benefits of a slightly better model. After a method is chosen, it is applied to the experimental data to generate a specific prediction tool. Several different terms are commonly used for this second step, such as \'supervised learning\', \'fitting the model\' or \'regression\'. Each method has its own formalism describing how this is done, but essentially a method is capable of generating a certain class of functions, of which one is chosen, that minimizes the difference between measured and predicted experimental outcomes for the training set T. Here we describe a computer program implementing the SMM prediction method, which can be applied to model the sequence specificity of quantifiable biological processes. In such experiments, the input parameter × corresponds to a sequences of amino-or nucleic acids, and the experimental outcome y~meas~is a real numbers measuring the process efficiency. The method utilizes training sets in which the examined sequences all have the same length. The SMM method was previously applied successfully to predictions of MHC binding \[[@B1]\], TAP transport \[[@B2]\] and proteasomal cleavage \[[@B3]\]. However, its software implementation was not made publicly available, because it relied on a commercial library with restrictive licensing terms. Also, applying the method to a new problem required manual changes in the source code, which is impracticable for an external user. Both of these issues are addressed in the software implementation made available here. In addition, this manuscript describes two specific features that we have found to be effective in generating high quality models and which can easily be utilized in other prediction methods, namely handling of bounded data and combinatorial peptide libraries. The SMM software is not meant for \'classical\' sequence analysis, which can be roughly defined as aligning related sequences in order to identify conserved residues or in order to generate classifiers which can identify additional related sequences. Rather, the typical application is the characterization of a sequence recognition event, such as the sequence specific cleavage efficiency of a protease. This characterization does not assume any evolutionary relationship between different recognized sequences. In terms of scientific fields, that makes the SMM software aimed at applications in biochemistry or molecular biology. Implementation ============== Code ---- The SMM algorithm is implemented in C++ code. Only standard libraries or freely available external libraries were used. The two external libraries are Tinyxml \[[@B4]\] to handle the XML input and output and the Gnu Scientific Library \[[@B5]\] for efficient vector and matrix operations. The source code has been compiled and tested using Visual C++ on a windows system and using g++ (1.5 or above) on a Debian and Suse Linux system. A windows executable is also available. The source code, documentation and examples are available as [additional files 1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"} of this manuscript, and on the project homepage at <http://www.mhc-pathway.net/smm>. On a standard 2.6 GHz Pentium 4 PC running windows XP, the creation of a prediction tool from a typical set of training data containing 300 8-mer peptides takes about 10 minutes. Input and output ---------------- The program expects an XML document as its standard input. A more sophisticated (graphical) user interface is not likely to be of great interest for the projected user community, most of which will probably call the executable file from a script or integrate the code into an existing program. The XML input document contains either training data to generate a new tool (Figure [1](#F1){ref-type="fig"}) or sequences for which a prediction should be made with a previously generated tool. The output of the program is again in XML format. For all types of input and output, XML schemas defining their exact structure are supplied. The schemas contain annotation for each input element, documenting their intended use. A simplified alternative input format exists for the most common application, namely the generation of a scoring matrix from a set of standard amino acid sequences. Several example input files are supplied with the program, which should make it easy for users to create similar input files with their own data. Core algorithm -------------- An amino acid can be encoded as a binary vector of length 20, with zeros at all positions except the one coding for the specific amino acid. Extending this notation, a peptide of length N can be encoded as an N\*20 binary vector. The same conversion can be made for nucleic residues, resulting in N\*4 length vectors. Any fixed length sequence can be converted into a fixed length binary vector following similar rules. The SMM algorithm expects such a vector as the experimental input parameter \'x\'. For a set of experiments, the vectors ^t^x (where ^t^x indicates the transposed vector x) can be stacked up resulting in a matrix H. An example for a set of nucleic sequences is shown in Figure [2](#F2){ref-type="fig"}. The matrix H is multiplied by a vector w which assigns a weight to each possible residue at each position in the sequence. This generates a vector of predictions y~pred~ H w = y~pred~ (1) From a given training set of sequences encoded in H with measured values y~meas~, the \'correct\' values for w can be found by minimizing the difference between the predicted values y~pred~and the measured values y~meas~according to a norm \|\| \|\|. To suppress the effect of noise in the experimental data, a second term is added to the minimization: \|\|H w - y~meas~\|\| + ^t^w Λ w → minimum, (2) in which Λ is a positive scalar or a diagonal matrix with positive entries. To better understand the effect of the Λ term, consider first minimizing (2) with Λ set to zero. In this case, the optimal entries for the weight vector w minimize the difference between predicted (y~pred~) and measured (y~meas~) values. Minimizing (2) with a non-zero value for Λ results in a shift of the optimal entries in w towards values closer to zero, especially for entries in w that do not significantly decrease the distance between predicted and measured values. This technique, generally called regularization, suppresses the effect of noise in the measured data y~meas~on the entries in the weight vector w. Refer to \[[@B6]\] or \[[@B7]\] for a general introduction to regularization. If the \|\| \|\| norm is simply the sum squared error (or L2 norm), equation (2) can be solved analytically for any given value of Λ to: w = (^t^HH + Λ)^-1t^H y~meas~ (3) The optimal value for Λ can then be determined by cross validation: The experimental data points corresponding to rows of (H, y~meas~) are randomly separated into training sets T~i~= (H, y~meas~)~train\ i~and blind sets (H, y~meas~)~blind\ i~. For a given Λ, equation (3) is used to calculate w~i~for each training set T~i~. These w~i~can then be used to make predictions for their corresponding blind sets. Summing the distances for all blind predictions gives the cross validated distance Φ: Φ(Λ) = Σ~i~\|\| H~blind\ i~w~i~(Λ) -y~blind\ i~\|\| (4) Minimizing Φ as a function of Λ therefore corresponds to minimizing the cross validated distance by \'damping\' the influence of coefficients in w which are susceptible to noise. As the resulting optimal value for Λ and the corresponding w~i~are somewhat influenced by the split into training and blind sets, the same procedure is repeated several times with different splits, which is called bagging \[[@B8]\]. The final w is generated by averaging over all optimal w~i~generated in the cross validation and bagging repeats. Results and discussion ====================== The following sections describe specific properties of the SMM method, and are meant to serve as a guideline when and how to apply it. Additional data validating the SMM algorithm and comparing it with other prediction methods can be found in previous publications \[[@B1]-[@B3]\]. Linear model ------------ If no pair coefficients are incorporated, the output vector w of the SMM method is a standard \'scoring matrix\', which quantifies the contribution of each residue at each position in the input sequence to the prediction. Such a matrix is easy to interpret and analyze without requiring any additional software or expert knowledge of how the matrix was generated, which is especially important when communicating results to experimentalists. Several methods predicting peptide binding to MHC molecules take this approach, e.g. \[[@B9]-[@B12]\], and a comparative study showed that simple statistical methods to generate matrices can perform better than more complex artificial neural networks if the amount of data is limited \[[@B13]\]. Using such a linear model implicitly assumes that the influence of residues at each position in the sequence on the measured value can be considered independent and additive. This has to be a reasonable first approximation in order to successfully apply the SMM method, even if pair coefficients are incorporated. This is the main difference to general learning algorithms such as neural networks, which can in principle model any functional relationship between sequences and measurements. Quantitative data ----------------- The experimental measurements that serve as input to the SMM method and the predicted output are quantitative, not binary. For example, in the case of peptide binding to MHC molecules, IC50 values quantifying binding affinities are used, and not a classification into binding and non-binding peptides. If different representations of the quantitative data are possible, such as either IC50 or log(IC50) values, a representation should be chosen in which the y~meas~values approximately follow a normal distribution. Otherwise the SMM predictions, which are sums of independent contributions and therefore roughly normally distributed themselves, will not be able to fit the experimental data well. In the case of binding affinities, this means log(IC50) values should be used, as IC50 values themselves are usually log normal distributed. Noisy data ---------- Experimental measurements inevitably contain noise. This will cause problems when building models that take the measured values to be exact. Accordingly, the SMM method incorporates a regularization parameter Λ (equation 2), which corresponds to preferring a simpler solution with \'smooth\' values for w to one that exactly reproduces observations. In the first SMM applications \[[@B1],[@B2]\], Λ was a scalar, in which case this approach is called Ridge regression or zero-order regularization. Choosing a scalar value implicitly assumes that the level of signal to noise is roughly the same at each position in the input sequence. In the current version, Λ can also be chosen as sequence position dependent, which is sometimes called local Ridge regression. As shown for an example in Figure [3](#F3){ref-type="fig"}, this makes Λ into a diagonal matrix in which all Λ~i~values belonging to residues at the same sequence position are set to the same value. For a sequence of length N, there are N different Λ~i~values. For a number of training sets containing peptide binding data to MHC molecules, we compared the prediction performance achieved using a position specific matrix Λ~i~to a scalar Λ. The position specific regularization nearly always resulted in better predictions. The difference was especially large if the influence of different sequence positions varied greatly (data not shown). Bounded data ------------ Any experimental technique generates measured values contained within a finite range. For example, in many biological experiments a \"zero\" measurement usually means that the actual value is below the experimental resolution, not that the actual value is 0. Similarly, very large values beyond the expected sensitivity limit are no longer quantitatively accurate. These data points at the upper or lower boundaries of the sensitivity range do not convey the same information as quantified values, but they still do contain information. In the case of MHC binding data available to us, approximately 20% of peptides fall in this category. The SMM method is, to the best of our knowledge, the only method designed to extract information from such boundary values. This is done by means of the novel L2~\<\>~norm, illustrated in Table [1](#T1){ref-type="table"}. For example, if an experimental measurement corresponds to an upper boundary y~meas~\> z, and the predicted value is greater than z, then the distance between y~meas~and y~pred~is zero. This norm has the useful property that any analytical solution according to the L2 norm can be converted into a solution according to the L2~\<\>~norm through an iterative process: First, all measurements including boundary values are treated as normal values, and the solution using the L2 norm is found. In a second step, for each y~pred~, y~meas~value pair for which the L2~\<\>~norm would be zero, the y~meas~value is set to its corresponding y~pred~value. For these y~meas~\* values, the distance L2(y~pred~, y~meas~\*) = L2~\<\>~(y~pred~, y~meas~). These y~meas~\* values are then used to solve again according to the L2 norm. This process is repeated until the y~meas~\* values no longer change, as illustrated in Figure [4](#F4){ref-type="fig"}. Randomized peptide library data ------------------------------- As stated before, the experimental data used as input for the SMM method consists of same length amino-or nucleic acid sequences associated with a quantitative measurement. When designing an experiment, the selection of sequences to test can introduce bias into the training data, for example by over-or under-representing residues at specific sequence positions. One way to avoid this is the use of randomized peptide libraries, also known as positional scanning combinatorial peptide libraries, which are mixtures of peptides of the same length. In a given library, all peptides have one residue in common at a fixed position and a random mixture of amino acids at all other positions. For example, the library XAXXXXXX contains 8-mer peptides with a common Alanine at position 2. Such libraries can be used to directly measure the influence of their common residue, by comparing their measured process efficiency to that of the completely randomized library XXXXXXXX. In the case of 8-mer peptides, 160 library experiments are sufficient to characterize the influence of each residue at each position. The results of such a complete scan can be summarized in a scoring matrix. This approach has been used successfully in many different experimental systems \[[@B14]-[@B17]\] A novel feature of the SMM method is that it can combine data from these two sources. When the SMM algorithm is given experimental data from individual peptides and from a randomized library summarized in a scoring matrix mat~lib~, it simply subtracts the values predicted by mat~lib~from each individual peptide measurement. These y\'~meas~= y~meas~- y~pred,lib~values are then used to generate a second scoring matrix mat\'. The final SMM scoring matrix is simply the sum of the two: mat~combined~= mat\' + mat~lib~. Figure [5](#F5){ref-type="fig"} compares the performance of this combined approach to that of a prediction based on peptide or library experiments alone. If enough peptide data is present (roughly the same number as matrix parameters), the combined prediction is better than that of the library matrix alone. At all data points, the combined prediction is better than that using the peptides alone. Importantly, this simple strategy of subtracting library predictions can be used in combination with any prediction method, and is likely to generate similar results, as it effectively increases the training set size. To visualize the prediction quality associated with the distances reported in Figure [5](#F5){ref-type="fig"}, Figure [6](#F6){ref-type="fig"} depicts a scatter plot of the predicted and measured binding affinity for individual peptides corresponding to the data point with the lowest distance in Figure [5](#F5){ref-type="fig"}. Introducing pair coefficients ----------------------------- Pair coefficients quantify the contribution of a pair of residues to the measured value that deviates from the sum of their individual contributions. The form of equation (1) remains unchanged if pair coefficients are introduced in the same binary notations as the individual coefficients. Figure [2](#F2){ref-type="fig"} gives an example how a set of nucleic sequences is transformed into a matrix H, if two such pair coefficients are taken into account. Note that the number of possible pair coefficients is very large. For a sequence of three nucleic acids, there are already (3\*4) \* (2\*4) / 2 = 48 pair coefficients. For a 9-mer peptide, (9\*20) \* (8 \* 20) / 2 = 14400 pair coefficients exist. To the best of our knowledge, only the SMM method and the additive method \[[@B18]\] explicitly quantify the influence of pair coefficients. Since most training sets contain only a few hundred measurements, determination of the exact values of all pair coefficients is not feasible. To overcome this difficulty, the SMM algorithm limits the number of pair coefficients to be determined. First, only coefficients for which sufficient training data exists are taken into account. As a rule of thumb, 5 sequences containing the same pair of residues at the same positions have to be present in the training set for a pair coefficient to be considered. In a second filtering step, only pair coefficients for which the information in the training set is reasonably consistent are retained. In the previous SMM version, this used to be determined by multiple fitting of pair coefficients and discarding those for which a sign change was observed. In the current version, a much faster approach is used. First, a scoring matrix is calculated for the training set without any pair coefficients. Then, for each pair coefficient the predicted and measured values for the sequences containing it are compared. Only if a large enough majority (\>60%) of measured values are above or below the matrix based predictions is the pair coefficient retained. The remaining pair coefficients are determined in complete analogy to the scoring matrix itself, but with a scalar Λ value. We tested the effect of incorporating pair coefficients on prediction quality compared to using a scoring matrix alone for a number of training sets containing peptide to MHC binding data. The pair coefficients showed a consistent positive contribution for large training sets, which comprise more measurements than 1.5 times the number of scoring matrix coefficients. However, the improvement is rather small, as reported before in \[[@B1]\]. This makes it reasonable to ignore pair coefficients if the simplicity of a scoring matrix is more valuable than a small improvement in prediction quality. If higher order sequence interactions such as those described by pair coefficients are expected to be the dominant influence on experimental outcomes, other prediction methods may be better suited than the SMM method. For example, by choosing a different sequence representation than the binary vectors, the information in the training set can be generalized, thereby effectively reducing the degrees of freedom in the input parameters \[[@B19]\]. This allows applying general higher order learning algorithms such as artificial neural networks even with limited input data. Conclusion ========== The SMM method generates quantitative models of the sequence specificity of biological processes, which in turn can be used to understand and predict these processes. It has previously been shown to perform very well compared to other prediction methods and tools for three specific types of experimental data \[[@B1]-[@B3]\]. However, it is difficult to generalize a comparison between different methods, due to two main problems. First, the training data sets utilized in different studies are often not available, so that when comparing tools generated by different methods it is often unclear when good performance is due to a superior method or a better (larger) set of training data. Second, generating tools from the same training set can be difficult, because publications that make the tools available, often only describe the basic principle of the method used. To overcome this second obstacle, we herein presented a computer program implementing the SMM method. Significant effort was devoted to ensuring that the program is robust, documented, cross platform compatible and generates reasonable output without requiring additional parameters. Also, any commercial libraries previously utilized were removed to allow free distribution of the code. This will permit any interested user to apply the SMM method with reasonable effort, allowing for the most important validation: application to scientific practice. Finally, we believe that two strategies demonstrated in this manuscript will be valuable in combination with other prediction methods as well. First, our strategy for the inclusion of experimental data gathered with randomized peptide libraries can be directly transferred to any prediction method. When other experimental data is limited and data from a combinatorial library is available, this should always have a positive effect on prediction quality. Secondly, the L2~\<\>~norm can be applied as an error function for other prediction methods. This will increase the amount of training data effectively available to prediction methods requiring quantitative input, by enabling them to handle experimental boundary values. Availability and requirements ============================= Project name: smm Project home page: <http://www.mhc-pathway.net/smm> Operating system(s): Platform independent Programming language: C++ Other requirements: The Gnu Scientific Library (GSL) has to be installed License: None for the smm code, but GSL requires the GNU GPL license Any restrictions to use by non-academics: None Authors\' contributions ======================= BP conceived of this study and implemented the software, AS participated in the design of the software and helped to draft the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A .zip file containing the SMM source code, documentation and examples for windows systems. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A .tar.gz file containing the SMM source code, documentation and examples for linux systems. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ The authors want to acknowledge the help of Emrah Kostem in adopting the source code to g++ and John Sidney for critically reading the manuscript. This work was supported by the National Institutes of Health Contract HHSN26620040006C Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Input training data.**The \<TrainingData\> element consists of a series of \<DataPoints\> (2). Each contains a sequence and a measurement value. The characters allowed in \<Sequence\> are specified in \<Alphabet\> (1), and the number of characters has to correspond to \<SequenceLength\> (1). In this example, \<Alphabet\> and \<SequenceLength\> specify 8-mer peptides in single letter amino acid code. Each measurement can optionally be associated with a threshold (3) that can either be \<Greater\> or \<Lesser\>, signaling that the measurement corresponds to an upper or lower boundary of measurable values. ::: ![](1471-2105-6-132-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Converting sequences into matrices.**Input sequences of three nucleic acids each are converted to rows of a matrix H. The first column of each row is set to 1, which serves as a constant offset added to each prediction. Columns A1 to T1 contain a binary representation of the first residue in the sequence, in which all columns are set to zero except the one corresponding to the residue. The same is repeated for the second and third residue in the sequence in columns A2 to T2 and A3 to T3. The two last columns G1A3 and A2A3 contain pair coefficients explained at the end of the results section. They are set to one if the two specified residues are present in the input sequence at the two specified positions and zero otherwise. Multiplying matrix H with the weight vector w results in a vector y~pred~of predicted values for the sequences. Rows A1 to T3 of vector w are commonly written as a \'scoring matrix\' which quantifies the contribution of each possible residue at each position to the prediction. Rows G1A3 and A2A3 of vector w quantify the impact of the pair coefficients. ::: ![](1471-2105-6-132-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Sequence position dependent regularization.**Example for the regularization term ^t^w Λ w in equation (2). The weight vector w corresponds to a scoring matrix for three nucleic acids as in Figure 2, but without pair coefficients. The diagonal matrix Λ has three different values Λ1, Λ2 and Λ3 effecting values in vector w corresponding to sequence positions 1, 2 and 3. There is no regularization penalty on the \'Offset\' value. ::: ![](1471-2105-6-132-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Iterative model fitting using the L2~\<\>~norm.**In this example, the model is a linear function which is fitted to a set of paired values (x, y~meas~). For two of the x values (x = 3 and x = 5), the measured values are thresholds (Greater 3). Fitting a linear function to paired values according to the L2 norm corresponds to the standard linear regression. **A**depicts the model fit (straight line) to the measured values (black boxes), ignoring any thresholds. For x = 5, the model value y~pred~taken from the regression curve is 3.4, above the measured threshold value 3. Therefore, in the next iteration the y~meas~\* value is set to the model value 3.4. **B**shows the new linear regression with the adjusted y~meas~\* values. This procedure is repeated until the y~meas~\* values no longer change (8 iterations, panel **C**). ::: ![](1471-2105-6-132-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Combining peptide and library data improves prediction quality.**A set of 449 9-mer peptides with measured affinities for TAP taken from \[20\] was split into 5 blind sets. For each of these blind sets, predictions where made from different size subsets of the remaining peptides. The x-axis depicts the number of peptides in these subsets used for generating predictions using either the peptides alone (circles) or in combination with data from a combinatorial peptide library (squares). The dashed line displays the prediction of the library alone, which was taken from \[15\]. The y-axis depicts the L2~\<\>~distance of the predictions for the combined 5 blind sets. ::: ![](1471-2105-6-132-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Visualization of prediction quality.**Scatter plot of predicted vs. measured affinity for peptide binding to TAP. The depicted prediction corresponds to the data point in figure 5 with the lowest cross-validated distance, in which 350 peptides and the peptide library were used for training. ::: ![](1471-2105-6-132-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The L2~\<\>~norm ::: **Measurement** **L2~\<\>~distance** ------------------------------------- ---------------------- ---------------------- Quantitative (no threshold) (y~pred~-y~meas~)^2^ (y~pred~-y~meas~)^2^ Upper boundary (threshold: greater) 0 (y~pred~-y~meas~)^2^ Lower boundary (threshold: lesser) (y~pred~-y~meas~)^2^ 0 :::

PubMed Central

2024-06-05T03:55:58.320608

2005-5-31

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173087/", "journal": "BMC Bioinformatics. 2005 May 31; 6:132", "authors": [ { "first": "Bjoern", "last": "Peters" }, { "first": "Alessandro", "last": "Sette" } ] }

PMC1173088

Background ========== Models for codon substitutions are used in computational biology for a wide range of applications such as reconstructing ancestral DNA sequences, determining Ka/Ks ratios to identify periods of adaptive evolution and aligning coding DNA. Methods for estimating mutation matrices from observed substitutions in sequence alignments of proteins were established by Dayhoff \[[@B1]\]. These matrices contain the probabilities of amino acid mutations for a given period of evolution and have long been used for scoring protein sequence alignments, evolutionary studies and homology searches. More than a decade ago, when large-scale protein databases became established, several amino acid substitution matrices based on observed mutation counts in protein alignments were constructed \[[@B2]-[@B4]\], replacing the original Dayhoff matrices that were based on relatively few alignments. However, to describe substitutions at the codon level, parameterized models have been developed \[[@B5],[@B6]\] and are widely used in the study of molecular evolution. In the same way that the growth of protein databases allowed refined construction of amino acid substitution matrices, the recent increase of nucleotide sequence data made it possible to apply these methods at the codon level. The matrices presented here were constructed using the approach of Gonnet \[[@B4]\]. This implies that pairwise alignments using full dynamic programming \[[@B7],[@B8]\] were used in order to count the observed transitions between codons. The sequence data was taken from the complete vertebrate genome databases of *ENSEMBL*\[[@B9]\]. Results ======= The additional files contain the 64 × 64 matrices presented here. Various aspects of the matrices will be discussed in this section. We present the matrix with the exact counts of the observed substitutions ([Additional file 1](#S1){ref-type="supplementary-material"}), the matrix containing the substitution probabilities derived therefrom ([Additional file 2](#S2){ref-type="supplementary-material"}) and a matrix containing similarity scores for all possible substitutions ([Additional file 3](#S3){ref-type="supplementary-material"}). Substitution counts ------------------- The 17,502 alignments that have been used to construct the matrix presented here contained 8.3 million aligned codon pairs. From each of these aligned pairs, the number of each of the 3730 (61 × 61 + 3 × 3) possible substitutions were counted. The difference between the numbers of frequent and rare substitution is high with the most frequent substitution (the GAG identity) being observed 153,040 times, and the rarest substitution, between TGG and GAG, being counted only 45 times, about 3400 times less often. To estimate the precision of the count of the rarest substitution, a binomial distribution of the counts can be assumed. The substitution with minimal count *c*~*min*~occurs with probability *p*= 45/(8.3·10^6^) = 5.4·10^-6^. The variance of a binomial distribution is *σ*^2^= *N*(1 - *p*)*p*and thus for very small *p*, the variance of *c*~*min*~is almost equal to *c*~*min*~= 45 and the standard deviation *σ*is 6.7. This means that although a very large amount of data is used to construct the matrix, it is just enough to produce codon counts that are of a tolerable accuracy for rare transitions. Only a further increase of high-quality genomic data will allow the clustering of the data into specific subsets. These possibilities will be discussed below. Substitution probabilities -------------------------- The mutation matrix *M*constructed from the counts contains the substitution probabilities for the individual codons. Entry *M*~*i*,*j*~gives the probability that codon *j*mutates to codon *i*. (As a consequence, each column of *M*sums to 1). A convenient measure to express the amount of mutations in a matrix is the percentage identity ![](1471-2105-6-134-i1.gif) with *f*~*i*~being the natural frequencies of the codons. For the matrix reported here, *p*is .35, meaning that in the alignments used, 65% of the codons have undergone substitution (to any other codon, thereby involving up to three nucleotide changes). It is also possible to calculate the percentage of identical amino acids resulting from this matrix: ![](1471-2105-6-134-i2.gif) In the second sum, *i*goes over all codons that code for the same amino acid as *j*does. The result for *p*~*AA*~is .69, therefore 31% of the amino acids are expected to mutate. This allows the determination of the relationship between the codon substitution matrices and amino acid substitution matrices, because the amino acid PAM distance can be derived from the percentage of amino acid identity. Analogously to the definition of 1 PAM, 1 Codon-PAM can be defined as the distance at which 1% of the codons undergo substitution. Again, a codon substitution can involve up to three nucleotide base changes. The substitution matrix for any distance *d*is approximated by raising the 1 CodonPAM matrix to the power of *d*. The relationship between CodonPAM, PAM and *f*2 is shown in Figure [2](#F2){ref-type="fig"}. It shows that amino acid PAM increases almost linearly with CodonPAM. The curve is slightly steeper for the low distances and flattens with increasing distance. The amount of synonymous substitution decays from 1 to .51 (f2 being a measure of synonymous mutation described in the Methods section). Mutation scores --------------- Since the substitution probabilities are influenced by the codon frequencies, it is not possible to see directly which substitutions occur more than expected and which occur less. This issue is corrected in the scoring matrix *D*, where entry *D*~*i*,*j*~expresses how much more likely it is that codons *i*and *j*were derived from a common ancestral codon compared to a random pairing of them. A higher score for a transition means that this transition is indeed more likely than one with a lower score. The scoring matrix is symmetric, i.e. the transition from codon *i*to *j*has the same score as the transition from *j*to *i*. Table [1](#T1){ref-type="table"} displays average scores for different categories of substitutions. It confirms the fact that synonymous substitution scores are generally higher than non-synonymous scores. But it can also be seen that as more nucleotides change, the scores become lower. Synonymous substitutions in which all three bases change, have lower scores than non-synonymous substitutions with only one base change. Discussion ========== Synonymous mutations -------------------- It has been observed that different genes have different Ka/Ks ratios and therefore the fraction of synonymous substitutions will differ between different gene pairs having a certain PAM distance. This is because there are no strong selective constraints on synonymous substitutions and therefore the number of these substitutions accumulates in a clock-like manner \[[@B10]\] while the number of nonsynonymous substitutions is governed by functional constraints. Increasing amounts of genomic data would allow the construction and comparison of matrices from alignments with differing amounts of synonymous and non-synonymous substitutions, representing a two-dimensional array of matrices, where one dimension is the evolutionary distance and the other corresponds to the amount of synonymous change. Unfortunately, the current size of the nucleotide databases does not yet allow such a clustering of the available data. Instead, the alignments selected to construct the matrices were filtered to fall within a window of synonymous mutations, thereby excluding the most extreme values. (see the Methods section for details). Figure [1](#F1){ref-type="fig"} shows the distribution of the alignments\' f2 values. Range of applicability ---------------------- One possible application of scoring matrices is protein and coding DNA alignment. In order to compare alignments based on amino acid substitution matrices and the codon matrices presented here, the likelihood scores are compared. Since these scores express the probability ratios of the two sequences having evolved from a common ancestor to them being aligned by random chance, they serve as a confidence measure of an alignment. The higher the score, the higher the likelihood that the alignment is by reasons of ancestry than by random chance. As the likelihood scores serve as an indicator of alignment quality, orthologous sequences for species pairs of various distance and classes were used to determine when codon matrix based alignments produced higher scores than amino-acid PAM matrices. Table [2](#T2){ref-type="table"} displays for several pairs of species, the number of orthologs used to perform the alignment analysis, the average PAM distance between these orthologs (found by selecting the highest-scoring PAM matrix) and the average ratio of codon based scores to amino acid based scores. A number greater than 1 means that on an average, the codon based scores were higher. The result is that for closely related species, the codon based scores are always higher, but the more distant two species are, the better the performance of the amino acid based alignments. An interesting point is that although codon mutations in different sets of species were found to be significantly different (χ^2^tests, data not shown here), the above finding holds not only for the vertebrates, from which the matrices were constructed, but also for the invertebrates, yeasts and even bacteria. From the results in Table [2](#T2){ref-type="table"}, a PAM distance smaller than 50 would favor the use of codon substitution matrices instead of amino acid based matrices. Conclusion ========== Because codon substitution matrices are substantially bigger than amino acid matrices and also because some of the substitutions are extremely rare compared to the most frequent ones, large amounts of genomic data are necessary to model the transitions accurately. The 17,502 alignments used here produce enough aligned codons to fulfill this criterium, but do not allow further clustering of the data set in order to create more specific matrices. The codon substitution matrix presented here is to our knowledge the first based entirely on empirical data and can serve in many fields of computational biology. We have found that at long distances, when the synonymous mutations have reached saturation, amino acid matrices are better suited for alignments and long-distance homology searching. Codon mutation matrices have advantages for alignment up to medium evolutionary distances and for usages that require DNA such as ancestral reconstruction of DNA sequences and the calculation of Ka/Ks ratios. Methods ======= The basic methods to create scoring matrices are well established. The construction of codon substitution matrices is analogous to that of amino acid transition matrices. The main difference lies in the fact that codon matrices are much larger (4096 elements (64 × 64) instead of 400 elements (20 × 20)). In addition, the stop codons need special consideration. Substitutions between stop codons and sense codons are assumed to be very rare because the effect of such substitutions on the function of the protein would probably be very serious thus the chance of acceptance is very small and usually limited to the 3\' end of the nucleotide sequence. This makes it almost impossible to observe such events. Therefore, these substitutions are not included in the matrices presented here. Substitutions between stop codons, however, are counted and thus also contained in the matrices. This means that the 64 × 64 matrices are block diagonal composed of a 61 × 61 matrix for the coding codons and a 3 × 3 matrix describing substitutions between stop codons. Using orthologs --------------- The matrices are constructed from pairwise alignments of orthologous sequences from five vertebrates -- human (*Homo sapiens*), mouse (*Mus musculus*), chicken (*Gallus gallus*), frog (*Xenopus tropicalis*) and zebrafish (*Brachydanio rerio*). The complete genome databases from *ENSEMBL*\[[@B9]\] were used for this purpose. Using only orthologs has the advantage that no gene is overrepresented in the data set. This is because a particular gene can have many paralogous genes in a genome, but we allowed at most one ortholog per other genome. Circular tours -------------- When counting the substitutions in alignments from all pairs of species, substitutions that occurred early in the tree are counted more often than those that happened later, because paths between two species include the branches near the root more often than those near the leaves. This bias can be prevented by using only species pairs along a circular tour. This way every branch of the tree (and therefore every substitution that ever happened in the history of the genes) examined, is counted at most twice. Concretely, this means that only the orthologs between human and mouse (3107 pairs), mouse and chicken (3691 pairs), chicken and frog (3671 pairs), frog and fish (3441 pairs) and fish and human (3592 pairs) are counted, resulting in 17,502 alignments. Counting substitutions ---------------------- These alignments must fulfill several criteria: 1) they should all be of similar evolutionary distance because substitution probability depends on evolutionary distance. A trade-off exists for the acceptable range of distances as including a broad range of distances, increases the amount of data but at the same time blurs the distance specific information. 2) The alignments must be of a distance that is high enough to allow the observation of rarer substitutions. 3) There must be enough alignments to have statistically significant data for the rare substitutions. After some observations performed on a subset of the data, a distance range of 25 to 60 PAM (57% to 78% identity) of the protein alignments was found to best satisfy these criteria. Another selection criterium was based on the amount of synonymous substitutions between the sequences to eliminate saturation effects from the observed synonymous substitutions. One way to estimate the amount of synonymous substitution is *f*2, the percentage of conserved synonymous codons at two-fold redundant amino acid sites. Two identical sequences have an *f*2 value of 1 and it will decay to a value near .5 for increasing amounts of substitutions. An *f*2 range between .50 and .95 was found to exclude the most extreme cases of synonymous substitutions while leaving enough alignments to fill the matrices. Figure [1](#F1){ref-type="fig"} shows the distribution of *f*2 values for all alignments within the PAM range of 25 to 60. The sequence pairs with *f*2 values between .50 and .95 are shown in green and were used to construct the matrices, while the alignments corresponding to the red bins were discarded. Full dynamic programming \[[@B4],[@B7],[@B8]\] was employed in order to construct the alignments. The DNA alignment was obtained by mapping the coding DNA to the aligned proteins. Directly aligning the DNA was not yet possible, because no a priori knowledge about codon similarities was assumed. However, in the refinement steps (see further below), the DNA itself was aligned using the codon substitution matrices from the previous refinement round. Once the sequences were aligned, the actual substitution matrices could be computed. This included counting the observed codon substitutions in the collected alignments and storing them in a 64 × 64 count matrix *C*. Since the direction of a substitution is not known, for each observed substitution between codons *i*and *j*, *C*~*i*,*j*~as well as *C*~*j*,*i*~is increased by 1/2. Insertion or deletion sites were ignored since they provide no information about actual substitutions. From the count matrix, the mutation matrix was derived according to equation 3: ![](1471-2105-6-134-i3.gif) ### Calculating similarity scores The transition scores for a given substitution matrix express the relative probabilities of two codons originating from a common ancestor compared to the probability of them being paired by random chance. The logarithm is taken to make the scores additive, thereby speeding up the computation of alignment scores. *D*is calculated as ![](1471-2105-6-134-i4.gif) where *f*~*i*~is the frequency of codon *i*in the observed data set. The factor 10 is used for purely historical reasons. In an iterative process, the sequence pairs were aligned again with dynamic programming \[[@B7],[@B8]\] but this time directly on the codon sequences using the substitution matrices obtained before. Exponentiation of the mutation matrix was used to approximate matrices for different evolutionary distances \[[@B12]\], allowing a maximum-likelihood estimation of the best-fitting matrix to align the sequences. From these new alignments new mutation matrices and finally scoring matrices were constructed in the way described above, until after six iterations a sufficient convergence of the matrix was reached. Authors\' contributions ======================= GMC initiated and guided this project and contributed to the programming. AS did the majority of the programming and analysis. Some of the work was based on previous research by GHG. All work was supervised by GHG. AS and GMC drafted the manuscript which was approved by GHG. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 The 64 × 64 matrix containing the observed substitution counts as described in the results section. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 The 64 × 64 matrix containing the substitution probabilities. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 The 64 × 64 matrix of the transition scores. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ GMC wishes to acknowledge Michael T. Hallett, Jan Norberg, and Xianghong Zhou for starting this project. The authors wish to thank Peter von Rohr, Markus Friberg, Daniel Margadant, Christophe Dessimoz and the anonymous reviewers for careful reading and helpful suggestions. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **f2 histogram.**Histogram of the f2 values from the 17,502 alignments used to construct the matrix. ::: ![](1471-2105-6-134-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### CodonPAM vs PAM and f2 ::: ![](1471-2105-6-134-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Analysis of the scores. Average scores for different categories of substitutions. The stop codons are excluded from this analysis. ::: n Substitutions Avg. Score ---------------------------- ----------------- ------------ Identity 61 12.9 Synonymous: all 87 8.7 Synonymous: 1 base change 67 10.0 Synonymous: 2 base changes 14 6.6 Synonymous: 3 base changes 6 -1.7 Non-syn.: all 1743 -7.3 Non-syn.: 1 base change 196 -1.3 Non-syn.: 2 base changes 770 -5.9 Non-syn.: 3 base changes 777 -10.3 All Substitutions 1891 -5.9 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Range of applicability. Ratios of likelihood scores for amino acid and codon based alignments for orthologs between several species pairs, where *N*is the number of orthologs used. ::: N Avg. PAM Scores ratio ---------------------------- ------- ---------- -------------- *Homo sapiens* vs. *Mus musculus* 14655 17.4 1.150 vs. *Gallus gallus* 9272 29.3 1.060 vs. *X. tropicalis* 9953 39.1 1.026 vs. *B. rerio* 7507 43.7 1.013 *Drosophila melanogaster* vs. *A. gambiae* 5059 57.3 .995 vs. *H. sapiens* 3371 77.5 .959 vs. *C. elegans* 2156 88.8 .945 *Saccharomyces cerevisiae* vs. *C. glabrata* 3467 52.7 1.002 vs. *A. gossypii* 2909 61.4 .978 vs. *H. sapiens* 1187 94.1 .931 *Escherichia coli* vs. *E. coli*strain O6 3156 2.0 1.323 vs. *Salmonella typhi* 2557 14.2 1.067 vs. *P. aeruginosa* 1234 71.6 .980 vs. *B. japonicum* 765 90.2 .959 :::

PubMed Central

2024-06-05T03:55:58.323706

2005-6-1

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173088/", "journal": "BMC Bioinformatics. 2005 Jun 1; 6:134", "authors": [ { "first": "Adrian", "last": "Schneider" }, { "first": "Gina M", "last": "Cannarozzi" }, { "first": "Gaston H", "last": "Gonnet" } ] }

PMC1173089

Text mining? \...\...\...\...\...\....Why bury it first and then mine it again? =============================================================================== Recently, Sir Tim Berners-Lee, the inventor of the Web, said: \'Life sciences are particularly suitable for pioneering the Semantic Web. For example, within drug discovery, many databases and information systems used by drug researchers are already in, or are ready to be transformed to, machine-readable formats\' \[[@B1]\]. Well, we need our breed of optimists to drive things, and he is obviously one of us. Too much to read ================ Bioinformatics increasingly consists of computer-aided meta-analysis of dispersed articles and database records to assist researchers in the interpretation of massive datasets. Epidemiological studies and high throughput technologies such as microarrays nowadays often lead to sets of potentially relevant papers being identified that surpass human capability for reading, interpretation and synthesis. Aggregating information from many records, followed by logical association of the concepts represented in the full dataset is what I generally refer to here as meta-analysis. Unfortunately, most of the information in the current information sources is not easily accessible, nor, due to ambiguity of the description of concepts in textual format, is it easily readable by computer programs. Semantics: a crucial addition to text ===================================== If all words had only one possible meaning, computers would be perfectly able to analyse texts. In reality however, words, terms and phrases in text are highly ambiguous. Knowledgeable people have few problems with these ambiguities when they read, because they use context to disambiguate \'on the fly\'. Even when fed a lot of semantically sound background information, however, computers currently lag far behind humans in their ability to interpret natural language. Therefore, proper semantic tagging of concepts in texts is crucial to make Computational Biology truly viable. Electronic Publishing has so far only scratched the surface of what is needed. Open Access publication shows great potential, andis essential for effective information mining, but it will not achieve its full potential if information continues to be buried in plain text. Having true semantic mark up combined with open access for mining is an urgent need to make possible a computational approach to life sciences. Databases and pathway tools are not enough ========================================== Of course, we have our curated databases, increasingly equipped with fancy analysis and visualisation tools. These databases hold established knowledge on molecules, their interactions and pathways. These databases are definitely useful tools to rapidly get a rough view on \'what may be switched on or involved\'. However, the inherent restrictions and downsides of such databases include the great efforts that are needed to keep the error level acceptably low and to keep them up-to-date. Furthermore, biological complexity is more than a composition of multiple, man-imagined pathways. Thus free text continues to be a crucial source of cutting edge information for scientists A further reason for the continued value of free text is that pre-digested information in databases only contains *explicit knowledge*and will therefore at best cough up \'what somebody already knew\', although I as an individual may not be aware of it. Real serendipitous finding of new things and the association between concepts beyond direct co-occurrence is not supported by most existing tools. However, text is a nightmare for computers ========================================== Unfortunately, free text records are a nightmare of ambiguity. Synonyms and homonyms riddle the records and, in particular, gene and protein names/symbols appear to be impossible to resolve with complete accuracy, based on their textual expression. In an ideal world, scientists would mention formal identifiers from recognized data bases such as EntrezGene or SwissProt in the text, rather than their favourite synonym of the molecule. With traditional search-based text retrieval tools, this problem was little more than a recurrent nuisance, and current traditional search engine providers probably couldn\'t care less, as their users are satisfied when they \'find what they want on page one\'. But now that computational meta-analysis of textual records is increasingly needed, and the underlying datasets may often consist of tens of thousands of papers, it becomes impossible to manually weed out problems relating to synonyms and homonyms. So, like it or not, we have to face the challenge of semantic enrichment ======================================================================== Before any meta-analysis algorithm can be meaningfully applied, semantic analysis and tagging of the underlying texts with unique identifiers for individual concepts is needed. Ideally, this process should be a one-off effort. In a perfect scenario, communication between scientists would take place entirely at the unique concept level. However, this is not reality. What was clear and straightforward in the mind of the researcher after completing the experiments gets lost in a variety of ambiguous expressions during the writing process and ambiguities are produced every minute. Therefore scientific writing can somewhat cynically be called *information burying.*Authors are actually inclined and stimulated to use variable expressions and synonyms, aphorisms and the like to make their article \'readable\'. This will not change in the foreseeable future and this means we have to face the challenge of semantic tagging of these texts. To keep the challenge manageable, I will restrain this argument to term identification and not include full analysis of the language structure of sentences as attempted by Natural Language Processing approaches. Will authors do it for us? ========================== Scientific text is dry enough as it stands. Should we therefore even contemplate attempts to force authors to make it semantically correct and uniform, according to strict nomenclature or worse, using structured data entry? I would argue that not only would this approach be completely unfeasible, given the creative and individualistic minds of the producers of scientific knowledge, but in addition, the result of hypothetical success would lead to a totally dull form of scientific literature. In the end, no matter what computational aids we offer, researchers will always keep reading for final evidence. So, we need to make things computer-digestible ***in the background.*** Will computers do it for us? ============================ With the emerging, improved named entity tagging and semantic tagging technologies, a far more elegant and practical solution is available or just around the corner. Correct (computer) recognition of the concepts denoted by words and phrases in free-text, and the semantic enrichment that comes with it, greatly facilitates direct meta-analysis of subsets of the literature. This pre-processing step also facilitates much more accurate cross-linking of information between articles and databases. If yes, why not? ================ It is surprising that publishers have so far taken almost no steps in this direction, even though it holds almost limitless potential, and it is clear that adding value to plain text is the only perspective publishers may have to charge for textual records in the future. Simple, technically feasible additions to current ontology-based semantic tagging software would allow the development of \'tag as you type\' tools that normalise and map ambiguous terms in text to the unique concept they denote and to the corresponding ontologies. The semantic tagging of new and existing text, notably the well over 14 million papers represented in Medline, seems a daunting perspective at first glance. Researchers are neither educated nor encouraged to take a semantic network approach when publishing their data. As a consequence, even decades after electronic publishing became commonplace, most papers are still ending up as dead, non-interactive pages rusting silently away in electronic archives until a search engine happens to be fed with the correct combination of keywords to retrieve them. Or\...will computers and authors do it together? ================================================ In my opinion, author-interactive publication tools should ***not***force the writing scientist to use specific terms from pre-structured lists and nomenclatures. Trying to do exactly that, is what made most efforts of nomenclature standardization ineffective. In contrast, the very valuable efforts of the HGNC\'s, Entrezgenes and SwissProts of this world should be used to *disambiguate terms on the fly*and only ask the author for assistance in the rare cases the ontology driven system can not make an informed decision about the meaning of a term, for example in case a homonym is used without sufficiently distinguishing context surrounding it. If a title such as: \'Epidemiological considerations of BSE\' is typed, the system is unlikely to be able to decide whether the author meant Breast Self Examination or Bovine Spongiform Encephalopathy. If no further context follows, the author could be prompted to resolve the ambiguity. However, if someone decides that he likes EBV more than Human Herpesvirus 4, or she prefers CD154 over TNFSF5, the author should not be forced to change that in the text, as long as the semantic tag added to that term in the background is linked to the unique concept identifier of the virus or the gene in the leading ontologies and nomenclature data bases (see figure [1](#F1){ref-type="fig"}). Obviously, some extra work will be asked from the author, but not anywhere near as much as most critics of the semantic web idea seem to expect. In general, the expectation would be that the vast majority of relevant concepts in biomedical text could be correctly tagged on the fly and in the background without bothering the author at all, other than to resolve the occasional ambiguity, and to review the overall markup once complete. Yes, but.. ========== Obviously, the critics of the Semantic Web approach will bring in their arguments about the difficulties with tagging the legacy documents and the issue of the rapidly developing knowledge, reflecting in ontology changes. Obviously they have a potential point: If only recently published documents are being tagged, computational tools that rely on the tagging will be very restricted in their resources. The technological approach proposed here will enable us to tag existing texts on the fly with high accuracy, by taking care of the \'easy tagging\' and interact with the author or the reader only to resolve the few remaining ambiguities. Moreover, as the tagging is not necessarily static, updated ontologies will lead to (optional) updated proposed tags each time the text is retrieved. Tagging would thus effectively be limited to texts that are still used. If an article is never retrieved by anyone it is probably of limited urgency to tag it. So, let\'s do it! ================= Given the enormous impact a collective semantic enrichment effort would make, particularly on many aspects of bioinformatics research and computational biology, this process should have started many years ago. The Open Access publishers, but *BMC Bioinformatics*in particular, should take the lead and promote the use of semantic tagging, probably starting at one of the time points at which stubborn scientists like me are most likely to accept any guidance, and that is when they submit a manuscript. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The part of text of this editorial that was marked in italics was fed to an existing semantic tagging tool and it can be seen in the figure how different expressions are mapped to the same concept number. ::: ![](1471-2105-6-142-1) :::

PubMed Central

2024-06-05T03:55:58.325923

2005-6-7

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173089/", "journal": "BMC Bioinformatics. 2005 Jun 7; 6:142", "authors": [ { "first": "Barend", "last": "Mons" } ] }

PMC1173090

Background ========== Dehydrogenases dedicated to L-tyrosine (TYR) biosynthesis comprise a family of TyrA homologs that have different specificities for the cyclohexadienyl substrate: ones specific for L-arogenate (AGN), ones specific for prephenate (PPA), and those that are able to use both \[[@B1],[@B2]\]. Figure [1](#F1){ref-type="fig"} illustrates the biochemical relationship of these specificities to divergent transformations beginning with chorismate (CHA) utilization and converging on TYR formation. Compounding this complexity, a given TyrA enzyme having any of the aforementioned cyclohexadienyl specificities may be specific for NAD^+^or NADP^+^, or may use both. This is consistent with a growing appreciation \[[@B3],[@B4]\] that different substrate specificities are often accommodated across a given protein family that nevertheless maintains a common scaffold of fundamental reaction chemistry. Even within the single category of broad TyrA specificity, there is a continuum ranging from examples where alternative substrates are accepted equally well to other cases where one substrate may be preferred by an order of magnitude or more. Table [1](#T1){ref-type="table"} provides a key to the nomenclature used to identify the various possible substrate-utilization combinations (both cyclohexadienyl and pyridine nucleotide) exhibited by TyrA proteins. The TyrA family is typical of many protein families in that its members have a relatively small core domain that is not highly conserved. As such, substantial challenges for bioinformatic analysis are posed. Here we have not only carried out a labor-intensive manual analysis, but we have also developed tools intended to facilitate and refine follow-on studies of this protein family in the genome era. The approaches implemented in this study with the TYR segment of aromatic biosynthesis hopefully can serve as a template for forthcoming integrant analyses of other pathway segments of aromatic biosynthesis, and indeed for metabolic subsystems in general. This manuscript contains three broad sections. First, the biochemical and enzymological complexity of the TyrA protein family is presented in terms of the diversity that exists in nature with respect to substrate specificity and the association of the core domain with other catalytic or regulatory domains. Secondly, the genomic colinear organization of *tyrA*genes with other genes is evaluated, i.e., *tyrA*is considered in its syntenic context. Thirdly, *tyrA*is evaluated in its context of regulation. These three sections are tied together in a framework of evolutionary perspective. Results and discussion ====================== Background of TyrA diversity ---------------------------- Our evolutionary analysis is limited by the amount of information that can be managed in a single study, with the focus fixed upon the domain *Bacteria*(due to the relative density of genome representation for *Bacteria*in the public databases). However, in order to show where future expansion of the analysis might lead, the selection of TyrA proteins in Fig. [2](#F2){ref-type="fig"} are from all three domains of life, i.e., *Bacteria*, *Archaea*, and *Eukarya*(lower eukaryotes and higher plants). For practicality of presentation, numerous orphan (i.e., without close relatives) TyrA sequences are not shown, and not all members of a given group are necessarily included. The main purposes of the radial tree shown in Fig. [2](#F2){ref-type="fig"} are: (i) to illustrate that TyrA proteins of major phylogenetic groupings are generally congruent with 16S rRNA groupings and (ii) to convey a snapshot visualization of the overall complexity of the TyrA protein family from the vantage point of its varied substrate specificities as well as its multiple fusion partners. As an illustration of the detailed information that follows, note that the TyrA sequences from the beta Proteobacteria at five o\'clock in Fig. [2](#F2){ref-type="fig"} form a cohesive cluster (termed a \'congruency group\'). In this clade there exists a proposed ancestral background of broad specificity where either AGN or PPA in combination with either NAD^+^or NADP^+^could be used. This profile of broad substrate use (which can be denoted as ~NAD(P)~TyrA~c~; see Table [1](#T1){ref-type="table"}) generally persists in the beta Proteobacteria. From this background, narrowed specificities for the AGN/NADP^+^couple emerged once in the lineage represented by *Nitrosomonas europaea*(Fig. [2](#F2){ref-type="fig"}; dark blue line), narrowed specificity for NAD^+^emerged once in species of *Neisseria*(orange line), and fusion of *tyrA*~*c*~with *aroF*(which encodes enolpyruvylshikimate-3-P synthase, the sixth enzyme in the common pathway of aromatic biosynthesis; see \[[@B5],[@B6]\] for nomenclature used) occurred recently within the *Burkholderia*lineage. These character-state transformations appear to occur with relative ease, and independent emergence of the same character states can be seen elsewhere in the tree. Phylogenetically congruent TyrA groupings ----------------------------------------- ### Multiple alignments of catalytic-core domains A phylogenetic tree is only as good as the input alignment. An optimal multiple alignment of TyrA homologs requires a trimmed set of sequences that corresponds to the catalytic-core domain. Alignment of sequences with non-homologous N-terminal fusions (such as with chorismate mutase• (AroQ•), HisH~b~•, or plant transit peptides•; note the convention of using a bullet to indicate the fusion point of one domain with another domain) will make them appear to be more closely related than they actually are because residues in the non-homologous N-terminal regions find matches at random. Likewise, those TyrA sequences with C-terminal fusions (such as with •AroF, •ACT, or •REG) will appear to be anomalously close to one another. Even enzyme proteins that have much greater sequence conservation and amino-acid lengths than TyrA proteins cannot reasonably be expected to yield a protein tree that would be congruent over an extensive phylogenetic range with the overall 16S rRNA tree. However, if genome representation is sufficiently dense within a range of closely related organisms, 16S rRNA congruency with a given protein can be expected within that range of organisms provided that (i) the particular functional role has been retained and (ii) lateral gene transfer has not occurred to obscure the relationship. This expectation follows from the outcome of a detailed analysis of tryptophan-pathway proteins in *Bacteria*\[[@B7],[@B8]\]. ### Congruency within major clades TyrA sequences from higher-plant and yeast *Eukarya*form cohesive clusters. Genome representation among *Archaea*is still relatively limited. (Fig. [2](#F2){ref-type="fig"} does reveal, however, that genes encoding TyrA proteins in *Archaea*have experienced various catalytic- and regulatory-domain fusions at least as frequently as those in *Bacteria*). Eventual expansion of both the tryptophan-pathway and tyrosine-pathway analyses to *Archaea*should be quite interesting. The great majority of TyrA sequences available are from *Bacteria*, and one can see (by inspection of the major clades supported by high bootstrap values in Fig. [2](#F2){ref-type="fig"}) a qualitatively apparent congruence of TyrA-tree sub-sections with 16S rRNA expectations of vertical genealogy. Thus, all cyanobacteria possess a ~NADP~TyrA~a~type of TyrA enzyme, and this is a very cohesive grouping. A few of the larger cyanobacterial genomes have a co-existing second enzyme of the TyrA~c\_Δ~type (discussed in detail later). The low-GC gram-positive bacteria (*Bacillus/Staphylococcus/Enterococcus/Listeria*) exhibit the ~NAD~TyrA~p~pattern of specificity and also possess a C-terminal domain (ACT) of allosteric regulation. It is interesting that the TyrA~p~•ACT proteins of the *Streptococcus*lineage (at eight o\'clock in Fig. [2](#F2){ref-type="fig"}) differ from the main low-GC clade in possessing broad specificity for pyridine nucleotides (as indicated with black line color). The most parsimonious evolutionary conclusion would be that in the low-GC gram-positive grouping, acquisition of the ACT domain and narrowed specificity for prephenate preceded narrowed specificity for NAD^+^. Thus, the latter event occurred after divergence of the *Streptococcus*lineage from the remainder of the low-GC clade. Members of the subclass taxon *Actinobacteridae*(mostly actinomycetes) possess AGN-specific TyrA enzymes (light blue fill color in Fig. [2](#F2){ref-type="fig"}), but they separate into two distinct groups that correlate either with broad specificity for pyridine nucleotides (*Actinobacteridae*\_1) or a NAD^+^-specific pattern (*Actinobacteridae*\_2). The Proteobacteria are discussed immediately below. ### Proteobacteria By far the greatest genomic density available is for Proteobacteria, the group of *Bacteria*that includes purple bacteria and their relatives. The various divisions of Proteobacteria, as currently named, lack hierarchical equivalence. For example, the epsilon and delta divisions branch from much deeper positions on the phylogenetic tree than do the alpha Proteobacteria. As genome representation expands for epsilon and delta Proteobacteria, it is probable that these will subdivide to newly named groupings of approximate hierarchical equivalence with alpha Proteobacteria. The most recently diverged Proteobacteria are the beta and gamma divisions. From the combination of our previous analysis of tryptophan biosynthesis \[[@B7],[@B8]\], TYR biosynthesis (this paper), and other segments of aromatic biosynthesis (unpublished data), we find it useful to separate \"upper-gamma\" Proteobacteria from \"lower-gamma\" Proteobacteria (an \"enteric lineage\" with *Shewanella oneidensis*as approximately the most divergent member). This separation is because the beta Proteobacteria and the upper-gamma Proteobacteria exhibit a smooth continuity of relatively few evolutionary events with respect to aromatic biosynthesis, in striking contrast to extraordinarily dynamic evolutionary events in the lower-gamma Proteobacteria. As a consequence, the lower-gamma Proteobacteria are much more distinct (in terms of aromatic biosynthesis) from the upper-gamma Proteobacteria than the upper-gamma are from the beta Proteobacteria. Figure [2](#F2){ref-type="fig"} shows that alpha, beta and epsilon divisions of Proteobacteria form phylogenetically coherent clusters with respect to their TyrA proteins. Although delta Proteobacteria fall into two well-separated groupings denoted as Delta\_1 and Delta\_2, this should not be surprising since these groupings diverge at a deep level on the 16S rRNA tree where genome representation is poor. In addition, the *Myxococcus xanthus*TyrA sequence, currently an orphan (three o\'clock in Fig. [2](#F2){ref-type="fig"}), represents a third divergent lineage in delta Proetobacteria. In contrast to delta Proteobacteria, genomic representation for the gamma Proteobacteria is relatively good. Nevertheless their TyrA sequences separate into several well-spaced groupings, albeit for entirely different reasons. In this case, the split seen between two clades of these fairly close relatives (upper-gamma and lower-gamma) is attributed to particularly dynamic evolutionary events compressed into a relatively short time span in the lower-gamma Proteobacteria. (We refer to such a dynamic divergence as an evolutionary jump; see the next section.) Note that the allocation of upper-gamma and lower-gamma Proteobacteria to separate TyrA congruency groups is not the same as being incongruent. It is quite possible that as new genomes come on line, new and intermediate TyrA sequences may result in the merging of the foregoing two congruency groups (currently tyrosine congruency group 1 (TyrCG-1) and tyrosine congruency group 2 (TyrCG-2)). ### Comparison of tryptophan and tyrosine congruency groups Although the true extent of lateral gene transfer (LGT) at present must be described as intensely controversial, there is little doubt that any given organism is mosaic with respect to some unknown fraction of its gene repertoire. Our \"accounting\" system for keeping track of proteins that are faithful to the vertical genealogy is to formulate congruency groupings that are defined by congruence of given protein-tree clusters to a section of the 16S rRNA tree. Ultimately this information will reveal which organisms are \"pure\" with respect to the vertical inheritance of a given pathway or pathway segment. Our congruency groups are intended to be fluid, in that with the continued availability of new sequences, a previous orphan sequence may very well become the seed for a new congruency group. On the other hand, previously separate congruency groups have the potential to merge. (See Methods for more information.) The present tyrosine congruency groups are listed on the AroPath website \[[@B9]\]. Seven tryptophan congruency groups in *Bacteria*were previously formulated \[[@B8]\] based upon the correspondence of cohesive clusters in trees of Trp-protein concatenates with sections of 16S rRNA trees. The information input for formulation of tryptophan congruency groups is of greater quality than for tyrosine congruency groups because seven-protein concatenates could be used for the former. On the other hand, the broad information input supporting tyrosine congruency groups in this study is more comprehensive because of greater genome availability. Tryptophan congruency group 1 (TrpCG-1) corresponds perfectly with the organisms represented in TyrCG-1, these being the lower-gamma Proteobacteria (enteric lineage). The upper-gamma Proteobacteria (TyrCG-2) and the beta Proteobacteria (tyrosine congruency group 3; TyrCG-3) are represented by different tyrosine congruency groups. In contrast, the membership of tryptophan congruency group 2 (TrpCG-2) includes both the upper-gamma Proteobacteria and the beta Proteobacteria. The latter merging probably reflects the advantage conferred by the greater information content of the concatenated sequences used to define tryptophan congruency groups. Species of *Xylella*and *Xanthomonas*are usually referred to as gamma Proteobacteria. They probably represent an outlying deeply branching lineage, although trees based on concatenated strings of proteins \[[@B10]\] or 16S rRNA \[[@B11]\] position them with beta Proteobacteria. In any event, Trp-protein concatenate trees placed *Xylella*and *Xanthomonas*within TrpCG-2, which contains both upper-gamma and beta Proteobacteria. In contrast, the TyrA domains from *Xylella*and *Xanthomonas*were well separated (at about two o\'clock in Fig. [2](#F2){ref-type="fig"}) from those of any other organism. This might simply be due to the limited resolving power of a single protein in combination with too few close relatives. (Note that single Trp-protein trees sometimes failed to achieve the congruency-group placements that were resolved by seven-protein Trp concatenates \[[@B8]\]). An additional clue may be relevant. The TyrA proteins from the *Xylella/Xanthomonas*genera possess an ACT domain, which has not been observed in any other proteobacterial TyrA proteins thus far. In view of this, origin by LGT seems to be a distinct possibility, but with the important caveat that no likely genome donors are yet obvious on the criterion of sequence similarity. Perhaps more likely is the following possible explanation that postulates a basis for accelerated divergence. The TyrA domains of *Xanthomonas/Xylella*proteins have an indel structuring (insertions and/or deletions) that places them within the TyrA~c\_Δ~specificity subclass (see below). We suggest (see below) that such indel structuring reflects interaction of the core TyrA domain with an extra-domain extension. Thus, selection for amino acid changes accomplishing a new domain-domain interaction could account for accelerated divergence of the *Xanthomonas/Xylella*sequences on the TyrA tree (Fig. [2](#F2){ref-type="fig"}). Cohesive tryptophan congruency groups of the alpha Proteobacteria (tryptophan congruency group 3; TrpCG-3) and the cyanobacteria (tryptophan congruency group 4; TrpCG-4) match up well with the corresponding tyrosine congruency groups (tyrosine congruency group 4 (TyrCG-4) and tyrosine congruency group 8 (TyrCG-8), respectively). The TyrA proteins of epsilon Proteobacteria define a cohesive tyrosine congruency group (tyrosine congruency group 5; TyrCG-5), whereas the Trp-protein concatenates of epsilon Proteobacteria did not exhibit a coherent congruency group, due at least in part to LGT \[[@B8]\]. The delta Proteobacteria separate into two distinct tyrosine congruency groups: Delta\_1 (tyrosine congruency group 6; TyrCG-6) and Delta\_2 (tyrosine congruency group 7; TyrCG-7), as shown in Fig. [2](#F2){ref-type="fig"}. It is likely that corresponding tryptophan congruency groups exist (work in progress), but at the time of the Xie et al. study \[[@B8]\] only Trp-pathway protein concatenates for *Desulfovibrio vulgaris*(Delta\_2) and *Geobacter sulfurreducens*(Delta\_1) were available, and they were provisionally listed as \"orphans\". In the present work TyrA sequences from *Deinococcus radiodurans*and *Thermus thermophilus*are the sole members of tyrosine congruency group 12 (TyrCG-12). At the time of the Trp-pathway work, the genome of *Thermus*was unavailable and the *Deinococcus*concatenate was listed as an orphan. It is expected that the *Deinococcus*and *Thermus*concatenates will now seed a new tryptophan congruency group. Whereas tryptophan congruency group 5 (TrpCG-5) is defined by cohesive concatenates from actinomycete bacteria, the TyrA proteins from the same organisms separated into two distinct congruency groups. It is intriguing that this partitioning into two congruency groups correlates with narrowed specificity for NAD^+^(indicating an evolutionary jump) in one of the groups. The latter group (tyrosine congruency group 11; TyrCG-11) is denoted *Actinobacteridae*\_2 in Fig. [2](#F2){ref-type="fig"}, whereas tyrosine congruency group 10 (TyrCG-10) is displayed as *Actinobacteridae*\_1. The opposite scenario whereby a single tyrosine congruency group corresponds to split tryptophan congruency groups applies in the case of low-GC gram-positive bacteria. Whereas TyrA proteins form a single congruency group in these organisms (tyrosine congruency group 9; TyrCG-9), a small cluster of Trp-pathway concatenates from *Bacillus subtilis*, *B. stearothermophilus*, and *B. halodurans*(tryptophan congruency group 6; TrpCG-6) separate distinctly from the remaining organisms (tryptophan congruency group 7; TrpCG-7). The latter evolutionary jump reflects a dynamic scenario of tryptophan-pathway evolutionary events that include loss of one gene from the *trp*operon, insertion of the *trp*operon into a 6-gene *aro*operon to produce a supraoperon, and acquisition of the TRAP (tryptophan-activated RNA-binding protein) mechanism of regulation by an RNA-binding protein \[[@B7]\]. Tyrosine congruency groups and tryptophan congruency groups are maintained and updated at the AroPath website \[[@B12]\]. Distribution in nature of TyrA specificity subclasses for the cyclohexadienyl substrate --------------------------------------------------------------------------------------- Four qualitative classes of specificity for the cyclohexadienyl substrate populate the TyrA superfamily of homologs (Fig. [1](#F1){ref-type="fig"}). These include PPA-specific (TyrA~p~), AGN-specific (TyrA~a~), the broad-specificity cyclohexadienyl (TyrA~c~) dehydrogenases and a fourth class represented by an enzyme of antibiotic biosynthesis (PapC) that converts 4-amino-4-deoxy-prephenate to 4-amino-phenylpyruvate \[[@B13]\]. Representatives of each specificity class have been studied at molecular and genetic levels. TyrA family members sharing a given substrate specificity do not necessarily cluster tightly together, and assignment of substrate specificity to experimentally uncharacterized TyrA homologs is uncertain unless they exhibit very high amino acid identities with experimentally characterized TyrA proteins. In some cases we do not accept older literature reports without more recent verification. For example, the yeast *Saccharomyces cerevisiae*TyrA~**x**~was characterized as a TyrA~**p**~protein \[[@B14]\] long before it was recognized \[[@B15]\] that PPA preparations were often contaminated with AGN (an unknown compound at that time). Our collection of curated TyrA sequences at AroPath (see Table [3](#T3){ref-type="table"}) contains trimmed sequences that comprise catalytic-core domains. This collection was divided into two groups based on whether the sequences contained the relatively short N-terminal pyridine-nucleotide discriminator segment or the longer C-terminal cyclohexadienyl-substrate core segment. The sequences in the latter group were assembled into subgroups representing established substrate specificities (TyrA~a~, TyrA~p~and TyrA~c~) and were aligned separately to obtain overall consensus sequences for cyclohexadienyl-substrate core segments. The TyrA~c~group members from the lower-gamma assemblage of Proteobacteria (as well as from a few other lineages) were so distinctive that a fourth group (TyrA~c\_Δ~) was defined. This latter group is, in fact, the most divergent of the four. Figure [3](#F3){ref-type="fig"} shows a comparison of the four consensus sequences, with invariant anchor residues shaded yellow and residues conserved across all groups shaded in gray. Residues within each group that are \>50% conserved are shown in capital letters. In pairwise BLAST (Basic Local Alignment Tool) \[[@B16]\]comparisons, TyrA~a~and TyrA~c~consensus sequences are most similar (47% identity), followed by the TyrA~c~/TyrA~p~pair (40% identity), with TyrA~a~and TyrA~p~exhibiting 34% identity. TyrA~c\_Δ~is quite distinct from the other three groupings, exhibiting only 27% identity with TyrA~c~, 23% identity with TyrA~c~, and 18% identity with TyrA~p~. ### Cyclohexadienyl dehydrogenases Many TyrA proteins (at least in the domain *Bacteria*) are of the TyrA~c~subclass. The cyclohexadienyl dehydrogenases commonly accept PPA or AGN about equally well, but various degrees of preference for one of the alternative substrates are also observed. Detailed molecular and genetic studies of TyrA~c~proteins from *Pseudomonas aeruginosa*, \[[@B17]\], *P. stutzeri*\[[@B1]\], and *Zymomonas mobilis*\[[@B18]\] have been carried out. The distinct variety of TyrA~c~mentioned above, which has been denoted TyrA~c\_Δ~exhibits a number of indels (mostly deletions) within the catalytic-core region when its consensus sequence is aligned with those of the other TyrA classes (Fig. [3](#F3){ref-type="fig"}). It is intriguing that the indel structuring of TyrA~c\_Δ~correlates with the presence of an extra-core extension. This extension is often AroQ, but not always. For example, in the genera *Nostoc*and *Anabaena*it appears to be a degraded, catalytically inactive AroQ, whereas in *Xanthomonas*or *Xylella*it is an ACT domain. Since the one large clade of TyrA~c\_Δ~proteins that has so far been studied prefers PPA over AGN by well over an order of magnitude, an evolutionary relationship of indel insertions to the narrowing of substrate preference for PPA might exist. If so, however, this cannot be the only molecular change to accomplish favored utilization of PPA over AGN since a number of TyrA~c~proteins, (e.g., TyrA~c~from *Neisseria gonorrhoeae*), also exhibits an overwhelming preference for PPA, even though this class lacks the indel structuring. ### Arogenate dehydrogenases The TyrA~**a**~class of specificity is currently represented by higher plants and at least three widely spaced bacterial lineages: cyanobacteria, actinomycetes and *Nitrosomonas europaea*. This discontinuity of phylogenetic spacing is consistent with a fundamental evolutionary scenario \[[@B19]\] whereby the ancestral dehydrogenase was a broad-specificity TyrA~**c**~that evolved narrowed substrate specificity (to yield either TyrA~**p**~or TyrA~**a**~) independently on multiple occasions in modern lineages. The ubiquitous presence of TyrA~a~in cyanobacteria has been heavily documented \[[@B20]\]. *Nitrosomonas europaea*currently (as of March, 2005) has no sufficiently close genome relatives that have been sequenced. The first BLAST hit returned from a ~NADP~TyrA~**a**~query from *N. europaea*(March,2005) is the protein from *Ralstonia solanacearum*(48% identity), which is known to possess broad specificity for both of its substrates (i.e., ~NAD(P)~TyrA~**c**~) \[[@B21],[@B22]\]. The TyrA sequences of *Actinobacteria*separate into two distinct groupings on the protein tree (Fig. [2](#F2){ref-type="fig"}). Coryneform bacteria in one sub-cluster have been rigorously characterized as the ~NAD(P)~TyrA~a~substrate specificity type. On the other hand, a variety of *Streptomyces*species have been shown \[[@B23],[@B24]\] to possess ~NAD~TyrA~a~, and TyrA proteins of these organisms populate the second *Actinobacteria*sub-cluster of Fig. [2](#F2){ref-type="fig"}. Figure [4](#F4){ref-type="fig"} shows sequence alignments of the N-terminal pyridine-nucleotide discriminator regions of currently available actinomycetes. The conserved \'D\' residue (highlighted in yellow) in the upper group is a reliable indicator of NAD^+^specificity, in part because NADP^+^is repelled by the negative charge at this position. The asparagine residue (highlighted in blue) in the corresponding position in members of the lower group indicates NAD(P)^+^specificity as discussed by Bonner et al. \[[@B25]\]. *Rubrobacter xylanophilus*is the most distant representative of the *Actinobacteria*, being the sole member of the subclass taxon *Rubrobacteridae*, and its protein (denoted Rxyl) appears as an orphan in Fig. [2](#F2){ref-type="fig"}. A similar relationship of phylogenetic separation associated with narrowed specificity for pyridine-nucleotide substrate exists for the low-GC gram-positive bacteria (eight o\'clock in Fig. [2](#F2){ref-type="fig"}). Here the major clade is NAD^+^-specific, whereas species of *Streptococcus*have retained the ancestral breadth of specificity for NAD^+^/NADP^+^. Alignments of the pyridine-nucleotide discriminator regions of these latter two groups match up extremely well with the upper alignment of Fig. [4](#F4){ref-type="fig"} where residue 32 of the Wierenga fingerprint \[[@B26]\] is \'D\' and with the lower alignment where residue 32 is \'N\' (data not shown). Recently, a plant *tyrA*~**a**~from *Arabidopsis thaliana*has been reported to consist of two near-identical domains that are fused \[[@B27]\]. The gene encoding this 68-kDa protein co-exists in the genome with a single-domain paralog \[[@B28]\] that encodes a predicted 37-kDa protein, somewhat larger than the catalytic-core domain of TyrA~**a**~from *Synechocystis*. TyrA~**a**~(known to be located in higher-plant chloroplasts \[[@B2]\]) may have originated from cyanobacteria via endosymbiosis. If so, however, the plant TyrA~a~sequences have diverged sufficiently that they no longer share a specific phylogenetic grouping with the cyanobacterial TyrA sequences. This is in marked contrast with the phylogenetic coherence of the tryptophan synthase subunit proteins (TrpEa and TrpEb\_1) from cyanobacteria and higher plants \[[@B29]\]. ### Prephenate dehydrogenases TyrA~p~is conspicuously represented by a large clade of low-GC gram-positive organisms, of which *Bacillus subtilis*TyrA~p~is the best studied \[[@B30]\]. Thus far, all TyrA~p~proteins are fused to a C-terminal ACT domain, and therefore no \"minimal\" TyrA~p~proteins that consist only of a catalytic core are available as yet. At the level of physiological function, it should be added that those cyclohexadienyl dehydrogenases that exhibit a very substantial preference for prephenate are for all practical purposes prephenate dehydrogenases, even though they carry a formal designation of TyrA~c~or TyrA~c\_Δ~. These include most, if not all, of the AroQ•TyrA~c\_Δ~enzymes of the enteric lineage (lower-gamma in Fig. [2](#F2){ref-type="fig"}). The TyrA~c~protein from *Neisseria gonorrhoeae*(and by inference, the closely related *N. meningitides*) is also a well-studied example of overwhelming preference for prephenate \[[@B21]\]. ### PapC dehydrogenases PapC participates in the formation of *p*-aminophenylalanine as a step in the synthesis of at least two antibiotics (see Fig. [1](#F1){ref-type="fig"}). It is so far represented by only a few sequences. The PapC specificity is strongly indicated by absence of the otherwise invariant residue H197 (*E. coli*numbering) that is associated with recognition of a 4-hydroxy moiety in the cyclohexadienyl substrates of the aforementioned dehydrogenases. This moiety, of course, differs in being a 4-amino substituent in the substrate used by the PapC dehydrogenase (Fig. [1](#F1){ref-type="fig"}). See Bonner et al. \[[@B25]\] for a more detailed overview. The \"redundant\" *trp/aro*supraoperon of *Nostoc/Anabaena* ----------------------------------------------------------- All cyanobacteria possess a highly conserved *tyrA*~*a*~gene, as well as a complete suite of tryptophan-pathway genes that are dispersed (unlinked) in the genome. The large-genome cyanobacterial lineage consisting of the *Nostoc*and *Anabaena*genera possess in addition a unique and seemingly redundant *trp*/*aro*supraoperon consisting of most of the aforementioned genes \[[@B31]\]. These include a second *tyrA*gene (curated as *tyrA*~*c\_Δ*~), six *trp*-pathway genes (all except *trpC*), and genes encoding the first two common-pathway steps of aromatic amino acid biosynthesis. All of these supraoperonic genes appear to be redundant in that they are represented by homologs (paralogs or xenologs) elsewhere in the *Nostoc*and *Anabaena*genomes at scattered loci. The closest BLAST hits for the *Nostoc/Anabaena*TyrA~c\_Δ~proteins are not the co-existing TyrA~a~homologs present in their own genomes (and universally present in cyanobacteria). Rather the closest BLAST hits are to the TyrA~c\_Δ~domains of the AroQ•TyrA~c\_Δ~fusions in the enteric lineage. Since the enteric proteins are NAD^+^-specific and strongly prefer prephenate, it is likely that the \"extra\" cyanobacterial proteins are also ~NAD~TyrA~c\_Δ~proteins. Indeed, this would be consistent with enzymological evidence provided in the literature for both *Nostoc*and *Anabaena*\[[@B20]\]. Concerning the evolutionary origin of the redundant block of linked genes found in the *Nostoc*and *Anabaena*genomes, at least two possibilities await further illumination. (i) These genes might have been acquired by a common ancestor of *Nostoc*and *Anabaena*via lateral gene transfer. This is consistent with the observation that biosynthetic-pathway operons are generally absent in the cyanobacteria, and all of the linked genes could have been recruited in a single event. However, at present no candidate donor genomes are known that possess this supraoperon combination of genes. If the TyrA~c\_Δ~proteins of *Nostoc*/*Anabaena*and the enteric lineage are possibly related by LGT, it is of interest that the N-terminal extension of TyrA~c\_Δ~from *Nostoc*/*Anabaena*resembles a degraded AroQ domain of AroQ•TyrA~c\_Δ~from enterics. In both cases the N-terminal residues may compensate for indel deletions within the catalytic core region of TyrA~c\_Δ~. Subsequently, AroQ function may have evolved in one lineage (or have been lost in the other). This possibility of domain-domain interaction is consistent with the established interdependence of the AroQ• and •TyrA~c\_Δ~domains from *E. coli*\[[@B32]\]. Alternatively, *tyrA*~*a*~and *tyrA*~*c\_Δ*~(and the duplicated *trp*and *aro*genes present in the supraoperon) might be ancient paralogs within the cyanobacterial lineage. If so, at a time following divergence of heterocystous cyanobacteria from the unicellular cyanobacteria, the latter may have lost the clustered block of aromatic-pathway genes in a single event of reductive evolution. The supraoperonic genes might be related to a specialized function associated with \"developmental\" physiological processes that typify the filamentous, heterocyst-forming cyanobacteria. This might be reminiscent of the nature of the phenazine-pigment operon of *Pseudomonas aeruginosa*. Here unique phenazine-pathway genes are combined with a redundant gene of common-pathway aromatic biosynthesis and two redundant (and fused) genes of tryptophan biosynthesis. This accomplishes the linkage of specific phenazine biosynthesis with a supply of 2-amino-2-deoxy-isochorismate, the branchpoint of divergence toward phenazine and tryptophan \[[@B33],[@B34]\]. This complexity in which multiple paralogs are differentially deployed is consistent with the large genome sizes of *Anabaena*(7.2 MB) and *Nostoc*(9.2 MB), compared with the much smaller unicellular genomes of *Prochlorococcus marinus*(1.7 MB), *Synechococcus*sp. WH8102 (2.4 MB), and *Synechocystis*sp. PCC6803 (3.6 MB). Profile hidden Markov models (HMMs) to distinguish specificity subfamilies for cyclohexadienyl substrate -------------------------------------------------------------------------------------------------------- The limited information thus far available about specific molecular roles of particular TyrA amino acid residues has been summarized recently \[[@B25]\]. The catalytic-core domains of known TyrA~a~, TyrA~p~, TyrA~c~, and TyrA~c\_Δ~proteins were selected from our files of TyrA catalytic-core domains \[[@B35]\], and a new subset of sequences was prepared that lacked the pyridine nucleotide discriminator segment, a glycine-rich βαβ region at the N terminus. Although the glycine-rich βαβ region is not the only segment that contacts pyridine nucleotide substrate, it is the sole region that discriminates between NAD^+^and NADP^+^. The resulting trimmed sequence is defined as the \"cyclohexadienyl-substrate core segment\". No distinctive motifs were found that, in isolation, would be a clear predictive indicator of specificity for cyclohexadienyl substrate. Similar substrate specificity profiles probably can be dictated by alternative patterns of interplay between different residue combinations. Because of the rapid accumulation of incorrectly annotated TyrA entries in GenBank and other databases, partly due to the complications of misnaming that are associated with gene fusions and partly to a failure to assimilate published substrate specificities, the use of BLAST does not return reliable annotations with respect to substrate specificity. Even the HMMs used in Pfam \[[@B36]\] and Interpro \[[@B37]\] were not helpful in this case because the HMM deployed in those databases was broadly but incorrectly defined as \'prephenate dehydrogenase (NADP^+^) activity\' for all TyrA dehydrogenases (accession number PF02153 in Pfam and entry IPR003099 in Interpro). However, Profile HMM is known to be well suited for modeling a particular sequence family of interest and for finding additional remote homologs \[[@B38]\]. It is reputed to outperform methods that rely only upon pair-wise alignment of homologous residues in predicting protein function \[[@B39]\]. Therefore, profile HMMs were constructed using our multiple sequence alignments of each curated TyrA specificity subfamily, using the HMMER package \[[@B38]\]. The profile HMMs obtained are only tentatively reliable for prediction of substrate specificity. To facilitate ongoing and future functional annotations, we have made our profile HMMs available as a working resource for \"specificity prediction\" at AroPath \[[@B40]\]. Users can match query sequences against the four profile HMMs to predict the subfamily to which a query sequence belongs. It is anticipated that future experimental data relevant to substrate specificity will facilitate refinement of the prediction program. For example, at present the program predicts that the TyrA sequences from organisms such as *Helicobacter pylori*and *Saccharomyces cerevisiae*belong to the TyrA~a~grouping, and it will be interesting to see whether this holds up to experimental confirmation. It is additionally fascinating that (i) the dehydrogenase from *Archaeoglobus fulgidus*is predicted to belong to the indel-containing TyrA~c\_Δ~grouping and (ii) that it possesses a possible cooperatively interacting extra-core domain extension (an AroQ fusion), just as occurs for the large clade of enteric bacteria. If this is relevant, it is even more fascinating that the *Archaeoglobus aroQ*is fused at the C-terminal side of *tyrA*~c\_Δ~, rather than at the N terminus as is the case with enteric bacteria. Users at AroPath \[[@B41]\] can enter query sequences into interactive multiple sequence alignments with any of the four sets of \"cyclohexadienyl-substrate core segments\" sequences that were used to train the profile HMMs. An ongoing effort is in process to extend the predictor capability to include the pyridine nucleotide substrate as well. One can also align query sequences of interest with either an assemblage of the complete set of curator-approved TyrA catalytic-core TyrA sequences or with any desired subset of seed sequences. The catalytic-core domain of TyrA proteins ------------------------------------------ The simplest set of fully functional TyrA proteins consists only of the catalytic-core domain (about 180 amino acids) \[[@B1]\] and includes the well-characterized TyrA~**c**~enzymes from *Neisseria gonorrhoeae*\[[@B21]\] and *Zymomonas mobilis*\[[@B18]\], as well as TyrA~**a**~from a cyanobacterium \[[@B25]\]. In addition the catalytic-core domain from *Pseudomonas stutzeri*has been engineered for study from a *tyrA*~*c*~•*aroF*fusion \[[@B1]\]. These model core proteins are roughly as divergent from one another on the TyrA protein tree as are the organisms that contain them (Fig. [2](#F2){ref-type="fig"}). In view of the possibility raised in this paper about inter-domain interactions, the single-domain TyrA proteins are undoubtedly the simplest sources for study of the fundamental properties of the catalytic-core domain. Xie et al. \[[@B1]\] suggested that in the set of catalytic-core TyrA proteins, inhibitors bind at the catalytic site and exhibit classical competitive inhibition with respect to the particular cyclohexadienyl substrates that can be accepted by a given organism. This model predicts that the specificity for the sidechains of substrates used would parallel the specificity for inhibitor sidechains. The information summarized in Table [4](#T4){ref-type="table"} supports this expectation. Thus, the TyrA~**c**~proteins of *P. stutzeri*and *P. aeruginosa*will accept either a pyruvyl (as with PPA) or an alanyl (as with AGN) sidechain in the alternative substrates used, and this is paralleled by recognition of either a pyruvyl (4-hydroxyphenylpyruvate) or an alanyl (TYR) sidechain in the competent inhibitor structures. In another case, the *N. gonorrhoeae*TyrA~c~exhibits an overwhelming substrate preference for PPA, and consistent with the foregoing, is subject to inhibition by 4-hydroxyphenylpyruvate but not by TYR. A variety of analog inhibitor structures were used by Xie et al. \[[@B1]\] to show that the minimal structure for binding at the substrate-binding site of *P. stutzeri*TyrA~**c**~is a six-membered ring with a 4-hydroxy substituent. In contrast to the TyrA~**c**~proteins just described, the *Z. mobilis*TyrA~**c**~is totally insensitive to inhibition by either 4-hydroxyphenylpyruvate or TYR. Since both of these compounds lack a 1-carboxy moiety, it is reasonable to assume that the 1-carboxy substituent present in the two substrates accepted may be required for binding at the catalytic center. Thus, although TyrA~c~from *Z. mobilis*will accept the same two substrates as does the TyrA~**c**~from *P. stutzeri*, the greatly different inhibition results suggest that *Z. mobilis*obeys more stringent rules for binding at the catalytic site (i.e., a ring carboxylate must be present). *Synechocystis*sp. and *Arabidopsis thaliana*TyrA~a~proteins accept as a substrate only AGN, which has an alanyl sidechain. The ring-carboxylate moiety is evidently not absolutely required for binding since these TyrA~a~proteins can recognize TYR (alanyl sidechain) as a competitive inhibitor. In contrast, since *N. europaea*TyrA~a~is not inhibited by TYR, it resembles the *Z. mobilis*TyrA~c~in the putative requirement for a 1-carboxy substituent to secure successful binding at the catalytic site. In summary, some TyrA proteins probably exercise greater discrimination in their requirement for a 1-carboxy moiety for binding at the catalytic site, and these are insensitive to competitive inhibition by the aromatic reaction products (which lack the 1-carboxy substituent). Other TyrA proteins that require the 1-carboxy moiety for the fundamental catalytic process, but presumably do not require it for binding, will recognize product inhibitors that have the same sidechain as any substrate recognized. Specificity for the pyridine nucleotide co-substrate within the TyrA superfamily -------------------------------------------------------------------------------- NAD^+^differs from NADP^+^only in that NADP^+^has a phosphate group esterified at the 2\'-position of adenosine ribose. Therefore, the ability of a dehydrogenase to discriminate between those two lies in the particular enzyme region that contacts the ribose moiety. The glycine-rich region known to constitute the ADP-binding βαβ fold is well known to be this point of contact \[[@B26]\]. This Rossmann β α β fold is inevitably positioned at the extreme N terminus of TyrA proteins, and the typical GXGXXG motif is almost always observed, as illustrated in Fig. [4](#F4){ref-type="fig"}. This region is helpful for assessment of probable specificities for pyridine nucleotide. One can be fairly sure that TyrA proteins possessing D-32 (*E. coli*numbering, reference \[[@B26]\]) are NAD^+^-specific. A negatively charged residue (D or E) at position 32 is critical for hydrogen binding to the diol group of the ribose near the adenine moiety in NAD^+^-specific enzymes. NADP^+^-specific dehydrogenases cannot tolerate a negatively charged residue at position 32. TyrA proteins that possess an asparagine residue in the corresponding position appear to be broadly specific for both NAD^+^and NADP^+^as discussed above. No clearcut motif has been identified for NADP^+^-specific TyrA proteins, although at least one positively charged residue is expected in the region just beyond residue 32. By elimination, those sequences lacking D-32 or N-32 are strong candidates for NADP^+^specificity. As with the cyclohexadienyl co-substrate, narrowed specificity for NAD^+^(or NADP^+^) also seems to have occurred independently on many occasions (some examples given earlier). The absolute specificity of TyrA~p~proteins for PPA tends to be accompanied by absolute specificity for NAD^+^, as illustrated by the large *Bacillus/Staphylococcus/Listeria/Enterococcus*clade at eight o\'clock in Fig. [2](#F2){ref-type="fig"}. However, it is interesting that species of *Streptococcus*have retained the presumed ancestral breadth of specificity for the pyridine nucleotide substrate. The opposite relationship, whereby absolute specificity for AGN tends to be accompanied by absolute specificity for NADP^+^, is also observed. Here three of the four TyrA~a~lineages described earlier exhibit this pattern. Exceptions, though, are the aforementioned TyrA~a~proteins of *Actinobacteridae\_*1 which accept either NAD^+^or NADP^+^, as well as the TyrA~a~proteins of the sister *Actinobacteridae*\_2 which are specialized for NAD^+^\[[@B42],[@B43]\]. The TyrA~c~proteins of most complete-genome organisms thus far have happened to be NAD^+^-specific, and this has been the property of the most rigorously characterized ones (from *Z. mobilis*, *P. stutzeri*, and *P. aeruginosa*). However, it is clear from extensive enzymological surveys \[[@B22]\] that TyrA~c~proteins having broad specificity for NAD^+^/NADP^+^are common, examples including species of *Ralstonia*and *Burkholderia*. The spectrum of variation that can exist, even within a clade of organisms that are of fairly close relationship, is illustrated by one striking example. In the pseudomonad clade marked by a common *tyrA•aroF*fusion, the *Acinetobacter*sp. TyrA~c~is NADP^+^-specific \[[@B44]\], whereas the sister subclade *Pseudomonas*/*Azotobacter*exhibits NAD^+^specificity (Fig. [2](#F2){ref-type="fig"}). Here the entire clade marked by a common ancestral fusion shares approximately the same profile of cyclohexadienyl substrate preference, but cofactor specificity has been narrowed in opposite directions. We had previously suggested that there might be a general structural relationship of substrate pairing that tends to favor interaction between PPA and NAD^+^, on the one hand, and, on the other hand, between the greater positive charge of AGN and the greater negative charge of NADP^+^. These relationships may indeed be favored, but it increasingly appears that any combination can occur. Beyond the catalytic core: allosteric domains --------------------------------------------- Various lineages have acquired an amino acid binding domain known as the ACT domain (pfam01842), which is known to bind a variety of amino acids, thus functioning as an allosteric domain for many proteins including phosphoglycerate dehydrogenase, aspartokinase, acetolactate synthase, phenylalanine hydroxylase, prephenate dehydratase and formyltetrahydrofolate deformylase. Recruitment of this domain by fusion with *tyrA*~**p**~appears to have occurred in a common ancestor of the large *Bacillus*/*Staphylococcus*/*Listeria*/*Enterococcus*/*Streptococcus*assemblage (Fig. [2](#F2){ref-type="fig"}). It is interesting that *B. subtilis*also possesses a gene encoding a free-standing ACT domain in its genome (incorrectly annotated as *pheB*). An additional fusion of genes encoding an ACT domain and *tyrA*(that arose independently, judging from the widely spaced tree positions) occurred in the common ancestor of *Xanthomonas*and *Xylella*. *Actinobacteria*usually possess a C-terminal extension that probably functions as an allosteric domain. The extension possessed by the *Actinobacteridae*\_2 assemblage, which includes *Streptomyces coelicolor*and its relatives, appears to be an ACT domain. On the other hand, it is not all all clear that the C-terminal extension of the *Actinobacteridae*\_2 assemblage is an ACT domain. This difference, in addition to the differing specificities for pyridine nucleotide substrate, may have contributed to the overall TyrA~a~divergence observed between the two *Actinobacteridae*groups. There is no correlation between presence of the ACT domain and specificity for cyclohexadienyl substrate since TyrA~**p**~from the *Bacillus*clade is PPA-specific, *Xanthomonas*/*Xylella*TyrA~c~is broadly specific, and *Streptomyces*TyrA~a~is AGN-specific. *B. subtilis*, which belongs to the large clade having an ACT domain as a carboxy extension, has been extensively characterized \[[@B30]\]. 4-Hydroxyphenylpyruvate is an effective competitive inhibitor, as would be consistent with our proposed effects at the catalytic core for a PPA-specific enzyme. However, TYR, phenylalanine (PHE) and tryptophan were also inhibitors. The violation of the rule that the latter three amino acid inhibitors would not be expected to bind the catalytic core region (because they have alanyl sidechains even though the substrate-binding site only recognizes the pyruvyl sidechain of prephenate) and the finding that some of these were not competitive inhibitors can now be accounted for by the presence of the allosteric ACT domain. A carboxy extension shared by a number of *Archaea*(denoted \'REG\' in Fig. [2](#F2){ref-type="fig"}) is presumably a regulatory domain as well. This is consistent with the recent result of Porat et al. \[[@B45]\] that not only 4-hydroxyphenylpyruvate, but also TYR, inhibited prephenate dehydrogenase activity of *Methanococcus maripaludis*. The tyrA gene is a popular fusion partner ----------------------------------------- ### Fusion with *aroQ* *tyrA*may be fused with a number of other catalytic domains, each of them relevant to aromatic biosynthesis (Fig. [2](#F2){ref-type="fig"}). *aroQ*(encoding chorismate mutase) is frequently fused with a number of other aromatic-pathway genes \[[@B46]\]. The lower-gamma Proteobacteria (enteric lineage) located at twelve o\'clock in Fig. [2](#F2){ref-type="fig"} possess an *aroQ•tyrA*~*c\_Δ*~fusion. The fusion physically links chorismate mutase (which forms PPA) with TyrA~c\_Δ~(which utilizes PPA). The two protein domains of AroQ•TyrA~c\_Δ~may have co-evolved to produce cooperative protein-protein interactions since physical separation of the domains evoked relatively low activities of both activities in *E. coli*\[[@B32]\]. Substantial comparative work shows that the *aroQ•tyrA*~*c\_Δ*~fusion has been stably maintained throughout the entire enteric lineage \[[@B47]\]. Exceptions in some genomes lacking this fusion altogether can be attributed to reductive evolutionary loss in pathogens (e.g., *Haemophilus ducreyi*) or endosymbionts (e.g., *Buchnera aphidicola*). An independent *aroQ•tyrA*fusion was generated in the common ancestor of *Sulfolobus solfataricus*and *S. tokodaii*(Fig. [2](#F2){ref-type="fig"}). Since the TyrA domain of *Sulfolobus*species lacks the indel structure of the TyrA~c\_Δ~class, it would be interesting to see whether physical separation of the two domains would yield evidence of independent function, in contrast to the results mentioned just above for *E. coli*. ### Fusion with *aroF* Secondly, *tyrA*~c~has been fused with *aroF*on at least two separate occasions in *Bacteria*. (The *aroF*gene encodes enolpyruvylshikimate-3-P synthase, the sixth enzyme in the common pathway of aromatic biosynthesis; see \[[@B5],[@B6]\] for nomenclature used.) One clade includes members of the upper-gamma Proteobacteria: *P. aeruginosa*, *P. syringae*, *P. putida*, *P. stutzeri*, *P. fluorescens*and *Azotobacter vinelandii*. It is interesting that *P. syringae*has experienced a deletion of about 200 residues at the N-terminal region of the AroF domain. This has been coupled with the acquisition of a stand-alone *aroF*gene that is absent in other members of the clade. Interestingly, the latter AroF shows high identity only with AroF from *Agrobacterium tumefaciens*, an alpha proteobacterium. The *A. tumefaciens aroF*, in turn, is unique compared to its α-subdivision relatives, both in having divergent sequence and in being unlinked to *cmk*and *rpsA*. Thus, it seems likely that the incongruence of AroF belonging to both *P. syringae*and *A. tumefaciens*reflects acquisition via LGT from some as yet unknown source. The disruption of the fused *aroF*domain in *P. syringae*is an unusual instance where the catalytic function of one fusion domain has become discarded while the function of the second domain has been retained. It is interesting to consider the possibility that the truncated remnant of the *aroF*fusion domain might be exploitable for use as an innovative source of a new regulatory domain. An additional fusion of *tyrA*with *aroF*has occurred independently within the beta Proteobacteria in the common ancestor of *Burkholderia pseudomallei*and *B. mallei*. This has been very recent since the closely related *B. fungorum*and *B. cepacia*organisms lack the fusion. It has been suggested that presence of a given fusion may be useful for sorting out clades that diverged from a common ancestor, independent of other methods \[[@B48]\]. Different fusions offer the power of discriminating clades at various hierarchical levels, i.e., nested clades discriminated by nested gene fusions. The *tyrA•aroF*fusion occurred in the common ancestor of the clade that includes the upper-gamma Proteobacteria shown in Fig. [2](#F2){ref-type="fig"}. One can reasonably assume that relatively close upper-gamma organisms lacking the *tyrA•aroF*fusion diverged from the common ancestor of the fusion clade prior to the fusion event. Such would appear to be the case, for example, with *Acidithiobacillus ferrooxidans*, an outlying member of the upper-gamma Proteobacteria that lacks the fusion. It is reasonable to conclude that the fusion event must have pre-dated the differential specialization for the pyridine nucleotide cosubstrate that distinguishes *Acinetobacter*sp. (NADP^+^-specific) from the large grouping of pseudomonads that are NAD^+^-specific. ### Fusion with *hisH*~b~ Thirdly, a single organism, *Rhodobacter sphaeroides*, possesses a *hisH*~b~•*tyrA*fusion that must have occurred very recently. *hisH*~b~encodes an aromatic aminotransferase that is closely related to (or sometimes even synonymous with) imidazole acetol phosphate aminotransferase \[[@B49]\]. The *hisH*~*b*~/*tyrA*/*aroF*linkage group is part of a supraoperon in some gram-negative bacteria in which a relatively conserved, yet frequently shuffled gene order is observed \[[@B5],[@B6]\]. Hence, it is reasonable to assume that at the time just prior to fusion, *hisH*~b~, *tyrA*and *aroF*were adjacent. Note that among the fusions currently known, *hisH*~b~and *aroF*are fused to the N-terminal and C-terminal ends of *tyrA*, respectively. It would be interesting to know the substrate specificity of the *R. sphaeroides*TyrA domain. If it is AGN-specific the significance of *hisH*~b~presumably would be to transaminate PPA to form AGN, the substrate used by TyrA~a~(see Fig. [1](#F1){ref-type="fig"}). On the other hand, if the dehydrogenase is PPA-specific, the significance of the HisH~b~domain would be to transaminate the product of the TyrA~p~reaction. If the enzyme is a TyrA~c~enzyme (as is probable), then HisH~b~likely is competent to catalyze either of the foregoing reactions. ### Fusion with ACT The widespread ACT regulatory domain appears to have been acquired by independent fusions at least three separate times judging from the widely separated lineages that possess a TyrA•ACT fusion (Fig. [2](#F2){ref-type="fig"}). Xie et al. \[[@B5]\] initially noted homologous domains positioned at the N terminus of mammalian phenylalanine hydroxylase and at the C terminus of most microbial prephenate dehydratases. This domain is responsible for phenylalanine-mediated activation and phenylalanine-mediated inhibition of the hydroxylase and dehydratase enzymes, respectively. This domain was later named the ACT domain \[[@B50]\] and shown to be a widely distributed domain family that shares a conserved overall fold. Members of the ACT-domain family possess a wide variety of different ligand-binding capabilities. For example, the ACT domain of 3-phosphoglycerate dehydrogenase binds *L*-serine as a allosteric inhibitor. ### Fusion with REG Another putative regulatory domain fused to tyrA (denoted *tyrA•REG*) is thus far restricted to some of the *Archaea*. This domain is a predicted regulatory domain, as described in COG4937. ### A novel 4-domain fusion *Archaeoglobus fulgidus*exhibits a striking four-domain fusion consisting of three catalytic domains and a regulatory ACT domain (TyrA•AroQ•PheA•ACT). The TyrA domain is predicted to belong to the TyrA~c\_Δ~class when used as a query input into the AroPath Specificity Predictor Tool \[[@B40]\]. We speculated earlier that the •AroQ fusion domain of *Archaeoglobus*may exercise cooperative interactions with TyrA~c\_Δ~, as appears to occur between the AroQ•TyrA~c\_Δ~domains of *E. coli*and its relatives. *tyrA*in its syntenic context ----------------------------- Although the genes of prokaryotes have clearly been subject to frequent scrambling, some gene-gene associations persist more tenaciously than others. Xie et al. \[[@B5],[@B6]\] asserted that one such ancestral gene string that has resisted scrambling forces is *hisH*~*b*~\> *tyrA*\> *aroF*. As suggested above, contemporary gene fusions can serve as frozen-in-time indicators of ancient gene organizations that were later obscured by gene-scrambling events. Another gene string that is often within the syntenic region of *hisH*~b~, *tyrA*, and *aroF*is *cmk*\> *rpsA*. Gene synteny in prokaryotes has not been easily recognized because substantial manual scrutiny in combination with a sufficient density of genomic representation on a given portion of the phylogenetic tree is necessary to detect patterns of synteny that are camouflaged by frequent scrambling events (inversion, deletion and transposition). The domain *Bacteria*is now represented by a collection of sequenced genomes that is progressively approaching the genomic densities needed for meaningful analysis. Figure [5](#F5){ref-type="fig"} provides a visual sense of the frequency with which *tyrA*is closely positioned with other genes of aromatic biosynthesis, as well as the underlying patterns of overall synteny. These patterns are unstable, and yet persistent traces of synteny can be seen where genomic representation is sufficiently dense. The four genes of particular emphasis in this paper are color coded. Other genes that are engaged in aromatic biosynthesis are colored grey, and any other genes are white. At a very deep level of phylogenetic branching, *Thermotoga*exhibits a *tyrA*gene flanked by seven genes encoding all of the common steps of aromatic biosynthesis (two of them being fused). Since closely related genomes are not yet available here, we cannot judge whether these genes came together recently or whether an ancient pattern of synteny has been retained. Although *tyrA*is not linked to any functionally relevant genes in *Aquifex*, representing another point of deep phylogenetic branching, this does not necessarily mean that *tyrA*was not already generally associated with other aromatic-pathway genes at an early time. For reasons that are totally mysterious, certain scattered lineages exhibit a total lack of operon organization for aromatic-pathway genes (and indeed for most other biosynthetic pathways, such as that for histidine biosynthesis). These lineages (Fig. [5](#F5){ref-type="fig"}) include, besides *Aquifex*, those of *Deinococcus*, the actinomycetes, the cyanobacteria, and *Chlorobium*. Except for the actinomycetes, this phenomenon of total gene dispersal also applies to genes of tryptophan biosynthesis \[[@B7],[@B8]\]. When the various examples of *hisH*~b~\> *tyrA*\> *aroF*linkage are mapped on a 16S rRNA tree, they first appear in gram-positive bacteria. In *Bacillus*and related organisms (such as *Listeria*), the *hisH*~*b*~\> *tyrA*\> *aroF*unit is associated with a large ancestral operon consisting of *aroG*\> *aroB*\> *aroH*\> *hisH*~*b*~\> *tyrA*~*p*~\> *aroF*. *Bacillus*additionally possesses the *cmk*\> *rpsA*unit, albeit in a separate location. Interestingly, in one narrow subclade (*B. subtilis*, *B. halodurans*and *B. stearothermophilus*) the *trp*operon has been inserted between *aroH*and *hisH*~*b*~to yield a supraoperon that has been fully characterized as a complex functional unit \[[@B51]\]. See Xie et al. \[[@B7]\] for a full presentation of evolutionary interpretation relevant to the latter. Though highly scrambled, a pattern of association of *pheA*with *hisH*~b~\> *tyrA*\>*aroF*is suggested by linkage patterns seen at the hierarchical level of C*ytophaga*and *Bacteroides*(Fig. [5](#F5){ref-type="fig"}). *aroQ*became associated with *pheA*through gene fusion as early as the divergence of the *Spirochaetes*to yield an *aroQ•pheA*\>*tyrA*\>*aroF*\>*cmk*\>*rpsA*linkage unit (*Leptospira interrogans*in Fig. [5](#F5){ref-type="fig"}). The *aroQ•pheA*gene associated with *tyrA*and *aroF*in *Clostridium difficile*appears to have arisen from a distinctly different fusion event than that present in delta, epsilon, beta and upper-gamma Proteobacteria or from that present in lower-gamma Proteobacteria (based upon analysis of inter-domain linker regions; unpublished data). Consensus ancestral gene organizations for the most densely represented divisions of Proteobacteria have been deduced as shown at the bottom of Fig. [5](#F5){ref-type="fig"}. Detailed information that supports a deduced consensus for ancestral gene organizations with respect to beta Proteobacteria, upper-gamma Proteobacteria, and lower-gamma Proteobacteria are shown later (Figs. [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). We suggest that the last common ancestor of all Proteobacteria possessed the gene organization *aroQ•pheA*\>*hisH*~b~\>*tyrA*\>*aroF*\>*cmk*\>*rpsA*. This is similar to the synteny that has been retained in general by the beta Proteobacteria and the upper-gamma Proteobacteria. The *aroQ•pheA*\>*hisH*~b~\>*tyrA*portion likely specified all the catalytic requirements for conversion of chorismate to PHE and conversion of chorismate to TYR. Chorismate mutase activity specified by the *aroQ*domain could supply PPA for both PHE and TYR biosynthesis. Likewise, HisH~b~, widely utilized as an aromatic aminotransferase \[[@B49]\], could also function for both PHE and TYR biosynthesis. Though currently available members of delta and epsilon Proteobacteria exhibit substantial gene scrambling, the various fragmentary linkage patterns seen provide support for the ancestor proposed. *Geobacter*(and other Delta\_1 members) has the *aroQ•pheA*\> *tyrA*\> *aroF*\> *cmk*\> *rpsA*linkage group (with *lytB*inserted between *cmk*and *rpsA*). *Desulfovibrio vulgaris*, another delta Proteobacterium (Delta\_2) that is highly divergent from *Geobacter*, has a very interesting pattern of conservation and scrambling. *aroQ•pheA*\> *aroF*\> *tyrA*has been attached to a complete 7-gene *trp*operon. *hisH*~b~\> *cmk*(not shown in Fig. [5](#F5){ref-type="fig"}) is completely separated from *rpsA*. The supraoperonic gene organization shown for *D. vulgaris*begins with two recently discovered genes, herein denoted *aroA*\' and *aroB*\', that encode enzymes specifying an alternative biochemical route to dehydroquinate \[[@B52]\]. The epsilon Proteobacteria all display significant gene scrambling, but piecemeal evidence for the unscrambled ancestor proposed is present. For example, *Campylobacter jejuni*possesses an *aroQ•pheA*\> *hisH*~b~unit, as well as *aroF \> lytB \> rpsA*(Fig. [5](#F5){ref-type="fig"}). *Wollinella succinogenes*and *Helicobacter hepaticus*both possesses an *aroF*\> *lytB*\> *rpsA*unit. The ancestor of alpha Proteobacteria has lost the *aroQ•pheA*fusion, and a stand-alone *pheA*is consistently observed. Members of this group are quite uniform in the stable possession of *hisH*~b~\> *tyrA*and *aroF*\> *cmk*\> *rpsA*as two separated linkage groups. The beta Proteobacteria are represented by members that have the gene organization: *serC*\> *aroQ•pheA*\> *hisH*~b~\> *tyrA*\> *aroF*\> *cmk*\> *rpsA*. This is also seen in the members of the upper-gamma Proteobacteria. Figure [5](#F5){ref-type="fig"} includes organisms that illustrate the traces of synteny that can be detected in *Bacteria*where overall genome representation is just barely adequate. The following two figures illustrate how syntenic patterns of more resolution and refinement become evident with denser genome representation. Zooming in on syntenic contexts of proteobacteria ------------------------------------------------- ### Beta proteobacteria and upper-gamma proteobacteria The beta Proteobacteria exhibit a dynamic but still interpretable pattern of altered synteny (Fig. [6](#F6){ref-type="fig"} and Table [5](#T5){ref-type="table"}). Species of *Ralstonia*have retained the proposed ancestral synteny that is marked with yellow highlighting in Fig. [6](#F6){ref-type="fig"}. This syntenic organization is such that the aromatic-gene unit *aroQ•pheA*\> *hisH*~b~\> *tyrA*\> *aroF*is nested between *gyrA*\> *serC*at the leftward flank and *cmk*\> *rpsA*\> *himD*at the rightward flank. Species of *Burkholderia*(the next closest lineage) are almost identical, but exhibit individual evolutionary events (marked by circled numbers on the left, which correspond to a description of the proposed evolutionary events given in companion Table [5](#T5){ref-type="table"}). These events include gene insertion, loss of *hisH*~b~, translocation of genes away from the ancestral supraoperon, and fusion of *tyrA*and *aroF*(in the common ancestor of *B. mallei*and *B. pseudomallei*). At a deeper level in the beta Proteobacteria section of the tree, *Nitrosomonas europaea*exhibits a separation of the ancestral supraoperon between *tyrA*and *aroF*. Either a very large insertion was made between *tyrA*and *aroF*, or one of the two gene clusters shown was transposed as part of a sufficiently large segment to include all of the conserved flanking genes. In *Chromobacterium violaceum tyrA*has become completely isolated from other gene members of the ancestral supraoperon, and *aroF*has assumed an inverted orientation with respect to *cmk*. Species of *Neisseria*exhibit no remnants of supraoperon synteny at all, and wholesale dispersal of all the supraoperon genes has occurred. (It is interesting that among the beta Proteobacteria, *Neisseria*species are also unique in that all of the *trp*-pathway genes are dispersed \[[@B7]\]). The gamma Proteobacteria have separated into two distinctly different synteny patterns. The lower-gamma Proteobacteria have undergone marked syntenic change (see below). The assemblage portrayed between *Acidithiobacillus*and *Microbulbifer*in the lower part of Fig. [6](#F6){ref-type="fig"} (termed the upper-gamma Proteobacteria) exhibit a strong overall syntenic resemblance of supraoperon genes to that of the beta Proteobacteria. *Acidithiobacillus*possesses a near-intact ancestral supraoperon, differing only in having two insertions: one gene encoding 3-deoxy-D-**arabino**-heptulosonate 7-phosphate (DAHP) synthase between *hisH*~*b*~and *tyrA*, and the other being the insertion of *serA*between *serC*and *aroQ•pheA*. *Pseudomonas aeruginosa*and *P. stutzeri*have also retained nearly intact ancestral supraoperons, differing only in the fusion of *tyrA*and aroF. The *serC*\> *aroQ•pheA*\> *hisH*~b~\> *tyrA•aroF*\> *cmk*\> *rpsA*supraoperon has been studied in *P. stutzeri*\[[@B5],[@B6]\]. The *tyrA•aroF*fusion occurred in the common ancestor of the clade shown between *Azotobacter*and *Microbulbifer*in Fig. [6](#F6){ref-type="fig"}. The supraoperons of *P. syringae, P. fluorescens*and *P. putida*lack *hisH*~b~. *P. syringae*exhibits a recent C-terminal truncation of the *aroF*domain, coupled with acquisition elsewhere in the genome of a free-standing •*aroF*that is not phylogenetically congruent (probably of LGT origin). *Acinetobacter*sp. and *Microbulbifer degradans*possess an *aroQ•pheA*\> *tyrA•aroF*unit that has become dissociated from *serC*at one end and from *cmk*on the other end. In *Xylella*and *Xanthomonas*, *hisH*~*b*~has been deleted from the genome and *tyrA*has been transposed away from *serC*\> *aroQ•pheA*\> *aroF*. The latter unit has been transposed away from *gyrA*, the ancestral flanking gene. On the other hand, *cmk*\> *rpsA*has remained next to *himD*, the gene usually flanking *rpsA*. ### The enteric lineage The lower-gamma Proteobacteria differ sharply from upper-gamma Proteobacteria in their possession of the *tyrA*~*c\_Δ*~class of *tyrA*and its fusion with *aroQ*. In Fig. [2](#F2){ref-type="fig"} this clade of AroQ•TyrA~c\_Δ~fusions was presented as one exhibiting absolute specificity for NAD^+^, combined with an overwhelming but not complete specificity for PPA. In Fig. [7](#F7){ref-type="fig"} the gene synteny associated with *tyrA*~*c\_Δ*~is profiled against the 16S rRNA phylogenetic trees of the lower-gamma Proteobacteria possessing these genes, and the proposed evolutionary events are summarized in the companion Table [6](#T6){ref-type="table"}. Figure [5](#F5){ref-type="fig"} has indicated a synteny consensus for the common ancestor at this hierarchical level whereby *gyrA*\> *serC*\> *hisH*~b~\> *aroF*\> *cmk*\> *rpsA*parallels the ancestral synteny of β-Proteobacteria, but without *aroQ•pheA*or *tyrA*in the middle of the linkage group. Many dynamic evolutionary events of altered aromatic biosynthesis have occurred within the lower-gamma Proteobacteria since their divergence from the upper-gamma Proteobacteria. This includes the emergence of three allosterically distinct DAHP synthases, one of which now comprises the two-gene, three-domain *tyr*operon (*aroA*~Iα\_*Y*~\> *aroQ•tyrA*~*c\_Δ*~). The upper-gamma Proteobacteria characteristically possess the *aroA*~*Iα*~paralogs encoding AroA~Iα\_H~(TRP-inhibited DAHP synthase) and AroA~Iα\_Y~(TYR-inhibited DAHP synthase). It has been asserted that AroA~Iα\_F~(PHE-inhibited DAHP synthase) was the most recent paralog, acquired just after divergence of the lower-gamma Proteobacteria \[[@B53]\]. It is bizarre that *Shewanella oneidensis*possesses a pseudogene of *aroA*~Iβ~fused to the C terminus of *aroQ•pheA*. The *aroA*~Iβ~subclass of Family-I DAHP synthases is not usually observed in gram-negative bacteria \[[@B54]\]. The dissociation of *tyrA*~*c\_Δ*~from the *serC*/*rpsA*linkage group correlates with the fusion of *aroQ*with *tyrA*~*c\_Δ*~. The *aroQ•pheA*fusion has also escaped from the ser*C*/*rpsA*linkage grouping and has become linked with the newly emerged *tyr*operon. Some sort of duplication and recombinational event between *aroQ•pheA*and *tyrA*~*c\_Δ*~may have led to the creation of *aroQ•tyrA*~*c\_Δ*~since the AroQ•PheA proteins of lower-gamma Proteobacteria are distinct from AroQ•PheA proteins of other Proteobacteria with respect to the inter-domain linker length and the indel content (data not shown). Although it usually is absent from the lower-gamma Proteobacteria, HisH~b~has persisted as the broad-specificity aromatic aminotransferase in the *Pasteurella/Haemophilus*grouping where two *hisH*paralogs are generally present, one of narrow specificity (denoted *hisH*~n~) being within the histidine operon. The *aspC*gene next to *aroF*in *Shewanella*is a paralog that probably functions as an aromatic aminotransferase, suggestive of the situation in the *E. coli*grouping where *tyrB*is a close paralog relative of *aspC*, tyrB having become specialized for aromatic biosynthesis \[[@B49]\]. Gene reduction associated with both endosymbiotic and pathogenic lifestyles are evident. Thus, *Buchnera*lacks *tyrA*, *cmk*, *hisH*, *tyrB*, and possesses only a single *aroA*~*Iα*~species (*aroA*~Iα\_*H*~). *Haemophilus ducreyi*also lacks *tyrA*, as well as aroA~Iα\_H~and the entire *trp*operon \[[@B5]\]. TyrA in its context of regulation --------------------------------- ### TyrR regulon Knowledge of the gene regulation impacting TyrA in prokaryotes is sparse, being limited to the lower-gamma Proteobacteria. Here, extensive information gathered from *E. coli*has revealed that *aroQ•tyrA*~*c\_Δ*~belongs to a large regulon controlled by the TyrR repressor. The limited phylogenetic distribution of TyrR, being present only in the lower-gamma Proteobacteria (Fig. [8](#F8){ref-type="fig"}), indicates that it is a recent evolutionary acquisition. In *E. coli*the regulon members that are under the control of *tyrR*are the *aroA*~Iα\_Y~\> *tyrA*operon, the *aroLM*operon, *tyrP*, *tyrB*, *aroP*, *mtr*, *aroA*~Iα\_F~and *tyrR*itself \[[@B55]\]. Thus, *tyrR*not only regulates the tyrosine branch of the pathway, but heavily impacts the common pathway and the transport of all three aromatic amino acids as well. Although outside the scope of this study, a logical expansion of it would be to examine the individual evolutionary histories of all the members of the contemporary *E. coli*TyrR regulon, i.e., asking when and in what order did these genes come under the influence of *tyrR*? Clearly, the recruitment of structural genes by *tyrR*has been recent, quite dynamic and even now, exhibits evidence of further ongoing change. For example, tyrosine phenol-lyase (a catabolic enzyme that is only sparsely present in gamma Proteobacteria) has been recruited to the TyrR regulons of *Erwinia herbicola*\[[@B56]\] and *Citrobacter freundii*\[[@B57]\]. In these cases, not only does TyrR perform as a transcriptional activator, but it requires cyclic AMP receptor protein and integration host factor to do so. As exemplified by *E. coli*, TyrR is generally a repressor. However, the transcriptional expression of *mtr*is activated by TyrR in the presence of TYR, and *tyrP*is activated in the presence of PHE (although it is repressed in the presence of TYR). The N-terminal domain of TyrR has been associated with the ability of TyrR to activate transcription in the case of *mtr*and *tyrP*\[[@B55]\]. Members of the *Haemophilus/Pasteurella*lineage have all lost the N-terminal domain and presumably all lack the ability to accomplish transcriptional activation, as has been demonstrated experimentally with *H. influenzae*TyrR \[[@B58]\]. In view of the interesting complexity that two operons (*mtr*and *aroLM*) in *E. coli*are regulated by both *tyrR*and *trpR*\[[@B55]\], it may be more than coincidental that *tyrR*and *trpR*seem to have emerged at about the same evolutionary time, i.e., coincident with the divergence of the upper-gamma Proteobacteria from the lower-gamma Proteobacteria (Fig. [7](#F7){ref-type="fig"}). A possible interaction between the TyrR and TrpR proteins has been noted \[[@B55]\]. ### PhhR in relationship to aromatic catabolism Arias-Barrau et al. \[[@B59]\] have recently characterized a central catabolic pathway (Hmg) that degrades homogentisate in three steps to fumarate and acetoacetate as a source of carbon and energy. One of several peripheral pathways feeding into the central pathway begins with PHE and produces homogentisate via the reactions of phenylalanine hydroxylase (Phh), aromatic aminotransferase, and 4-hydroxyphenylpyruvate dioxygenase (Hpd). In the absence of Phh, a shorter version of the peripheral pathway is one that can use TYR, but not PHE, as a source of carbon and energy. In Fig. [8](#F8){ref-type="fig"} the presence of Phh, Hpd, and Hmg segments of catabolism are mapped on a 16S rRNA tree. (The aromatic aminotransferase distribution is not shown since a multiplicity of aromatic aminotransferases having overlapping substrate specificities makes it particularly challenging to identify the functional role \[[@B49]\].) The cyanobacteria are unique among *Bacteria*in the use of Hpd for a completely different metabolic role unrelated to aromatic catabolism, i.e., the synthesis of vitamin E derivatives \[[@B60]\]. PhhR is a homolog of TyrR that has been shown in *P. aeruginosa*to be a divergently transcribed activator of a 3-gene operon needed for PHE and TYR catabolism \[[@B61]\]. The structural genes encode phenylalanine hydroxylase (*phhA*), carbinolamine dehydratase (*phhB*) and 4-hydroxyphenylpyruvate aminotransferase (*phhC*), and are powered by a σ^54^promoter \[[@B61],[@B62]\]. PhhR evolved relatively recently since it is only present in some gamma Proteobacteria (Fig. [8](#F8){ref-type="fig"}). The ancestral regulatory gene for the Phh peripheral pathway may have been a member of the leucine-responsive regulatory protein/asparagine synthase C (Lrp/AsnC) family judging from the adjacent and divergently oriented position of *asnC*genes to *phhA*in organisms such as *Xanthomonas axonopodis*and *Mesorhizobium loti*. A recent overview of the many different regulator families involved in the control of aromatic catabolism conveys an emerging sense of the variety and dynamic evolutionary processes that underlie aromatic catabolism \[[@B63]\]. Occasional distant homologs of *phhR*that appear in erratic fashion (see Fig. [9](#F9){ref-type="fig"}) may have some other regulatory function. For example, *Clostridium tetani*may use its PhhR homolog as a transcriptional activator of the gene encoding tyrosine phenol-lyase, as occurs in species of *Erwinia*\[[@B56]\] and Citrobacter \[[@B57]\]. ### Relationship of TyrR and PhhR What might be of origin of TyrR? TyrR is an anomalous member of the large prokaryote family of σ^54^enhancer-binding proteins that activate promoters dependent upon σ^54^. TyrR is unique within its homology grouping in that it targets σ^70^promoters for regulation, usually (but not always) being a repressor. Its closest homolog relative is PhhR, a canonical member of σ^54^enhancer-binding proteins. σ^54^-dependent enhancer proteins possess a highly conserved σ^54^-contact motif, GAFTGA, that is intimately involved in formation of the ternary complex of enhancer and σ^54^-RNA polymerase holoenzyme \[[@B64]\]. This is perfectly or nearly perfectly retained in the upper clades shown in Fig. [9](#F9){ref-type="fig"}, but is disrupted or completely absent in the clades between *Shewanella oneidensis*and *Pasteurella multocida*. The deeper phylogenetic distribution of PhhR (Fig. [8](#F8){ref-type="fig"}) suggests that TyrR evolved as a variant of PhhR. If correct, a regulatory gene that is oriented to catabolism (*phhR*), and itself of relatively recent origin, was conscripted even more recently for a completely new role in the regulation of primary biosynthesis (*tyrR*). Consistent with the latter supposition, the gain of TyrR generally correlates with the loss of competence for aromatic catabolism (Fig. [8](#F8){ref-type="fig"}). In contrast to the *Citrobacter*/*Salmonella*/*Escherichia*/*Shigella*and the *Pasteurella*/*Haemophilus*clades (whose TyrR homologs completely lack the GAFTGA motif), the remaining enteric clades have retained some residues in this region. These residues appear to be more than random remnants. It would be interesting to know if these residues have any functional significance. Indeed, the *Photobacterium/Vibrio*clade has retained the ancestral catabolic capabilities (Fig. [8](#F8){ref-type="fig"}) that would appear to demand retention of regulation via PhhR; yet the parallelism of the overall features of biosynthesis that are shared with other lower-gamma Proteobacteria would seem, on the other hand, to demand TyrR-mediated regulation. Perhaps this \"TyrR\" species participates in the regulation of both catabolic and biosynthetic genes. In this connection, it is interesting that Chaney et al. \[[@B64]\] found that a change in the GAFTGA motif of NifA could be partially \"suppressed\" by mutational changes in the N-terminal region of σ^54^. Even more striking as a possible evolutionary intermediate is the most outlying member of the lower- gamma Proteobacteria, *Shewanella oneidensis*. The position of its TyrR on the protein tree parallels expectations based on the 16S rRNA tree. This, plus the conservation of the TyrR regulon features and the overall gene synteny suggest *E. coli*-like function as TyrR, i.e. acting as a general repressor of regulon-member σ^70^promoters engaged in aromatic biosynthesis. However, the location of \"*tyrR*\" in *S. oneidensis*between *phhA*and *phhB*on one side, and *hmgB*and *hmgC*on the other side, strongly implies some kind of regulatory relationship with the catabolic genes. It would be quite interesting to determine experimentally whether \"TyrR\" in *S. oneidensis*(and maybe *Vibrio*, as well) can function as a repressor of the usual suite of σ^70^promoters, as well as an activator of σ^54^promoters for *phhA*/*phhB*and/or *hmgB*/*hmgC*. We suggest that TyrR evolved as a modified version of PhhR as follows. In view of the distribution of genes encoding PhhR and TyrR, as well as the aforementioned catabolic enzymes, the most parsimonious evolutionary scenario may be that central and peripheral catabolic pathways depicted in Fig. [8](#F8){ref-type="fig"} are quite ancient, but acquisition of PhhR as a σ^54^-dependent activator of phenylalanine hydroxylase was quite recent, originating about the time of divergence of gamma Proteobacteria. The clade defined by *Shewanella/Vibrio/Photobacterium*retained the catabolic pathway, whereas the other enteric lineages discarded the catabolic pathway, but retained PhhR, which was then recruited as a σ^70^-dependent regulator of aromatic biosynthesis (TyrR). ### Regulation by attenuation A widespread mechanism of regulation is via an attenuation mechanism whereby transcripts initiated at given promoters can be terminated prior to reaching the structural genes of an operon. Whether termination occurs usually depends on the balance (dictated by a variety of mechanisms) between mutually exclusive terminator and anti-terminator structures \[[@B65]\]. Merino has developed a website \[[@B66]\] to provide a database of putative attenuators ahead of various operons in *Bacteria*. We screened this database for likely attenuators relevant to the regulation of *tyrA*. Table [7](#T7){ref-type="table"} shows intriguing results that point to significant experimental work that would be desirable. *tyrA*is frequently a member of apparent supraoperons, as alluded to elsewhere in this paper, and some of these appear to be large gene clusters controlled by attenuation. Substantial work is needed to establish the depth of clades possessing a given attenuator. For example, the *hisH*~*b*~\> *tyrA*operon is reliably present throughout all alpha Proteobacteria. Since *Agrobacterium tumefaciens*has been found to possess an attenuator ahead of the *hisH*~*b*~\> *tyrA*operon, one might reasonably expect that most of the alpha Proteobacteria would possess the attenuator as well. If not, this attenuator would have been a very recent evolutionary innovation. Likewise, since the *aroA*~Iα\_Y~\> *tyrA*operon is widely present throughout the lower-gamma Proteobacteria, it would be interesting to confirm whether only the several species of *Vibrio*identified on the Merino website have an attenuator ahead of this operon (or whether other attenuators present are too weak to exceed the threshold imposed for preliminary detection). Some of the supraoperons that appear to be controlled by attenuation are interesting in that they contain the majority of genes needed for both PHE and TYR biosynthesis, e.g., the supraoperons in *Enterococcus faecalis*and *Streptococcus pneumoniae*. The latter organism displays two attenuator units. The supraoperon of *Desulfovibrio vulgaris*is novel in that it begins with two relatively rare genes encoding alternative enzyme steps for aromatic biosynthesis \[[@B52]\], denoted here as *aroA*\' and *aroB*\'. The leading five genes are adjacent to the seven-gene *trp*operon. Conclusion ========== Protein divergence within a vertical genealogy is not necessarily smooth and progressive. Qualitative biochemical innovations can result in a barrage of new selective pressures that result in evolutionary jumps. The consequent incongruence might easily be mistaken for LGT. The basis for evolutionary jumps will usually only be recognized by detailed and comprehensive analyses of any given subsystem. Examples in this study are as follows. (**i**) The *tyrA*~c\_Δ~gene of the lower-gamma Proteobacteria has diverged markedly from *tyrA*~c~of the upper-gamma Proteobacteria. Here the milestone event was fusion of *aroQ*to a putative *tyrA*~c~in the ancestor of lower-gamma Proteobacteria to produce *aroQ•tyrA*~c\_Δ~. Indels within the •*tyrA*~c\_Δ~domain presumably reflect a multiplicity of selections for functional interactions known to exist between the two fused domains as discussed earlier. (**ii**) Members of the subclass taxon *Actinobacteridae*possess TyrA~a~proteins that separate into two distinct groupings. The presumed ancestral ~NAD(p)~TyrA~a~that is still present in the *Actinobacteridae*\_1 clade very likely spawned the divergent NAD^+^-specific variety of TyrA~a~to yield the contemporary *Actinobacteridae*\_2 clade. The previous evolutionary analysis of *trp-*pathway genes \[[@B7],[@B8]\] can be viewed as a model for comparable studies with other gene systems. Expansion to the greater aromatic pathway is a logical extension. The dynamics of evolutionary change for *tyrA*can be matched to the dynamics exhibited by the *trp*system. For example, the lower-gamma Proteobacteria separate as a distinct phylogenetic unit from beta Proteobacteria and upper-gamma Proteobacteria on criteria defined by milestone evolutionary events that altered many character states of both tryptophan and tyrosine biosynthesis in the lower-gamma Proteobacteria. In the future one can anticipate that comprehensive and systematic phylogenetic analysis of each protein member of the TYR, PHE and TRP branches, the common aromatic-pathway trunk, and minor vitamin-like branches (such as the 4-aminobenzoate/folate branch) will accommodate a progressively integrated picture of the entire aromatic network, including catabolic pathways and many other specialized pathways. Methods ======= TyrA sequences -------------- Most TyrA sequences were obtained from the National Center for Biotechnology Information (NCBI) \[[@B16]\]. TyrA sequences from incomplete genomes were retrieved from the PEDANT database \[[@B67]\]. Several sequences in our curated TyrA collection have been corrected for incorrect translation start sites. Various curated TyrA sequence files can be downloaded from our website. These files include complete sequences, trimmed catalytic-core domains, and amino-acid sequence segments that are relevant to specificity for pyridine nucleotide or to specificity for the cyclohexadienyl substrate. The sequence files are summarized in Table [3](#T3){ref-type="table"}. Congruency groupings -------------------- TyrA proteins that cluster together on the TyrA protein tree in congruence with the 16S rRNA tree are called congruency groups. Exact correspondence of branching orders is not necessarily observed. So far, congruency groupings have been assembled for tryptophan-protein concatenates \[[@B8]\] and for TyrA proteins. Completion of equivalent work with the remaining aromatic-pathway segments will identify the repertoire of bacterial organisms in possession of a \"pure\" vertical genealogy with respect to aromatic biosynthesis. Congruency groups for TyrA can be accessed at our AroPath website \[[@B9]\], where a listing of the membership of congruency groups is maintained and updated. Any members of congruency-group clusters, whose position there is incongruent with 16S rRNA expectations, probably (but not necessarily) originated by LGT. The donor lineage may not be obvious, but as more genomes come on line, many cases where donor identities are currently unknown may become revealed. A listing of \"orphan\" TyrA proteins that belong to no current congruency group is given. Such orphans reflect the lack of sufficient genome representation in particular phylogenetic regions and undoubtedly will become the nucleus for additional congruency groups in due course. Alignments ---------- Multiple alignments were obtained by use of the ClustalW or ClustalX programs (Version 1.83) \[[@B68]\]. Manual adjustments were needed in the region of the GxGxxG motif for binding of pyridine nucleotide cofactor in the N-terminal region of TyrA proteins. Guidance for alignment was assisted by maximizing conformation with the Wierenga fingerprint, making allowance for a variable loop of 2--5 residues \[[@B26]\]. This was done with the assistance of the BioEdit multiple alignment tool of Hall (5.0.9 Edition) \[[@B69]\]. The refined multiple alignment was used as input for generation of a phylogenetic tree using the phylogeny inference package (Version 3.2), PHYLIP \[[@B70]\]. The neighbor-joining program was used to obtain a distance-based tree. The distance matrix was obtained by use of Protdist with a Dayhoff Pam matrix. The Seqboot and Consense programs were then applied to assess the statistical support of the tree using bootstrap resampling (1,000 replications). We also used the ANCESCON package \[[@B71]\], which produced similar results as shown in Fig. [2](#F2){ref-type="fig"} (albeit with even wider separation of many groups). The presence of regulatory domains (ACT and REG) was accepted when indicated by the Domain Architecture Retrieval Tool (DART) on the BLAST menu at NCBI \[[@B16]\]. Profile HMMs ------------ Profile hidden Markov models for each of the four TyrA subfamilies, TyrA~a~, TyrA~c~, TyrA~p~and *tyrA*~*c\_Δ*~, were built using Sean Eddy\'s HMMER package \[[@B72]\]. The HMMs were generated from our file of curated cyclohexadienyl-substrate core segments (see Table [3](#T3){ref-type="table"}). The seed sequences for each subfamily were first aligned using ClustalW \[[@B68]\]. The resulting multiple sequence alignments were then manually edited to produce more accurate alignment of the seed sequences. Finally, the edited multiple sequence alignments were used to generate the profile HMMs for each TyrA subfamily. Appraisal of gene fusions as one-time or multiple events -------------------------------------------------------- Whether any given contemporary gene fusions tracked back to a fusion event in a common ancestor or whether they occurred independently was evaluated by phylogenetic analysis of the individual protein domains and by inspection of the inter-domain linker region. Linker regions were determined by multiple alignments of fusion sequences with corresponding free-standing domains present in the closest relatives to organisms that lack the gene fusions. Authors\' contributions ======================= JS and MW integrated this specific effort with the broader and ongoing objective of implementing a dynamic and progressively updateable website (AroPath). JS also made substantial contributions to the bioinformatic work. CB did all of the art work and a majority of the bioinformatic analyses. RJ provided initial guiding concepts, a general organizational overview, and assembled the initial manuscript draft. CB, RJ, and JS contributed to the formulation of conclusions made, and all of the authors read and approved the final version of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table S1, entitled \"Key to organism acronyms and sequence identifiers\", is provided as supplementary material in an html document. This table contains the full collection of sequence data and annotations contained in this paper, and gene identification (gi) numbers are included and hyperlinked to facilitate access to the corresponding GenBank records. For future reference to a progressively updated table, refer to the AroPath website \[[@B73]\]. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ R. Jensen thanks the National Library of Medicine (Grant G13 LM008297) for partial support. This research is partially supported by the U. S. Army Research Institute of Infectious Diseases (USAMRIID). This analysis would not have been possible were it not for the yeoman efforts in comparative enzymology carried out over a period of more than 25 years by many students and postdoctoral fellows, most notably Graham S. Byng, Robert Whitaker, Alan X. Berry and Suhail Ahmad. This has produced an invaluable resource of comprehensive data, some of it unpublished. This paper is dedicated to our colleague and collaborator, John E. Gander, on the occasion of his 80^th^birthday. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Composite of alternative biochemical routes from chorismate (CHA) to L-tyrosine (TYR) in nature. An antibiotic synthesis branch from CHA is also shown (dimmed). Here the intermediates shown to intervene between chorismate and pristinamycin or chloramphenicol are *p*-aminochorismate (ADC), *p*-aminoprephenate (ADP), *p*-aminophenylpyruvate (APP), and *p*-aminophenylalanine (APA). PPA may be transaminated by prephenate aminotransferase (PAT) to yield *L*-arogenate (AGN). The four TyrA homologs and the reactions they catalyze are colored differently. Arogenate dehydrogenase (TyrA~**a**~) converts AGN to TYR. Alternatively, prephenate dehydrogenase (TyrA~**p**~) converts PPA to 4-hydroxyphenylpyruvate (HPP) which is then transaminated to TYR via an homolog of TyrB, AspC, HisH, or Tat \[49\]. A broad-specificity cyclohexadienyl dehydrogenase (TyrA~**c**~) is competent to catalyze either the TyrA~**a**~or the TyrA~**p**~reaction. PapC converts the 4-amino analog of PPA to the 4-amino analog of HPP. AroQ, AroH, and AroR are distinct homologs known to exist in nature for performance of the chorismate mutase reaction. Other abbreviations: AA, amino acid donor, KA, keto-acid accepter. ::: ![](1741-7007-3-13-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Phylogenetic tree for trimmed core domains of selected members of the TyrA Superfamily. Acronyms used for the various organisms are given in alphabetical order in Table 2. (A more extensive listing that includes organisms not shown in Fig. 2 and which also is hyperlinked to all of the individual GenBank records is given in Table S1. A similar table that also includes compilation of known and predicted substrate specificities is maintained at AroPath \[73\]. Lineages possessing experimentally established TyrA~**a**~, TyrA~p~, TyrA~c~or TyrA~c\_Δ~proteins are indicated by fill colors specified in the legend. Three specificity patterns for the pyridine nucleotide substrate are shown by line colors (see figure box). Although the cyanobacteria are depicted as having NADP^+^-specific TyrA proteins, some of them can also accept NAD, albeit to a lesser degree. All proteins having an aspartate residue homologous to D-32 of the *E. coli*~NAD~TyrA~c\_Δ~domain are presumed specific for NAD^+^. Fusion of TyrA domains with other catalytic domains is indicated within grey boxes (AroQ•TyrA, TyrA•AroF, HisH~b~•TyrA, and TyrA•AroQ•PheA•ACT) using the convention of a bullet to represent the interdomain area. The boxes overlap any relevant lineages. TyrA proteins having carboxy-terminal fusions with regulatory domains (TyrA•ACT and TyrA•REG) are also shown. The distance scale bar at the bottom left represents substitutions per site. ::: ![](1741-7007-3-13-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Multiple alignment of the HMM consensus sequences obtained for different substrate-specificity groupings within cyclohexadienyl-substrate core segments (see Table 3). Invariant anchor residues are highlighted in yellow, conserved residues in grey. These consensus sequences will change continuously as corrections and refinements are made. The version shown was current as of April, 2005. ::: ![](1741-7007-3-13-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Alignment of the N-terminal glycine-rich P-loop of TyrA•ACT proteins from the Class *Actinobacteria*. These are specific for *L*-arogenate as substrate, but fall into two groups with respect to the pyridine nucleotide co-substrate. The top NAD^+^-specific group possesses an aspartate (D) at position 32 (*E. coli*numbering), whereas the bottom NAD^+^/NADP^+^group possesses an asparagine at the homologous position. Residue numbers are shown at the left. The species in the middle are color coded to match the hierarchical taxon positions obtained from NCBI. The variable loop of the Wierenga fingerprint \[26\], which in *E. coli*contains five residues (22--26), contains the minimal two residues in all of the *Actinobacteria*shown. The organisms on the right are color coded according to the taxonomic position indicated on the left (NCBI). The *Rubrobacter xylanophilus*TyrA~a~sequence is an orphan in the tree displayed in Fig. 2, as consistent with its outlying position in the taxonomy scheme. ::: ![](1741-7007-3-13-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Context of gene organization for *tyrA*, profiled against the 16S rRNA tree of the domain *Bacteria*. *pheA*, *hisH*~b~, *tyrA*, and *aroF*are color coded. Lineages typified by complete dispersal of aromatic-pathway genes are indicated by \"GENE DISPERSAL\". Gmet refers to *Geobacter metallireducens*; Dace refers to *Desulfuromonas acetoxidans*; and Ddes refers to *Desulfovibrio desulfuricans*. Consensus gene organizations are shown for the alpha and beta divisions of the Proteobacteria. The gamma division is subdivided to yield consensus gene organizations for the upper- and lower-gamma (enteric lineage) organisms. Genes that are adjacent and share a common transcriptional direction appear to reside in operons (or supraoperons). Any white spacing indicates substantial separation of the gene clusters shown in the genome. Genes of special interest are color coded, other genes of aromatic biosynthesis are shown in gray and all other genes are shown in white. ::: ![](1741-7007-3-13-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Zoom-in from Fig. 5 showing *tyrA*synteny for the beta Proteobacteria and the upper-gamma Proteobacteria. The tree shown, based upon 16S rRNA sequences of the indicated organisms, indicates correct branching orders, but (to facilitate presentation) is not strictly correct in proportion. Circled numbers (in violet) indicate deduced evolutionary events for the beta Proteobacteria (see top of Table 5), whereas circled numbers (in pink; see bottom of Table 5) correspond to deduced evolutionary events for the upper-gamma Proteobacteria. Gene organizations of organisms indicated are shown on the right. The dotted outlining of some gene boxes in *Coxiella burnetii*and in *P. syringae*indicates pseudogene status. ::: ![](1741-7007-3-13-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### Zoom-in from Fig. 5 showing *tyrA*synteny for the lower-gamma Proteobacteria (enteric lineage). Deduced phylogenetic events numbered on the left are described in Table 6. The branching position for *Buchnera*is as suggested in ref. \[7\]. Dotted horizontal lines near the top of the tree indicate branch lengths that were shortened for convenience of presentation. Dotted outlining of boxes around some genes indicates their pseudogene status. It is unknown if the various open reading frame (ORF) insertions are functional. ::: ![](1741-7007-3-13-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Distribution of modules of aromatic catabolism mapped on a 16S rRNA tree. In this figure, only presence or absence (not gene order) is indicated. The Phh module (orange) consists of phenylalanine hydroxylase (PhhA), carbinolamine dehydratase (PhhB), and tyrosine aminotransferase (not shown, see Text), and accomplishes the overall conversion of PHE to 4-hydroxyphenylpyruvate. The Hpd module (yellow) is 4-hydroxyphenylpyruvate dioxygenase, which converts 4-hydroxyphenylpyruvate to homogentisate. The Hmg module (blue) catalyzes the 3-step conversion of homogentisate to acetoacetate and fumarate. The distribution of PhhR and TyrR is shown in boxes. In some cases the HmgC member is shaded light blue to indicate that the gene encoding this isomerase could not be found and is probably encoded by an as yet unknown analog. Some long branches are drawn with gaps that represent 25% of the length of the scale bar. ::: ![](1741-7007-3-13-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### Protein tree of TyrR homologs. Nodes supported by bootstrap values of 998 or more are marked with solid circles, and the bootstrap values for nodes internal to these are shown. Generic names relevant to the organism abbreviations can be viewed in Fig. 8. A conserved region containing the **σ**^54^contact motif **GAFTGA**is highlighted as an orange band. \"Imperfect\" residues in this region are shown in lower-case fonts. Residue numbers are shown at the right. TyrR and PhhR are regulators of **σ**^70^and **σ**^54^promoters, respectively. Four **σ**^54^proteins of unknown function have very long branches, and to facilitate the visual presentation, the gaps in branch continuity shown represent a scale-bar distance of 0.1. Clades possessing **σ**^54^regulators are indicated with blue stripes, and **σ**^70^regulators are indicated with green stripes. ::: ![](1741-7007-3-13-9) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Abbreviations used to designate substrate specificities of *tyrA/*TyrA homologs ::: Abbreviation^a^ ------------------------- ----------------- -------------------------------------------------------------------------------------------------------------------------- Gene Gene Product Description of specificity^b^ *tyrA*~x~ TyrA~x~ Specificity for cyclohexadienyl substrate is unknown *tyrA*~c~ TyrA~c~ Broad-specificity cyclohexadienyl dehydrogenase (CDH) *tyrA*~p~ TyrA~p~ Narrow-specificity prephenate dehydrogenase (PDH) TyrA~c\_Δ~ TyrA~c\_Δ~ Broad-specificity cyclohexadienyl dehydrogenase having catalytic-core indels in correlation with an extra-core extension *tyrA*~a~ TyrA~a~ Narrow-specificity arogenate dehydrogenase (ADH) ~*NAD*~*tyrA*~a~ ~NAD~TyrA~a~ TyrA homolog is AGN-specific and NAD^+^-specific ~*NADP*~*tyrA*~a~ ~NADP~TyrA~a~ TyrA homolog is AGN-specific and NADP^+^-specific ~*NAD*(*P*)~*tyrA*~*a*~ ~NAD(P)~TyrA~a~ TyrA homolog is AGN-specific but utilizes either NAD^+^or NADP^+^ ~x~*tyrA*~x~ ~x~TyrA~x~ Specificity for both the cyclohexadienyl and pyridine nucleotide substrates is unknown ^a^Abbreviations in the upper-table (upper 5 rows) indicate the specificities for the cyclohexadienyl substrate. Abbreviations in the lower-table (lower 4 rows) indicate specificities for both cyclohexadienyl (right subscripts) and pyridine nucleotide substrates (left subscripts). Combinations not shown can be deduced from the examples given, e.g., a TyrA homolog specific for prephenate and NAD^+^would be designated ~NAD~TyrA~p~. ^b^The abbreviations CDH, PDH, and ADH (shown parenthetically) have been used frequently in the literature. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Key to organism acronyms ::: ------------------------------------------------------------------------------------------------------- **Organism** **Abbreviation**\ **Abbreviation**\ **in Paper** **on website^b^** --------------------------------------------------------------- ------------------- ------------------- *Acidithiobacillus ferrooxidans*ATCC 23270 Aferr *Acinetobacter*sp. ADP1 ACIN *Actinobacillus actinomycetemcomitans*HK1651 Aact *Actinomyces naeslundii*MG1 Anae *Actinoplanes teichomyceticus* Atei *Agrobacterium tumefaciens*strain C58 Atum *Amycolatopsis balhimycina* Abal *Amycolatopsis orientalis* Aori *Anabaena*sp. PCC 7120 ANAB *Arabidopsis thaliana* Atha *Archaeoglobus fulgidus*DSM 4304 Aful Aful\_1 *Azotobacter vinelandii* Avin Avin\_1 *Bacillis anthracis*str. A2012 Bant *Bacillus cereus*ATCC 14579 Bcer *Bacillus halodurans*C-125 Bhal Bhal\_2 *Bacillus stearothermophilus* Bste *Bacillis subtilis* Bsub *Bacillus thuringiensis*israelensis Bthu *Bifidobacterium longum*NCC2705 Blon Blon\_1 *Blochmannia floridanus* Bflo *Bordetella bronchisepticus* Bbro *Burkholderia cepacia*J2315 Bcep *Burkholderia fungorum*LB400 Bfun *Burkholderia mallei*ATCC 23344 Bmal *Burkholderia pseudomallei*K96243 Bpse Bpse\_6 *Campylobacter jejuni* Cjej *Chromobacterium violaceum*ATCC 12472 Cvio *Corynebacterium diphtheriae*NCTC 13129 Cdip *Corynebacterium efficiens*YS-314 Ceff *Corynebacterium glutamicum*ATCC 13032 Cglu Cglu\_1 *Desulfovibrio desulfuricans*G20 Ddes *Desulfovibrio vulgaris*subsp. *vulgaris*strain Hildenborough Dvul *Desulfuromonas acetoxidans* Dace Dace\_5 *Enterococcus faecalis*V583 Efae\_2 *Enterococcus faecium* Efae\_1 Efae\_1 *Erwinia carotovoa*subsp.*atroseptica*SCRI1043 Ecar *Escherichia coli*K12 Ecol *Geobacter metallireducens*GS-15 Gmet *Geobacter sulfurreducens*PCA Gsul *Gloeobacter violaceus*PCC 7421 Gvio *Haemophilus influenzae*Rd KW20 Hinf *Helicobacter hepaticus*ATCC 51449 Hhep *Helicobacter pylori*26695 Hpyl *Klebsiella pneumoniae*subsp. *pneumoniae*MGH 78578 Kpne *Leifsonia xyli*subsp. *Xyli*strain CTCB07 Lxyl *Listeria innocua*Clip 11262 Linn *Listeria monocytogenes*EGD-e Lmon *Lotus corniculatus*var. *japonicus* Lcor Lcor\_3 *Lycopersicon esculentum* Lesc *Methanococcus jannaschii* Mjan *Methanopyrus kandleri*AV19 Mkan Mkan\_1 *Methanosarcina barkeri*strain Fusaro Mbar *Methanothermobacter thermoautotrophicus*strain Delta H Mthe Mthe\_7 *Microbulbifer degradans*2--40 Mdeg *Mycobacterium avium*subsp. *paratuberculosis*strain k10 Mavi *Mycobacterium bovis*TrEMBL Mbov Mbov\_2 *Mycobacterium leprae*TN Mlep *Mycobacterium tuberculosis*CDC1551 Mtub *Myxococcus xanthus*DK 1622 Mxan *Neisseria gonorrhoeae*FA 1090 Ngon *Nitrosomonas europaea*ATCC 19718 Neur *Nocardia farcinica*IFM 10152 Nfar *Nonomuraea*sp. NONO *Nostoc punctiforme*PCC73102 Npun Npun\_1 *Novosphingomonas aromaticivorans*DSM 12444 Naro *Oceanobacillus iheyensis*THE831 Oihe *Oryza sativa*ssp. *japonica* Osat *Pantoea agglomerans* Pagg *Pasteurella multocida*subsp. *multocida*strain Pm70 Pmul *Photorhabdus luminescens*subsp. *laumondii*TT01 Plum *Prochlorococcus marinus*subsp. *pastoris*strain CCMP1378 Pmar\_1 Pmar\_3 *Prochlorococcus marinus*MIT9313 Pmar\_2 Pmar\_10 *Propionibacterium acnes*KPA171202 Pacn *Pseudomonas aeruginosa*PAO1 Paer Paer\_1 *Pseudomonas fluorescens*PfO-1 Pflu *Pseudomonas putida*KT2440 Pput *Pseudomonas stutzeri* Pstu *Ralstonia eutropha*JMP134 Reut *Ralstonia solanacearum*GMI1000 Rsol *Rhodobacter capsulatus* Rcap *Rhodobacter sphaeroides*2.4.1 Rsph *Rhodopseudomonas palustris*CGA009 Rpal *Rhodospirillum rubrum* Rrub Rrub\_1 *Rubrobacter xylanophilus*DSM 9941 Rxyl *Saccharomyces cerevisiae* Scer *Salmonella typhimurium*LT2 Styp Styp\_1 *Schizosaccharomyces pombe* Spom *Shewanella oneidensis*MR-1 Sone *Shewanella putrifacians* Spu *Staphylococcus aureus*subsp. Aureus MW2 Saur Saur\_2 *Streptococcus gordonii*str. Challis Sgor *Streptococcus pneumoniae*R6 Spne *Streptomyces avermitilis*MA-4680 Save *Streptomyces caeruleus* Scae Scae\_2 *Streptomyces coelicolor*A3(2) Scoe Scoe\_1 *Streptomyces lavendulae* Slav *Streptomyces pristinaespiralis* Spri *Streptomyces roseochromogenes*subsp. Oscitans Sros Sros\_1 *Streptomyces toyocaensis*strain 7 Stoy *Sulfolobus solfataricus*P2 Ssol *Sulfolobus tokodaii*strain 7 Stok *Synechococcus*sp. WH8102 SYNE\_1 SYNE\_1 *Synechococcus*sp. PCC7002 SYNE\_2 *Synechocystis*sp. PCC6803 SYNE\_3 SYNE\_3 *Thermobifida fusca* Tfus *Thermosynechococcus elongates*BP-1 Telo *Trichodesmium erythraeum*IMS101 Tery *Tropheryma whipplei*TW08/27 Twhi *Vibrio cholerae*O1 biovar eltor strain N16961 Vcho *Vibrio parahaemolyticus*RIMD 2210633 Vpar *Wolinella succinogenes*DSM 1740 Wsu *Xanthomonas campestris*pv. *campestris*strain ATCC 33913 Xcam *Xylella fastidiosa*9a5c Xfas *Yersinia enterocolitica*(type 0:8) Yent *Zymomonas mobilis*subsp. *mobilis*ZM4 Zmob ------------------------------------------------------------------------------------------------------- ^a^The system of acronym usage is: the first letter (capital) is the first letter of the genus followed by the first three letters (lower-case) of the species. If there is no species designation the first four letters of the genu*s*are used (all in capitals). Redundant 4-letter acronyms are distinguished by unique following numbers. See \[74\] for a comprehensive listing with hyperlinks to the Taxonomy database records and the GenBank records at NCBI. ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Curated TyrA amino-acid sequence files at AroPath \[35\] ::: ------------------------------------------------------ Complete TyrA sequences Catalytic-core domains^a^ Pyridine-nucleotide discriminator segments^b^ NAD^+^-specific NADP^+^-specific Broad specificity Cyclohexadienyl-substrate core segments Arogenate-specific (TyrA~a~) Prephenate-specific (TyrA~p~) Broad specificity TyrA~c~ TyrA~c\_Δ~ Pseudogene TyrA sequences ------------------------------------------------------ ^a^Trimmed free of N-terminal or C-terminal extensions, including any fusions with regulatory domains or other catalytic domains. ^b^High-glycine βαβ Rossmann fold at the N-terminus. ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Cyclohexadienyl substrates and inhibitors of TyrA proteins possess identical sidechains ::: **Organism** **Co-substrate** **Substrate(s)** **Inhibitor(s)**^a^ **Reference** ------------------------------ ------------------ ------------------ --------------------- --------------- *Synechocystis*sp. NADP^+^ AGN TYR \[25\] *Arabidopsis thaliana* NADP^+^ AGN TYR \[27, 28\] *Nitrosomonas europaea* NADP^+^ AGN None \[21\] *Corynebacterium glutamicum* NAD(P)^+^ AGN None \[42, 43\] *Neisseria gonorrhoeae* NAD^+^ PPA^b^ HPP \[21\] *Pseudomonas stutzeri* NAD^+^ PPA/AGN HPP/TYR \[1\] *Pseudomonas aeruginosa* NAD^+^ PPA/AGN HPP/TYR \[17\] *Zymomonas mobilis* NAD^+^ PPA/AGN None \[18\] ^a^Abbreviation: HPP, 4-hydroxyphenylpyruvate. ^b^This TyrA~c~enzyme has an overwhelming preference for PPA, but will use AGN poorly. ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Key to evolutionary events asserted in Figure 6 ::: **Group** **Evolutionary event(s) proposed** ------------- ------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Beta 1 Dispersal of *aroQ•pheA*\> *hisH*~*b*~\> *tyrA*away from one another and away from *gyrA*\> *serC*and from *cmk*\> *rspA*\> *himD*; inversion of *aroF*with respect to *cmk*. 2 Complete dispersal of all nine genes originally in the *gyrA*/*himD*linkage group. 3 Insertion of *serA*after *serC*^*a*^; separation of *tyrA*and *aroF*to yield the separated 6-gene unit and 4-gene unit shown. 4 Expulsion of *hisH*~b~from the genome; insertion of \'ORF\' after *serC*. 5 Fusion of *tyrA*with *aroF*. 6 Loss of *hisH*~b~from genome. Upper-Gamma 1 Insertion of *serA*after *serC*^*a*^; insertion of *aroA*~Iα~after *hisH*~b~. 2 Translocation of *hisH*~*b*~and *tyrA*to other regions, leaving two separated 3-gene units. 3 Fusion of *tyrA*with *aroF*. 4 Loss of *hisH*~b~. 5 N-terminal deletion of •*aroF*domain, and acquisition of new *aroF*gene (probable LGT). 6 Separation of *cmk*\> *rpsA*\> *himD*from *aroQ•pheA*\> *tyrA•aroF*. 7 Insertion of 4 unknown genes between *gyrA*and *serC*in opposite orientation and separation of *gyrA*\> ORF \> ORF \> ORF \> *serC*from *aroQ•pheA*\> *tyrA•aroF*. 8 Loss of *himD*; translocation of *serC*away from *gyrA*and *aroQ•pheA*. ^a^Since both *Nitrosomonas*(beta Proteobacteria) and *Acidothiobacillus*(upper-gamma Proteobacteria) emerge at deep positions in the tree of Fig. 5, an almost equally parsimonius possibility is that the ancestral *serA*was retained in this syntenic position in these two genera, but was transposed elsewhere shortly after early divergence. ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Key to evolutionary events asserted in Figure 7 ::: **Number** **Evolutionary events proposed** ------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 1 Escape of *aroQ•pheA*and *tyrA*from the ancestral *gyrA*\> *serC*\> *aroQ•pheA*\> *hisH*~*b*~\> *tyrA*\> *aroF*\> *cmk*\> *rpsA*\> *himD*supraoperon. Origin of an *aroQ•tyrA*fusion. Origin of the *aroA*~Iα\_Y~\> *aroQ•tyrA*operon. Addition of *tyrR*. Addition of third *aroA*~Iα~species: *aroA*~Iα\_F~. 2 Fusion of *aroQ•pheA*with *aroA*~Iβ~pseudogene of unknown origin. Replacement of *hisH*~b~by *aspC*duplicate linked with three ORFs. 3 Dissociation of *gyrA*and *serC*. 4 Removal of all genes intervening between *aroQ•pheA*and *aroQ•tyrA*. 5 Dissociation of *aroF*from both *serC*and *cmk*\> *rpsA*\> *himD*. Insertion of *trpR*within the intervening region between *aroQ•pheA*and *aroQ•tyrA*. 6 Dissociation of *serC*\> *hisH*~b~\> *aroF*from *cmk*\> *rpsA*\> *himD*. 7 Loss of *aroA*~Iα\_Y~from *tyr*operon. 8 *aroF*becomes dissociated from *hisH*~b~, and *aroA*~Iα\_Y~is removed from the *tyrA*operon. 9 *ORF*\> *gyrA*is inserted after *aroF*. 10 *aroQ•tyrA*becomes a pseudogene. 11 *hisH*~b~is lost. 12 *himD*is lost. 13 *cmk*, *himD*and *aroA*~Iα\_Y~\> *aroQ•tyrA*are lost. 14 *aroF, himD*, *aroQ•pheA*, and *aroA*~Iα\_Y~\> *aroQ•tyrA*are lost. 15 All intervening genes between *aroQ•pheA*and *aroQ•tyrA*are eliminated. 16 *pheA*domain of *aroQ•pheA*becomes a pseudogene. 17 Insertion of *ycaL*between *aroF*and *cmk*. 18 Insertion of ORF between *aroF*and *ycaL*. 19 Insertion of ORF between *aroQ•pheA*and *aroQ•tyrA*. ::: ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Putative attenuators^a^associated with *tyrA* ::: **Organism** **Gene organization**^b^ **Fig**^c^ ----------------------------------- ------------------------------------------------------------------------------------ ------------ *Agrobacterium tumefaciens* ^¬^*hisH*~*b*~\> ***tyrA*** 4^d^ *Bacillus anthracis* ^¬^*aroG*\> *hisH*~*b*~\> ***tyrA***\> *aroF* (11) *Bacillus cereus* ^¬^*aroG*\> *hisH*~*b*~\> ***tyrA***\> *aroF* (11) *Bacillus halodurans* ^¬^***tyrA***\> *aroF* (11) *Bacteroides thetaiotaomicron* ^¬^*pheA*\> *hisH*~*b*~\> *aroA*~*Iβ*~•aroQ \> ***tyrA*** 4 *Bordetella parapertussis* ^¬^*gyrA*\> *serC*\> *aroQ•pheA*\> ***tyrA***\> *aroF*\> *cm k*\> *rpsA*\> *himD* 5 *Desulfovibrio vulgaris* ^¬^*aroA*\' \> *aroB*\' \> *aroQ•pheA*\> *aroF*\> ***tyrA***\> \[*trp*operon\] 4 *Enterococcus faecalis* ^¬^*aroD*\> *aroA*~*Iβ*~\> *aroB*\> *aroG*\> ***tyrA***\> *aroF*\> *aroE*\> *pheA* 4 *Lactococcus lactis* ^¬^*ysaA*\> *blrG*\> *kinG*\> ***tyrA***\> *aroF*\> *aroE*\> *pheA* 4 *Lactobacillus plantarum* ^¬^ORF \> *aroG*\> *ORF*\> *aroF*\> ***tyrA***\> *aroE* *Listeria innocua* ^¬^*aroG \> aroB \> aroH \> hisH*~*b*~\> ***tyrA***\> *aroF* 4,(11) *Streptococcus pneumoniae* ^¬^*ORF\>aroC*~*I*~*\>aroD\>aroB\>aroG\>****tyrA***\> ^¬^*ORF\>aroF\>aroE\>pheA* 4 *Thermoanaerobacter tencongensis* ^¬^*pheA \> aroA*~*I*β~\> ***tyrA***\> *aroF*\> *ORF*\> *ORF* *Thermus thermophilus* ^¬^*aroA*~*Iβ*~\> ***tyrA*** *Vibrio parahaemolyticus* ^¬^*aroA*~*Iα\_Y*~\> ***tyrA*** 6 *Vibrio vulnificus* ^¬^*aroA*~Iα\_Y~\> ***tyrA*** 6 ^a^Attenuators were extracted from the website of Merino \[66\]. Links are provided for viewing the complete data, including a visualization of the putative attenuator structures. ^b^The symbol ^¬^is used for attenuators. Genes encoding the alternative biochemical steps that were recently reported for formation of dehydroquinate from aspartate semialdehyde and ketohexose 1-phosphate \[52\] are designated *aroA*\' and *aroB*\'. ^c^Refers to figures within this manuscript or, if enclosed within parentheses, to the figure in ref. \[7\]. ^d^See the consensus gene organization for α Proteobacteria. :::

PubMed Central

2024-06-05T03:55:58.327311

2005-5-12

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173090/", "journal": "BMC Biol. 2005 May 12; 3:13", "authors": [ { "first": "Jian", "last": "Song" }, { "first": "Carol A", "last": "Bonner" }, { "first": "Murray", "last": "Wolinsky" }, { "first": "Roy A", "last": "Jensen" } ] }

PMC1173091

Background ========== Three protein complexes that are conserved from yeast to humans, the 19S proteasome lid, the CSN, and eIF3, contain subunits characterized by two protein motifs: the MPN ([M]{.underline}pr1/ [P]{.underline}ad1 [N]{.underline}-terminal) and the PCI ([p]{.underline}roteasome/[C]{.underline}SN/e[I]{.underline}F3) domains \[[@B1]\]. The proteasome 20S catalytic particle and the 19S regulatory subunit cooperate in degrading polyubiquitylated proteins (reviewed in \[[@B2]\]). The 19S proteasome can be separated into the base complex, which binds and unfolds substrates \[[@B3]\], and the eight subunit lid complex, which cleaves ubiquitin from substrates, thus apparently facilitating the entry of substrates into the catalytic proteasome barrel \[[@B4],[@B5]\]. In higher eukaryotes, the subunits of the 19S lid show pair-wise similarity to the eight subunits of the CSN \[[@B6]-[@B9]\]. In vivo, CSN promotes the activity of cullin-RING ubiquitin ligases \[[@B10]-[@B16]\], multiprotein complexes containing cullins, the RING protein RBX1, and one of several hundred substrate-specific adaptors \[[@B17]-[@B22]\]. The MPN domain containing CSN subunit 5 harbors a protease motif \[[@B23]\] that cleaves the ubiquitin-related peptide NEDD8 from cullins \[[@B24],[@B25]\]. This activity, acting in concert with the CSN-associated deubiquitylation enzyme Ubp12, was proposed to promote cullin function by facilitating the recruitment of labile substrate adaptors \[[@B11],[@B16],[@B26],[@B27]\]. The third PCI/MPN complex, eIF3, is more distantly related to CSN and the 19S lid (reviewed in \[[@B28]\]). Whereas human eIF3 consists of up to 13 subunits, consecutively named eIF3a -- l and GA17 \[[@B29],[@B30]\], budding yeast contains only six to eight subunits (depending on purification conditions). Five of these subunits are orthologs of human eIF3a, b, c, g, and eIF3i \[[@B31],[@B32]\] and appear to constitute a conserved core complex \[[@B31],[@B33]\]. Fission yeast contains the same five core subunits, in addition to the non-core subunits eIF3d/Moe1p, eIF3e/Int6p, and eIF3h \[[@B34]-[@B36]\]. A putative eIF3f ortholog was also identified, but biochemical evidence confirming it as an authentic eIF3 subunit functioning in protein synthesis is still outstanding \[[@B34],[@B35],[@B37],[@B38]\]. eIF3 is the most complex translation initiation factor and plays at least two important roles in protein synthesis. First, eIF3 binds to the 40S ribosome and facilitates loading of the Met-tRNA/eIF2· GTP ternary complex to form the 43S preinitiation complex. Subsequently, eIF3 apparently assists eIF4 in recruiting mRNAs to the 43S complex. A critical in vivo function of eIF3 core subunits in these processes was indicated by the lethality of the respective budding yeast deletion strains \[[@B31]\]. In contrast, in fission yeast, the non-core subunits eIF3d and eIF3e are not essential for viability or global protein synthesis \[[@B34],[@B36],[@B38],[@B39]\]. It was therefore proposed that distinct subclasses of eIF3 complexes, containing different combinations of core and non-core subunits, may regulate specific subsets of mRNAs in fission yeast \[[@B34],[@B36],[@B38],[@B39]\]. Our study provides the first experimental evidence substantiating this hypothesis by demonstrating that biochemically distinct eIF3 complexes defined by the PCI domain proteins eIF3e and eIF3m (a novel eIF3 protein) associate with different sets of mRNAs. Results ======= Both *csn6*and *csn7b*are essential genes ----------------------------------------- CSN complexes of higher eukaryotes typically contain eight distinct subunits (two MPN and six PCI proteins). However, in fission yeast only six subunits are known (Csn1p, 2p, 3p, 4p, 5p, 7Ap; Ref. \[[@B13]\]), none of which are essential for viability \[[@B40]-[@B42]\]. We noticed two genes in the *Schizosaccharomyces pombe*genome database, originally annotated as *csn6*(SPBC4C3.07) and *csn7b*(SPAC1751.03), which encode MPN and PCI domain containing proteins with considerable similarity to metazoan CSN6 and CSN7B, respectively (data not shown). In order to determine whether these genes might function in the known biochemical pathways regulated by CSN, cullin deneddylation and Ubp12p-mediated deubiquitylation \[[@B24],[@B26]\], we deleted *csn6*and *csn7b*in wild-type diploid cells. Upon sporulation and tetrad analysis, only two viable spores could be recovered in each case, both of which were uracil auxotroph (Fig. [1A](#F1){ref-type="fig"}, and data not shown). Thus, unlike the genes encoding the six known subunits of the fission yeast CSN \[[@B41],[@B42]\], *csn6*and *csn7b*are essential. Csn6p and Csn7Bp associate with eIF3 components ----------------------------------------------- To elucidate the essential functions of Csn6p and Csn7Bp, we sought to identify their interacting proteins. To this end, we attempted to modify *csn6*and *csn7b*at their endogenous genomic loci with five consecutive protein A epitope tags (proA). The C-terminal tag also contained a cleavage site for Tobacco Etch Virus (TEV) protease upstream of the protein A moieties. Diploid analysis indicated that C-terminal tagging destroyed the essential function of *csn6*(data not shown). In contrast, a *csn7b-proA*haploid strain was viable and did not display any obvious phenotypes (data not shown). This indicated that the tagged allele was functional. Total protein lysate from the *csn7b-proA*haploid strain was absorbed to Immunoglobulin G (IgG) resin and specifically retained proteins were eluted with sodium dodecyl sulfate (SDS) or by cleavage with TEV protease. The two eluates were resolved by gel electrophoresis and Csn7Bp-proA interacting proteins were identified by peptide mass fingerprinting using matrix-assisted laser desorption/ionization-time-of-flight (MALDI-TOF) mass spectrometry. Both eluates contained an identical set of proteins. In addition to Csn6p, the Csn7Bp complex contained the eIF3 core subunits eIF3a, b, c, and i, as well as the non-core subunit eIF3h (Fig. [1B](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). The eIF3g core subunit was not detected using MALDI-TOF in this purification, but was detected in an independent purification of the Csn7Bp complex using the more sensitive method of nano-liquid chromatography and tandem mass spectrometry (LC-MS/MS, see Fig. [5A](#F5){ref-type="fig"}). However, eIF3e and eIF3d, two previously described components of *S. pombe*eIF3 \[[@B34],[@B36],[@B38],[@B39]\] were consistently absent from the Csn7Bp complex when analyzed either by MALDI-TOF or by LC-MS/MS (Fig. [1B](#F1){ref-type="fig"} and [5A](#F5){ref-type="fig"}). Notably, no authentic CSN subunits were found in the Csn7Bp complex, consistent with the previous demonstration that *S. pombe*CSN contains only six non-essential subunits \[[@B13]\]. Structural analysis indicated that Csn6p is not only related to human CSN6, but shows a slightly greater overall similarity to human eIF3f (data not shown). Based on the functional characterization described below (Fig. [2](#F2){ref-type="fig"}), fission yeast Csn6p therefore appears to be the ortholog of human eIF3f. A Csn7Bp ortholog is not known in the human eIF3, although the recently identified subunit GA17 \[[@B30]\] shows considerable similarity to Csn7Bp (K. Hofmann, personal communication). In addition, interactions of metazoan CSN7 with eIF3 subunits have previously been observed \[[@B43],[@B44]\]. With regards to the unified nomenclature of eIF3 subunits \[[@B29]\], we propose eIF3m as a more appropriate name for the novel fission yeast eIF3 subunit encoded by *csn7b*(see Table [1](#T1){ref-type="table"}). Throughout this manuscript, we will use the proposed new names for the *eif3f*and *eif3m*genes and their products. The original annotations have recently been revised by the curator of the *S. pombe*genome database, in order to reflect the findings presented here. The eIF3f and eIF3m proteins are primarily cytoplasmic ------------------------------------------------------ The evidence presented above suggested that eIF3f and eIF3m are essential subunits of *S. pombe*eIF3. To substantiate this possibility, we determined the subcellular localization of eIF3f and eIF3m. If these proteins were authentic eIF3 subunits involved in protein synthesis, they should assume a subcellular localization similar to known eIF3 subunits. We therefore prepared three strains with green fluorescent protein (GFP) tagged alleles of eIF3b, eIF3e, and eIF3m integrated into their respective genomic loci. In addition, since C-terminal GFP-tagging destroyed the essential function of the small eIF3f protein, we prepared a strain mildly overexpressing N-terminally GFP-tagged eIF3f from a pREP81 plasmid. Fluorescence microscopy of live cells revealed that all four proteins were primarily localized in the cytoplasm (Fig. [2A](#F2){ref-type="fig"}), which would be consistent with a function in protein translation. In contrast, CSN subunits are largely nuclear in fission yeast (\[[@B41]\], and our own unpublished observation). eIF3f and eIF3m are required for global protein synthesis --------------------------------------------------------- To directly assess the potential role of eIF3f and eIF3m in protein synthesis, we prepared haploid *eif3f*and *eif3m*deletion strains maintained viable by plasmids driving the regulated expression of *eif3f*and *eif3m*, respectively. Expression from these plasmids could be turned off by the addition of thiamin to the growth media, resulting in a strong reduction in protein levels within 20 to 32 hours (Fig. [2B](#F2){ref-type="fig"}). Coincident with shut-off of eIF3f and eIF3m expression, global cellular protein synthesis was diminished by \~80%, as determined by metabolic labeling with ^35^S-methionine (Fig. [2C](#F2){ref-type="fig"}). Further biochemical analysis by sucrose gradient velocity centrifugation revealed a strong reduction in the formation of polysomes in *eif3f*and *eif3m*mutants following promoter shut-off (Fig. [2D](#F2){ref-type="fig"}). In contrast, as described previously \[[@B34],[@B36],[@B38],[@B39]\], *eif3e*mutants showed only a minor reduction in total protein synthesis (data not shown) and essentially normal polysomes (Fig. [2D](#F2){ref-type="fig"}). The conserved cullin deneddylation function of CSN was not impaired either in *eif3f*and *eif3m*mutants or in *eif3e*mutants, since cullin 1 (Cul1p) did not accumulate exclusively in the neddylated form as it does in *csn5*deletion strains (Fig. [2E](#F2){ref-type="fig"}). Similarly, unlike in *csn5*mutants (Fig. [2E](#F2){ref-type="fig"}; \[[@B26]\]), CSN/Ubp12p-mediated inhibition of Cul1p in vitro ubiquitin ligase activity was not impaired upon shut-off of either *eif3f*or *eif3m*expression. Cul1p neddylation and activity were also unaffected in *eif3e*mutants (Fig. [2E](#F2){ref-type="fig"}). These data suggested that eIF3f and eIF3m are bona fide subunits of fission yeast eIF3 but not CSN, performing essential functions in protein synthesis similar to eIF3 core subunits. The eIF3m and eIF3e proteins define distinct eIF3 complexes ----------------------------------------------------------- Although the purification of the eIF3m complex robustly retrieved roughly stoichiometric amounts of known eIF3 subunits, two subunits, eIF3d and eIF3e, were conspicuously absent (Fig. [1B](#F1){ref-type="fig"}, Table [1](#T1){ref-type="table"}). To exclude the possibility that we had overlooked these subunits because they were obscured by the IgG bands, we affinity-purified eIF3e-associated proteins from an *eif3e-proA*strain. The retrieved proteins were identified using MALDI-TOF mass spectrometry as before. Like the eIF3m complex, the eIF3e complex contained roughly stoichiometric amounts of the eIF3 core subunits a, b, c, g, and i (Fig. [3A](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}). In addition, eIF3f was present while, unexpectedly, eIF3m and eIF3h were missing. We did not detect the proteasome lid subunit Rpn5p, which previously had been shown to associate with eIF3e and eIF3d \[[@B45]\]. This observation indicated that the described eIF3d/eIF3e/Rpn5p interaction is either unstable under our purification conditions or contains only a minor fraction of the total eIF3e engaged in protein interactions. These results raised the intriguing possibility that fission yeast contains two distinct eIF3 complexes that comprise an overlapping set of core subunits, but are distinguished by the presence of either the essential PCI protein eIF3m and the MPN domain protein eIF3h or the non-essential PCI protein eIF3e and its binding partner eIF3d. To exclude the possibility that our mass spectrometry analysis missed substoichiometric amounts of eIF3d in the eIF3m complex, we analyzed eIF3m complexes by immunoblotting with eIF3d antisera. Whereas eIF3d was readily detectable in the eIF3e complex, it was undetectable in the eIF3m complex (Fig. [3B](#F3){ref-type="fig"}). In the reciprocal experiment, eIF3d antibodies co-immunoprecipitated eIF3e-13Myc, but not eIF3m-13Myc or Csn5p-13Myc (Fig. [3C](#F3){ref-type="fig"}). To exclude the possibility that our cell lysis or affinity purification conditions led to uncontrolled release of eIF3 subunits, thus mimicking the existence of distinct subcomplexes, we performed a purification using the protein A-tagged eIF3b core subunit as bait. As determined by LC-MS/MS performed on excised bands detected in the complex upon PAGE separation, the eIF3b complex contained all known eIF3 core components as well as eIF3f and eIF3m (Fig. [3D](#F3){ref-type="fig"}). The eIF3e and eIF3d proteins were also present, albeit in clearly substoichiometric amounts. These findings indicate that purification of eIF3b resulted in the copurification of the distinct eIF3m- and eIF3e-containing complexes. Taken together, these data strongly suggested the existence of two distinct eIF3 complexes defined by the PCI proteins eIF3m and eIF3e. The eIF3m and eIF3e proteins associate with distinct sets of mRNAs ------------------------------------------------------------------ The finding that *eif3m*is essential, whereas *eif3e*is dispensable, suggested the possibility that the different eIF3 complexes they define regulate different subsets of mRNAs. To approach this, we sought to identify mRNAs specifically associated with the eIF3m and eIF3e complexes. The complexes were affinity-purified as described above in the presence of RNAse inhibitors, and the associated RNA was extracted, amplified, and converted to Cy3-labeled cDNA as described in the Methods section. The Cy3-labeled cDNA was hybridized competitively with Cy5-labeled cDNA prepared from total RNA onto microarrays representing all 4988 predicted *S. pombe*open reading frames (ORFs) and RNAs. Based on the background subtracted hybridization signals, eIF3-associated mRNAs were ranked according to their factor of enrichment in the eIF3m and eIF3e complexes over a mock purified sample (\[see [additional file 1](#S1){ref-type="supplementary-material"}\]). The microarray analysis suggested a global role for eIF3m and a more restricted function of eIF3e in translation. Using an arbitrary cut-off value of three-fold enrichment over mock, eIF3m associated with 2464 different mRNAs, whereas eIF3e associated with only 520 distinct species (Fig. [4A](#F4){ref-type="fig"}). We observed 414 mRNAs enriched more than three-fold in both samples. (Fig. [4A](#F4){ref-type="fig"}). In addition, our analysis revealed 106 mRNAs uniquely enriched in the eIF3e complex, whereas 2050 transcripts were uniquely enriched in the eIF3m complex (Fig. [4A,C](#F4){ref-type="fig"}). A ranked list of all 106 mRNAs that were uniquely enriched more than three-fold in the eIF3e complex was assembled (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]). Approximately 75% of these mRNAs fell into one of three categories: mRNAs encoding proteins involved in intermediary metabolism (25.5%), those encoding proteins involved in protein metabolism (14.2%), and those encoding proteins with unknown functions (35.8%; Fig. [4B](#F4){ref-type="fig"}). The remainder was distributed roughly equally among four categories: mRNAs encoding proteins involved in nucleic acid metabolism, those encoding proteins involved in transcription, those encoding transporter proteins, and those encoding unclassified proteins. A corresponding list of the 106 most highly enriched mRNAs exclusively present in the eIF3m complex (out of a total of 2050 enriched by a factor greater than three) revealed that \~58% also fell into three categories: mRNAs encoding proteins involved in nucleic acid metabolism (9.4%), those encoding transporter proteins (18.9%), and those encoding proteins of unknown function (20.8%). The remainder was roughly equally distributed among four categories: mRNAs encoding proteins involved in intermediary metabolism, those mRNAs encoding proteins involved in protein metabolism, those encoding proteins involved in transcription, and those encoding unclassified proteins (Fig. [4B](#F4){ref-type="fig"}, \[see [additional file 2](#S2){ref-type="supplementary-material"}\]). To exclude the possibility that our affinity purification protocol unspecifically enriched highly expressed mRNAs, we averaged the data from the three competively hybridized samples of total *S. pombe*cDNA to determine the relative expression rank of each mRNA (rank 1 = most abundant mRNA, rank 5407 = least abundant mRNA; see legend to Fig. [4C](#F4){ref-type="fig"} for detail). A plot of expression rank vs. factor of enrichment of the 106 mRNAs most enriched in the eIF3e and eIF3m complexes revealed that the eIF3m complex contained mRNAs that were distributed equally over the entire range of expression levels. In contrast, the eIF3e complex was enriched for rare mRNAs (\~65 % of eIF3e-associated mRNAs had an expression rank \> 4000; see Fig. [4C](#F4){ref-type="fig"}). The lack of a linear correlation between mRNA expression level and enrichment in the eIF3e and eIF3m complexes argues against unspecific copurification of abundant mRNAs. Specific association of mRNAs with eIF3m and eIF3e complexes ------------------------------------------------------------ To validate the results from the microarray hybridizations, we chose several mRNAs at random from each list of 106 and used RT-PCR to confirm their interactions with the respective eIF3 complexes. eIF3e-proA- and eIF3m-proA-associated complexes were affinity-purified as before and separated by PAGE. Both complexes showed the same differential subunit composition as described above (Fig. [5A](#F5){ref-type="fig"}). The associated RNA was purified, split into equal portions, and employed in RT-PCR reactions with gene-specific primer pairs. The RT-PCR analysis of a total of fourteen mRNAs included eight mRNAs identified in the eIF3e complex, five mRNAs identified in the eIF3m complex, and one mRNA found to be associated with both complexes with similar enrichment factors. These mRNAs varied widely in their factors of enrichment and their enrichment ranks as indicated in Fig. [5B](#F5){ref-type="fig"}. Without exception, all eight mRNAs found to be enriched in the eIF3e complex by microarray analysis were also found preferentially associated with this complex by RT-PCR (Fig. [5B](#F5){ref-type="fig"}). Based on this result, we calculate a false positive rate of less than 0.11 with a 95% confidence interval of 0.00 -- 0.31. Whereas mRNAs with a high enrichment factor (\>10, \[see [additional file 2](#S2){ref-type="supplementary-material"}\]) appeared to be exclusively associated with eIF3e-proA (within the limit of detection), those with a lower enrichment factor (\<10) also displayed a low amount of association with the eIF3m complex (Fig. [5B](#F5){ref-type="fig"}). Conversely, all five mRNAs highly enriched in the eIF3m complex as determined by microarray analysis (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]) were almost exclusively bound to this complex when analyzed by RT-PCR (Fig. [5B](#F5){ref-type="fig"}). Notably, the mRNA encoding the proteasome subunit Rpn5p, which was found in both complexes by microarray analysis with similar enrichment factors (eIF3e = 5.1; eIF3m = 8.1) was also confirmed as being associated with both eIF3 complexes by RT-PCR (Fig. [5B](#F5){ref-type="fig"}). These results strongly suggest that the eIF3m and eIF3e complexes specifically associate with distinct sets of mRNAs. Discussion ========== PCI proteins define distinct eIF3 complexes ------------------------------------------- Our results show that Csn6p and Csn7Bp are subunits of fission yeast eIF3. Csn6p was shown to be the eIF3f subunit conserved in all eukaryotes except budding yeast. Consistent with this finding, eIF3f has previously been found to be associated with eIF3g, although the role of eIF3f in protein synthesis had not been examined \[[@B34]\]. Our study further identified two distinct eIF3 complexes defined by the PCI proteins eIF3e and eIF3m (Fig. [3A](#F3){ref-type="fig"}). No direct equivalent of eIF3m is apparent in eIF3 preparations from other eukaryotes. Considering the essential function of eIF3m in protein synthesis in fission yeast, it is possible that other PCI proteins substitute for eIF3m in other organisms. Consistent with this idea is the finding that the PCI domain containing GA17, a recently identified protein copurifying with human eIF3 \[[@B30]\], shows more similarity to eIF3m than to any other PCI protein (K. Hofmann, personal communication). In addition, the PCI domain containing CSN subunits 1, 3, 7, and 8 were found to interact with eIF3 subunits in *Arabidopsis thaliana*and in human cells \[[@B43],[@B44],[@B46]\]. In addition, Pci8p/Csn11p, a non-essential PCI subunit of the budding yeast CSN \[[@B47],[@B48]\], copurifies with essential eIF3 core subunits \[[@B49]\]. Although the functional significance of these interactions is still unclear, our findings raise the intriguing possibility that they define multiple subclasses of distinct eIF3 complexes. Whereas fission yeast appears to rely on an essential eIF3 complex classified by eIF3m and a non-essential complex specified by eIF3e, higher eukaryotes may divide the task of mRNA translation among a multitude of eIF3 complexes. Since the individual PCI proteins defining these subcomplexes are expected to be present in substoichiometric amounts relative to core components, as is eIF3e in the *S. pombe*core complex, they may have escaped detection in bulk eIF3 preparations from higher eukaryotes. Distinct eIF3 complexes and translational specificity ----------------------------------------------------- Although our finding of distinct eIF3 complexes that associate with different sets of mRNAs was surprising, circumstantial evidence pointing to their existence was provided in previous studies. For example, fission yeast eIF3e copurifies with eIF3 core subunits, and yet, unlike core subunits, eIF3e is not required for global protein synthesis or viability. Two competing models were proposed to explain the pleiotropic phenotype of *eif3e*deletion mutants, which includes slow growth in minimal media and meiotic defects \[[@B34],[@B36],[@B38],[@B39],[@B50],[@B51]\]. In the first model, eIF3e regulates the translation of all transcripts, but rare mRNAs may be more affected by a fractional reduction in translation efficiency and thus give rise to distinct phenotypes. In the second model, eIF3e regulates the translation of a specific subset of mRNAs encoding proteins whose depletion leads to distinct phenotypes but not lethality. The findings presented here strongly favor the second model. Our microarray analysis revealed 106 mRNAs uniquely enriched in the eIF3e complex, many being rare, whereas 2050 transcripts were uniquely enriched in the eIF3m complex. The latter finding is consistent with the severe protein synthesis defect and lethality of *eif3m*deletion mutants. Whereas these mRNA lists were based on an arbitrary cut-off value of three-fold enrichment over mock, this criterion proved stringent enough to confirm specific association with their respective eIF3 complexes of all 14 mRNAs retested by RT-PCR. Thus, since we calculate the false positive rate to be less than 0.11 with 95% confidence interval 0.00 -- 0.31, the maximum number of mRNAs listed potentially false positively as enriched in the eIF3e complex would be 33 out of 106 (with 95% confidence). However, we consider it unlikely that all mRNAs found enriched in the eIF3e complex are exclusive translational targets of this complex, because the cellular phenotypes of *eif3e*mutants are rather discrete. Although many of the eIF3e-associated mRNAs encode essential proteins, viability, global protein synthesis, and polysome formation are not affected in these mutants when they are grown in rich media (Fig. [2D](#F2){ref-type="fig"}, and data not shown). It is therefore likely that eIF3m or other PCI domain proteins can deputize for eIF3e in many cases. In fact, our microarray study revealed that a substantial portion of eIF3e-associated mRNAs was also recovered in the eIF3m complex (Fig. [4A](#F4){ref-type="fig"}). Nonetheless, we strongly suspect that inefficient translation of a limited set of eIF3e-associated mRNAs contributes to distinct cellular phenotypes found in *eif3e*mutants (see below). The eIF3e protein and translational control during stress response ------------------------------------------------------------------ The *eif3e*mutants display marked sensitivity to a wide variety of cellular stresses, such as osmotic stress, nutrient starvation, and low temperature \[[@B34],[@B38],[@B39],[@B50],[@B52]\] In addition, *eif3e*mutants show synthetic lethality with proteasome mutants, which are presumed to accumulate a large load of misfolded proteins \[[@B45]\]. Conversely, overexpression of *eif3e*confers resistance to a broad spectrum of unrelated drugs, whose only commonality appears to be that they induce cellular stress \[[@B38]\]. Overexpression of eIF3e also activates genes involved in stress defense, such as thioredoxin \[[@B38]\]. Interestingly, the mRNA encoding thioredoxin was also identified as a putative translational target of eIF3e (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]). Finally, eIF3e undergoes relocalization into cytoplasmic foci in response to heat and osmotic stress, thus implicating it in stress regulation \[[@B35]\]. Global protein synthesis is actively switched off in response to cellular stress and nutrient deprivation (reviewed in Refs. \[[@B53],[@B54]\]). This shut-off protects cells from proteotoxicity by relieving chaperones of their load of unfolded client proteins, conserving amino acids for other essential functions, and attenuating the metabolic consequences of protein synthesis. However, long-term adaptation to stress conditions, repair and recovery require synthesis of new stress-induced proteins. During the stress response, some mRNAs must be translated in the presence of a repressed general translation machinery, suggesting that select eIFs can escape the global repression. Our studies raise the intriguing possibility that the eIF3e complex has such a specialized function during stress response. Lack of *eif3e*may prevent stress-induced synthesis of its critical translational targets, thus leading to the known stress sensitivity of the mutant. Substantial additional work will be required to address this possibility. Potential mechanisms underlying mRNA discrimination by eIF3 complexes --------------------------------------------------------------------- Several lines of evidence have begun to implicate other general translation factors, including eIF2 and eIF4, in translational specificity \[[@B55]\]. Importantly, these factors are downstream of important signaling cascades, including the nutrient-sensing TOR pathway and a variety of stress-induced kinases (reviewed in Refs. \[[@B55],[@B56]\]). Although the details of these regulatory mechanisms are beginning to emerge, it is still unclear exactly how eIF2 and eIF4 contribute to mRNA specificity. Since eIF3 cooperates with the cap-binding eIF4 complex in directing the 5\' end of mRNAs to the 40S ribosome, and several eIF3 subunits contain RNA binding motifs, our data is consistent with a model in which eIF3 complexes defined by distinct PCI proteins recognize determining features of mRNAs presented to the 40S subunit in concert with eIF4. In agreement with this idea is the recent finding that the human PCI protein eIF3a specifically facilitates the translation of the mRNA encoding ribonucleotide reductase M2 \[[@B57]\]. In addition, *A. thaliana*eIF3h, a non-core subunit, was recently shown to be specifically required for efficient translation of the transcription factor ATB2 \[[@B46]\]. While we have so far been unsuccessful in identifying common sequence elements within the upstream regions of mRNAs enriched in the eIF3e complex (data not shown), this analysis was severely hampered by the fact that 5\'UTRs are poorly defined in *S. pombe*. It is therefore still possible that critical features, including secondary structure determinants, lie within the 5\'UTR or other regions of eIF3e-associated mRNAs. Substantial additional work will be required to delineate the exact mechanism of how eIF3e may contribute to the translation of specific mRNAs and the potential impact of this mechanism on human cancers defective in INT6/eIF3e \[[@B58],[@B59]\]. Conclusion ========== We provide biochemical evidence for the existence of two distinct eIF3 complexes in fission yeast. These complexes contain an overlapping set of subunits, but are distinguished by the PCI proteins eIF3e and eIF3m. Based on the finding that the distinct eIF3 complexes associate with different mRNAs, we propose that they have different translational specificities. Methods ======= Fission yeast techniques ------------------------ Preparation of fission yeast cultures, cDNA cloning, yeast transformation, PCR-based genomic epitope tagging and gene deletion, mating, and tetrad analysis were carried out using standard methods as described \[[@B26]\]. The strains used in this study are listed in Table [2](#T2){ref-type="table"}. Affinity purification and mass spectrometry ------------------------------------------- Epitope-tagged strains (*eif3m-proA*and *eif3e-proA*strains) were grown in 12 L yeast extract and supplements (YES) medium to 1.5 optical density (OD) at 595nm wavelength and harvested by centrifugation. Cell lysates were prepared by bead lysis in a lysis buffer containing 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.5% Triton X-100, 1 mM dithiothreitol, supplemented with protease inhibitors (10 ug/ml leupeptin, 10 ug/ml pepstatin, 5 ug/ml aprotinin and 1mM phenylmethylsulfonylfluoride). Upon centrifugation at 17,000 rpm for 50 min at 4°C, 300 mg total protein in a volume of 50 ml was absorbed to 100 ul Dyna beads (Dynal Biotech, Oslo, Norway) coupled to whole rabbit IgG (Jackson Immunochemicals, West Grove, Pennsylvania, USA). Mock purifications were performed in parallel by absorbing lysate from cells lacking any tagged proteins to IgG beads. Beads were collected by a magnetic device and washed five times in 5 ml lysis buffer. TEV protease (Invitrogen, Carlsbad, California, USA) cleavage or elution in 5% SDS solution were used to collect bound proteins. Eluted proteins were separated on trycine gels or by conventional SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and visualized by Coomassie G250 staining. Protein bands were excised from gels, digested with trypsin (Promega Corporation, Madison, Wisconsin, USA), and subjected to mass spectrometry. The eIF3 complexes were analyzed by MALDI-TOF, and by LC-MS/MS at the Harvard NIEHS Center Proteomics Facility, according to standard procedures. Protein synthesis and polysome analysis --------------------------------------- The *eif3f*and *eif3m*conditional shut-off strains were inoculated in 10 ml Edinburgh Minimal Media (EMM) with or without thiamin and cultured at 30°C for 30 hrs. 100 uCi ^35^S-methionine were added to the cells, which were labeled for 30 min and then harvested by centrifugation. Cell lysates were prepared by bead disruption and equal amounts of the lysates were analyzed by SDS-PAGE. ^35^S-methionine incorporation was determined by autoradiography and quantified by PhosphorImager analysis (Bio-Rad, Hercules, California, USA). Polysome analysis was performed exactly as described in Ref. \[[@B60]\]. Briefly, cells were grown in EMM with or without thiamin for 30 hrs. Cycloheximide (100 ug/ml) was added to each culture for 20 min. Cells were harvested and lysed in breaking buffer (20 mM Tris/HCl pH 7.4, 50 mM NaCl, 30 mM MgCl~2~, 1 mM DTT, 50 ug/ml cycloheximide, 0.2 mg/heparin, 50 U/ml SuperRNasein, and protease inhibitors), followed by centrifugation at 13,000 × g for 5 min. 20 A~260\ nm~units of each supernatant in a volume of 500 ul were fractionated on 5-45 % sucrose gradients for 3 hrs at 30,000 rpm using a SW40-Ti rotor in a Beckman L70 centrifuge. Polysome profiles were obtained by monitoring the absorbance at 254 nm along the gradient using a Bio-Rad fractionator, and the output was recorded using a Bio-Rad UV detector. RT-PCR ------ RNAs associated with the eIF3e and eIF3m complexes were purified as described above. Primer design and RT-PCR conditions were according to the manual supplied with the Platinum Quantitative RT-PCR Thermoscript kit (Invitrogen). Primers used for RT-PCR are listed in Table [3](#T3){ref-type="table"}. The primer concentration for actin amplification was 0.25 uM. Other primers were used at 1 uM. Total RNA was obtained as described \[[@B19]\]. Immunological techniques ------------------------ Immunoprecipitation, immunoblotting, and in vitro ubiquitination assays were carried out exactly as described previously \[[@B26]\]. Antibodies against protein A and ubiquitin were purchased from Sigma and Genzyme (Cambridge, Massachusetts, USA), respectively. Anti-eIF3d/Moe1p polyclonal antibody was provided by E. Chang (Baylor College of Medicine, Houston, Texas, USA) Microarray analysis ------------------- The eIF3 complexes were affinity-purified as described above, except that 50 U/ml SuperRNasein (Ambion, Austin, Texas, USA) and 5 mM ribonucleoside vanadyl complex (New England Biolab, Beverly, Massachusetts, USA) were added to the lysis buffer. A mock purification was performed under identical conditions with lysate from cells lacking any tagged proteins. The methods for extraction of RNA, cDNA synthesis, RNA amplification, and aminoallyl labeling were adopted from previous publications \[[@B61],[@B62]\]. Briefly, after binding and washing, beads were treated with 3 U/ul DNase I, followed by elution with 5 % SDS, extraction with phenol/chloroform and then ethanol precipitation. The associated RNA was reversely transcribed using Superscript II reverse transcriptase (Gibco-BRL Division of Invitrogen) with 1 uM oligo-dT(15)-T7 primer (5\'-AAA CGA CGG CCA GTG AAT TGT AAT ACG ACT CAC TAT AGG CGC-3\') and 1 uM template switch primer (5\'-AAG CAG TGG TAT CAA CGC AGA GTA CGC GGG-3\'). Full-length double stranded (ds)-cDNA was synthesized using Advantage Polymerase (Clontech, BD Biosciences, Mountain View, California, USA). Newly synthesized ds-cDNA was passed through Bio-6 columns (Bio-Rad) and dried. RNA amplification was performed with the T7 megascript kit (Ambion). Amplified RNA was purified using the RNeasy kit (Qiagen, Hilden, Germany) and reversely transcribed with Superscript II reverse transcriptase in the presence of aminoallyl-dUTP in reactions primed with oligo-dT (Invitrogen). After RNAse H digestion and purification on QIAquick columns (Qiagen), the aminoallyl-cDNA was labeled with Cy3, and purified on QIAquick columns. Control Cy5-labeled aminoallyl-cDNA was made by oligo dT primed reverse transcription of total *S. pombe*RNA. Labeled cDNAs were hybridized onto glass slide microarrays. Microarrays were made by spotting purified ds-PCR products 500 to 1200 bp in length onto glass slides coated with aminopropylsilane (Erie Scientific, Portsmouth, New Hampshire, USA). The PCR products used represented 4988 predicted ORFs and transcripts as annotated by the Sanger Centre \[[@B63]\] and are located at the 3\' end of predicted genes to maximize sensitivity in detecting oligo-dT primed reverse transcription products. Descriptions of primer pairs and the features on the microarrays are available at the Longhorn Array Database \[[@B64]\]. Median pixel intensities for each spot obtained from hybridizations of eIF3-associated RNA (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 1\]) were corrected by local background subtraction and divided by the corresponding values obtained from a mock purification. This resulted in the factor of enrichment of mRNAs bound to eIF3 complexes over mock. The mRNAs were ranked according to their factor of enrichment (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 2\]). An enrichment factor of three was chosen as an arbitrary threshold. We found 2464 mRNAs were enriched greater than three-fold over mock in the eIF3m complex, while we observed 520 mRNAs enriched in the eIF3e complex (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 3\]). Comparison of the data sets revealed 414 mRNAs that occurred in both complexes, 2050 mRNAs uniquely enriched in the eIF3m complex and 106 mRNAs that were exclusively enriched in the eIF3e complex (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 4; [additional file 2](#S2){ref-type="supplementary-material"}\]). A corresponding list of the 106 mRNAs most highly and uniquely enriched in the eIF3m complex was assembled (\[see [additional file 2](#S2){ref-type="supplementary-material"}\]) and used to select specific mRNAs for RT-PCR analysis. The competitive Cy5-labeled total cDNA provided clear microarray feature identification and a guide to relative sequence abundance in the pre-purification sample from which the relative expression rank of each mRNA was determined (\[see [additional file 1](#S1){ref-type="supplementary-material"}, sheet 2\]). In this study, data were not adjusted by ratio normalization to total RNA signals or local area normalization functions. The microarray data were deposited with the public ArrayExpress (E-MEXP-208) \[[@B65]\] and GEO \[[@B66]\] databases. List of abbreviations ===================== CSN COP9 signalosome Cul Cullin eIF Eukaryotic initiation factor EMM Edinburgh Minimal Media GFP Green fluorescent protein IgG Immunoglobulin G LC MS/MS Liquid chromatography tandem mass spectrometry MALDI-TOF Matrix-assisted laser desorption ionization time-of-flight MPN Mpr1/Pad1 N-terminal OD Optical density PAGE Polyacrylamide gel electrophoresis PCI Proteasome, COP9 Signalosome, eIF3 ProA Protein A RT-PCR Reverse transcription-polymerase chain reaction TEV Tobacco etch virus YES Yeast extract and supplements (growth media) Authors\' contributions ======================= CZ performed the bulk of the experiments shown in this study. FA and SW prepared various strains used in this study. SK and ARI performed mass spectrometry. AO performed the microarray hybridizations, and JL participated in the design of the microarray studies and in data analysis. DAW contributed to study design and drafting of the manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 This article is accompanied by a Microsoft Excel file (.xls) containing microarray data (see Methods for description). ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 A Microsoft Excel file (.xls) is included containing the lists of the 106 mRNAs most highly and uniquely enriched in the eIF3e and eIF3m complexes. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank E. Chang for eIF3d/Moe1p antisera, V. Leautaud, G. Dittmar, F. Bachand, and P. Silver for help with polysome analyses, and M. Schmidt and A. Houseman for assistance with microarray data analysis. This study was funded by National Institute of Health (NIH) grant GM59780 to DAW, NIH grant P40 RR16320 to JL, by the NIEHS Center Grant ES-00002, and the NIEHS Training Grant ES07155. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Analysis of *csn6*and *csn7b*genes and proteins**. (**A**) The *csn6*and *csn7b*genes were disrupted in diploid *S. pombe*cells by inserting the *ura4*marker. Diploids were sporulated and spore viability was examined by tetrad analysis. Only two spores were viable, indicating that *csn6*and *csn7b*are essential. (**B**) Cell lysate from a strain carrying *csn7b*modified with five C-terminal pro-A tags and a TEV cleavage site at the endogenous genomic locus was absorbed to IgG resin, followed by elution of bound proteins with TEV protease (left panel) or SDS (right panel). Gels were stained with Coomassie Brilliant Blue, and proteins were identified by MALDI-TOF mass spectrometry. The asterisks denote degradation products of elF3c. ::: ![](1741-7007-3-14-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Roles of *eif3e*and *eif3m*in protein synthesis**. (**A**) Subcellular localization. Live cells expressing GFP-tagged alleles of *eif3b*, *eif3e*, and *eif3m*at the endogenous genomic loci were examined by fluorescence microscopy. N-terminally GFP-tagged *eif3f*was expressed at low levels from the pREP81 plasmid. (**B**) Shut-off strains. Diploid heterozygous *eif3f*and *eif3m*deletion strains were transformed with pREP81 plasmids driving the thiamin-repressible expression of Myc-tagged eIF3f and eIF3m, respectively. Diploids were sporulated and homozygous disruptants carrying the *eif3f*and *eif3m*plasmids were selected. Cells were grown in liquid medium to an OD~595~of 0.3 in the absence of thiamin, followed by promoter shut-off by the addition of thiamin. Samples were taken at the indicated time points after promoter shut-off and analyzed for the expression of plasmid borne eIF3f and eIF3m by immunoblotting with anti-Myc antibodies. (**C**) Effect on total protein synthesis. The *eif3f*and *eif3m*shut-off strains were maintained in the absence or presence of thiamin as indicated. Strains transformed with empty pREP81 plasmid were included as control. Cells were metabolically labeled with ^35^S-methionine, and aliquots of total cellular proteins were separated by SDS-PAGE and analyzed by autoradiography. The Coomassie Blue stained gel is shown to document equal protein loading. Data were quantified by PhosphorImager analysis and results are displayed in a bar graph. (**D**) Effect on polysomes. Polysome profiles were determined for the indicated strains as described in the Methods section. An *eif3e*deletion strain is shown for reference. (**E**) Effect on CSN-mediated regulation of cullin-RING ubiquitin ligases. Cul1p complexes were immunopurified from the indicated strains and employed in substrate-independent in vitro ubiquitylation reactions with purified E1, the E2 Cdc34p, ubiquitin, and ATP \[26\]. Polyubiquitin chains formed in the reaction are indicated (top panel). In contrast to *csn5*mutants, derepression of Cul1p activity is not observed in cells lacking *eif3f*, *eif3m*, or *eif3e*. The neddylation state of Cul1p was determined by immunoblotting (middle panel). Hyperneddylated Cul1p only accumulates in *csn5*mutants. Shut-off of *eif3f*and *eif3m*expression was verified by immunoblotting with anti-Myc antibodies (lower panel). ::: ![](1741-7007-3-14-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Distinct eIF3 complexes**. (**A**) eIF3 complexes associated with eIF3e and eIF3m were affinity-purified by absorption of the respective pro-A-tagged proteins to IgG beads. Bound proteins were eluted by cleavage with SDS (eIF3e) or TEV protease (eIF3m). (**B**) The eluates described in (A) were analyzed by immunoblotting with eIF3d antibodies. The asterisk refers to cross-reactivity of the secondary antibody with Ig light chains. (**C**) Lysates from cells expressing Myc-tagged Csn5p, eIF3e, and eIF3m from their endogenous genomic loci were immunoprecipitated with anti-eIF3d antisera. Coprecipitated proteins were identified by immunoblotting as indicated. Total cell lysates are shown to document expression levels of the endogenously tagged proteins. (**D**) Protein lysate from cells expressing protein A-tagged eIF3b from the endogenous promoter was absorbed to IgG beads, and specifically retained proteins were identified by LC-MS/MS. ::: ![](1741-7007-3-14-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Association of mRNAs with eIF3 complexes**. (**A**) Messenger RNAs associated with eIF3e and eIF3m complexes were identified by microarray hybridization as described in the Methods section. The graph indicates the numbers of mRNAs enriched more than three-fold over mock in both complexes. (**B**) Classification into functional groups of the 106 mRNAs most highly and uniquely enriched in complexes with eIF3e or eIF3m. (**C**) The relative expression ranks of the 106 most highly and uniquely enriched mRNAs in the eIF3e and eIF3m complexeswere determined using the relative hybridization signals obtained with total *S. pombe*cDNA (\[see [additional file 1](#S1){ref-type="supplementary-material"}\]). The expression ranks were blotted against the factor of enrichment of each mRNA in the eIF3 complexes. No correlation between expression rank and enrichment is apparent, indicating that highly expressed mRNAs were not unspecifically enriched in eIF3 complexes. The 5407 individual features on the microarray slides represent the 4988 ORFs and transcripts predicted in the Sanger Centre *S. pombe*genome database and various controls. To simplify the analysis, expression ranks were also assigned to these control spots, thus resulting in expression ranks for some ORFs higher than the theoretically possible number of 4988. ::: ![](1741-7007-3-14-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Specific association of mRNAs with the eIF3e and eIF3m complexes**. (**A**) The eIF3 complexes shown were affinity-purified as described in the Methods section, and separated by SDS-PAGE followed by staining with Coomassie Brilliant Blue. The labeled bands were identified by LC-MS/MS. eIF3f\* refers to a band in the eIF3m complex that was not identified by mass spectrometry in this experiment, and is therefore assigned based on its comigration with eIF3f identified in the eIF3e complex (right lane). eIF3f was positively identified by MALDI-TOF mass spectrometry as a subunit of the eIF3m complex in Fig. 1B. (**B**) The indicated eIF3e- and eIF3m-associated mRNAs were extracted from the purified complexes and employed in RT-PCR reactions using primers specific for each mRNA (see Table 3). PCR products obtained with total *S. pombe*RNA are shown for reference in the left panel. The factor of enrichment of each mRNA in the respective eIF3 complex and the enrichment rank (out of all 106 mRNAs enriched in either complex) are indicated below the gels. ::: ![](1741-7007-3-14-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of eIF3 subunits in various eukaryotes ::: ***Unified Nomenclature*** ***Domain*** ***Human*** ***S. cerevisiae (Core subunits)*** ***S. pombe Csn7Bp complex*** ***S. pombe Int6p complex*** ***S. pombe eIF3b complex*** ---------------------------- --------------- ------------- ------------------------------------- ------------------------------- ------------------------------ ------------------------------ eIF3a PCI p170 Tif32p p107 p107 p107 eIF3b RRM p116 Prt1p p84 p84 p84 eIF3c PCI p110 Nip1p p104 p104 p104 eIF3d \- p66 \- \- Moe1p Moe1p eIF3e PCI p48 \- \- Int6p Int6p eIF3f MPN p47 \- Csn6p Csn6p Csn6p eIF3g RRM/Zn finger p44 Tif35p Tif35p Tif35p Tif35p eIF3h MPN p40 \- p40 \- p40 eIF3i WD repeat p36 Tif34p Sum1p Sum1p Sum1p eIF3j \- p35 \- \- \- \- eIF3k PCI p28 \- \- \- \- eIF3l PCI p67 \- \- \- \- eIF3m PCI GA17 (?) \- Csn7Bp \- Csn7Bp ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Strains used in this study ::: **Name** **Genotype** **Source** ---------- ----------------------------- ----------------------------------------- ------------ DS448/1 *leu1-32 ura4-d18 ade6-704* *h+* Lab stock DS448/2 *leu1-32 ura4-d18 ade6-704* *h-* Lab stock C399/3 *leu1-32 ura4-d18 ade6-704* *csn5.13myc kan h-* Lab stock C485/3 *leu1-32 ura4-d18 ade6-704* *eif3m.13myc kan h-* Lab stock C617/1 *leu1-32 ura4-d18 ade6-704* *eif3m.CBP.tev.5proA kan h-* Lab stock C642/1 *leu1-32 ura4-d18* *diploid, eif3f/eif3f::ura4* This study C642/2 *leu1-32 ura4-d18* *diploid, eif3f/eif3f::ura4* This study C648 *leu1-32 ura4-d18 ade6-704* *eif3e.cbp.tev.5proA kan h+* This study C650/1 *leu1-32 ura4-d18 ade6-704* *eif3f::ura4 pRep81.6xhis.myc.eif3f h-* This study C650/2 *leu1-32 ura4-d18 ade6-704* *eif3m::ura4 pRep81.6xhis.myc.eif3m h-* This study C652/1 *leu1-32 ura4-d18 ade6-704* *eif3e::ura4 h+* This study C663/1 *leu1-32 ura4-d18 ade6-704* *eif3m.gfp kan h-* This study C663/2 *leu1-32 ura4-d18 ade6-704* *eif3e.gfp kan h-* This study C665 *leu1-32 ura4-d18 ade6-704* *eif3e.13myc kan h-* This study C701 *leu1-32 ura4-d18 ade6-704* *eif3b.gfp kan, h-* This study C702 *leu1-32 ura4-d18 ade6-704* *eif3b.cbp.tev.5proA kan h-* This study ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### RT-PCR primers ::: **Gene Name** **Sequence** **Product Size (bp)** --------------- ---------------------- ----------------------- *actin*F GAGCTTCCTGATGGTCAAGT 200 *actin*R GGATACATAGTGGTACCACC *cdc16*F AAGTTAGCGCCCAAATCACC 150 *cdc16*R TTTTGAATGCCCCCCCACGG *gar2*F GGTGCCATTGAGAAACCTTC 150 *gar2*R CCGAACCTTCAAAGAAATCA *pof6*F TCTGATCGGCCTAAGCTGTC 150 *pof6*R CAATCTTGCAAATTCAACTC *rpn5*F TGAGAAGCAAGTTCGTCAGG 160 *rpn5*R TGAAACAATCGATGACAAGT *rps8*-1F AAGCGTATTCACGAGGTCCG 150 *rps8*-1R CAACTCGTTGTTAGAAGGGT *scd1*F TCAGAGTTGGCTGCTTTCTT 150 *scd1*R ATCCATTGTGTGCCCTGTTC SPAC1348.05F CTTAGTGAACAGTTTGGAAG 150 SPAC1348.05R TGATAAACCAACGGATCCGA SPAC21E11.04F GAGACATCACCTGCTCCAGA 150 SPAC21E11.04R TTTGCTCCGGTGACTAGGTG SPBC12C2.06F TCTGTTCCCAAACCTCAAGC 150 SPBC12C2.06R ATCGATTTTTGCACCTTTAG SPBC1683.01F GGCCGTAAATTTGTCTACGG 150 SPBC1683.01R ACCACCAATACCAACACCAA SPBC36B7.03F CTCTCAATTAAATTTCACCC 150 SPBC36B7.03R AGGAGTACCGTATAAAGCAT SPCC70.05cF GTACCTGGAAATAACTCTCC 150 SPCC70.05cR CATAGCCTTTTCTAAGAGAT *tf2-12*F AAGCATGTACCAGAGATAGG 150 *tf2-12*R TGAATCACCTAGAAGAATTA *ung1*F ACTTTGGAGAGTTCTTGGTT 150 *ung1*R TGGAGTATGATGTGACCATG F = Forward; R = Reverse :::

PubMed Central

2024-06-05T03:55:58.339959

2005-5-17

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173091/", "journal": "BMC Biol. 2005 May 17; 3:14", "authors": [ { "first": "Chunshui", "last": "Zhou" }, { "first": "Fatih", "last": "Arslan" }, { "first": "Susan", "last": "Wee" }, { "first": "Srinivasan", "last": "Krishnan" }, { "first": "Alexander R", "last": "Ivanov" }, { "first": "Anna", "last": "Oliva" }, { "first": "Janet", "last": "Leatherwood" }, { "first": "Dieter A", "last": "Wolf" } ] }

PMC1173092

Background ========== Prostate cancer is the most common cancer in men resulting in over 30,000 deaths annually \[[@B1]\]. Early detection and treatment has the potential to markedly reduce the morbidity and mortality associated with the disease. While elevated Prostate Specific Antigen (PSA) \[[@B2]\] is the best available indicator of men with cancer \[[@B3]\], its diagnostic utility is limited due to elevated PSA levels in other non-malignant prostate conditions, varying levels in advanced disease and poor correlation between PSA levels and extent of disease. Furthermore, the variable course of prostate cancer -- many patients will not die of the disease -- means that radical therapy for all early cases would result in over treatment of significant number of patients. High throughput genomic technologies, by simultaneously interrogating the expression levels of thousands of genes, offers the potential to identify new biomarkers for early detection, prognosis, targets for therapy and for reclassification of prostate tumors. Using expression microarrays, a number of studies have characterized expression profiles for prostate cancer and other tumors. In some cases, correlations between tumor expression signatures, clinical parameters and outcome \[[@B4]-[@B12]\] have been identified. While potentially powerful, such studies can be significantly impacted by the choice of baseline or \"normal\" tissue used to detect tumor related expression changes. Most prostate expression studies to date have utilized normal appearing tissue adjacent to tumor as the tissue for comparison. However, a variety of methods such as chromosomal analysis \[[@B13]\], SAGE \[[@B14]\] and ploidy analysis \[[@B15],[@B16]\] have shown molecular abnormalities in normal appearing prostate adjacent to tumor. Even the term \"normal appearing\" prostate tissue adjacent to tumor may be misleading, as morphologic researchers using quantitative imaging analysis \[[@B17]-[@B19]\], have identified morphologic changes in the epithelial nuclei and blood vessels architecture in prostate tissue adjacent to tumor that are not routinely commented upon by pathologists. This suggests that, in some cases, tissues adjacent to cancer, although appearing morphologically normal by traditional microscopic examination, may contain genetic changes associated with the genesis of or reaction to cancer. Therefore, the use of adjacent normal as the baseline tissue for comparative gene expression studies may mask tumor related molecular changes preceding the appearance of histological tumor. More recently, a microarray study from our institution \[[@B12]\], using adjacent normal and tumor samples, describes a potential field effect around prostate cancers and regulation of selected genes in both adjacent normals and tumors. Using the same microarray data set for our analysis, we have compared the gene expression profiles of prostate cancer, normal appearing prostate tissue adjacent to tumor, and normal appearing prostate tissue from cancer free tissue donors with the aim of identifying the optimal baseline tissue for expression studies and the gene expression changes between the three specimen types. Methods ======= Clinical profile of cases ------------------------- The 60 tumor samples used in this study consisted of 2, 13, 27, 6 and 12 cases of primary prostatic adenocarcinoma of Gleason grade 5, 6, 7, 8 and 9 respectively. There were 4, 20, 23 and 13 cases spanning the age groups 40--49, 50--59, 60--69 and 70--79 respectively. Of the cases, 36 were stage T3 or higher with 2, 22, 23, 11 and 2 cases of stage T2a, T2b, T3a, T3b and 4 respectively. The 63 adjacent normal samples consisted of 2, 11, 29, 8, and 13 cases of Gleason grade 5, 6, 7, 8 and 9 respectively. There were 4, 21, 25 and 12 cases spanning the age groups 40--49, 50--59, 60--69 and 70--79 respectively. There were 2, 21, 26, 12 and 2 cases of Stage T2a, T2b, T3a, T3b and T4 respectively. Of the donors, 11 are under and 7 are over the age of 40. Samples and sample procurement ------------------------------ The tumor and adjacent normal tissue samples were acquired from the University of Pittsburgh Medical Center under stringent Institutional Review Board guidelines with appropriate informed consent. Specimens were received directly from the operating room. Samples (\>500 mg) were excised and snap frozen in liquid nitrogen within 30 min of excision and stored at -80°C in the University of Pittsburgh Pathology Tissue Bank until extraction of RNA. All samples were submitted for pathology evaluation. In every case, the tissue was excised from the junction between the ejaculatory duct and the prostatic urethra in the transition zone of the prostate. In particular, adjacent normal tissue was excised away from the cancer lesion macroscopically, and their histological diagnosis was confirmed microscopically. Donor tissue specimens were received through a collaborative arrangement with the Center for Organ Recovery and Education (CORE), the local organ procurement agency. The arrangement allows the University of Pittsburgh Pathology Tissue Bank to acquire normal prostates and associated serum/plasma specimens from healthy individuals who have donated their organs for transplant. There is extensive collaborative support from CORE. The donor prostatectomies harvested from brain dead, perfused donors and are bathed in Ringer\'s Lactate solution and transported on wet ice. These donor prostates are transported and handled with the harvested \"transplant\" organs. This significantly reduces transit time and minimizes the degradation of RNA. The processing methodologies used consist of snap freezing tissues in bulk, freezing in OCT and processing the tissues for routine histology (paraffin embedded tissues. For microarray analysis, the donor samples were excised from the same zone as the tumor and adjacent normal samples. cRNA preparation and Affymetrix chip hybridization -------------------------------------------------- cRNA was prepared and hybridized to Affymetrix oligonucleotide arrays as previously described \[[@B12]\] Statistical analysis -------------------- We analyzed prostate tissue samples from 18 donors and 63 prostate cancer patients. From the prostate cancer patients we took samples from the histologic tumor as well as normal appearing tissue adjacent to the tumor. High quality RNA and chip data were obtained from 60 cancer and 63 adjacent to tumor samples. In total 141 samples were run against the Affymetrix U95A chip and analyzed. The raw scanned array images were first processed through the Affymetrix Microarray Analysis Suite 5.0 (Affymetrix Corporation, Santa Clara, Ca) to generate probe cel intensity (\*.cel) files. The \*.cel files were then analyzed using both MAS 5.0 and dChip software from Harvard University \[[@B20]\], to generate gene expression signal values for each probe set. Data normalization to remove variation in overall chip intensities was perfumed by global scaling to a chip mean target intensity of 200 (MAS 5.0) or by the rank-invariant method (dCHIP). The MAS 5.0 with global scaling data and dChip with rank-invariant normalization data gave similar results in the subsequent analysis. Therefore, in the interests of clarity, we will focus on the MAS 5.0 results in the remainder this paper. In the next phase of analysis, the donors, adjacent normals and tumors were compared for differences in gene expression by using signal values for all 12625 probe sets for each sample. For statistical analysis, we used the Significance Analysis of Microarray (SAM) software package from Stanford University \[[@B21]\]. This method was chosen over conventional statistical tests because of its acceptance in the microarray community, its general simplicity and its ability to provide an estimate of the false discovery rate (the ratio of false positives to total positives). The false discovery rate is particularly important when comparing the expression of thousands of genes simultaneously. For example, when using the Student\'s t test at a P value of 0.05 to examine a population of 10,000 genes, one would expect 500 false positives. If there were in fact 100 true positives, the false positive rate would be an unacceptably high 0.83 (500/600). Briefly, SAM calculates a value for each probe set on the array. This value represents the observed difference in mean expression levels between the specimen classes being compared (i.e. tumor and donor) divided by the variance in the data and a fudge factor (see the original paper for details \[[@B21]\]. The resulting value is called the \"observed d value\". To determine the significance of this value, SAM estimates the \"expected\" d value if there were no difference between the specimen classes. This is done by permutating (randomly changing) the class labels without changing the data and recalculating the SAM value for each probe set. After thousands of permutations, the result estimates the value that would be obtained if the difference in gene expression were due to chance alone. This is the \"expected d value\". The significance of the observed differential gene expression can be estimated by comparing the observed and expected d values. A user defined threshold or \"delta\" (observed d value -- expected d value) can be adjusted to select only those genes observed d value exceeds (for up regulated genes) or is lower (for down regulated genes) than delta. The greater the \"delta\", the greater the stringency of the result and lower the false discovery rate. For each delta value, the SAM output consists of a gene (probe set) list and an associated false discovery rate. The false discovery rate is estimated from the distribution of expected and observed d values. Probe sets are ordered on the basis of observed d value metric, probe sets with high (or low) values represent genes with relatively high differential expression. The \"SAM Plots\" are also very useful in visualizing differences in overall differential gene expression between specimen classes. Results ======= Expression analysis of tumor, adjacent normal and donor tissue -------------------------------------------------------------- The prostate tumors analyzed in this study consisted of Gleason grades 5, 6, 7, 8, 9 and patients spanned the ages of 40 through 79. The goal of our research was to examine the differential gene expression patterns observed when comparing our three specimen classes: tumor versus adjacent normal, tumor versus donor and adjacent normal versus donor. The comparison was made at three points in the analytical process: 1) after normalization to remove variation in overall chip intensity 2) after statistical analysis of the data and, 3) after examining the differentially expressed gene lists. To examine differences in normalized gene expression between tumors, adjacent normals and donors the mean MAS 5.0 and dChip generated signal of each probe set for each specimen class (60 tumors, 63 adjacent normals and 19 donors), was calculated and plotted on a series of scatter plots (Fig [1](#F1){ref-type="fig"}). Figure [1](#F1){ref-type="fig"} shows the scatter plots and Pearson correlations of the normalized expression data analyzed using both MAS 5.0 and dChip as described above (vide supra, Methods). Data scatter is maximum in the tumor versus donor comparison, intermediate in adjacent normal versus donor and minimal in tumor versus adjacent normal. These findings are suggestive of more differential gene expression in tumor versus donor than tumor versus adjacent normal. In other words, donor normal tissue and adjacent normal tissue do not show the same degree of differential gene expression when paired with tumor tissue. Another striking result apparent in Figure [1](#F1){ref-type="fig"} is the close correlation and limited scatter of the tumor versus adjacent plot, even at low levels of signal. Tumor and adjacent normal specimens came from the same population of patients while donor specimens were received from a different set of individuals. To examine potential patient specific expression effects, the 60 tumors and 63 adjacent normal cases were randomly segmented into two groups, one group provided just tumor data and the other just adjacent normal data and a scatter plot of expression was generated. Since the segmentation of 63 cases can be performed in many different ways (permutations), the scatter analysis was performed 1000 times and the correlation between the sample groups determined by obtaining the mean correlation coefficient of the 1000 permutations. (Figure [2](#F2){ref-type="fig"}). In this analysis, the close correlation in expression between tumor and adjacent normal specimens persisted even when tumor and adjacent normal samples were taken from different patients. To determine the statistical significance of the observed differential expression between the three specimen groups, SAM analysis was performed. From each comparison (tumor v adjacent normal, tumor v donor and adjacent normal v donor), a SAM plot was generated and the plots for the three comparisons were overlaid (Fig [3](#F3){ref-type="fig"}). The diagonal line in Figure [3](#F3){ref-type="fig"} represents no differential expression (identical observed and expected d values, for further details see Materials and Methods) with points displaced from the diagonal representing differential expression. Figure [3](#F3){ref-type="fig"} shows that each of the comparisons yields a distinct expression profile with donor v tumor exhibiting more differential expression than adjacent normal v tumor or donor v adjacent normal. To further characterize the expression profiles from these comparisons, differentially expressed gene lists were created from each comparison by selecting genes whose d values (for details, see Materials and Methods) exceed a given threshold. False discovery rates (false positives/ total number of genes in gene list) is no greater than 2.5% at the deltas chosen for this analysis (Table [1](#T1){ref-type="table"}). At a delta of 2.0, when tumor expression is compared to donor expression (Table [1](#T1){ref-type="table"}), 474 differentially regulated genes can be detected. At the same delta, when tumor expression is compared to adjacent normals, only 92 genes are differentially regulated between these two tissues. Furthermore, at this delta, comparison of tumor expression with adjacent normals does not yield any genes up-regulated in tumors whereas the comparison with donors demonstrates up-regulation of 121 genes. Similarly at other deltas, approximately three times more differentially regulated genes can be detected when tumors are compared to donors than to adjacent normals. As was discussed above, tumors and adjacent normal tissues are obtained from the same patients and donor tissues from a different sample population. Therefore the larger gene expression differences between tumors and donors may represent underlying patient specific (genetic, demographic or handling) differences in patient (tumor and adjacent normal) and donor prostates rather than intrinsic differences between tumor, adjacent normal and donor normal tissues. It is significant however, that SAM analysis indicates that adjacent normals exhibit far less differential regulation than tumors when both are compared to donors. At all deltas (Table [1](#T1){ref-type="table"}), tumors v donors exhibit greater differential expression than adjacent normals v donor implying that tumors and adjacent normals are not identical in gene expression. Therefore, tumor specific, and not patient specific, expression changes can indeed by detected by comparing tumors to donor prostates. Significantly, these results establish the presence of unique gene expression profiles for prostate tissue from donors, adjacent normals and tumors (see Fig [3](#F3){ref-type="fig"}) with tumors differing more from donors than from adjacent normals. A potential limitation of our data is that donors span the ages of 5 to 60 and all tumor patients are older than 40. Therefore, the differential gene expression between donors and patients may be due to age specific differences in their prostates. To examine this, we segmented the donors into different age groups and compared only the 40 to 60 year old donors with tumors of the same age group. Although the number of cases in the study were small, the expression pattern observed in this age matched analysis is identical (data not shown) to that when all donors are included suggesting that potential age related differences in donor prostates do not contribute to the results of the donor v tumor analysis. GO annotation of differential gene expression --------------------------------------------- We examined the gene lists produced by SAM analysis of tumor, adjacent normal and donor tissue with two objectives: 1) to identify and functionally annotate some of the genes that contribute to the unique expression signatures of these tissues and 2) to determine whether adjacent normals or donors are the more appropriate baseline tissue for detecting differentially expressed genes in tumors. Functional annotation and comparison of the gene lists was performed using Gene Ontology terms \[[@B22]\], for biological processes and Affymetrix\'s Gene Ontology Mining Tool <http://www.affymetrix.com>. When donors are used as the baseline for comparison, tumors exhibit up-regulation of proliferation related genes including transcription factors, signal transducers and growth regulators (see [additional file 1](#S1){ref-type="supplementary-material"}). This list includes putative oncogenes, signal transducers and growth regulators. Some of the most up-regulated genes are *v-fos*, *jun B*, *jun D*, *c-src tyrosine kinase*, *FGF receptor activating protein, immediate early protein*and *early growth response 1*. The most down-regulated genes in tumors include those involved in immune response and signal transduction. Some of the genes in this list are the *interferon induced transmembrane proteins, Duffy blood group antigen*and *tumor necrosis factor*. In contrast, when adjacent normal tissue is used as the baseline for comparison, tumor tissue exhibits far fewer differentially expressed genes and the genes themselves are less compelling. The list up regulated genes is dominated by ribosomal proteins and metabolic enzymes, while the down-regulated list includes muscle related genes such as *tropomysin*, *actin*and *actinin*. When expression in adjacent normal is compared to donors, an up-regulation pattern remarkably similar to tumors is seen ([additional file 1](#S1){ref-type="supplementary-material"}). Adjacent normals also exhibit up-regulation of putative oncogenes, signal transducers and growth regulators with an almost 70% overlap of the 50 and 100 most up-regulated genes in tumors and adjacent normals, respectively. Similarly there is almost 60% overlap between the most down-regulated genes in tumors and adjacent normals that includes genes involved in immune response. The biological processes regulated in tumors and adjacent normals were also studied using Affymetrix\'s Gene Ontology Mining tool. The up regulated gene lists obtained at a SAM delta of 2.0 (Table [1](#T1){ref-type="table"}) were uploaded to the tool and the resulting annotations examined. Comparison of tumor gene expression to donor expression reveals up-regulation of genes involved in a number of biological processes (Figure [4a](#F4){ref-type="fig"}). Amongst these are genes involved in apoptosis, cell cycle, cell proliferation, immune response, protein phosphorylation, protein biosynthesis and transcription. A subset of these including genes involved in immune response and transcription are also up-regulated in adjacent normals (Figure [4b](#F4){ref-type="fig"}). In contrast when tumor expression is compared to adjacent normals, up-regulation of majority of these processes, except protein metabolism, is not detected (Figure [4c](#F4){ref-type="fig"}). Two important conclusions can be derived from the gene annotations, 1) though there are large number of genes regulated in tumors, there is a relatively small subset of genes including oncogenes and signal transducers that are highly regulated in both adjacent normal and tumor tissues and 2) regulation of a number of potentially important biological processes in tumors can be detected from using donors as the baseline tissues. The common regulation of oncogenes, signal transducers and immune response genes in adjacent normals is a striking result in that it suggests that adjacent normal tissue although appearing morphologically normal, undergo gene expression changes that may be important in tumorigenesis or as a reaction to tumor. Since these genes are regulated in both tumor and adjacent normal, they are not picked up on a direct comparison of the two tissues. While it is possible that donors are different from both adjacent normals and tumors due to processing artifacts -- the tumor and adjacent tissues were taken at surgery and donors at harvesting -- it is unlikely that the large differences seen in donor v tumor are all due to processing differences. This issue is examined further in the discussion section. The up regulation of proliferation markers in both adjacent normals and tumors coupled with the result that more differential regulation is detected when tumors are compared to donors than to adjacent normals suggests that donor prostates may be the more appropriate tissue for expression studies. Regulation of critical of biological processes and pathways may remain undetected if tissue adjacent to tumors is used for comparison. Discussion ========== There is a growing interest in the use of high throughput microarray analysis for the molecular reclassification of diseases. This interest appears to be well founded, as many groups have reported consistent patterns of gene expression associated with pathologic phenotypes, clinical behaviors and outcomes \[[@B4]-[@B11]\]. In the area of prostate cancer numerous groups \[[@B23]-[@B29]\] have all reported significant differential gene expression between histologic tumor specimens and normal appearing prostate tissue from patients with tumor present elsewhere in the prostate. Recently, a group from our institution reported a 70 gene signature that may predict aggressiveness of prostate cancer \[[@B12]\]. Comparison of the gene lists from published data sets with the results of our tumor versus adjacent normal analysis is complicated by the heterogeneity in samples, analysis platforms and analysis methods. Nevertheless, our study is qualitatively similar to other studies in the expression profile of tumors compared to adjacent normal tissue. A number of genes including *hepsin*, *myc*, *fatty acid synthase SPARC1*and *EBNA-2*coactivator show similar expression patterns across multiple prostate cancer studies \[[@B30]\], and are also regulated in our study. Our donors did not have prostate cancer or prostatic intraepithelial neoplasia (PIN) identified in their prostate and as such are good candidates for \"true normals\". Differential expression was much greater between tumor and donor tissue than between tumor and adjacent normal. The fact that tumor and adjacent specimens come from the same patients could possibly explain this difference but this was ruled out by our analysis. Another possibility is that tissue handing and processing differences could account for some or all of the differential expression seen when donor tissue is use as a baseline. In fact, data in the literature does suggest that tissue processing could effect the expression of genes such as *fos*, *jun*and *egr*in prostate tissue \[[@B29]\]. However, the same literature indicates that the effect warm ischemic time is limited to specific genes and in general, involves less than 1% of the regulated genes \[[@B29]\]. Our studies emphasize the need for documentation and quality of all experimental processing steps, from sample acquisition to sample hybridization, in order to completely characterize gene expression differences between prostate donors, tumors and normal tissue adjacent to tumors. In our experiment, tumor and adjacent normal specimens where taken from the same prostates and handled the same way. If differences in patient and donor tissue handling was the major issue driving differential expression in the tumor v donor and adjacent normal v donor comparisons, one would expect tumor v donor and adjacent normal to result in very similar expression profiles. However, we have shown that tumor v donor exhibits far greater differential expression than adjacent normal v donor (see Results). Furthermore, the differentially expressed genes seen in both tumor and adjacent normal include proto-oncogene and transcription factors that one might rationally expect to see in expectation of or in response to a local tumor. Therefore, while the possibility that some expression changes are due to differences in tissue handling cannot be formally ruled out, it is unlikely that the large and specific differences we observe in tumor v donor, tumor v adjacent normal and adjacent normal v donor are entirely due to processing differences. Clearly additional studies, including examination of patient process specimens that do not host prostate cancer (such as cystoprostatectomy for bladder cancer or prostates removed for benign hypertrophy) to examine this process further. The most important finding from our analysis is the potential importance of the donor specimens and the possibility that a field effect exists around prostate tumors, resulting in significant molecular changes in histologically normal appearing tissue adjacent to prostate cancer. Significantly, evidence for such malignancy associated changes have been presented in other organs such as the cervix, bladder and breast \[[@B31]-[@B33]\] Furthermore, a variety of methods such as chromosomal analysis \[[@B13]\], SAGE \[[@B14]\], ploidy analysis \[[@B15],[@B16]\] have shown molecular abnormalities in normal appearing prostate adjacent to tumor. Image analysis has also been employed to identify consistent changes in \"normal appearing\" prostate tissue adjacent to tumor \[[@B17],[@B18]\]. In one study cases of prostatic adenocarcinoma was consistently detected by examining histologically normal tissue using high-resolution image cytometry \[[@B18]\], and in another, combined highly sensitive and discriminating Fourier transform-infrared spectroscopy with statistical analysis was used to detect damaged DNA in normal appearing prostate tissue adjacent to cancer \[[@B34]\]. In expression analysis, while most published prostate studies have used adjacent normals as the baseline tissue, Dhansekaran \[[@B23]\], used both commercially available pooled donor normal tissue and adjacent normal tissue and noted differences in expression profile between the two specimen types. Genes that were differentially expressed in adjacent normals when compared to the pooled donor normals included signal transducers and transcription factors; and expression of these genes in adjacent normals was attributed to a field effect around tumors. Similarly, Yu \[[@B12]\] have noted dysregulation of selected genes in both adjacent normals and tumors when compared to donors. Prakash \[[@B27]\], found that gene expression in asymptomatic benign prostatic hyperplasia adjacent to tumors was different from asymptomatic BPH or symptomatic BPH not associated with tumors. The unique expression signature of BPH next to tumors included fos, jun, immediate early genes and this list was remarkably similar to the most up-regulated genes in the adjacent normals tissue in our study (see adjacent normal v donor, [additional file 1](#S1){ref-type="supplementary-material"}). Finally, within archives of the University of Pittsburgh Pathology Tissue Bank, there was a donor prostate, which was found to harbor prostate cancer. When run on the Affymetrix arrays, the tumor classified with the tumors samples rather than the donor samples. Although this is clearly no more than an anecdotal event, it is an interesting finding. Though microarray technology represents a major advance and provides a powerful tool for high-throughput expression analysis, the most effective use of this technology requires careful consideration of baseline normal tissue. Our results here emphasize the need for careful examination of what constitutes normal tissue and the importance of future studies to fully characterize normal appearing tissue adjacent to prostate cancers. Conclusions =========== Prostate tumor tissue, histologically normal tissue adjacent to tumors and donor normal prostate tissue exhibit unique gene expression profiles with tumor and adjacent normal profiles more similar to each other than to the donors. These results suggest that normal appearing tissue around prostate tumors may also be undergoing tumor related changes and that careful characterization of these different tissues is necessary to understand molecular changes in leading up to prostate cancer. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= URC was involved in the data analysis and interpretation and manuscript writing and revision. RD, as director of the Tissue Bank was involved in obtaining IRB approval and providing deidentified tissue for this research study. CM was involved in data analysis and interpretation. GM and MB as principal investigator and co-investigators of the grant titled \"the Molecular Reclassification of Cancer\" were responsible for design of the study, providing intellectual input and final approval of the version submitted for review. JG was responsible for providing intellectual direction, guidance for the analysis team, review and final approval of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/45/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 **The fifty most up regulated and down regulated genes from the tumor v donor, tumor v adjacent normal and adjacent normal v donor comparison**. The Affymetrix probe set id, gene names and assignment of biological process for each gene are shown. The biological process annotation includes information from Affymetrix, gene ontology and literature references. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was funded by 1 U01 CA86735-01 (to MB)\"A Shared Resource for the Molecular Classification of Prostate Cancer Comprehensive Prostate Cancer Tissue Resource (CPCTR) Consortium, National Cancer Institute, 5 U01 CA88110-02 (to GM, MB is Co-Principal Investigator) \"Molecular Reclassification of Prostate Cancer\" Director\'s Challenge for the Molecular Classification of Cancer Consortium National Cancer Institute and 2000-98 (to JG) DOD National Medical Technology Test Bed An Advanced Tissue Banking Information System Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Differential gene expression analysis of donor, tumor and adjacent normal protate cancer samples.**Scatter plot of MAS 5.0 derived tumor vs donor, adjacent normal v donor and tumor v adjacent normal samples. For each probe set, the mean MAS 5.0 expression values of all the samples in each specimen group was calculated. Scatter plot were constructed using the mean values for each specimen group. ::: ![](1471-2407-5-45-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Regression analysis of permuted donors, adjacent normal and tumor samples.**The 60 tumors and 63 adjacent normal tissues were segmented so that tumors and adjacent normal samples in each comparison were selected from different patients. The resulting tumor and adjacent normal samples were then subjected to regression analysis. Donor v tumor, donor v adjacent normal and adjacent normal comparisons were performed. Since the segmentation can be performed in many different ways, the analysis was performed 1000 times. The mean correlation coeifficient and standard deviation from each of these comparisons were plotted as box plots. ::: ![](1471-2407-5-45-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Overlayed SAM plots (for details, see Materials and Methods) from the donor v tumor, donor v adjacent normal and tumor v adjacent normal analyses.**Each of the SAM plots was overlayed to direct comparison of the plots. The diagonal line represents no differential gene expression where the observed d value equals the expected d value after 1000 permutations of the class labels. Genes that are differentially expressed are displaced from the diagonal (greater than 0 for up regulation and less than 0 for down regulation). Genes that are more differentially expressed are more displaced from the diagonal than those that are closer to the diagonal. For each of the comparisons, a plot is generated from the d values of the 12625 probe sets in the two specimen groups. Red = donor v tumor plot; green = adjacent normal v tumor plot; black = adjacent normal v donor plot. ::: ![](1471-2407-5-45-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Gene Ontology annotation of differentially expressed gene lists.**The fifty most upregulated genes from the donor v tumor, adjacent normal v tumor and tumor v adjacent normal comparisons were uploaded to Affymetrix\'s Gene Ontology Mining Tool, *a*, donor v tumor; *b*, donor v adjacent normal; *c*adjacent normal v tumor; The annotations is presented as a hierarchy of terms, from general to most specific terms (from left to right). The numbers in parenthesis indicate the number of genes that are annotated with the term. In all of the analysis, annotation of all the submitted probe sets is not achieved. Typically, annotation exists for approximately 60% of the probe sets. ::: ![](1471-2407-5-45-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Differential gene expression in the tumor v donor, tumor v adjacent normal and adjacent normal v donor comparisons. The number of genes identified as differentially regulated at each delta (observed d value -- expected d value; for details, see Materials and Methods) are shown. Also, shown are the number of up and down regulated genes at each delta. For each of these deltas, the false discovery rate was no greater than 2.5%. ::: **Delta 3.0** **Delta 2.0** **Delta 1.56** ----------------------------- --------------- --------------- ---------------- **Tumor v Donor** 65 474 928 Up Regulated 10 121 305 Down Regulated 55 353 623 **Tumor v Adjacent Normal** 0 92 382 Up Regulated 0 58 Down Regulated 92 324 **Adjacent Normal v Donor** 12 86 254 Up Regulated 5 25 63 Down Regulated 7 61 191 :::

PubMed Central

2024-06-05T03:55:58.344404

2005-5-13

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173092/", "journal": "BMC Cancer. 2005 May 13; 5:45", "authors": [ { "first": "Uma R", "last": "Chandran" }, { "first": "Rajiv", "last": "Dhir" }, { "first": "Changqing", "last": "Ma" }, { "first": "George", "last": "Michalopoulos" }, { "first": "Michael", "last": "Becich" }, { "first": "John", "last": "Gilbertson" } ] }

PMC1173093

Background ========== Breast cancer is the most frequently diagnosed cancer among women in the United States \[[@B1]\]. Of 668,470 expected female cancer cases in 2004, 215,991 (32.3%) were breast cancer \[[@B1]\]. The incidence of female breast cancer is most common in developed countries such as the U.S. and Western Europe \[[@B2]\], and is higher among whites than other racial and ethnic groups \[[@B3]\]. Among the registries in the Surveillance, Epidemiology, and End Results (SEER) Program of the U.S. National Cancer Institute, female malignant breast cancer incidence rates are lowest in Utah \[[@B3]\], despite approximately 85% of the population being white, non-Hispanic \[[@B4]\]. About 70% of the 2.3-million population in Utah is affiliated with the Latter-day Saint (LDS or Mormon) religion \[[@B4],[@B5]\]. Hence, female breast cancer incidence rates in Utah are largely representative of LDS women and their lifestyle behaviors. A study based on Utah data from 1971 through 1985 showed that LDS women had significantly lower age-adjusted breast cancer incidence rates than did non-LDS women \[[@B6]\]. A recent cross-sectional study involving non-Hispanic white females in Utah compared breast cancer risk factor behaviors between LDS and non-LDS \[[@B7]\]. This study focused on many factors previously shown to influence breast cancer, including age at first birth \[[@B8],[@B9]\], number of pregnancies and number of births (parity) \[[@B10]\], lifetime duration of breastfeeding \[[@B10],[@B11]\], and alcohol consumption \[[@B12]\]. LDS women compared with non-LDS women had a slightly older age at first birth, higher average number of pregnancies and children (parity), and more total years of breastfeeding. LDS women also displayed lower levels of alcohol consumption. A cross-sectional study conducted in Utah in the late 1970s likewise found LDS women to have a later age at first pregnancy, more pregnancies, and lower prevalence of alcohol consumption \[[@B13]\]. However, research has also shown that reproductive factors which may decrease the risk of breast cancer development may explain poorer breast cancer survival \[[@B14]\]. The current study updates the findings of Lyon et al. \[[@B6]\] to the years 1985--99 and also examines female breast cancer survival in Utah according to LDS status and level of religiosity for LDS. Methods ======= Utah Cancer Registry (UCR) data were linked to LDS Church membership records to estimate breast cancer survival according to LDS (religiously active and less active) and non-LDS populations in Utah during 1985--99. Utah Cancer Registry -------------------- The UCR, established in 1966, has continuously participated in the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute since 1973. UCR staff members and local cancer registrars identify incident cases of cancer among Utah residents through routine and systematic review of pathology reports, medical records, radiation therapy records, hospital discharge lists, and vital records. Tumor characteristics including histology, grade, and primary site are coded according to the International Classification of Disease for Oncology-Second Edition (ICDO-2) \[[@B15]\]. Breast cancer-specific death is coded according to the International Classification of Diseases, 9^th^Revision, as 174.0 -- 174.9 \[[@B16]\], and 10^th^Revision, which uses an alphanumeric system and codes them as C50 \[[@B17]\]. Categories of stage of disease at diagnosis are documented in the Summary Staging Guide of the SEER Program of the National Cancer Institute \[[@B18]\]. Registry records also include information regarding treatment, survival, and patient characteristics such as age at diagnosis, gender, race/ethnicity, and place of residence at diagnosis. Such information is ascertained from specific statements in medical records, reports from private pathology laboratories and radiotherapy units, and death certificates. Cancer surveillance in Utah is conducted in accordance with standards instituted by the SEER Program and the North American Association of Central Cancer Registries \[[@B19],[@B20]\]. In order to provide valid estimates of cancer survival, a high percentage of cancer cases must be routinely followed to ascertain both vital status and date of last contact. UCR records are linked four times each year with records of death certificates from the Office of Vital Records and Health Statistics from the Utah Department of Health. Results from these routine linkages identify cancer patients who have died, regardless of their cause of death. Registry staff members work closely with cancer registrars in local hospitals to document, through systematic review of medical records, the date of last contact for those cancer patients not known to be deceased. Records for these patients are also linked annually with administrative databases, including Medicaid reimbursement claims, Utah driver license records, and Utah voter registration lists; these databases are often a source of updated information regarding vital status and/or date of last contact. According to SEER Program standards, follow-up is considered to be current when the Registry documents that a patient has died, or when the Registry documents that a patient was alive within 18 months of the annual anniversary of the date of original cancer diagnosis. By this definition, 97.9% of the patients in the present study were determined to have complete follow-up at the time data were captured for analysis. Linkage of cancer data with LDS church records ---------------------------------------------- UCR records were linked to LDS church membership records to determine membership in the LDS church. A person was identified as a member of the LDS church if the membership record included a baptismal date. The linkage process took place under direct supervision of the Church and the UCR and was conducted in the Church\'s Member and Statistical Records Department. After the records were linked, all personal identifying information was stripped from the database to ensure confidentiality. Records were linked using the probabilistic linking program LinkPro \[[@B21]\]. The program calculates probabilities to identify whether a pair of records refers to the same person. Ten variables that were common to both sets of records were used to link the records: first, middle, and last names; birth day; birth month; birth year; gender; zip code; vital status; and maiden name. SOUNDEX versions of the names were used in the matching process, while actual names were used in the review process. Records were linked if they matched on at least seven of the ten variables. Ambiguous links and records that matched on six of the ten variables were manually reviewed. There were approximately 120,000 incident cases of cancer identified in the UCR database from 1973 to 1999 and approximately 6.6 million records in the LDS Church database. There were 81,617 (68%) UCR records linked to a Church membership record. Of these linked records, 74,829 (92%) matched on at least seven of the ten variables. In addition to information about baptism, LDS church membership records include information on whether born in the church and whether endowed in the temple. To receive one\'s temple endowment requires conscientious effort toward obedience to the lifestyle prescribed by LDS doctrine \[[@B22]\]. This includes adherence to certain health practices such as not smoking or consuming alcohol, abstention from sexual relationships outside of marriage, and being honest in ones dealings with others \[[@B23]\]. A man or woman aged 19 year or older who is considered worthy in a personal interview with a church authority may receive a temple endowment. Generally LDS men and women receive their endowment prior to serving an LDS mission or at the time of a temple marriage. Previous research has considered the temple endowment as an appropriate surrogate marker for religious commitment among adults in the LDS church \[[@B24]\]. Although the data allowed us to identify LDS church membership and religiosity among LDS women, information about religious preference and religiosity was not available among non-LDS women. Data ---- The present study was based on 9,619 female microscopically confirmed, actively followed malignant breast cancer cases (ICDO-2 site code C50.0:C50.9) diagnosed in Utah from 1985 through 1999. There were 5,234 (54.4%) religiously active LDS, 1,369 (14.2%) less religiously active LDS, and 3,015 (31.4%) non-LDS. In addition to LDS membership and level of religiosity, the following variables were considered: age, marital status, summary stage (pathological), histological grade, number of primary cancers in one individual, and treatment. Because of the high percentage of whites in Utah \[[@B4]\], analyses are limited to whites only. Age was categorized as 0--34, 35--44, 45--54, 55--64, 65--74, and 75 years and older; marital status was categorized into four groups: never married, married/cohabitating, previously married, and unknown; and calendar years were categorized into four groups: 1985--88, 1989--92, 1993--96, and 1997--99. Number of primary cancers in an individual describes the chronology of diagnosis of all primary malignant cancers over the entire lifetime of the person \[[@B19]\]. Localized tumors are confined to breast, regional tumors are spread to contiguous organs or lymph nodes, and distant tumors are spread to remote organs. Histological grade is defined by the SEER Program Coding Manual and by the International Classification of Diseases for Oncology, Second Edition in increasing level of severity as either low grade (well differentiated), medium grade (moderately differentiated), or high grade (poorly differentiated) \[[@B15],[@B19]\]. These categories are approximately equivalent to the more widely used Gleason grading system as Gleason score 2--4 (low), 5--7 (medium), and 8--10 (high) \[[@B19]\]. Radiation therapy was identified if radiation was a first course of cancer-directed therapy. Surgery was classified as conservative surgery \[none; no cancer-directed surgery of primary site; partial mastectomy, not otherwise specified (NOS); less than total mastectomy, NOS\], mastectomy \[total (simple) mastectomy, NOS; modified mastectomy; radical mastectomy, NOS; extended radical mastectomy\], or other surgery (mastectomy, NOS; surgery, NOS) \[[@B19]\]. Cases who had died but had unknown date of death, whose cause of death was determined by death certificate or autopsy only, or who represented a second or later primary cancer were excluded from the survival analysis. After these exclusions, there were 8,731 cases available for this portion of the study. SEER registries were included in the study if they provided data to the National Cancer Institute during all the years from 1985 through 1999. These registries are as follows: San Francisco-Oakland, Connecticut, Metropolitan Detroit, Hawaii, Iowa, New Mexico, Seattle (Puget Sound), Utah, and Metropolitan Atlanta. Statistical analyses -------------------- Age-adjusted rates were derived using the direct method and standardized to the 2000 U.S. population. Counts used in the rate calculations were microscopically confirmed. LDS and non-LDS breast cancer incidence rates were compared with SEER incidence rates. Ninety-five percent confidence intervals for the incidence rate calculations were based on standard error estimates from the formula SE(rate) ≈ rate/(cancer cases)^1/2^\[[@B25]\]. Population attributable-risk percent is used to estimate the impact of having a certain breast cancer risk profile in Utah. Among LDS, population values were not available by level of religiosity. Hence, rates are presented for LDS and non-LDS. However, level of religiosity among LDS could be estimated among the cases such that the survival analysis considered religiously active LDS, religiously less active LDS, and non-LDS. Survival time was calculated as the time interval between diagnosis and date of last information. For deceased cases, the date of last information was the date of death. For cases not known to be deceased, the date of last information was the date that the case was last known to be alive. Cases with death from breast cancer (the outcome of interest), and all other cases were censored at the time of last information. We estimated breast cancer-specific survival for cases where breast cancer is a single cancer primary or the first of multiple cancer primaries. Bivariate comparisons were evaluated using the chi-square test. Survival estimates were calculated by the Cox proportional hazards method with one-month intervals \[[@B26],[@B27]\]. Statistical significance based on Cox proportional hazards was determined using 95% confidence intervals. The appropriateness of the Cox proportional hazards model was assessed graphically. Statistical analyses were derived in Statistical Analysis System (SAS) software, version 9.0 (SAS Institute Inc., Cary, NC, USA, 2003). Results ======= Bivariate analyses of the association between religion and selected variables for female malignant breast cancer patients in Utah, 1985--99 are displayed in Table [1](#T1){ref-type="table"}. Significant differences in the distribution of patients among these groups were observed across age categories, marital status, cancer primaries, and histological grade. Religiously active LDS patients tended to be older and were more likely married or previously married. Religiously less active LDS were more likely to have multiple cancer primaries. The high percentage of unknown histological grade at diagnosis makes this variable impossible to evaluate. The median difference in rates between LDS and non-LDS women across the study years was 20.2 per 100,000. The smallest difference in rates was 8.5 in 1992 and the largest difference in rates was 43 in 1991. Age-adjusted female malignant breast cancer incidence rates among white women in Utah are presented by LDS status and calendar year in Figure [1](#F1){ref-type="fig"}. The comparatively low rates in Utah compared with SEER (without Utah) exist across calendar years. The lower rates are primarily explained by LDS women, although non-LDS women also contributed to the lower rates. For the combined years 1995 through 1999, lower female malignant breast cancer incidence rates among LDS compared with non-LDS women in Utah were observed across the age span (Figure [2](#F2){ref-type="fig"}). Age-adjusted breast cancer incidence rates were 107.6 (95% CI: 103.9 -- 111.3) for LDS women and 130.5 (123.2 -- 137.9) for non-LDS women. In SEER (without Utah) the age-adjusted breast cancer incidence rate was 141.9 (140.9 -- 142.9). Hypothetically speaking, if non-LDS women in Utah had the same breast cancer risk profile as LDS women, 214 (4.8%) fewer malignant breast cancer cases would have occurred during 1995--99. If the 9 SEER regions had the same breast cancer risk profile as women in Utah, 12,261 (16.2%) fewer malignant breast cancer cases would have occurred. If the 9 SEER regions had the same breast cancer risk profile as LDS women in Utah, 16,947 (22.4%) fewer malignant breast cancer cases would have occurred. A proportional hazards model was used to estimate the death hazard for female malignant breast cancer among patients, with religiously active LDS as the referent group. The death hazard ratio for religiously less active LDS was 1.09 (0.94 -- 1.27) and for non-LDS was 0.86 (0.75 -- 0.98). Adjusted hazard ratios for age and marital status were also computed for patients based on their already having survived 0 months, 6 months, 12 months, 18 months, and 24 months (Figure [3](#F3){ref-type="fig"}). Religiously less active LDS compared with religiously active LDS showed no significant difference in the death hazards due to breast cancer across each of the conditioned time periods. On the other hand, non-LDS compared with religiously active LDS had significantly lower death hazard rates for breast cancer for each of the conditioned time periods. Discussion ========== Utah presents a unique area for breast cancer studies because of the relatively low incidence of breast cancer among its residents, and the lower risk of breast cancer among members of the LDS church compared with non-LDS in the state. Lifestyle behaviors common among members of the LDS church, such as higher birthrates, higher prevalence and duration of breastfeeding, and lower alcohol consumption have been associated with lower incidence rates of breast cancer. The significantly lower risk of breast cancer among LDS shows the potential preventive effect certain behaviors may have against breast cancer. Yet this study also showed poorer survivability among a population with a traditionally lower incidence rate of breast cancer. It appears that the same factors that are associated with a lower incidence rate of breast cancer are also associated with poorer prognosis once breast cancer is diagnosed. This paradoxical finding is consistent with the results of a study by Korzeniowski and Dyba who reported that \"reproductive factors known to decrease the risk of breast cancer development have an adverse effect on prognosis\" \[[@B14]\]. Results in the current paper confirm those found by Lyon et al. \[[@B6]\]. Specifically, age-adjusted rates are consistently lower for LDS than non-LDS women. In addition, both LDS and non-LDS women displayed lower breast cancer incidence rates than in SEER. Lower rates among LDS women were observed across the age span. The lower rates among LDS women were likely primarily explained by their having significantly higher average number of pregnancies and children (parity), total years breastfeeding, and lower alcohol consumption \[[@B7],[@B13]\]. Not only did LDS women in Utah experienced lower breast cancer incidence rates than did women in SEER, non-LDS women in Utah also had lower rates. This may be attributed to a conforming effect wherein the non-LDS minority approximates the lifestyle behaviors of the LDS majority to increase their social acceptability \[[@B28]\]. Social and cultural factors influence behaviors such as breastfeeding \[[@B29]\]. Religious preference and the dominant religious preference of one\'s residence are two independent factors associated with differences in desired number of children \[[@B30]\]. A religious group may have certain reproductive standards, but when they come in contact with a religious group with different standards, their reproductive health attitudes may change \[[@B31]\]. When a majority group exerts moderate social pressure, especially if greater social acceptance is achievable, the minority group is more likely to conform to the majority\'s values \[[@B32]\]. Each of these factors fits the relationship between LDS and non-LDS in Utah and may explain non-LDS conformity to selected LDS practices. It should be noted that the linked data did not provide information about religious preference or religiosity among the nearly 30% non-LDS women. However, if the women in the study represent a similar representation as those identified in a recent cross-sectional survey of Utah adult women, approximately 26% of the non-LDS women were religiously active, 41% were less religiously active, and 34% had no religious preference \[[@B7]\]. Daniels et al. showed that the significantly lower average number of pregnancies and children (parity), total years breastfeeding, and higher alcohol consumption in non-LDS women were primarily in less religiously active women and those with no religious preference \[[@B7]\]. The population attributable-risk percent was presented to indicate the amount of breast cancer that could have been avoided under the hypothetical situation that everyone in Utah experienced the same breast cancer risk profile as LDS women. If everyone in SEER had the same breast cancer risk profile as LDS women in Utah, this percentage was 22.4%. In other words, approximately 22.4% of all breast cancer in SEER could have been avoided had women in SEER experienced a similar breast cancer risk profile as LDS women in Utah. Applying this percentage to the U.S. female population in 2004, an estimated 48,382 new cases could have been avoided had women in the U.S. displayed similar breast cancer risk behaviors as LDS women in Utah. There was not a significant difference in death hazard rates between religiously active and less active LDS. However, there was a significant difference in death hazard rates between religiously active LDS and non-LDS. Lower death hazard rates in non-LDS became more pronounced as we conditioned on months already survived from diagnosis. Two possible explanations may help explain this, proximity to last childbirth and diagnosis during the lactation period. New studies have indicated that a more recent birth prior to the diagnosis of breast cancer is associated with poorer survivability \[[@B30]-[@B33]\]. Older studies had failed to show an association, arguing that poorer survivability was coincidental and likely the result of significant delays in diagnosis \[[@B34]-[@B37]\]. Religiously active LDS women generally have more children than their non-LDS counterparts. While a greater number of childbirths is not associated with a worse prognosis in breast cancer \[[@B32]\], this higher parity does raise the likelihood that religiously active LDS women will have a more recent birth than non-LDS at the time of breast cancer diagnosis. A less conclusive, yet possible contributor to poorer survivability among active LDS, is diagnosis during the lactation period. One study noted that breast cancer diagnosed during the lactation period has been associated with poorer survival among women younger than 45 years \[[@B35]\]. This relationship persisted despite adjustment for nodal status, tumor size, and age. A more recent literature review, however, concluded that no epidemiologic evidence exists to indicate that breastfeeding increases the risk of breast cancer recurring or of a second breast cancer developing \[[@B41]\]. It is biologically plausible that breastfeeding may contribute to a reduction in the development and growth risk of breast cancer. Breastfeeding reduces the number of ovulations proportionally to its intensity, and maintaining a lower estrogen level than that observed during the menstrual cycle \[[@B42]\]; it mobilizes endogenous and exogenous carcinogens present in the ductal and lobular epithelial cell environment \[[@B43]\]; and it reduces pH, levels of estrogens, and local carcinogens of the lobules and ducts \[[@B44],[@B45]\]. Due to the limited and conflicting evidence supporting an association between diagnosis during lactation and poorer prognosis of breast cancer, and in light of biological mechanisms during lactation which may actually limit cancer growth, the possibility should not be ruled out that poorer prognosis associated with diagnosis during lactation may actually be attributable to the childbirth which most lactating women would have recently experienced. Further research is needed to clarify this association. Limitations =========== The temple endowment was used as a surrogate marker for religiosity among LDS women. Although receiving one\'s temple endowment requires conscientious effort toward obedience to the lifestyle prescribed by LDS doctrine, which includes certain health practices like not smoking or consuming alcohol and abstention from sexual relationships outside of marriage, the data did not allow us to identify whether a previously endowed individual continued to adhere to the doctrine of the church. Yet we assume that given the high level of religious commitment required to receive one\'s temple endowment, the tendency was for these individuals to continue adherence to the health practices of the LDS religion. A related religiosity measure to the temple endowment was not available for non-LDS women. We were concerned with the degree of accurate assignment of the cause of death, but the influence of inaccurate specification of cancer death appeared to be very small. Bias may result when considering multiple cancer primaries because religiously active LDS are less likely to smoke cigarettes. Hence, they are less likely to get a cancer related to smoking, where these cancers tend to be more lethal (e.g., lung and pancreatic cancers). There would have been a selection bias if we had considered death from any cancer. However, limiting disease-specific death to breast cancer made the groups more comparable with respect to smoking history. It was not possible to explore the association between histological grade and survivability because of the large number of patients in the \"unknown\" category. Nevertheless, the difference in distribution of histological grade among patients with known grade information was very similar and not statistically different (χ^2^(4) = 1.71, p = .789). There is no reason to believe that the \"unknown\" grade would have been distributed differently among the religious groups. Finally, specific data on breast cancer risk behaviors was not available on the patient level. However, information was available on whether or not a person had previously received their temple endowment. This measure was used as a surrogate for a host of lifestyle behaviors previously observed among religiously active LDS in Utah, such as higher parity and total time breast feeding and lower prevalence of alcohol consumption. Conclusion ========== LDS lifestyle is associated with lower incidence of breast cancer. The age distribution among LDS breast cancer patients in this study strengthens the argument that higher parity, lifetime breast feeding, and lower alcohol consumption have a preventive effect against breast cancer. Non-LDS in Utah approximate LDS behavior to increase social acceptability, serendipitously contributing to a relatively low breast cancer incidence. LDS showed poorer survival from breast cancer than non-LDS in Utah. Parity and breastfeeding, associated with lower risk of breast cancer, may contribute to poorer prognosis once breast cancer is diagnosed. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Ray M. Merrill conceptualized the study, analyzed the data, and drafted the manuscript. Jeffrey A. Folsom assisted in interpreting the data and writing the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/49/prepub> Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Malignant breast cancer incidence in white women in Utah and SEER (without Utah) ::: ![](1471-2407-5-49-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Age-specific malignant breast cancer incidence among white women in Utah according to LDS status, 1995--99 ::: ![](1471-2407-5-49-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Hazard ratios (95% confidence intervals) of death from breast cancer for religiously less active LDS and non-LDS compared with religiously active LDS conditioned on having already survived *x*months ::: ![](1471-2407-5-49-3) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Bivariate analyses of selected factors among breast cancer patients by LDS status and level of religiosity in Utah, 1985--99 ::: Religiously Active LDS Religiously Less Active LDS Non-LDS χ^2^P value ------------------------------ ------------------------ ----------------------------- --------- ------------- ------ ------ --------- No. \% No. \% No. \% Age at Diagnosis 00--34 119 2.3 55 4.1 91 3.1 35--44 463 9.1 165 12.3 465 15.6 45--54 962 18.8 250 18.6 639 21.5 55--64 1167 22.9 312 23.2 674 22.6 65--74 1299 25.5 351 26.1 668 22.4 149.7 75+ 1093 21.4 211 15.7 441 14.8 \<.0001 Marital Status at Diagnosis Never Married 123 2.3 110 8.0 217 7.2 Married (cohabitating) 3044 58.2 626 45.7 1944 64.4 Previously Married 2067 39.5 633 46.2 810 26.9 414.9 Unknown 0 0.0 0 0.0 45 1.5 \<.0001 Breast Cancer Primaries Single 4742 90.6 1200 87.7 2735 90.7 11.8 First of Multiple 492 9.4 169 12.3 281 9.3 .0028 Summary Stage (Pathological) Localized 3208 61.3 839 61.3 1859 61.6 Regional 1756 33.5 443 32.4 982 32.6 Distant 166 3.2 54 3.9 106 3.5 4.3 Unknown 104 2.0 33 2.4 69 2.3 .6340 Histological Grade Low (well-differentiated) 576 11.0 152 11.1 336 11.1 Medium (moderately diff.) 1796 34.3 463 33.8 1076 35.7 High (poorly diff./undiff.) 1272 24.3 357 26.1 796 26.4 13.7 Unknown 1590 30.4 397 29.0 808 26.8 .0337 Radiation Therapy No 3504 67.0 896 65.5 1961 65.0 3.5 Yes 1730 33.0 473 34.5 1055 35.0 .1736 Surgery Conservative Management 1835 35.1 508 37.1 1144 37.9 Mastectomy 3368 64.3 856 62.5 1858 61.6 8.4 Other Surgery 31 0.6 5 0.4 14 0.5 .0775 Year of Diagnosis 1985--88 1139 21.8 306 22.4 693 23.0 1989--92 1282 24.5 374 27.3 732 24.3 1993--96 1525 29.1 371 27.1 855 28.3 7.8 1997--99 1288 24.6 318 23.2 736 24.4 .2498 :::

PubMed Central

2024-06-05T03:55:58.347376

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173093/", "journal": "BMC Cancer. 2005 May 18; 5:49", "authors": [ { "first": "Ray M", "last": "Merrill" }, { "first": "Jeffrey A", "last": "Folsom" } ] }

PMC1173094

Background ========== Since 1994, colorectal adenocarcinoma has been the third most frequent cancer in the Grand-Duchy of Luxembourg (Western Europe), preceded by breast and prostatic cancer \[[@B1]-[@B3]\]. In the United States of America (USA), Japan and France, there have been changes in the site distribution of large bowel cancers over the last decades \[[@B4]-[@B9]\]. Because of the lack of a systematic, organized screening for colo-rectal cancer in Luxembourg, the data presented here may help to define new preventive screening approaches, to increase funding and hasten the introduction of screening for this tumour. The aim of this study was to investigate changes in incidence, stage, anatomical site distribution, and the observed survival rates of colon cancer over two 5-year periods, 1988--1992 and 1994--1998. Analogous to the report on rectal cancer published earlier \[[@B10]\], descriptive epidemiological data on colon cancer were collected to underline the need for the initiation of a nation-wide strategy for the early detection of colorectal cancer. Methods ======= Between January 1^st^, 1988 and December 31, 1998, all consecutive new cases of high-grade adenoma and all new cases of invasive colonic adenocarcinoma (COAC) were registered by the national Morphologic Tumour Registry (MTR) in Luxembourg. These cases had been exclusively diagnosed and \"double-read\" in the central department of pathology by nine senior pathologists. Over the decade the population increased from 374,900 to 429,000, -- an average increase of 1.3% per year. In the elderly age group \>69 years of age, there was a continuous increase by 1.8% (0.5 -- 4.4%) \[[@B11]\]. Patients of all nationalities, race or ethnical origin living in Luxembourg were considered. Disease histology was limited to high-grade adenomas (i.e. adenomas with severe dysplasia) and invasive adenocarcinomas of the colon. Tumours recorded as other carcinomas, metastatic or recurrent disease, metachronous adenomas or invasive adenocarcinomas, malignant carcinoid and mesenchymal malignant tumours were excluded from the current review. For patients with synchronous colon neoplasias, only cases with the worst stage category were taken in account. Data on patients with high-grade adenomas (i.e. stage 0 patients), who by definition could not have metastases in regional lymph nodes or distant metastases, were obtained from 254 biopsy-forceps and polypectomy specimens removed by 26 gastroenterologists and 11 senior-surgeons. Adenomatous proliferations in marginal zones of invasive adenocarcinomas and adenomas with severe dysplasia appearing as recurrent invasive disease within 2 years at the same site were not taken into account. Data on 1158 surgical resection specimens allowed an interpretation in relation to the local and regional spread, the lymph node and the residual tumour status. Hence the stratification of the invasive colonic adenocarcinomas by stage (stage I-IV) led to the exclusion of patients with biopsy diagnosis alone and patients who had only undergone polypectomy without segmental resection (i.e. without regional lymphadenectomy). Patients with known preoperative debulking radiotherapy were not considered. The surgical resection specimens were analysed with respect to tumour-stages (stage I to IV), according to the guidelines of the Union Internationale Contre le Cancer (UICC) and the American Joint Committee on Cancer (AJCC) \[[@B12]-[@B14]\]. We defined stage I and stage II as early cancers and stage III and stage IV as advanced colon cancers. To evaluate the influence of the number of colonoscopic procedures on the incidence rates and the stage distribution of colonic cancers in Luxembourg, the data on colonoscopies (1992--1998), gratefully made available by the National Health Found (NHF), were analysed \[[@B15]\]. The R-classification according to the TNM-system was performed by conventional methods \[[@B16],[@B17]\]. In our series the histological examination of the proximal, distal and lateral (deep) line of resection was mandatory. R0 indicates no residual tumour; R1 corresponds to microscopically and R2 to macroscopically residual tumour, while Rx did not allow a statement on residual tumour. The histopathological diagnoses were made according to the WHO-classification \[[@B18]\]. On this basis, we defined \'high grade\' -- adenomas, the equivalent of stage 0 or former adenocarcinoma in situ, as adenomas of tubular, villous or tubulo-villous histological type with high-grade intraepithelial neoplasia (i.e. severe dysplasia). This category includes cancer cells confined within the glandular basement membrane (intraepithelial) or lamina propria (intramucosal) with no extension through the muscularis mucosae into the submucosa. We considered all colon segments as large bowel, with the exception of the rectum and the recto-sigmoid junction. Anatomically the rectum was defined as a 16 cm long segment between the ano-cutaneous line and the sigmoid colon. Cancers described at the recto-sigmoidal junction without statement on the centimetres from the anal margin were considered as rectal adenocarcinomas and were excluded. Left-sided colon cancers were defined as adenocarcinomas of the descending colon and the sigmoid colon. Right-sided colon cancers included adenocarcinomas situated above the splenic flexure (i.e. splenic flexure, transverse colon, hepatic flexure, ascending colon and caecum with its prolongation, the appendix vermicularis). Beside the evaluation of the frequency of invasive COCA by the 17 \'classic\' 5-year age categories (0--4 years to 80 years and above) the changes of anatomic sub-site distribution in relation to 3 age groups young (\<40 years of age), middle aged (40--69 years) and elderly (70 years and above) patients were analyzed. The observed survival rates of patients with colon cancer were measured from the time of diagnosis and calculated by the actuarial method (life-table) with a 95% confidence interval (c.i.). To evaluate the observed survival rates in relation to the anatomical extent (stage I-IV) and the surgical resection margins (R-status), 164 patients with colon adenocarcinoma diagnosed only by biopsy-forceps without consecutive surgical resection data in our files, who were without final stage and who are part of the 1379 patients with an invasive COCA, were analysed separately. The mortality rates of colon cancer were issued from the anonymous files of the death certificates published by the National Health Direction (Ministry of Health) \[[@B19]\]. The age-standardized incidence rates of the colon cancers diagnosed in the Grand-Duchy of Luxembourg during the period 1983--1997 were compared with the data of other geographical European regions with similar population density and socio-economic characteristics, published by the WHO in \"Cancer in Five Continents, volumes VI, VII, VIII\" \[[@B1]-[@B3],[@B20]-[@B22]\]. The statistical evaluations include the chi-square test (χ^2^) with a level of significance p \< 0.05 and the life-table survival analysis. The age-standardized incidence rates were calculated by the direct method, the standard error of the age-standardized rates by the Poisson approximation \[[@B23]\]. Results ======= The histopathological diagnoses of 1379 consecutive invasive colon adenocarcinomas (COAC) were provided from 164 samples obtained by biopsy only (11.9%), 57 polypectomy specimens (4.1%) and 1158 surgical resection specimens (84.0%). The latter allowed a stratification by stage (UICC/AJCC, 1997) and had reliable follow-up data. The 1379 invasive adenocarcinomas of the colon involved 658 males (47.7%) and 721 females (52.3%); M/F-ratio of 1: 1.1. The mean age was 69.7 years (range: 26--97). During the study period, 254 patients were diagnosed with high-grade adenomas. These patients constitute a separate group from the 1379 invasive COAC. The average, annual, age-specific rates of colon AC in Luxembourg over the period 1988--1998 were 31.1 per 100,000 for all ages and both genders, for males 30.3 per 100,000 and for females 31.9 per 100,000 (Figure [1](#F1){ref-type="fig"}). The comparison of the crude incidence rates of patients with colon AC diagnosed in the 5-year periods 1988--1992 and 1994--1998 revealed an increase by 13.5% for both genders from 28.9 to 32.8, for females by 15.5% from 29.6 to 34.2 and for males by 11.3% from 28.2 to 31.4 per 100,000. The average, annual, age-standardized incidence rates (ASR-world population) of colon cancer (period 1988--1998) were 17.5 per 100,000, for males 20.5 per 100,000 and for females 15.4 per 100,000. Table [1](#T1){ref-type="table"} documents the world age-standardized incidence rates (ASR/W) of the invasive COAC in the Luxembourgish population (females and males) over the period 1983--1997 in comparison to other European countries. The time trend (1988--1998) of the frequency of invasive COAC showed that the total number of invasive COAC rose significantly from 109 cases in the year 1988 to 162 cases in 1998 (in females from 61 to 93, in males from 48 to 69). Comparing the two 5-year periods 1988--1992 and 1994--1998, there was a significant increase in the absolute number of all COAC from 557 to 687 cases (p \< 0.001). The absolute number of COAC in males increased significantly from 266 to 323 cases (p \< 0.02), and in females significantly from 291 to 364 cases (p \< 0.01). Figure [2](#F2){ref-type="fig"} shows the age-distribution of all patients with histologically confirmed invasive colonic adenocarcinoma (n = 1379 cases). Of these patients 1.7% were \<40 years of age; 3.8% between 40 and 49 years; 41.0% between 50 and 69 years; 31.5% between 70 and 79 years and 22.0% of the patients were 80 years and above. Concerning the COAC stage distribution, two peculiarities appeared at the extremes of patient age scales. Below the age of 40, none of the patients (n = 23) had early stage cancer (i.e. stage I or stage II). On the other hand, out of the 303 patients aged 80 years and above, most had stage II (39.0 %) and stage III (38.2 %) disease, whereas stage I (10.6 %) and stage IV (8.3 %) were much less frequent. Time trends in relation to the tumour-stages \[TNM-System (UICC/AJCC\] are represented in Figure [3](#F3){ref-type="fig"}. During the two 5-year periods 1988--1992 and 1994--1998, there was a highly significant increase (p \< 0.001) in the crude incidence rates of the stage 0 cases (i.e. high-grade colonic adenomas) by 190%, an increase in stage I cases by 14.3%, in stage II cases by 12.9%, in stage III cases by 38.5%, and a decrease in stage IV cases by 11.8%. The relation of the 254 new colonic high-grade adenomas to the 1379 new invasive adenocarcinomas stratified by years is shown in Figure [4](#F4){ref-type="fig"}. The comparison of the average of crude incidences of the high-grade adenomas to the invasive adenocarcinomas for the two 5-year periods indicates a highly significant improvement of the ratio by 148% in the second period. There were no significant changes in the number or crude incidences of colorectal endoscopies or colonoscopies with biopsy examinations. The distribution of the 1244 new invasive colon adenocarcinomas by anatomical site and two 5-year periods (1988--1992: n = 557 cases and 1994--1998: n = 687 cases) are given in Figure [5](#F5){ref-type="fig"}. Twelve cases (1.0%), which were not otherwise specified, had to be excluded. In relation to three age groups (\< 40 years of age, 40 -- 69 years and \>69 years) the anatomical sub-site distribution shows a shift to right-sided colon adenocarcinomas by 9.2% in the 40 to 69 years age group (n = 553) and by 76.2% in the group of patients of \>70 years of age (n = 658). The number of the left-sided colon cancers decreased by 14.5% in the 40--69 years age group and increased by 22.9% in those \>70 years. Patients with no precise site and those of the year 1993 were excluded for statistical (two 5-year periods) comparison reasons. In Table [2](#T2){ref-type="table"} the anatomical sub-site distribution in relation to the three age groups and their age-specific rates per 100,000 is represented. The overall observed survival rates of the 1379 patients with an invasive COAC, diagnosed by biopsy or polypectomy or surgical resection specimen were calculated and stratified by years: 1-year: 72 ± 2% (n = 997/1379); 2-year: 60 ± 3% (833/1379); 3-year: 54 ± 3% (n = 740/1379); 4-year: 48 ± 3% (n = 662/1379) and 5-year 44 ± 3% (n = 604/1379). In Table [3](#T3){ref-type="table"}, the observed 5-year survival rates of 931 invasive COAC stratified by stage and operation for cure (R0), and those of 58 patients with COAC with microscopically (R1) or macroscopically (R2) residual tumour after surgical resection are represented. The analysis of the observed survival rates of the 164 patients with colon AC, diagnosed by biopsy-forceps only (i.e. without known final stage), revealed that 70 patients (42.7%) survived 12 months, whereas 20 (12.2%) died in the first month after diagnosis, 46 (28.1%) within the next 6 months and 28 (17.1%) in the period from 6 to 12 months. The age-distribution of these patients showed that 10 (6.1%) patients were \<50 years, 105 (64.0%) were aged between 50 and 79 years and 49 (29.9%) were \>80 years of age. The mortality rates for colon cancer in the Grand-Duchy of Luxembourg over the last 15 years (1984--1998), grouped by 5-year periods, did not change significantly (Table [4](#T4){ref-type="table"}) \[[@B19]\]. There was a slight, insignificant (p \< 0.05) decrease in the crude death rates from 26.3 per 100,000 in the period 1984--1988 to 23.8 per 100,000 in the period 1989--1993 and 23.9 per 100,00 in 1994--1998 \[[@B19]\]. Discussion ========== In contrast to the USA and Canada, the age-standardized incidence rates of colon cancer in Luxembourg steadly increased from 1983--1997 for both genders (Table [1](#T1){ref-type="table"}). Our findings are in concordance with those of many other European countries, where the increase occurs mainly in the male population \[[@B20]-[@B22],[@B24]\]. Some authors believe that the decreasing incidence rates in the USA are due to preventive cancer screening \[[@B25]\]. We found 3.2% cases with synchronous colon adenocarcinomas, which is comparable to the 3.0% and 4.8% of synchronous cancers identified by authors in other countries \[[@B26],[@B27]\]. The effect of age was examined in relation to the classic age groups and in relation to three cohorts of patients, (a) under 40 years of age, (b) between 40 and 69 years, and (c) 70 years and older. The number of colon cancers increases around the age of 45 (Figure [2](#F2){ref-type="fig"}). This is consistent with the findings of Imperiale et al., who found a low incidence of colon adenomas in the age groups 40 to 49 years \[[@B28]\]. Patients under 40 years of age had a higher incidence of advanced stage colon cancers. Similar observations were reported by Mostafa et al. \[[@B29]\]. Considering that high-grade adenomas are obligate precancerous lesions with a slow progression, that polypectomy can mean prevention of cancer, and that the high-grade adenoma/invasive adenocarcinoma ratio (Figure [4](#F4){ref-type="fig"}) improved in the late 1990\'s, we observed evidence of earlier diagnoses \[[@B30]\]. We consider the continous registration of high-grade adenomas and determination of the adenoma-adenocarcinoma ratio as a useful additional quality assurance marker in evaluating clinical and histopathological diagnostic procedures. To explain the unexpected increase of precancerous colonic lesions, we analysed our files in order to identify parameters that might have changed, such as staff composition or other factors. Over the period 1992--1995 the number of physicians (n = 38) practicing endoscopy did not change and the number of high-grade adenomas (n = 109 cases) removed by each endoscopist varied between one and six cases. Considering that, the number and the crude annual incidence rate of the colonoscopies did not rise significantly over the time, that the staff of pathologists remained the same, the histopathological criteria and coding did not change, and that a systematic organized screening programme until the year 2000 was not launched and a spontanous, unorganized screening is not known, we believe that a more selective indication for colonoscopy examinations may be one of the causes of this increase. The abrupt change after 1993 in the number of stage 0-cases may be the consequence of the changes in the statute of the National Health Fund. In 1992 the notions of patients and medical care providers \'profiling\' at the national level were introduced. At the same time, the media campaign for preventive medical care strategies against cancer (breast, malignant melanoma) launched by the governmental institutions in collaboration with the NHF and the Luxembourg Foundation against Cancer may have had a beneficial (secondary) effect. Not only the incidence of stage 0-cases rose, but we found also a positive influence on the crude incidence rates of stage I and stage II patients increasing over the two 5-year period (1988--1992) and (1994--1998) by 14.3 and 12.9% respectively. Unfortunately there has been a highly significant (p \< 0.001) increase in stage III patients by 38.5% too. Colon AC in young patients (40 years of age or less) seems to be more aggressive, since those patients were diagnosed exclusively in stage III and stage IV. Since these patients often present a positive family history of colorectal cancer, there is an obvious need for health-care providers to heighten awareness, and to identify high risk persons in this population group \[[@B25],[@B31],[@B32]\]. In North America, Japan, France and other high-risk areas, there has been a proximal shift in the sub-site distribution of colorectal cancer \[[@B4],[@B29],[@B32]-[@B34]\]. Similar changes have occurred in Luxembourg. This proximal shift was found in both sexes, especially in patients aged 70 and above. In males the increase was 14.0%, in females 73.6%. Mitry et al. \[[@B33]\] found that right colon cancer rates had increased by 21.6% in males and 10.4% in females. This underlines the need to discuss the opportunity for total colonoscopy in patients of higher risk or elderly age groups \[[@B30]\]. In our cohort, the overall 5-year survival rates of the patients with surgery for colon cancer without distinction of the residual tumour-status were 46 ± 3% (95% c.i.). Because the data on age-specific risk of death from other causes than colon cancer during the 1990\'s are not yet available, the relative survival rates could not be calculated. It must be admitted that relative survival rates will give better data for colon cancer in Luxembourg than indicated by the observed survival rates reported here. The relative 5-year survival rates reported in the literature vary between 30.6 and 55.3% \[[@B6],[@B35]\]. Because COAC incidence increases at the age 50 approximately in our series, because a significant shift to right-sided colon cancer has been detected in patients of 70 years of age and above and because survival rates would probably improve if the disease is treated in its early stages, a strategy to perform twice-lifetime colonoscopy (at ages 50 and 60 years or at the transition to the pension age with 65 years) for asymptomatic patients without family history or other risk factors, as discussed by others, needs further investigations \[[@B4],[@B29],[@B34],[@B36],[@B37]\] Conclusion ========== In contrast to the time trends in the USA and Canada, we found increasing age-standardized incidence rates in colonic adenocarcinomas and an increase of the crude incidence rates especially in stage III patients. Despite the 4-fold increase in high-grade adenoma diagnoses (i.e. stage 0 cases), efforts should be reinforced in preventive strategies to detect more precancerous lesions and/or early tumour-stages, to reduce the true incidence of advanced tumour-stages at diagnosis, and by this the mortality rates. The findings of this study should provide baseline information and scientific evidence for the elaboration and implementation of cancer preventive and intervention strategies to those populations at average and high-risk for colorectal cancers. As long as new diagnostic procedures, such as stool-based DNA-test for colorectal tumours, are not yet available for mass-screening purposes, and as long as the effectiveness of a complete colonoscopy screening for colorectal cancer in asymptomatic patients beginning at the age of 50 has not been proven by population-based controlled trials, the data underline the need of at least a systematic faecal occult blood test (FOBT) screening for colorectal cancer beginning at the age of 50 \[[@B30],[@B38],[@B39]\]. Our data support mainly that screening could be considered in preventing deaths from colo(rectal) cancer. The findings can be used by the national health care authorities to define a screening strategy and to help in deciding the target age-groups in Luxembourg. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors (RS, PP, YW, NK, CC) collaborated intensely on all aspects of the manuscript, from research design to data preparation and presentation. RS wrote, and all authors approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/52/prepub> Acknowledgements ================ We thank Mrs Mireille Braun, Mrs Martine Becker and M. Guy Weber for their helpful assistance during the preparation of this manuscript. The study received financial support from the national Morphologic Tumour Registry (RMT a.s.b.l.). The opinions stated in this document are those of the authors and do not necessarily represent the official position of the RMT. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Time trend of the crude incidence rates of the invasive colon adenocarcinomas (n = 1379) by gender; 1988--1998. ::: ![](1471-2407-5-52-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Age-distribution of the invasive colonic adenocarcinomas (n = 1379 cases); 1988--1998. ::: ![](1471-2407-5-52-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Crude incidence rates of the high-grade adenomas (n = 254) and the invasive adenocarcinomas of the colon (n = 1379) stratified by stage (TNM/UICC-AJCC, 1997) \[14\]. ::: ![](1471-2407-5-52-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Crude incidence rates and colonic high-grade adenoma / invasive adenocarcinoma-ratio in comparison to colonoscopies: 1988--1998; both genders; n = 254\*/1379 cases (\*diagnoses by biopsy only included). ::: ![](1471-2407-5-52-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Number of invasive colonic adenocarcinomas (n = 1244; cases of the year 1993 excluded) by anatomic site; 1988--1992 and 1994--1998. (N.O.S.\* = not otherwise specified) ::: ![](1471-2407-5-52-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Colon cancer in the European Community: world age-standardizedincidence rates \[ASR(W)\*\], time trends: 1983--1987; 1988--1992; 1993--1997. ::: **Males** **83--87** **88--92** **93--97** **Females** **83--87** **88--92** **93--97** --------------------- ------------ ------------ ------------ --- --------------------- ------------ ------------ ------------ --- F/Bas Rhin 27.2 30.2 31.3 ↑ D/Saarland 19.4 20.4 21.5 ↑ D/Saarland 23.7 25.5 27.3 ↑ NL/Maastricht 17.1 17.9 19.4 ↑ I/Parma 23.0 24.4 27.0 ↑ DK/Denmark 19.3 19.9 18.6 ↓ IRL/Southern 21.8 24.2 25.3 ↑ IRL/Southern 19.7 20.3 18.3 ↓ E/Navarra 13.8 19.2 22.1 ↑ I/Parma 16.1 17.8 17.9 ↑ NL/Maastricht 19.7 21.9 22.1 ↑ UK/South Western 15.9 16.2 17.3 ↑ UK/South Western 18.2 19.2 22.1 ↑ F/Bas Rhin 17.4 18.8 16.9 ↓ **L/Luxembourg** 15.5 20.2 **21.9** ↑ **L/Luxembourg** 11.3 14.9 **15.0** ↑ DK/Denmark 20.3 20.6 20.7 ↑ S/Sweden 15.8 15.9 15.0 ↓ CH/St.Gall 16.4 16.5 19.7 ↑ E/Navarra 9.6 13.0 12.4 ↑ S/Sweden 17.5 17.7 17.7 ↔ CH/St.Gall 11.7 12.0 12.1 ↑ SF/Finland 11.9 12.8 14.5 ↑ SF/Finland 10.6 11.6 12.1 ↑ CND/Canada 27.8 26.9 25.9 ↓ CND/Canada 23.6 21.3 20.0 ↓ USA/Seer\*\*, white 31.1 28.1 25.5 ↓ USA/Seer\*\*, white 23.6 20.8 19.4 ↓ \* Cancer incidence in five continents. Vol VI; Vol VII; Vol VIII \[20, 21, 22\] \*\* SEER: Surveillance, Epidemiology and End Results Program ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Invasive colon adenocarcinomas (n = 1232\*) by anatomic site (left and right colon), by age groups (\<40 years; 40--69 years; \>69 years) and age-specific incidence rates; 1988--1992 and 1994--1998. ::: **Age groups** **Number of incident cases (ri)** **Person- years of obervation (ni)** **Age-specific rates per 100,000 (ri/ni)** **p-values \*\*** ------------------------ ----------------------------------- -------------------------------------- -------------------------------------------- ------------------- **\<40 years** colon right 1988--1992 5 1 072 099 0.5 p = n.s. colon right 1994--1998 9 1 151 878 0.8 colon left 1988--1992 2 1072 099 0.2 p = n.s. colon left 1994--1998 5 1151 878 0.4 **40--69 years** colon right 1988--1992 130 676 772 19.2 p = n.s. colon right 1994--1998 142 742 577 19.1 colon left 1988--1992 150 676 772 22.2 p = n.s. colon left 1994--1998 131 742 577 17.6 **\>69 years** colon right 1988--1992 122 175 629 69.5 p \< 0.001 colon right 1994--1998 215 196 145 109.6 colon left 1988--1992 144 175 629 82.0 p = n.s. colon left 1994--1998 177 196 145 90.2 \* excluded colon cancers N.O.S.(n = 12) and those of 1993 (n = 135) \*\* level of significance p \< 0.05; n.s. = not significant ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Prognoses of invasive colonic adenocarcinomas resected for cure (R0) or with residual tumour (R1+R2) and stratified by stage-TNM/UICC \[12, 13, 14, 23\]. ::: --------------------------------------------------------------------------------------------------------------------------------------------------------- **Stage** **TNM** **Observed 5-year survival rates (actuarial method, 95% confidence interval)** ------------------------ -------------- -------------------------------------------------------------------------------- --------------------- ---------- **UICC, 5^th^edition** **R0** **R1+R2** **p-values \*\*\*** **I \*** T1 N0 M0\ 71 +/- 8%\ 100% \*\*\ \<0.5188 T2 N0 M0 (n = 129) (n = 1) **II \*** T3 N0 M0\ 63 +/- 5%\ 45 +/- 30%\ \<0.2392 T4 N0 M0 (n = 388) (n = 11) **III \*** anyT N1 M0\ 38 +/- 5%\ 4% \*\*\ \<0.0007 anyT N2 M0 (n = 331) (n = 25) **IV \*** anyT anyN M1 17 +/- 8%\ 5% \*\*\ \<0.1584 (n = 83) (n = 21) **I-IV \*** 51 +/- 3%\ 14 +/- 9%\ \<0.0001 (n = 931) (n = 58) --------------------------------------------------------------------------------------------------------------------------------------------------------- \* only surgical specimens with available survival data \*\* number of cases too low for calculating the 95% confidence interval \*\*\* level of significance p \< 0.05 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Analysis of the mortality rates from all colon adenocarcinomas of the last 15 years (1984--1998) declared in Luxembourg by death certificate only \[19\]. ::: **Males** **Females** **both genders** **p-values\*** ----------------- ------------ ------------- ------------------ ---------------- **1984--1988:** 26.5/10^5^ 26.2/10^5^ 26.3/10^5^ p = n.s. **1989--1993:** 22.5/10^5^ 25.0/10^5^ 23.8/10^5^ p = n.s. **1994--1998:** 23.6/10^5^ 24.3/10^5^ 23.9/10^5^ \* level of significance p \< 0.05 ; n.s. = not significant for both genders :::

PubMed Central

2024-06-05T03:55:58.351136

2005-5-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173094/", "journal": "BMC Cancer. 2005 May 24; 5:52", "authors": [ { "first": "René", "last": "Scheiden" }, { "first": "Paul", "last": "Pescatore" }, { "first": "Yolande", "last": "Wagener" }, { "first": "Nelly", "last": "Kieffer" }, { "first": "Catherine", "last": "Capesius" } ] }

PMC1173095

Background ========== The *WWOX*gene, originally cloned by our laboratory, spans a genomic region greater than 1 Mb in size and is the second most common chromosomal fragile site, FRA16D (16q23) \[[@B1],[@B2]\]. Abnormalities affecting *WWOX*at the genomic and expression level have been reported in numerous neoplasias and cancer derived cell lines including, breast, ovarian, esophageal, lung, stomach, liver, pancreas and hematological malignancies \[[@B3]-[@B12]\]. We observed that ectopic WWOX expression inhibited anchorage independent growth and *in vivo*tumorigenicity of highly aggressive breast carcinoma lines, suggesting a putative tumor suppressor role for this novel protein \[[@B13],[@B2]\]. *WWOX*encodes a 46 KD, 414-amino acid protein that contains two WW domains at the NH~2~terminus and a short chain oxidoreductase (SDR) central domain \[[@B1]\]. The first WW domain- is involved in protein-protein interactions by binding the specific proline rich motif PPXY and several potential candidate partner proteins have been postulated \[[@B14],[@B15]\]. Within the SDR domain, the presence of WWOX amino acid residues, serine 281, and 293-**[Y]{.underline}**NRS**[K]{.underline}**-297 make up a catalytic signature motif conserved in short-chain steroid dehydrogenases \[[@B16]\]. We originally reported high *WWOX*mRNA expression levels in ovary, prostate, testis and breast \[[@B1]\]. In this study we analyzed WWOX protein expression pattern in normal ovary and ovarian carcinomas. We correlated WWOX protein expression with ovarian carcinoma histotypes and clinico-pathological parameters. In addition, since we recently observed a strong association between loss of WWOX expression and estrogen and progesterone receptor (ER and PR) status in breast cancer \[[@B12]\], we also investigated any potential association between expression of sex steroid hormone receptors and WWOX in ovarian cancer. Methods ======= Western blot analysis --------------------- Total protein extracts were prepared from snap frozen tissue of 38 human ovarian carcinomas and 5 normal human ovarian tissues. As negative control for WWOX protein expression we used protein extracts from the ovarian cell line PEO1 that does not express WWOX due to a homozygous deletion affecting exons 4--8 of this gene \[[@B4]\], a kind gift of Dr. Hani Gabra at Imperial College London, UK. As positive control we used the same cell line stably transfected with a WWOX expressing vector (PE01-WWOX) \[[@B10],[@B12]\]. Total cell protein lysates were made using RIPA buffer (50 mM Tris pH7.5, 150 mM NaCl, 0.5% sodium deoxycholate, 1% Triton X-100, 0.1% SDS) containing protease inhibitor cocktail (Roche, Mannheim, Germany). For western blotting 50 ug of total protein was separated by 12.5% SDS-PAGE and transferred to PVDF membranes (Millipore, Billerica, MA). Immunodetection was performed using Protein Detector™ (KPL, Gaithersburg, MD) western blotting reagents as described by the manufacturer. WWOX protein was detected using affinity-purified anti-WWOX rabbit polyclonal primary antibodies developed in our laboratory (final concentration 280 ng/ml) \[[@B12]\] and HRP conjugated anti-rabbit secondary antibody (KPL, city, state, 1:2000 dilution) followed by chemiluminescence autoradiography. Actin was used as the protein loading control and it was detected using monoclonal anti-actin antibody (ICN biomedicals, Burlingame, CA, 1:1000 dilution) and HRP conjugated anti-mouse secondary antibody (KPL, 1:5000). Quantitation of X-ray films exposed to western blot chemiluminescence-emitting membranes was performed using a Kodak digital science Image Station 440CF. Protein loading and signal intensity was controlled by normalizing each sample to the aforementioned positive control, PE01-WWOX as previously described \[[@B10],[@B12]\]. Ovarian tissue microarrays -------------------------- The MD Anderson Cancer Center ovarian TMA was prepared by D.R. at the J.L laboratory \[[@B17]\] with samples from 441 patients with primary epithelial ovarian cancer that had undergone surgery at M. D. Anderson Cancer Center between 1990 and 2001. Follow-up information was updated through June 2003. Histopathologic diagnoses assigned at the time of treatment were based on World Health Organization criteria, each sample was reclassified by grade based on the Gynecologic Oncology Group (GOG) criteria \[[@B18]\] and each case was also classified by disease stage according to the International Federation of Gynecology and Obstetrics (FIGO) system \[[@B19]\]. The appropriate institutional committee approved use of human tissue blocks and clinical records reviews. Each core was scored individually and the results are reported as the mean of at least two replicate core sample measurements \[[@B20]\]. We also used a second ovarian TMA set generated at the Fox Chase Cancer Center (FCCC) tumor bank by R.P. at the A.J.P.K laboratory. This TMA set included a total of 86 ovarian tumors and 21 normal ovarian samples. Cores of normal ovaries were used as positive controls. Only primary mucinous carcinomas of the ovary where included on this cohort of patients. Selection criteria was based on histopathological diagnosis made by a trained pathologist on whole sections, histochemical stainings, immuhohistochemical staining, and clinical assessment of the patient. By pooling the TMA sets from both institutions we were able to successfully analyze in duplicate WWOX protein staining from a total of 444 invasive surface epithelial derived primary ovarian carcinomas that included Serous (n = 375), Endometroid (n = 40), Mucinous (n = 10) and Clear Cell Carcinoma (CCC) (n = 19) histological subtypes. A subset of these cases (n = 323) was successfully processed and analyzed for allowing comparative analysis of WWOX protein expression with ER and PR status. Immunostaining method --------------------- Anti-WWOX immunostaining was performed as previously reported \[[@B12]\]. Mouse monoclonal antibodies for ER, NCL-ER-6F11 (Novocastra, UK) and PR, NCL-PGR-312 (Novocastra, UK) were used according to manufacturer recommendations. The anti-PR antibody used in these studies is able to recognize only the A isoform of this receptor \[[@B21]\]. Evaluation of immunohistochemical staining ------------------------------------------ Staining intensity was measured using a Chromavision Automated Cellular Imaging System (Chromavision^®^ACIS^®^, San Juan Capistrano, CA) as previously described \[[@B12]\]. The methods for scoring cytoplasmic or nuclear staining are based on different program functions of the ACIS^®^and are the recommended by the manufacture. For anti-WWOX staining measurements we employed the generic DAB software application provided by ACIS. The area of each individual TMA core was considered for the measurements in its totality. The software determines brown intensity (i.e. positive stained cells) regardless of the area covered by the positive cells. The cutoff to differentiate positive and negative staining was determined to be a mean intensity of 63 (arbitrary staining intensity units, s.i.u.) over a total of 255 (color saturation scale). Therefore, cores with values = 63 s.i.u were \'negative\' for WWOX immunostaining, i.e. no brown staining observable. Values between 63 and = 65 s.i.u. were considered to be in the \'weak/low\' staining category, i.e. barely visible brown staining, 65 to 81 s.i.u were considered \'moderate\' and cores with staining intensity values \> 81 s.i.u were considered to fall in the \'strong\' staining category. For sake of simplicity cases were classified in two groups: a) negative + low = Negative-Low WWOX group, i.e. any core with a value = 65 in staining intensity and b) moderate + strong = High WWOX group, i.e. any core with values above 65 in staining intensity. The Negative/Low WWOX group was composed of 79% (105/133) completely negative cases and 21% (28/133) weak cases. The mean WWOX intensity staining for normal ovarian surface epithelium (OSE) fell in the moderate category (mean value 70 ± 4). Anti-ER and PR antibodies ------------------------- For performing anti-ER and anti-PR determinations we utilized the nuclear antigen software application provided by ACIS. Only areas of tumor in each TMA core were considered for measurements. The free form drawing tool provided by the software was used to select tumoral areas and to exclude other structures. A final score was determined by the percentage of brown stained nuclei (i.e. positive cells) over the total of tumor cell nuclei measured (i.e. positive plus negative cells). Only cores with at least 10% tumor were included in the analysis. For the receptor analyses, positivity of nuclear staining was defined as follows: weak \> 5% stained nuclei, moderate 5--40% stained nuclei and strong \>50% stained nuclei. Statistical methods ------------------- Univariate analysis of clinical-pathological parameters based on WWOX staining was determined by χ^2^test with Yates\' correction. The basic significance level was fixed at *p*\< 0.05 and all data was analyzed using SPSS statistical software (Version 11.0; SPSS Inc., Chicago, IL). Results ======= Western blot analysis --------------------- Immunoblotting analysis demonstrated that full length WWOX protein was the predominant WWOX isoform expressed in both normal ovary and ovarian carcinomas (Figure [1B](#F1){ref-type="fig"}). The anti-WWOX antibody used was raised against the WW domains \[[@B2]\] and should detect all potential WWOX isoforms. The specificity of this WWOX antibody in the immunoblotting analyses was demonstrated by observing no immunoreactive products in cell extracts from the PEO1 ovarian carcinoma cell line used as negative control (Figure [1A](#F1){ref-type="fig"}). In addition, as a positive control PEO1 cells were stably transfected with a WWOX expression vector (Figure [1A](#F1){ref-type="fig"}). Normal ovarian tissues displayed a consistently strong WWOX specific signal while WWOX protein expression levels were extremely variable among tumor samples. Some tumors displayed barely detectable, if any, WWOX protein when compared to normal tissue, e.g. T108 while other samples had significantly higher WWOX levels e.g. T578 in Figure [1](#F1){ref-type="fig"}. We concluded that 26% (10/38) of tumors over-expressed WWOX, 34% (13/38) expressed normal WWOX levels and 40% (15/38) of tumors had levels lower than 50% that observed in normal ovary. Pattern of WWOX staining in normal ovary ---------------------------------------- Initially we characterized the cellular localization of WWOX protein expression in normal ovaries by IHC using the antibody described in the previous section. The specificity of the WWOX antibody for IHC analysis was validated by observing that the immunoreactivity was abolished by pre-absortion to the recombinant GST-WWOX fusion protein used to raise the antibody (data not shown). The analysis of TMA cores and whole sections of normal human ovary showed that WWOX protein is constitutively expressed at high levels in normal OSE cells (Figure [2A](#F2){ref-type="fig"}), in epithelial inclusion cysts within the cortex (Figure [2B](#F2){ref-type="fig"}) and in scattered granulosa-lutein cells surrounding aged corpora lutea or corpora albicans (not shown). Positive staining was also detected in one observed Walthard nest within the ovarian hilus and in a few stromal cell patches likely related with hormone production. In all cases the inmunostaining had a homogeneous and diffuse pattern that was localized to the cytoplasm. WWOX staining in ovarian carcinomas ----------------------------------- WWOX expression was highly variable among ovarian tumors (n = 444). WWOX staining intensity was determined in all four major surface epithelium derived ovarian carcinoma histotypes, i.e. Serous, Endometroid, Mucinous and Clear Cell (Table [1](#T1){ref-type="table"}). Thirty percent of ovarian carcinomas (133/444) showed loss of WWOX protein expression while 70% (311/444) demonstrated positive staining including cases with very strong staining (Figure [3A--I](#F3){ref-type="fig"}) (Table [1](#T1){ref-type="table"}). Among serous carcinomas, 29% of the cases (109/375) fell in the negative/low group and 71% (266/375) stained moderately or strongly positive. The most typical WWOX staining pattern observed for all histotypes was cytoplasmic and diffuse, no tissues demonstrated any nuclear staining. Interestingly, some of the carcinomas showed a distinctive strong cell membrane staining pattern (Figure [3](#F3){ref-type="fig"}). Other tumors displayed a predominantly apical border staining, in conjunction with staining of what appeared to be luminal secretions (Figure [3B](#F3){ref-type="fig"}). Among endometroid carcinomas 23% of the cases (9/40) were negative/low and 77% (31/40) stained moderately or strongly positive for WWOX expression. This histological subtype showed a diffuse and apical cytoplasmic pattern of protein expression (Figure [3D, E](#F3){ref-type="fig"}). Among the CCC group, 42% of the cases (8/19) were negative/weak while 58% (11/19) stained moderately or strongly for WWOX (Figure [3G--I](#F3){ref-type="fig"}). Within the mucinous carcinoma group, we observed that most tumors (7 out of 10 cases) did not stain for WWOX expression. Furthermore the three positive mucinous carcinomas were clearly below the mean value for normal OSE. The immunoreactivity localized to the cytoplasm as in the other histotypes but in a very specific perinuclear fashion (Figure [3J](#F3){ref-type="fig"}). The high rate of absence of WWOX protein expression among the mucinous carcinoma group was statistically different compared to the other three histotypes (p = 0.0131, Table [1](#T1){ref-type="table"}). WWOX staining analysis according to stage and grade --------------------------------------------------- WWOX expression was next correlated with tumor Stages (FIGO). We observed that 23% of Stage I (7/31), 29% of Stage II (8/28), 31% of Stage III (74/242) and 48% of Stage IV (31/65) tumors showed negative/low WWOX protein expression. We detected a statistically significant trend between the decrease in WWOX expression and more advanced FIGO stages (-b = -0.135; *p trend=*0.007, Table [2](#T2){ref-type="table"}). No correlation was found between tumor grade (GOG) and WWOX staining (p = 0.707, Table [2](#T2){ref-type="table"}). WWOX staining analysis according to relapse and survival -------------------------------------------------------- An analysis of WWOX protein expression relative to disease relapse was performed in a subset of cases from the MDACC TMA (n = 354). Between relapsing and non-relapsing cases no statistical difference was observed regarding WWOX expression (p = 0.094) (Table [2](#T2){ref-type="table"}). Among cases with progressive disease, 46% (27/59) showed negative/low WWOX levels of expression while 54% (32/59) fell in the positive category. It was possible to evaluate survival in cases from the MDACC group as shown by the Kaplan-Meier plot (Figure [4](#F4){ref-type="fig"}). Importantly, we observed a statistically significant correlation of WWOX expressing cases with longer overall survival and loss of WWOX protein expression correlated with shorter overall survival (p = 0.03). WWOX staining analysis in correlation with ER and PR status ----------------------------------------------------------- We analyzed the association of WWOX with ERα and PR status in ovarian carcinoma cases (n = 323). When ER status was correlated with WWOX expression levels no significant differences were found (Table [2](#T2){ref-type="table"}). On the other hand, a statistically significant correlation was observed between WWOX expression and PR expression (p = 0.008, Table [2](#T2){ref-type="table"}). In the WWOX negative cases 85% (70/82) were PR negative while only 15% (12/83) were PR positive. Discussion ========== Since our original cloning of *WWOX*\[[@B1]\] abundant evidence has accumulated indicating that this gene likely plays a role in either tumor initiation or progression in various neoplasias including ovarian cancer \[[@B3]-[@B12]\]. WWOX spans the second most common chromosomal fragile site, FRA16D. This genomic region is prone to chromosome breakage, recombination and gross rearrangements. Multiple ovarian carcinoma studies have demonstrated high frequency of allelic losses for specific chromosomal regions on 16q \[[@B22]\] and since the losses have been found both in low and high-grade tumors it has been proposed to be an early event in ovarian carcinoma development \[[@B23]\]. Loss of heterozygosity at 16q23.2-q24.2 has been correlated with ovarian carcinoma metastasis and advanced tumor stages \[[@B24]\]. Comparative Genomic Hybridization (CGH) studies confirmed those observations also indicating high frequency of 16q21-q24 genomic losses in advanced invasive ovarian carcinoma stages \[[@B25],[@B26]\]. Taken together the data indicates that a gene, likely *WWOX*, located at 16q23-24 is associated with a worse prognosis in ovarian cancer. In this study WWOX protein expression in normal ovary was predominantly observed in the OSE and epithelial inclusions, both sites proposed to be where ovarian epithelial neoplastic transformation originates. Because of this finding together with the observations of Paige (2001) \[[@B4]\], that ovarian carcinoma cell lines have high frequency of genomic losses affecting the *WWOX*gene including homozygous deletions, we considered it of relevance to analyze WWOX protein expression among the four major surface epithelial derived ovarian carcinoma histotypes. Our analysis by immunoblotting and IHC demonstrated that about one third of ovarian carcinomas displayed extremely reduced or absent WWOX protein expression. It is worth noting that in all instances, normal and tumor, WWOX staining was exclusively cytoplasmic and in some serous carcinomas staining was also localized to the apical cell membrane or luminal border. This is similar to our observations in breast tumors \[[@B12]\], where WWOX was always cytoplasmic without nuclear staining. This is also in agreement with observations from other groups \[[@B11]\]. This is in stark contrast with Watanabe (2003) \[[@B27]\] who observed that some cells of normal and tumor breast and gastric cases had WWOX protein expression localized to the nuclei. Possibly this discrepancy is due to the use of different antibodies. In this ovarian cancer study, we also observed that Serous and Endometroid histotypes express normal or high WWOX but two rare but well described ovarian carcinoma histotypes, mucinous and clear cell (28,29,30), demonstrated a higher frequency of loss of WWOX protein expression. This association was more significant for the mucinous type, for which we observed that 70% (7/10) of cases were devoid of WWOX protein expression. Eventhough a small number of mucinous cases were studied the overall high frequency of decreased or loss of WWOX in these tumors as a group is intriguing. Borderline significance was found for the pure CCC for which 42% of cases demonstrated no or very low WWOX protein expression (Figure [3H](#F3){ref-type="fig"} versus [3G](#F3){ref-type="fig"}). Primary mucinous ovarian tumors have been shown to carry a higher frequency of K-ras mutations as one of the few genetic hallmarks able to distinguish it from other histotypes, while CCC are characterized by a lack of p53 mutations. Mucinous carcinomas and CCC, has been associated with a higher frequency of resistance to chemotherapy (31). Interestingly, in our study, we observed a statistically significant trend between loss of WWOX expression and worse patient outcome. Patients whose tumors expressed WWOX at moderate or high levels fared better than those with low expression of this enzyme (Figure [4](#F4){ref-type="fig"}). We also found a statistically significant correlation with loss of WWOX and stage IV at initial diagnosis (FIGO). The association with overall survival could be due to an increased frequency of loss in higher stage tumors that are associated with a poorer outcome or alternatively due, in part, to an increased frequency of clear cell and mucinous tumors which have a lower likehood of responding to therapy \[[@B31]\]. It was also of interest that we observed a positive correlation between WWOX and PR expression. In contrast to our previous observations in breast cancer in which we observed a strong correlation between ER and PR with WWOX expression \[[@B12]\], in this ovarian study, only PR was associated with WWOX expression (Table [2](#T2){ref-type="table"}). Specifically PR negative ovarian carcinomas also lacked WWOX expression. Progestins acting through the PR have been postulated to be protective for ovarian carcinoma development \[[@B32],[@B33]\] due to a postulated ability of inhibiting cell proliferation and inducing apoptosis (34,35). We noticed that all (100%) of our analyzed mucinous carcinomas were PR negative which agrees with the literature regarding absence of PR expression in mucinous tumors \[[@B36],[@B37]\] while 90% of CCC were PR negative. These correlations raise the question of whether the observed positive association between PR loss and WWOX loss is a consequence of the predominance of WWOX loss in the two aforementioned histotypes, rather than a direct mechanistic association between WWOX and PR. Changes in progesterone signaling and regulation of progesterone-responsive genes can be of critical significance in the ovarian tumorigenic process by itself \[[@B38]\]. The opposite could be true as well i.e. that high levels of WWOX expression is associated with PR expression and as a consequence related to the putative protective effect of progestins that in turn could be associated with more favorable patient outcomes. Conclusion ========== We provide evidence that WWOX protein expression is frequently altered and highly variable in ovarian carcinomas. The loss of WWOX expression shows association with Mucinous and CCC histotypes more than the others and also shows a tendency with PR negative expression. Significantly, reduced WWOX protein expression correlates with less favorable outcome. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= CMA conceived, coordinated, designed the study, interpreted the data and revised the manuscript. MIN coordinated, performed pathological and immunohistochemical study, interpreted the data, drafted and write the report.DR constructed M.D. Anderson Cancer Center Tissue Microarray and interpreted the data. JHLM participated in preparation of the antibody, Western Blot analyses, interpretation of the data and critically reviewed the manuscript. MCA performed statistical analysis, figures and tables and interpreted the data. HK carried out Western Blot analyses. RP constructed Fox Chase Cancer Center Tissue Microarray. AJPKS, AG, JL and GM provided tumor samples, interpretation of the data and critically reviewed the manuscript. All of the authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2407/5/64/prepub> Acknowledgements ================ The authors gratefully acknowledge Nancy W Abbey for her outstanding technical assistance. The authors acknowledge NCI RO1 CA102444 (C. M. Aldaz), SPORE in Ovarian Cancer P50 CA83639, tissue pathology core grant NCI PO1 CA64602 and ACS RSG-04-028-CCE (J. Liu), and center grant ES-07784. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### WWOX protein expression in normal ovary and ovarian adenocarcinomas as determined by immunoblot analysis. A) Total protein extracts from PEO1 cell lines transfected with an empty vector or a WWOX expressing vector were analyzed by immunobloting with the anti-WWOX antibody. Note: no immunoreactive bands are observed in the vector transfected cell line. B) WWOX protein expression was determined by immunoblotting of total protein extracts from 38 ovarian carcinoma samples. Five normal ovarian tissue extracts are shown on the first five lanes. Quantitation of WWOX protein expression. Autoradiographs of WWOX and actin were digitized using the Kodak digital science Image Station 440CF. WWOX expression in each sample was normalizated to actin to correct for loading differences. In turn these numbers obtained from each tumor were normalized and expressed as relative to the normal ovarian values (i.e. Relative Expression). ::: ![](1471-2407-5-64-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### WWOX immunohistochemical staining of normal ovary. A) Representative photomicrograph (20X) of normal ovary displaying positive staining in ovarian surface epithelial cells. B) Photomicrograph (20X) showing strong WWOX inmunostaining localizing to the cytoplasm of inclusion cyst epithelial cells. ::: ![](1471-2407-5-64-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### WWOX inmunohistochemical staining in ovarian carcinoma samples by histotypes. A-C. *Serous ovarian carcinomas*. A, Strong-Moderate. Note heterogeneity in staining intensity pattern in this tumor sample; B, Moderate. Note predominance of apical staining in this papillary serous ovarian carcinoma; C Weak-Negative, WWOX lack of staining observed in approximately 30% of serous carcinoma cases. D-F. *Endometroid ovarian carcinomas*. D and E positive WWOX cytoplasmic staining and F negative staining. G-I. *Clear Cell ovarian Carcinomas*. G, representative photomicrograph of one of a moderately WWOX positive CCC case, while H and I, represent typical negative cases. J-L. *Mucinous ovarian carcinomas*, J, representative mildly positive case and K representative mucinous carcinoma of the endocervicoid subtype with demonstrating no WWOX staining. L, mucinous carcinoma of the intestinal subtype also negative for WWOX staining. ::: ![](1471-2407-5-64-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Kaplan Meier Plot analysis showing the pattern of Overall Survival relative to WWOX protein expression. ::: ![](1471-2407-5-64-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### WWOX intensity staining according to the four major epithelial derived ovarian carcinoma histotypes. ::: ***Histology*** ***WWOX intensity*** ***Total*** --------------------------- ---------------------- ------------------- --------- **Serous carcinoma** 109/375 (29%) 266/375 (71%) **375** **Endometroid carcinoma** 9/40 (23%) 31/40 (77%) **40** **Clear cell carcinoma** 8/19 (42%) 11/19 (58%) **19** **Mucinous carcinoma \*** 7/10 (70%) 3/10 (30%) **10** **Total** **133/444 (30%)** **311/444 (70%)** **444** *\* (χ*^2^= *8.67; p = 0.0131)* ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Clinical parameters and steroid receptor status analyzed in correlation with WWOX staining. ::: ***Clinical parameters*** ***WWOX intensity*** ***Total*** ***p value*** --------------------------- ---------------------- ------------- --------------- ------------- **Grade (GOG)** 1 4 (23%) 13 (77%) **17** *p = 0.707* 2 9 (33%) 18 (67%) **27** 3 107 (33%) 215 (67%) **322** **Stage (FIGO)** I 7 (23%) 24 (77%) **31** *p = 0.007* II 8 (29%) 20 (71%) **28** III 74 (31%) 168 (69%) **242** IV 31(48%) 34 (52%) **65** **Relapse** No 31(32%) 66 (68%) **97** *p = 0.09* Yes 61(31%) 137(69%) **198** Progressive disease 27(46%) 32 (54%) **59** **ER** Negative 22 (25%) 66 (75%) **88** *p = 0.984* Positive 59 (25%) 176 (75%) **235** **PR** Negative 70 (29%) 170 (71%) **240** *p = 0.00* Positive 12 (14%) 71 (86%) **83** :::

PubMed Central

2024-06-05T03:55:58.354966

2005-6-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173095/", "journal": "BMC Cancer. 2005 Jun 27; 5:64", "authors": [ { "first": "María I", "last": "Nunez" }, { "first": "Daniel G", "last": "Rosen" }, { "first": "John H", "last": "Ludes-Meyers" }, { "first": "Martín C", "last": "Abba" }, { "first": "Hyunsuk", "last": "Kil" }, { "first": "Robert", "last": "Page" }, { "first": "Andres JP", "last": "Klein-Szanto" }, { "first": "Andrew K", "last": "Godwin" }, { "first": "Jinsong", "last": "Liu" }, { "first": "Gordon B", "last": "Mills" }, { "first": "C Marcelo", "last": "Aldaz" } ] }

PMC1173096

Background ========== Myocarditis is an inflammatory cardiomyopathy that occurs in all age groups \[[@B1]\]. Most cases are thought to be sub-clinical, but myocarditis can manifest as fulminant or chronic heart failure \[[@B2]\]. The inflammation is presumed to most commonly start as an infectious process, although autoimmune and idiopathic forms also occur \[[@B1]\]. Bacterial or protozoal myocarditis is rare in the developed world, with the vast majority of cases being viral \[[@B3]\]. Although most viruses that are pathogenic for humans can cause myocarditis, common etiologic agents are enteroviruses and adenoviruses \[[@B4]\]. It remains unclear if the primary problem in severe disease is ongoing damage from a virus, a post-infectious inflammatory reaction, or a combination of both factors. If ongoing infection is the primary problem, IVGG could be efficacious if it contains antibodies to the microbe. IVGG also has anti-inflammatory properties that prevent the formation of and downregulate cytokines \[[@B5]\] so could be efficacious even if the primary problem is a post-infectious inflammatory reaction. The objective of this study was to review the evidence evaluating the use of IVGG in the treatment of patients with acute myocarditis. A secondary objective was to determine if there is an identifiable group of acute myocarditis patients (based on age, duration of symptoms, acuity of onset of symptoms, cardiac function at presentation, virologic results, or the presence or absence of histologic evidence of acute myocarditis on cardiac biopsy in patients where a biopsy was performed) who would benefit from IVGG. The review of randomized controlled trials has been registered with the Cochrane Collaboration and regular updates will be available in the Cochrane Library. Methods ======= Searching --------- We systematically searched the following electronic databases: Cochrane Central Register of Controlled Trials (4^th^quarter, 2004), MEDLINE (1966 to January Week 4, 2005), EMBASE (1988 to Week 6, 2005), CINAHL (1982 to January Week 3, 2005), Web of Science (1975 to February 5, 2005), and several trial registries in addition to the Cochrane Heart Group\'s Trial Registry \[[@B6]-[@B11]\]. A comprehensive search strategy was developed by a medical research librarian based on the following terms: immunoglobulin, gammaglobulin, ivig, immunoglobulin, igg, immune, serum, myocarditis, cardiomyopathy, myocardiopathy, carditis, heart, inflammation. The complete search strategy is available from the authors on request. We also contacted the primary author of the RCT, and reviewed the reference lists of all selected articles. In addition, we handsearched proceedings from the following meetings for RCTs: American Heart Association (1999--2004), American College of Cardiology (1998--2005), European Society of Cardiology (1998--2004), and International Heart and Lung Transplantation Society (1998--2004). The search was not limited by language or publication status. Study selection --------------- Two reviewers independently screened the searches for potentially relevant studies. The full manuscript of all potentially relevant studies was retrieved and each study was assessed independently by two reviewers for inclusion using predetermined eligibility criteria. Primary studies of any design were eligible for inclusion. Inclusion criteria included reports of patients given IVGG for possible viral myocarditis within six months of onset of symptoms. Because of the poor sensitivity of cardiac biopsy as a diagnostic tool for acute myocarditis, a histologic diagnosis was not required. Exclusion criteria included reports of patients with evidence of ischemic heart disease, rheumatic heart disease, or any other non-viral etiology, onset of myocarditis less than six months postpartum (as the pathogenesis of postpartum myocarditis may differ from that of presumed viral myocarditis), or receipt of immunosuppressive therapy (unless this was not possible as they were reported as part of a case series). Reports of use of high-dose immunoglobulin directed at specific organisms were excluded as efficacy may be influenced by the formulation available and the accuracy of the etiologic diagnosis. Assessment of methodological quality ------------------------------------ Study quality was assessed independently by two reviewers and discrepancies were resolved by consensus. RCTs were evaluated using the previously validated Jadad 5-point score to assess randomization, double blinding, and losses to follow-up \[[@B12]\]. Allocation concealment was assessed as adequate, inadequate, or unclear \[[@B13]\]. Non-RCTs were graded according to the levels of evidence presented in Table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Levels of evidence for studies of therapeutic interventions from the Oxford Centre for Evidence-based Medicine (website title <http://www.cebm.net/levels_of_evidence.asp>) \[14\] ::: Level Study design ------- -------------- -------------------------------------------------------------------------------------------------------------------- 1 A Systematic review (with homogeneity) of randomized controlled trials B Individual randomized controlled trial (with narrow confidence interval) C All or none 2 A Systematic review (with homogeneity) of cohort studies B Individual cohort study (including low quality randomized controlled trial, e.g., \< 80% follow-up) C \"Outcomes\" research; ecological studies 3 A Systematic review (with homogeneity) of case-control studies B Individual case-control study 4 Case series (and poor quality cohort and case-control studies) 5 Expert opinion without explicit critical appraisal, or based on physiology, bench research or \"first principles\" ::: Data extraction --------------- Data were extracted by one reviewer using a standard form and checked for accuracy and completeness by a second. The primary outcome was rate of survival without a transplant or requirement for placement of a left ventricular assist device. Secondary outcomes included change in echocardiographic measures of cardiac function, duration of hospitalization, and improvement in functional symptoms (determined by increased exercise tolerance as measured by any objective test and New York Heart Association Functional Capacity) when available. Data were collected on complications and adverse events. Data analysis ------------- Data from studies that were not RCTs were reported but not analyzed further. We planned to combine data from RCTs, to do sub-group analyses, and to test for publication bias, however these were not possible as only one RCT was identified. For the RCT, dichotomous data on efficacy (e.g. event-free survival) were expressed as an odds ratio (OR) with 95% confidence interval. Dichotomous data on adverse events were reported as a risk difference and number needed to harm with 95% confidence intervals. Continuous data (change in left ventricular ejection fraction and peak oxygen consumption) were converted to the mean difference with 95% confidence intervals. For peak oxygen consumption, we assumed a correlation of 0.5 and used the methods of Follmann \[[@B15]\] to calculate the standard deviations of the change from baseline estimates. Since the number per group was not given for this variable, an estimate of the sample size in each group had to be estimated by pro-rating the original group sample sizes to the new total sample size. Results ======= Figure [1](#F1){ref-type="fig"} presents a flow diagram of studies considered for inclusion in the review. Only one RCT evaluating IVGG for acute myocarditis has been reported to date \[[@B16]\]. The trial was of moderate methodological quality according to the Jadad scale (3 out of maximum 5 points, where 5 indicates highest quality); allocation concealment was unclear. The study enrolled 62 adults (mean age 43.0 +/- 12.3 years; 37 men), of which only ten had cellular inflammation on endomyocardial biopsy (four fulfilled the Dallas criteria for acute myocarditis, and three for borderline myocarditis). The authors reported that the given sample size would provide 80% power to detect a difference between groups of ≥ 8% in ejection fraction (EF) change scores. Patients were randomized to receive either 2 g/kg IVGG or an equivalent volume of 0.1% albumin in a blinded fashion. The incidence of death or requirement for cardiac transplant or placement of a left ventricular assist device was low in both groups. Event-free survival was not significantly different but favored the control group (OR 0.52, 95% CI 0.12, 2.30). Follow-up at 6 and 12 months showed no significant difference in improvement in left ventricular ejection fraction for cases and controls (mean difference 0.00, 95% CI -0.07, 0.07 at six months; 0.01, 95% CI -0.06, 0.08 at 12 months). Functional capacity as assessed by peak oxygen consumption was not significantly different in the two groups at 12 months (mean difference -0.80, 95% CI -4.57, 2.97). Infusion-related side effects occurred significantly more often in the treated group (RD 0.33, 95% CI 0.17, 0.50; number needed to harm = 3, 95% CI 2, 6), but all appeared to be mild. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow diagram of studies considered for inclusion in the review ::: ![](1471-2261-5-12-1) ::: Sixteen additional studies of various designs met the inclusion criteria \[see [Additional file 1](#S1){ref-type="supplementary-material"}\]. These included a retrospective case series \[[@B17]\], an uncontrolled trial \[[@B18]\], two case series \[[@B5],[@B19]\], and twelve case reports \[[@B20]-[@B31]\]. In a retrospective cohort study, Drucker compared children who were treated with IVGG (n = 21) to historic controls (n = 26); 65% of the sample was less than two years of age \[[@B17]\]. The treatment group showed improved cardiac function and a trend (though not statistically significant) towards improved survival. McNamara conducted an uncontrolled trial involving 10 adults (mean age 35.8 +/- 15 years) \[[@B18]\]. One patient with refractory arrhythmias and severe heart failure died during the infusion; the other nine patients showed an impressive improvement in cardiac function following administration of IVGG, with a significant increase in left ventricular ejection fraction at 12-month follow-up (p = 0.008). In ten case reports and both case series, improvement in cardiac function was described after use of IVGG in a total of 21 cases, although improvement was considered to be minimal in one case \[[@B26]\]. In the other two case reports, there was no benefit of IVGG in a patient presenting with hand, foot and mouth disease and acute myocarditis from coxsackievirus A16 \[[@B27]\] and in a patient with fulminant myocarditis of unclear etiology \[[@B28]\]. Hemolytic anemia was attributed to use of IVGG in one case report \[[@B26]\]. We could find no ongoing trials of the use of IVGG in acute myocarditis, although the European Study of Epidemiology and Treatment of Cardiac Inflammatory Diseases (ESETCID) is evaluating hyperimmunoglobulin for patients with cytomegalovirus-positive myocarditis \[[@B32]\]. Discussion ========== The small number of studies evaluating IVGG in the treatment of acute myocarditis reflects the immature state of this body of literature. We identified only one RCT (Level 1 evidence), which involved 62 adult patients with idiopathic cardiomyopathy (of which only 10 had histologic evidence of myocarditis) and showed no apparent benefit \[[@B9]\]. This is in contrast to an uncontrolled trial and numerous observational studies (Level 4 evidence), most of which suggested potential benefit. This indicates a possible bias towards publication of case reports or case series with positive results. The validity of evidence was generally poor, with the majority of included studies ranking low in terms of level of evidence. The most important threat to validity for fifteen of the seventeen included studies was the lack of controls, which could result in an overestimate of the benefit of IVGG. Spontaneous improvement in cardiac function is common with acute myocarditis and can be rapid or gradual, so it is possible that the improvement noted in these cases was part of the natural history of the disease. The retrospective cohort study demonstrated a greater improvement in cardiac function in patients given IVGG compared to historic controls \[[@B17]\]; this type of study is more susceptible to bias because of the inability to fully control for all potential confounders, such as other changes in patient management over the eight-year study period. Acute myocarditis is a relatively non-specific entity, as the diagnosis is often clinical. Laboratory confirmation consists of microbiologic, histologic, and immunohistochemical methods. With regard to microbiologic confirmation, it is rare to isolate the etiologic agent from a myocardial biopsy (probably because the biopsy is usually done too late in the course of the illness), and identification of organisms by molecular techniques is in its infancy \[[@B4]\]. However, with the development of new effective antivirals, there should be increased efforts towards making a virologic diagnosis early in the course of myocarditis as it is possible that antivirals would improve the prognosis. Furthermore, there are case reports of hyperimmunoglobulin showing an apparent effect in cases of varicella \[[@B33]\], cytomegalovirus \[[@B34]\], and parvovirus \[[@B35]\] myocarditis. With regard to histologic confirmation, the Dallas criteria (lymphocyte infiltration with myocyte necrosis) \[[@B36]\] and more recently the World Heart Federation criteria (\>14 leukocytes/mm^2^with necrosis or degeneration) \[[@B37]\] have been used for histologic diagnosis, but the sensitivity and specificity of these criteria are not known. There are no standardized values for immunohistochemical markers \[[@B38]\]. Our aim in this study was to analyze reports of use of IVGG for presumed viral myocarditis, but it is possible that many of the patients in the reports did not have viral myocarditis (only 10 of the 62 patients in the RTC had cellular inflammation on endomyocardial biopsy). Once our ability to accurately diagnose viral myocarditis improves, it may be possible to identify a subset of patients who will respond to IVGG. This might be patients whose disease was precipitated by a specific virus, or patients who are treated with IVGG early in the course of their illness when they have ongoing viral replication in the myocardium. Perhaps pediatric patients are more likely to respond. Children are thought to be more likely to present in the acute inflammatory stage of illness \[[@B16]\], and may have a worse prognosis than do adults when they present with fulminant myocarditis \[[@B39],[@B40]\] In conclusion, the value of IVGG in patients with acute myocarditis is obscured by the poor quality of evidence. We were not able to identify a subgroup of patients who appear to be more likely to respond to IVGG. A large RCT is required to evaluate the efficacy of IVGG for acute myocarditis with emphasis on the etiology of the myocarditis. Until there are RCTs demonstrating benefit, use of IVGG for acute myocarditis should not be part of routine practice. Moreover, there is a great need for further studies of the pathophysiology of acute myocarditis, which would allow for a better understanding of the etiology and the natural history of the disease. This might allow for improved diagnostic criteria, which would make it much easier to design studies of treatment options. This may also assist in identifying sub-groups of patients where IVGG or other therapies have a greater potential to confer clinical benefit. List of abbreviations ===================== IVGG -- intravenous gamma globulin RCT -- Randomized controlled trial Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= The idea for the study originated with JLR. The protocol was written by LH with help from JLR and EC. The literature review was completed by EC, and LH and JLR reviewed the papers and derived the data. Statistical analysis was done by BV. The manuscript was written by JLR and LH with input from TPK, EC and BV. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/12/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Observational studies and nonrandomized trials examining the use of IVGG in acute myocarditis. This table summarizes all studies reporting the use of IVGG for acute myocarditis. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We thank Philip Berry and Marilyn Josefsson, from the Alberta Research Centre for Child Health Evidence, for administrative support and collection of articles.

PubMed Central

2024-06-05T03:55:58.357326

2005-6-2

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173096/", "journal": "BMC Cardiovasc Disord. 2005 Jun 2; 5:12", "authors": [ { "first": "Joan L", "last": "Robinson" }, { "first": "Lisa", "last": "Hartling" }, { "first": "Ellen", "last": "Crumley" }, { "first": "Ben", "last": "Vandermeer" }, { "first": "Terry P", "last": "Klassen" } ] }

PMC1173097

Background ========== Hypertension is among the most prevalent chronic diseases in the United States \[[@B1]\]. Despite the availability of effective medications and well-published guidelines for the treatment of hypertension \[[@B1]-[@B4]\], the majority (approximately 75%) of hypertension in the United States remains poorly controlled \[[@B5]\]. Hypertension is particularly burdensome among racial/ethnic minority groups and hypertension-related cardiovascular disease has been shown to be the greatest contributor to previously documented racial differences in mortality \[[@B5]-[@B12]\]. Although higher rates of hypertension control and a reduction in racial differences in outcomes from hypertension may be obtained by increasing providers\' aggressiveness in intensifying therapy when indicated \[[@B2],[@B12]\], several studies have demonstrated that providers often allow their patients to remain poorly controlled \[[@B12],[@B13]\]. We previously examined the association of patients\' race/ethnicity with processes of hypertension care \[[@B13]\]. We found that, among a cohort of hypertension patients with repeatedly elevated blood pressures, Hispanics were significantly less likely to have therapy intensified and were more likely to have uncontrolled blood pressure (BP) than were other racial and ethnic groups. In an effort to identify potential targets for interventions to improve hypertension care among our patients, we further examined which patient and provider characteristics may explain racial differences in rates of therapy intensification. Methods ======= Study sample and procedures --------------------------- To determine which patient-centered characteristics are associated with providers intensifying drug therapy for hypertension and whether provider experience is related to differences in intensification of therapy, we studied a random subset of 1,205 patients who had a minimum of two hypertension-related outpatient visits to one of twelve general medicine clinics in community health centers and community practices affiliated with a large urban academic medical center from July 1, 2001 through June 30, 2002 (totaling 3,257 visits). To determine hypertension-related visits, we reviewed the electronic medical record (EMR) for all clinic visits with a primary or secondary diagnostic code of hypertension (HTN) (ICD-9 401- 401.9, 405- 405.99). The study protocol was approved by the Human Studies Committee at the Brigham and Women\'s Hospital. Medical record and administrative data -------------------------------------- We hired three abstractors who were trained by the primary investigator to review the EMR of each patient in our study sample, each abstractor reviewed records for approximately 400 individual patients. From each hypertension-related visit note in the EMR they collected the following data: patient race/ethnicity, name of the patient\'s primary provider, patient age at time of initial study period visit, sex, primary insurer at time of initial study period visit, presence of comorbid disease (diabetes, congestive heart failure (CHF), coronary artery disease (CAD), or renal failure) listed on the patient problem list, BP control (defined using cut point of \<130/85 for diabetic or renal failure patients and \<140/90 for all others), and any changes made to antihypertensive drug therapy during visit (decrease or discontinuation of drug, change to another class of drug, or increase of drug dose or addition of new drug). The EMR of each patient contains racial/ethnic data obtained via self-report the first time each patient registers in his/her clinic that then is classified by the data entry clerk as Hispanic ethnicity or not and race is then classified as Black, White, other, or unknown. Hence, it is impossible to know the country of origin of our Hispanic participants and we could not determine the exact racial/ethnic mix of patients classified as \"other\" or \"unknown.\" For these reasons, we excluded patients who\'s EMR listed their racial/ethnic classification as \"other\" (N = 25) or \"unknown\" (N = 67) and we limited our analyses to White patients (who identify themselves as non-Hispanic), Black patients (who identify themselves as non-Hispanic), and self-identified Hispanic patients. From the medical center\'s administrative database, we abstracted each provider\'s level of experience (intern, resident, or attending). Patient zip code was also obtained and linked to 2000 U.S. Census data to obtain the median annual household income in each patient\'s zip code. Intensity of therapy -------------------- Our methods for identifying intensified cases among our cohort have been previously described \[[@B13]\]. Because we were examining the quality of care delivered to hypertension patients prior to July of 2002, approximately one year before the release of the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation and Treatment of High Blood Pressure (JNC VII) \[[@B14]\], we used JNC VI definitions for blood pressure control in order to examine whether therapy was appropriately intensified. We classified each visit into two categories, intensified visits (an increase in intensity of drug therapy) versus non-intensified visits (a decrease or no change in intensity of drug therapy) according to previously published definitions of changes in medications \[[@B2]\]. We developed an algorithm to determine whether a patient received at least one increase in drug therapy (an increase in drug dose or addition of new medication) in response to a repeatedly elevated BP during the study period (Figure [1](#F1){ref-type="fig"}). Each patient with fewer than two visits with an elevated BP (N = 356) was excluded from the algorithm. Each patient with an uncontrolled BP at more than one visit was identified as either an intensified case (at least one drug increase) or a non-intensified case (no drug increases). Each reviewer examined a subset of 30 records; we then tested for inter-rater reliability and found excellent agreement among reviewers (kappa = 0.90). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Flow diagram of algorithm for determining an \"intensified case.\" Inter-rater reliability was high (kappa 0.90). We identified a total of 782 cases as either an \"intensified case (N = 600) or a \"non-intensified case\" (N = 157). ::: ![](1471-2261-5-16-1) ::: Our intensification algorithm is potentially limited because it does not account for differences in the number of times medication is increased in comparison to the number of uncontrolled BP visits for each patient. To address this issue, we created a secondary measure of intensification by calculating the proportion of times each patient had their therapy increased when their BP was uncontrolled (0 representing no increases and 1 representing an increase in therapy with every uncontrolled visit). We then determined whether the two intensification measures were correlated in order to further validate our algorithm and found that the two measures were highly correlated (coefficient 0.71, P \<0.001). Because our algorithm-derived method has the additional advantage of using more than a single visit\'s BP in determining whether therapy intensification is indicated, we used this measure of intensification for all subsequent analyses. Data analysis ------------- We compared patients\' demographic, clinical, and provider\'s characteristics by patient race/ethnicity (Table [1](#T1){ref-type="table"}) using chi-squared tests for categorical and Student\'s t-test for continuous variables. We also estimated the association of being an intensified case with patients\' demographic, clinical, and provider\'s characteristics (Table [2](#T2){ref-type="table"}). We report two-tailed P values with statistical significance set at P≤0.05 for all analyses. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic and clinical characteristics of population by race and ethnicity ::: **Variable** **White (N = 304)** **Black (N = 309)** **Hispanic (N = 144)** **P value\*** ------------------------------------ --------------------- --------------------- ------------------------ --------------- Men (%) 27.3 24.3 35.4 0.05 Aged \<65 years (%) 54.6 65.1 66.7 0.009 Annual household income^†^: \<0.001 = \$35,500 11.9 37.8 29.4 \$35,501--\$43,140 11.9 40.8 21.0 \$43,141--\$55,365 24.7 15.5 42.7 \> \$55,365 51.6 5.9 7.0 Insurance: \<0.001 Private/Medicare (%) 87.7 69.1 40.3 Medicaid (%) 4.3 17.6 28.5 Free care/Self-pay (%) 8.0 13.4 31.3 Mean number of hypertension visits 2.8 3.0 2.8 0.06 Diabetic (%) 14.5 27.5 27.8 \<0.001 Coronary artery disease (%) 5.3 4.2 2.1 0.02 Provider experience (%)^‡^: \<0.001 Intern 2.1 9.3 10.1 Resident 4.8 23.0 10.1 Attending 93.2 67.8 79.8 \*Using chi-square tests for categorical variables and ANOVA for mean number of hypertension visits. ^†^N = 734 for those patients with zip code available. ^‡^N = 691 for those patients with provider information available. ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Demographic and clinical characteristics of population by intensification of therapy ::: **Variable** **Non-Intensified Case (N = 157)** **Intensified Case (N = 600)** **P value\*** ------------------------------------ ------------------------------------ -------------------------------- --------------- Patient race/ethnicity (%): 0.03 White 19.7 80.3 Black 18.1 81.9 Hispanic 28.5 71.5 Men (%) 29.1 27.7 0.74 Age: 0.10 \<65 years 18.8 81.2 ≥65 years 23.8 76.2 Annual household income^†^: 0.93 ≤\$35,500 20.4 79.6 \$35,501--\$43,140 19.7 80.3 \$43,141--\$55,365 22.4 77.6 \> \$55,365 20.5 79.5 Insurance: 0.15 Private/Medicare (%) 19.5 80.5 Medicaid (%) 27.8 72.2 Free care/Self-pay (%) 20.0 80.0 Mean number of hypertension visits 2.6 3.0 \<0.001 Diabetic (%): 0.01 Yes 27.8 72.2 No 18.7 81.3 Coronary artery disease (%): 0.24 Yes 12.5 87.5 No 21.1 78.9 Provider training level (%)^‡^: 0.91 Intern 22.7 77.3 Resident 21.4 78.6 Attending 20.3 79.7 \*Using chi-square tests for categorical variables and Student\'s *t*-test to compare mean hypertension visits. ^†^N = 734 for those patients with zip code available. ^‡^N = 691 for those patients with provider information available. ::: Using logistic regression, we assessed whether race/ethnicity was associated with intensification of therapy after adjusting for all measured confounders. Data were available on every variable for 685 of the 757 patients (90.5%) for multivariable analyses. We report adjusted odds ratios with 95% confidence intervals for the intensified cases. In secondary analyses, we included interaction terms for patient race/ethnicity and provider experience level to determine whether racial/ethnic differences in intensification differed by provider experience level. All non-significant interaction terms were removed from the final model. All models were estimated using SUDAAN statistical software to adjust for within clinic correlation of visits \[[@B15]\]. Results ======= Patient, clinical, and provider characteristics ----------------------------------------------- Of the 757 patients, 304 (40%) were White, 309 (41%) were Black, and 144 (19%) were Hispanic (Table [1](#T1){ref-type="table"}). The majority of patients in our cohort were women (72%) and most were either privately insured or had Medicare (71%). There were 169 (22%) patients in our cohort with diabetes. We were able to determine the primary provider for 691 patients, of who 44 (6.4%) received care from an intern, 89 (12.9%) received care from a resident, and 558 (80.8%) received care from an attending. Demographic and clinical differences between patients\' racial/ethnic groups are presented in Table [1](#T1){ref-type="table"}. Intensification of therapy -------------------------- Of the 757 patients with multiple visits whose BP was uncontrolled at two or more visits, there were 600 (79.3%) who had their medications intensified. Hispanic patients were significantly less likely to have their medications intensified than White and Black patients (Table [2](#T2){ref-type="table"}). Diabetic patients and those with fewer visits during the study period were also less likely to have their medications intensified. After adjustment for age, insurance status, number of hypertension-related visits, diabetes, and physician experience, there were no remaining racial/ethnic differences in rates of intensification (Table [3](#T3){ref-type="table"}). Compared to patients aged = 65 years, younger patients were more likely to have their therapy intensified, and the odds of having therapy intensified increased with each visit. There were no significant interactions between race/ethnicity and provider experience in the multivariable model. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Adjusted odds ratios\* \[95% C.I.\] of being an intensified case ::: **Variable** **Odds Ratios of being Intensified Case** **95 % Confidence interval** ------------------------------------- ------------------------------------------- ------------------------------ Black^†^ 1.09 \[0.71--1.67\] Hispanic^†^ 0.78 \[0.58--1.06\] Age \< 65 years^‡^ 1.61 \[1.38--1.88\] Medicaid^§^ 0.76 \[0.57--1.02\] Free care or self pay^§^ 1.22 \[0.73--2.03\] Number of hypertension visits^\|\|^ 1.37 \[1.17--1.61\] Diabetes\*\* 0.61 \[0.34--1.08\] Intern provider^††^ 0.86 \[0.39--1.88\] Resident provider^††^ 0.89 \[0.55--1.46\] \* Adjusted for patient race/ethnicity, age, insurance status, number of hypertension-related visits, diabetes, and physician experience. ^†^Whites use as reference group ^‡^Aged ≥ 65 years used as reference group ^§^Privately insured or Medicare used as reference group ^\|\|^Represents increase in odds with each additional visit \*\*No diabetes used as reference group ^††^Attending provider used as reference group ::: Discussion ========== In our prior work, we found that Hispanic patients were more likely to receive the JNC recommended antihypertensive drug class then Whites, however, Hispanics were also significantly less likely than Whites to have their therapy appropriately intensified in response to an uncontrolled BP \[[@B13]\]. Furthermore, we found appropriate intensification of anti-hypertensive therapy was associated with subsequent BP control for all racial/ethnic groups, suggesting that the poorer rates of BP control among Hispanics in our prior study may have been due to significantly lower rates of antihypertensive medication intensification \[[@B13]\]. In this study, we found that these previously noted racial/ethnic disparities in rates of antihypertensive therapy intensification may be due to differences in visit patterns among patients and in physicians\' aggressiveness in managing BP in diabetic patients, suggesting that racial/ethnic differences in disease severity are likely determinants of unequal treatment of uncontrolled hypertension. There is substantial literature, that suggests that racial/ethnic minority groups are less likely to have their antihypertensive therapy appropriately intensified \[[@B16]-[@B19]\]. However, these studies were limited because the investigators were not able to assess practice patterns such as the frequency with which individual anti-hypertensive drugs were intensified in response to uncontrolled BP. We found that Hispanic patients in our cohort were also less likely to have their anti-hypertensive medications intensified at least once in response to repeatedly uncontrolled BP than were other racial/ethnic groups and our findings also expand beyond documenting racial/ethnic disparities in aggressiveness of therapy by determining the roles racial/ethnic differences in clinic utilization among patients and racial/ethnic differences in the prevalence of diabetes play in confounding differences in providers\' aggressiveness in treating hypertension. An important national priority in health care is the elimination of racial/ethnic disparities in healthcare; however, a better understanding of the determinants of disparities is needed to address this issue \[[@B20],[@B21]\]. Our finding that higher rates of diabetes among Hispanics in our cohort play a major role in the insufficient management of their uncontrolled BP is of particular concern given that Hispanics suffer a disproportionately larger burden from hypertension and diabetes compared to Whites; and that Hispanics are at a higher risk of having hypertension and diabetes, are less likely to be aware that they are hypertensive, are more likely to have target organ damage, and have significantly higher age-adjusted diabetes and hypertension-related mortality compared to whites in the U.S \[[@B5],[@B6],[@B22],[@B23]\]. Our findings are supported by several studies that have documented lower rates of BP control among diabetic patients \[[@B24],[@B25]\], and others showing that providers often do not attain adequate BP control for patients, even after multiple opportunities to do so \[[@B2],[@B24]\]. Our study has several limitations. We were unable to collect measures of patient adherence to prescribed therapy from the EMR. Providers may not want to intensify therapy at the same rate for patients they know are less compliant with therapy, although it is hard to identify these patients. We were also unable to determine English proficiency of each patient from the medical records. When a language barrier exists providers may be less likely to intensify therapy. We examined disparities in quality of hypertension care, as measured by intensification of therapy, among a cohort of patients and physicians during 2001--2002 using the established guidelines available during that time period, the JNC VI, as the reference for our analyses. In comparison to JNC VI, the JNC VII guidelines recommend much more aggressive management of hypertension both in terms of the accepted level of BP control (130/80 for diabetic and renal failure patients and 140/90 for all others) and the recommendations to providers to intensify antihypertensive therapy more rapidly when BP is uncontrolled \[[@B14]\]. However, since the JNC VII guidelines were not available until 2003, it would have been impossible for providers in our study sample to have been fully aware of them and it would have been misleading to assess the quality of hypertension care among our cohort using quality guidelines that were not in existence during that time. By using the less aggressive guidelines that were in existence during the study period, we may have underestimated the gap in rates of intensification between racial/ethnic groups and between patients with diabetes and those without. Lastly, we examined patients who received their care at primary care practices affiliated with a single large urban teaching hospital, and although there was substantial socioeconomic diversity in our sample our results may not be generalizable to smaller, rural, or non-teaching hospitals. Conclusion ========== We found significant racial/ethnic differences in intensification of drug therapy, and that these differences were largely accounted for by differences in frequency of clinic visits and in the prevalence of diabetes. Future interventions should focus on increasing physician awareness of the need to intensify drug therapy more, particulary among Hispanic patients, and on encouraging providers to treat hypertension more intensively in diabetic patients. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= LH conceived of the study, and participated in its design and coordination and helped to draft the manuscript. SS participated in the design of the study, performed the statistical analysis, and helped to draft the manuscript. DWB and JZA participated in the design of the study, and helped to draft the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/16/prepub> Acknowledgements ================ The authors would like to thank Deborah H. Williams for statistical programming assistance. This study was supported by a grant from the Agency of Healthcare Research and Quality (\#3U18 HS11046). Dr. Hicks was supported by the Robert Wood Johnson Foundation\'s Harold Amos Medical Faculty Development Program (\#043486).

PubMed Central

2024-06-05T03:55:58.358889

2005-6-22

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173097/", "journal": "BMC Cardiovasc Disord. 2005 Jun 22; 5:16", "authors": [ { "first": "LeRoi S", "last": "Hicks" }, { "first": "Shimon", "last": "Shaykevich" }, { "first": "David W", "last": "Bates" }, { "first": "John Z", "last": "Ayanian" } ] }

PMC1173098

Background ========== Hypertension is a continuous, independent, yet modifiable risk factor for cardiovascular, cerebrovascular and renal disease. It has been estimated that 62% of cerebrovascular disease and 49% of ischemic heart disease can be attributed to suboptimal blood pressure(BP) control \[[@B1]\]. Since 24 hour ambulatory blood pressure monitoring (ABPM) is currently recommended as the measurement of choice in difficult cases (uncertainty of diagnoses, fluctuating office visits, unresponsiveness to treatment, white coat affect(WCE)) \[[@B1]-[@B5]\] it is, in effect, the \'final arbiter\' or \'gold standard\' for the diagnosis of hypertension and the assessment of whether target has been achieved. The generally accepted blood pressure target on ABPM is a daytime average pressure of \< 135/85 mmHG \[[@B2]-[@B4]\]; the target blood pressure when measured in the doctors office is \< 140/90 mmHG. Increasing evidence suggests that ABPM more closely predicts target organ damage than does clinic measurement. The Ohsama study found that cardiovascular mortality, but not all cause mortality, was more closely predicted by the daytime systolic blood pressure than clinic pressures \[[@B6]\]. ABPM also better predicted cardiovascular events such as MI, CHF, stroke and TIA \[[@B4],[@B5]\] as well as other target organ damage such as ventricular hypertrophy, proteinuria, plasma creatinine, and stroke \[[@B5],[@B7]-[@B14]\]. Many patients have clinically significant WCE, making clinic measurements an unreliable indicator of true blood pressure control \[[@B11]\]. Future guidelines may suggest ABPM as routine in the diagnosis and monitoring of hypertension. This would create difficulties as this test is expensive and often difficult to obtain. Recently, automatic blood pressure measuring devices have been used either at home or in the clinic setting \[[@B16]\]. It has been found that the measurements taken at home, or while not in the presence of clinic staff, have a better correlation with daytime ABPM than do nurse or physician measurements \[[@B15]-[@B17]\]. Home or clinic self measurement is also preferred by patients over 24 hour monitoring \[[@B18]\]. The purpose of this study is to determine the possible clinical utility of the BpTRU automated blood pressure monitor in the diagnosis and monitoring of hypertension in the primary care clinic setting. Specifically, we set out i) to determine how BpTRU measurements related to ABPM measurements and ii) to determine the level of BpTRU measurement that best predicted a mean daytime blood pressure of \<135 mmHG systolic and \< 85 mmHG diastolic. The BpTRU monitor has been developed by VSM MedTech Ltd of Vancouver Canada specifically for the clinician\'s office. Methods ======= The data used in this study is a subset of the data collected for two ongoing RCTs being conducted at the Centre for Studies in Primary Care at Queen\'s University, Kingston, Ontario, Canada -- the Home Monitoring of Blood Pressure Study (ISRCTN25105161) and the Intensive Scheduled Management of Hypertension Study (ISRCTN05874865) which are funded by the Heart and Stroke Foundation of Ontario. A group of 481 subjects for which all relevant data were available were used for analyses of correlations. Subjects were recruited from 51 family practices in eastern Ontario. All patients with a diagnosis of hypertension who were being treated with antihypertensive medications were identified in each practice. Each patient\'s chart was reviewed to abstract the blood pressures recorded on the patient\'s chart at the last three office visits where blood pressure was measured. Only one recording was used from any single office visit. If there was more than one recording at a given visit, the last measurement recorded for that visit was used. These visits ranged from several weeks to several months apart, depending on the practice of the physician regarding follow-up of hypertensive patients. If the mean of these three readings, taken at different visits, was ≥140/90, the patient was labeled as \'uncontrolled\' by office measurement, meaning that the patient had not achieved the treatment target (≥130/80 was used for diabetics). The research nurse then contacted these uncontrolled patients and invited them to participate in the study. Subjects were excluded if they were \<18 years of age, pregnant, or had a known secondary cause for their hypertension. BpTRU measurements ------------------ The BpTRU device uses the oscillometric technique used by most ambulatory and home blood pressure measuring devices \[[@B19],[@B20]\]. It is designed to take an initial reading while the clinician is present, and then with the patient alone in the room, proceeds to take 5 more measurements at intervals of 1--5 minutes and then provides an average of these five readings. The specific steps we used were: i) the subjects were seated for at least 5 minutes, ii) the BpTRU cuff was applied to the non-dominant arm by the research nurse, iii) the initial BpTRU blood pressure reading was taken and recorded, iv) the staff then left the room while the BpTRU device took a minimum further 5 readings at intervals of either one minute or two minutes, v) these five readings were averaged by the device and this average was recorded. The BpTRU device has passed the standards of the British Hypertension Society and the Association of Advancement of Medical Instrumentation \[[@B20],[@B21]\]. Ambulatory monitoring --------------------- All patients then had ABPM monitoring using oscillometric A&D Model TM2430 equipment (A&D Medical, Milpitas, California, USA). The ABPM equipment was applied at the same visit as the other research measurements. The cuff was fixed to the non-dominant arm and the device was set to obtain automatic readings every 15 minutes during the day (0600--2200) and every 30 minutes at night (2200--0600). This monitoring took place on working days and subjects were instructed to behave and work as usual. The A&D Model TM2430 has been clinically validated according to the British Hypertension Society protocols \[[@B22]\]. Results ======= Of the 481 subjects, 210 (43.6 %) were male and 271 (56.3 %) female. The average age was 64.9 (SD 11.59) with a range of 33 to 92 and average BMI was 30.6 (SD 5.22). All but 2 subjects were on antihypertensive medications. In essence, this was a population of treated known hypertensive patients. Comparison of group mean blood pressures ---------------------------------------- Table [1](#T1){ref-type="table"} shows the mean and standard deviations(SD) of systolic and diastolic BP as measured in the family physicians\' offices (the average of the blood pressure measurements recorded on the patient chart at the last three visits), the initial BpTRU reading, the average BpTRU reading, and the 24 hour daytime average. There was no statistical difference in group mean between clinic measurement and initial BpTRU readings, and no statistical difference between average BpTRU readings and ABPM daytime average. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Mean systolic and diastolic BPs ::: ------------------------------------------------------------------------- ----------------------------------- Average of the blood pressures measured at the last three office visits 150.8 (SD10.26) / 82.9 (SD 8.44) BpTRU initial reading 150.0 (SD21.33) / 83.3 (SD12.00) BpTRU Average 140.0 (SD17.71) / 79.8 (SD 10.46) 24 hour daytime average 141.5 (SD 13.25) / 79.7 (SD 7.79) ------------------------------------------------------------------------- ----------------------------------- ::: Comparison of BP target achievement ----------------------------------- A large proportion of those subjects who had uncontrolled hypertension according to the mean of the three last clinic BPs had lower blood pressure on either or both of BpTRU or 24 hour monitor. In 393 (81.7%) of the 481 subjects the systolic blood pressure was less with the BpTRU than with the clinic readings and in 312 (64.8%) of the 481 subjects the systolic blood pressure was less with daytime ABPM than with the clinic readings. Diastolic pressure similarly decreased in 332 (69.0%) using the BpTRU average and 292 (60.7%) with daytime ABPM. In all, 250 of the 470 (53.2%) were found by BpTRU monitor to have achieved the accepted clinic measurement target of \<140/90 mmHG. ABPM revealed that 162 subjects (33.4%) were actually normotensive with a daytime average of \<135/85 mmHG. Correlations ------------ Pearson correlations (Table [2](#T2){ref-type="table"}) were calculated comparing the mean daytime systolic and diastolic ABPM results with the average of the blood pressure measurements recorded on the patient\'s office chart, the first BpTRU measurement taken in the presence of the staff, and with the average of five readings on the BpTRU device. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Pearson Correlation coefficients ::: Mean 24 hr daytime [systolic]{.underline} Mean 24 hr daytime [diastolic]{.underline} ------------------------------------------------------------------------- ------------------------------------------- -------------------------------------------- Mean of the blood pressure measurements at the last three office visits r = 0.145 r = 0.316 1st BpTRU measurement r = 0.473 r = 0.554 Average BpTRU(out of 5) r = 0.571 r = 0.610 ::: Using BpTRU to predict achievement of targets --------------------------------------------- ABPM is accepted as the most reliable way to determine if blood pressure targets are achieved in hypertensive patients, but using ABPM after every medication adjustment is impractical and costly. To determine which level of BpTRU blood pressure best predicted achievement of target (i.e. \<135/85) on ABPM, various prediction estimates (sensitivity, specificity, and predictive values) were calculated for different levels of BpTRU readings (Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Two-By-Two Tables And BpTRU Predictive Characteristics For Achievement Of ABPM Systolic Target ::: Systolic Target Achieved on ABPM (\<135 mmHG mean daytime blood pressure) ---------------------- ----- --------------------------------------------------------------------------- ----- --------------------------------------------- Yes NO \< 140 mmHG on BpTRU Yes 131 137 Sens = 80%; Spec = 55% PPV = 49%; NPV = 84% No 32 170 \< 135 mmHG on BpTRU Yes 111 101 Sens = 68%; Spec = 67% PPV = 52%; NPV = 80% No 52 206 \< 130 mmHG on BpTRU Yes 86 59 Sens = 53%; Spec = 81% PPV = 59%; NPV = 76% No 77 248 \< 125 mmHG on BpTRU Yes 62 31 Sens = 39%; Spec = 90% PPV = 67%; NPV = 73% No 101 276 \< 120 mmHG on BpTRU Yes 42 12 Sens = 26%; Spec = 96% PPV = 78%; NPV = 71% No 121 295 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Two-by-Two Tables and BpTRU predictive characteristics for achievement of ABPM diastolic target ::: Diastolic Target Achieved on ABPM (\<85 mmHG mean daytime blood pressure) --------------------- ----- --------------------------------------------------------------------------- ----- --------------------------------------------- Yes NO \< 90 mmHG on BpTRU Yes 341 58 Sens = 92%; Spec = 43% PPV = 86%; NPV = 61% No 28 43 \< 85 mmHG on BpTRU Yes 299 36 Sens = 81%; Spec = 64% PPV = 89%; NPV = 48% No 70 65 \< 80 mmHG on BpTRU Yes 238 13 Sens = 65%; Spec = 87% PPV = 95%; NPV = 40% No 131 88 \< 75 mmHG on BpTRU Yes 161 5 Sens = 44%; Spec = 95% PPV = 97%; NPV = 32% No 208 96 \< 70 mmHG on BpTRU Yes 82 1 Sens = 22%; Spec = 99% PPV = 99%; NPV = 26% No 287 100 ::: Discussion ========== The measurements of blood pressure done with a provider present, that is the mean of the physician office measurements, and the observed first BpTRU measurements were not significantly different from each other and probably are both affected by the white coat phenomenon. This also suggests that the mean of the last three office blood pressure measurements is a good proxy for the patient\'s current blood pressure, taken with a health care professional present (an observed blood pressure measurement). Both of these measures, that is, the mean of the last three office blood pressures and the first, observed reading from the BpTRU, were significantly higher than multiple measurements taken when a provider is not present, that is the mean of the five BpTRU measurements and the daytime mean on ABPM. The mean of the BpTRU and the daytime mean on ABPM were not significantly different. Recommended BpTRU Thresholds ---------------------------- Our data suggest that the degree of agreement between the BpTRU and 24 hour ABPM is not sufficient for clinicians to use BpTRU alone to determine if BP targets have been achieved. However, since the agreement between the BpTRU and ABPM is so much better than the usual sphygmomanometer-based, observed, clinic measures, it seems reasonable to use the BpTRU to make treatment adjustment decisions to a predetermined BpTRU level and then confirm it with a 24 hour ABPM. If one considers Table [3](#T3){ref-type="table"}, it would seem that a systolic BP level of \<135 mmHG provides the best overall agreement with the 24 hour ABPM. Similarly, for the diastolic two-by-two tables in Table [4](#T4){ref-type="table"}, one would likely choose \<85 mmHG diastolic as the most appropriate BpTRU level for diastolic pressure. Clinicians should consider treating patients to a level of \<135/85 mmHG using the BpTRU or similar automated multiple reading device, and at that point conduct a 24 hour ABPM to confirm that a daytime mean pressure of \<135/85 mmHG has been achieved. If it has not been achieved, the clinician should further treat to a BpTRU level of \<130/80 mmHG or \<125/75 mmHG before re-assessing with 24 hour ABPM. Limitations ----------- Our data are the best currently available to provide some sense of what a BP target level should be when using the BpTRU in office practice. However there are some limitations. The BpTRU and ABPM were carried out in a research setting not a true clinical setting. All the patients had uncontrolled blood pressure according to office BP measurements, so it does not represent a full spectrum; it does not include normotensive patients or patients where the chart records suggest target has been achieved. The data was collected in the course of another study. There is a need for a study that sets out, a priori, to compared BpTRU with ABPM in the full spectrum of patients, and which follows those patients for a longer term to assess clinical outcomes. Conclusion ========== The control of hypertension is vital to decrease cardiovascular morbidity and mortality, however the current recommended BP targets are often not being met. There is also growing evidence that office measured sphygmomanometer-based blood pressures are unreliable and do not predict outcomes as well as ABPM, whereas there is increasing evidence from prospective trials that 24 hour monitoring has prognostic significance. We have shown that the BpTRU has potential to be used in the clinic setting to help overcome the difficulties caused by the WCE without the cost of having to conduct frequent 24 hour ABPMs. Although this study demonstrates that the BpTRU has a sensitivity and specificity that are not ideal when compared to the ABPM device we used, it is superior to usual office measurement and can be used by clinicians as part of their strategy for determining whether BP target has been achieved. We suggest that the BpTRU be used to adjust treatment until patient\'s BpTRU pressure is below 135/85 mmHg and then a 24 hour ABPM be conducted to confirm. ABPM measurement is not practical as a means of monitoring targets after each medication adjustment but in conjunction with the BpTRU can form the basis for clinical decisions that will promote more effective control of hypertension. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= LB abstracted the data from the larger baseline dataset from the randomized controlled trials. She analysed the data, did the initial interpretation and wrote the initial draft of the article. MG designed and is conducting the randomized controlled trials from which the data for this study was drawn. He worked with LB on the analysis and the data interpretation and helped with the substantial re-writing of the original draft of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/5/18/prepub>

PubMed Central

2024-06-05T03:55:58.361194

2005-6-28

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173098/", "journal": "BMC Cardiovasc Disord. 2005 Jun 28; 5:18", "authors": [ { "first": "Linda", "last": "Beckett" }, { "first": "Marshall", "last": "Godwin" } ] }

PMC1173099

Background ========== The quality of health care is determined by two main factors: the quality of the judgements and decisions that determine what actions are to be taken, and the quality with which those actions are executed \[[@B1]\]. Within health care, there are wide variations in clinicians\' judgements on diagnosis and in the management of patients with the same symptoms and diagnosis \[[@B2]\]. These variations are seen within different disciplines, among both experts and novices. In spite of this evidence, clinicians generally believe that the decisions made by their colleagues would be similar to their own, and hence they assume that there is a broad consensus in medical practice \[[@B2]\]. Patients with symptoms from the upper gastrointestinal tract are regularly seen in clinical practice. The symptoms may be either of organic origin -- e.g. ulcer disease, oesophagitis, and malignancies in the oesophagus and the ventricle -- or of functional origin \[[@B3]\]. Dyspepsia is a collective term and includes conditions in both categories. Dyspepsia is common, and the subjective symptoms in either category varybetween patients. During a 3-month period, about 30% of the adult population suffers from dyspepsia \[[@B4]\]. Among these sufferers, only a minority has ulcer diseases (10%) \[[@B5]\] or reflux (12%) \[[@B6]\]. According to the literature, certain management strategies are recommended for some of these conditions, for example reflux \[[@B7]\]. For other conditions, for example functional dyspepsia, the evidence is more ambiguous \[[@B8]-[@B11]\]. One may expect a wider variation in the latter than in the former treatment strategies. The aim of this study was to describe, using a qualitative approach, the variation in the management strategies and treatments suggested by clinicians in three different disciplines for patients with symptoms from the upper gastrointestinal tract. According to this aim, the hypotheses that guided our design were: 1\. Management strategies and treatments suggested by different clinicians vary considerably, even when optimal treatment is clear cut, as documented by evidence in the literature. 2\. Clinicians believe that the management strategies of their colleagues will be similar to their own. Methods ======= Informants ---------- Informants were selected to represent clinicians who regularly encountered patients with dyspeptic symptoms in their daily practice and who thus were expected to have a treatment policy for these patients. We therefore invited Swedish clinicians who were specialists in medical gastroenterology, surgery, and general practice to participate in this study. All clinicians gave their consent to be interviewed. They also suggested other clinicians who might be interested in participating in the study. Altogether, 27 clinicians participated, nine from each of the three specialities. This number is the number commonly used in studies that use judgement analysis to ensure a variation in answers \[[@B12]\]. The specialists in medical gastroenterology (four women and five men) all worked at Lund University Hospital at the time of the study. The surgeons (one woman and eight men) worked at Lund University Hospital or Helsingborg Hospital. The general practitioners (two women and seven men) worked in the Public Health Service in the southern part of Sweden. The clinicians were 35--56 years (average 47 years) of age and their experience in their speciality ranged from 1 to 22 years (average 10 years). Interviews and case scenarios ----------------------------- A useful approach for studying clinicians\' strategies is judgement analysis \[[@B14]\], which describes the cognitive process involved in making a decision. It focuses on the actual decision made by the clinician and on which information, for example symptoms and clinical findings, that the clinician uses to reach that decision. In our study, each of the clinicians was interviewed separately. The interviews were based on four case histories. The histories described frequent symptoms and findings in 30-year-old men seen in general practice and in specialist out-patient clinics (Table [1](#T1){ref-type="table"}). Besides age and sex, all histories contained information on the localisation of symptoms, but information on subjective symptoms and findings from the examination varied from history to history. The questions, listed in Table [1](#T1){ref-type="table"}, focused on how the clinician viewed whether there was a need for treatment, what management the clinician would suggest, which factors the clinician thought were most important to consider, and what decisions the clinician thought colleagues in their own speciality or in other specialities would make. The same questions were asked for each patient. The case histories were presented in the sequence described in Table [1](#T1){ref-type="table"}. Thus, the questions were answered four times by each informant. One of the authors (KK) with experience in judgement and decision analysis interviewed each informant at her or his office. Each interview began with the interviewer reading the first history. The interviewer then asked the questions aloud and wrote down the informant\'s answers after each question. If the informant wished, the history and questions were reread. At the end of the question-and-answer session after each case history, the interviewer read aloud the participant\'s answers and in cases of ambiguity, adjusted the answers. The interviews thus took the form of an oral questionnaire, but with the possibility to explain the answers more fully. Content analysis ---------------- The analysis was performed after all data had been collected from the informants. Two of the authors (KK, MT) with experience in qualitative studies read the interview notes. In a first step, each of the clinicians\' answers was scrutinised to find content or evaluations in common, which were then coded as concepts. The following is an example of answers coded as a concept: To question A (\"Do you think that it is a problem to decide whether there is a need for treatment in a case like this?\" and if yes: \"What problems?\") informants gave answers like \"there is no demonstrable disease, but the patient has symptoms\" and \"there is no diagnosis but a suffering patient\" (patients 1 and 2). These answers were coded as the concept \"The patient suffers, but there is no distinct diagnosis\". In the second step, related concepts were combined into categories, common to all the case histories. The authors compared their coding and resolved any differences through discussion to reach a consensus. The concepts and categories were compared with the original answers to ensure that the answers were covered by the classifications. Results ======= The varied answers of the clinicians are presented in Table [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}. Different categories of problems existed that could explain the variation in the decision-making process. When answering the question \"Do you think there is a problem to decide whether there is a need for treatment in a case like this?\", five different categories of answers were given. For example, the basis for deciding whether or not to treat and what would be optimal treatment was ambiguous and, further, the evidence for treatment was poor. Clinicians also felt that they lacked competence in evaluating test results when making treatment decisions and that patient expectations and requirements were problems with patients 1--3. In these cases, the patients sometimes requested investigations and treatment, even though the clinician informed them that there was no reported evidence of effect in the literature for the requested treatment or investigation. For the same patient, the management strategies considered by the different clinicians varied widely, including extensive examinations, non-pharmacological treatment, drug treatment, and surgery (Table [3](#T3){ref-type="table"}). For patients 2 and 3, who had tested positive for Helicobacter pylori (H. pylori), some clinicians stated that they would suggest eradication treatment only if the patient had an ulcer, whereas other clinicians said they would suggest this treatment irrespective of the presence of ulcers. For patient 4, who had a duodenal ulcer, some clinicians suggested eradication treatment only if the patient tested positive for H. pylori, whereas others recommended that treatment irrespective of such test results. This variation was also observed among clinicians in the same speciality. Variation was more pronounced if the case history noted symptoms but no organic findings, as in patients 1 and 2, than if the case history noted unambiguous findings and symptoms, as in patients 3 and 4. When asked \"Which factors are most important to consider in your decision?\" (Table [4](#T4){ref-type="table"}), the clinicians defined three categories of information. For patients 1 and 2, whose symptoms were diffuse, several clinicians answered that the medical history was the most important piece of information to consider. They also claimed to need a more comprehensive medical history for patients 1 and 2 than for patients 3 and 4, whose symptoms were more obviously related to an organic diagnosis. A majority of the clinicians believed that their treatment decisions would be similar to those of most of their colleagues in their speciality. Some gastroenterologists believed that surgeons would prefer surgery to medication, given the same test results. Several gastroenterologists and surgeons believed that family clinicians investigated less, informed more, and prescribed acid inhibitors on wider indications than gastroenterologists and surgeons. Discussion ========== Methodological and results approaches ------------------------------------- The four case histories in this study are representative of patients with conditions frequently seen in medical practice. Diseases of the gastrointestinal tract account for about 6% of all reported consultations in Sweden \[[@B13]\]. Gastritis, dyspepsia, and unspecified diseases in the ventricle and duodenum (patients 1 and 2) are diagnosed in 1.6% of the population; oesophagitis and reflux (patient 3) in 0.9%; and different ulcers (patient 4) in 0.6%. The clinicians who participated in this study were not representative of a randomised selection of clinicians in Sweden but rather of a select group with a special interest in patients with symptoms from the gastrointestinal tract. Since their experience with these patients was extensive, they should have been able to develop a treatment policy. One would expect a variation in treatment strategy in this group to be more limited than among a randomised selection of clinicians. The results underpin existing evidence that variations in medical practice exist \[[@B14]-[@B16]\]. This study presents variations both in treatment strategy and in what information is considered important for making treatment decisions. The histories of patients with diffuse symptoms but no objective findings (patients 1 and 2) gave rise to more extensive variations than did the case histories of patients 3 and 4 where the symptoms were more obviously related to a diagnosis (patient 3) or where more obvious organic changes existed (patient 4). Uncertainty in diagnosis could lead to uncertainty regarding which outcome of alternative interventions is optimal \[[@B17],[@B18]\]. A meta-analysis that evaluates the most effective treatment in patients with functional dyspepsia \[[@B10]\] recommended the eradication of H. pylori if the treatment is to be effective from the patient\'s perspective, whereas other randomised, double-blind, controlled studies find that eradication has no beneficial effect \[[@B11]\]. These results were underpinned by the variations observed in this study. Uncertainty due to lack of knowledge or professional competence may lead to qualitative differences. In this study, different management strategies were recommended for patients with reflux disease (patient 3), for example, eradication treatment as one strategy and changes in life-style as another. However, evidence in the literature presenting eradication as an optimal management of reflux is lacking \[[@B17]\]. Instead, this management strategy raises the cost to society and to the patient and causes unnecessary antibiotic pressure. Randomised, double-blinded, prospective studies conclude that the only indication for treating dyspepsia with proton pump inhibitors (PPIs) is the presence of an ulcer or reflux \[[@B7]\]. However, the majority of clinicians prescribed PPIs and H~2~-receptor antagonists for the treatment of dyspepsia, although no evidence of an effect with this treatment compared to placebo has been documented \[[@B9],[@B10]\]. Thus, this regimen has led to high costs for society without any benefits. The drug costs for treatment of ventricular and duodenal ulcers and of reflux, including eradication treatment, were SEK 1.6 billion \[USD 210 million\] in 1998 in Sweden \[[@B13]\], while The Swedish Council on Technology Assessment in Health Care reports that society\'s direct and indirect costs for dyspepsia were between SEK 3.7 and 4.4 billion \[USD 490 and 590 million\] in 2000 \[[@B3]\]. From the informants\' answers to the question \"Which factors are most important to consider in your decision?\", the completeness of some of the answers given in the interviews could be questioned. Most likely, the clinicians failed to answer the question in full as they left out many factors worthy of consideration. Other factors in the patient\'s life such as stress, mental mood, working situation, and diet were not mentioned by the clinicians in the interviews but should often be considered in treatment strategies in an actual clinical situation. Within each of the three specialities, clinicians believed that their colleagues would treat patients in the same way as they themselves did. This aspect of \"professional certainty\" implies that clinicians believe that their practice is correct, irrespective of how much it in fact differs from that of other clinicians \[[@B18],[@B19]\]. This unawareness of variation in the management of frequently seen patients may indicate a lack of communication and discussion about everyday cases. Such discussions are perhaps reserved for more \"complex\" and rare cases. Opportunities to change clinicians\' practice --------------------------------------------- Among the causes of variation in medical practice, the influence of factors like patient characteristics (e.g. age, sex, morbidity, and personal preferences) could be more or less regarded as legitimate to explain variations in practice \[[@B20]\]. Other factors, like resource capacity, could be influenced by, for example, budget restrictions while management policy and practice style are more resistant to change \[[@B20]\]. These latter factors should therefore be the targets of efforts to change. However, many attempts to implement evidence by information alone, for example in the form of clinical practice guidelines, have failed to change management strategies \[[@B20]-[@B23]\]. Instead, a combination of methods is most likely needed if a permanent change is to occur \[[@B23]\]. Furthermore, it is of utmost importance that potential barriers are identified and that the clinicians who will be affected support the clinical practice guidelines to be implemented. In recent years, the individual autonomy of patients and letting patients\' preferences influence the choice of intervention have been emphasised \[[@B24]\]. In this study, such a strategy could explain why some clinicians mentioned that they felt pressured by patients to perform an intervention, even when the optimal strategy was non-intervention. Conclusion ========== In conclusion, our study adds to the present scientific literature by showing that, even in cases with a consensus in the scientific literature on treatment, clinicians can differ in their opinion of which management is optimal. Despite these variations, the clinicians believed that the decisions made by their colleagues would be similar to their own. Overall, we as researchers must make scientific evidence comprehensible and communicate evidence in such a way that clinicians are easily able to interpret and implement it in practice. Of particular significance is that scientific evidence leads to an evidence-based care, which is effective clinical practice, and to the promotion of health from the perspective of the patient, together with cost-effectiveness as a priority. Cost-effectiveness is a vital concern. Targeting high patient satisfaction may lead to ineffectiveness if the economic consequences of a treatment strategy are ignored. Declaration of competing interests ================================== The authors declare that they have no competing interests. Authors\' contributions ======================= Kerstin Knutsson, DDS, Odont dr, has special competence in judgement and decision-making within health care. KK contributed to the overall concept and design of the study and conducted the interviews. Grants from the Swedish Research Council (grant no. 521-2001-6341) supported this study. Bodil Ohlsson, MD, PhD, is a specialist in gastroenterology and contributed with her knowledge in this area to the design of the study and the interpretation of the data. Margareta Troein, MD, PhD, has special competence in qualitative methods in medical care and contributed to the qualitative analysis and interpretation of the data. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-230X/5/15/prepub> Acknowledgements ================ We thank Professor Åke Nilsson, Lund University Hospital, Lund University, Lund and Professor Stefan Lindgren, Malmö University Hospital, Lund University, Lund for their valuable opinions in the initial phase of this study and on the manuscript. Figures and Tables ================== ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Case histories and questions presented to the informants. ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Patient 1. A 30-year-old man with symptoms from the upper part of the abdomen. The patient has had intermittent pain, made worse by stress, for a few years. One year ago an ultrasound examination and an oesophagogastroduodenoscopy were performed. No abnormal findings were registered. H. pylori serology was negative. Patient 2. A 30-year-old man with symptoms from the upper part of the abdomen. The patient has had intermittent pain, made worse by stress, for a few years. One year ago an ultrasound examination and an oesophagogastroduodenoscopy were performed. No abnormal findings were registered. H. pylori serology was positive. Patient 3. A 30-year-old overweight man with symptoms from the upper part of the abdomen and retrosternally. The pain worsens when he leans forward and when he rests in a prone position. He sometimes experiences a sour or bitter taste in his mouth. H. pylori serology was positive. Patient 4. A 30-year-old man with symptoms from the upper part of the abdomen. The ultrasound examination revealed nothing abnormal, but the oesophagogastroduodenoscopy indicated a duodenal ulcer. After each case the following questions were asked:\ A. Do you think that it is a problem to decide whether there is a need for treatment in a case like this? If yes: What problems?\ B. How would you manage a case like this?\ C. Which factors do you think are most important to consider in your decision?\ D. What do you believe your colleagues would decide in a case like this? ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Clinicians\' answers coded as concepts and organised into categories. Answers of 27 clinicians to the question \"Do you think that it is a problem to decide whether there is a need for treatment in a case like this?\" If the physicians admitted that it was a problem, they were then asked \"What problems?\" Answers were coded as concepts and organised into categories for each patient. Numbers of answers are presented within parentheses. Each clinician may give several answers that could be included in different categories. ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *Categories* *Concepts* ----------------------------------------------------- ------------------------------------------------------------- ---------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------- --------------------------------------------------------- There is no problem to decide (16) (20) (19) (22) Basis for treatment decision is insufficient There is no diagnosis (5) There is no diagnosis (3) The amount of discomfort is not clear (1)\ No examination of Helicobacter pylori has been made (3) More examinations are needed (2) Evidence for treatment is poor There is hardly any evidence for benefits of treatment (1)\ Patient suffers, but there is no distinct diagnosis (1) Recurrent condition (2) It is difficult to treat dyspepsia (1)\ Patient suffers, but there is no distinct diagnosis (2)\ Psychosomatic problems (2) Clinician is uncertain of her or his own competence Uncertainty about indication for treatment when Helicobacter pylori serology is positive (1) Uncertainty about indication for treatment when Helicobacter pylori serology is positive (1) Patient\'s expectations Patient wishes treatment (1) Patient requires treatment due to positive Helicobacter pylori serology (2) Patient is not motivated to reduce weight (1)\ Patient requires treatment due to positive Helicobacter pylori serology (2) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Clinicians\' answers coded as concepts and organised into categories. Answers of 27 clinicians to the question \"How would you manage a case like this? \" Answers were coded as concepts and organised into categories for each patient. Numbers of answers are presented within parentheses. Each clinician may give several answers that could be included in different categories. ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *Categories* *Concepts* ---------------------------------------------------------------------------------- ------------------------------------------------------------------ ----------------------------------------------------------------------------- ---------------------------------------------------------------------------- ---------------------------------------------------------------------------- Extend the examination Exclude coeliac disease, examine colon, liver, gall bladder (6)\ Oesophagogastroduodenoscopy, expiration tests for Helicobacter pylori (10)\ Oesophagogastroduodenoscopy, expiration tests for Helicobacter pylori (18) Diagnostic tests for Helicobacter pylori (8) Evaluate patient\'s psychosocial conditions (4)\ Evaluate patient\'s psychosocial condition (2)\ Evaluate outcome of previous treatment (3) Evaluate outcome of previous treatment (1) Prescribe non-pharmacological treatment Reassuring information (8)\ Reassuring information (7)\ Smoking cessation (2)\ Unspecific life-style changes (1) Expectancy (3)\ Expectancy (4)\ Weight reduction (3)\ Dietary advice, reduction of alcohol intake (11)\ Dietary advice (6)\ Dietary advice (3)\ Reduction of stress (7) Reduction of stress (3) Unspecific life-style changes (6) Prescribe drugs against acidity Antacids (7)\ Antacids (7)\ Antacids (1)\ Proton pump inhibitors (11) H2-receptor antagonists (13)\ H2-receptor antagonists (9)\ H2-receptor antagonists (8)\ Proton pump inhibitors (5)\ Proton pump inhibitors (6)\ Proton pump inhibitors (15) Sucralphate (1) Sucralphate (1) Prescribe triple treatment (antacid and antibiotics against Helicobacter pylori) Only if patient has an ulcer (7)\ Only if patient has oesophagitis or an ulcer (1)\ Only if patient has a positive test for Helicobacter pylori (11)\ Yes, irrespective of whether patient has an ulcer (4) Yes, irrespective of whether patient has oesophagitis or an ulcer (1) Yes, irrespective of whether Helicobacter pylori serology is positive (14) Prescribe other drugs Drugs which increase gut motility (2)\ Drugs which increase gut motility (1) Alginate (6) NSAID (1)\ Bulking agent (1) Recommend surgery If patient has hiatus hernia (1) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Clinicians\' answers coded as concepts and organised into categories. Answers of 27 clinicians to the question \"Which factors are most important to consider in your decision?\" Answers were coded as concepts and organised into categories for each patient. Numbers of answers are presented within parentheses. Each clinician may give several answers that could be included in different categories. ::: ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *Categories* *Concepts* --------------------------- ------------------------------------------------- ----------------------------------------------------------------------------- ----------------------------------------------------------------------------- ----------------------------------- Medical history Patient\'s age (4)\ Patient\'s age (5)\ Patient\'s age (2)\ General state of health (1)\ Symptoms including localisation of pain (12)\ Symptoms including relation to meals (12)\ Symptoms generally and in relation to position (21) Symptoms (1) Life-style/stress (12)\ Effects of previous medication (1)\ Effects of previous medication (1)\ Similar problems for a long time (2) Similar problems for a long time (3) Results from examinations Oesophagogastroduodenoscopy and ultrasound (6)\ Oesophagogastroduodenoscopy and ultrasound (1)\ Oesophagogastroduodenoscopy (3)\ Oesophagogastroduodenoscopy (26)\ Test for Helicobacter pylori (5) Test for Helicobacter pylori (7) Test for Helicobacter pylori (1)\ Test for Helicobacter pylori (3) Weight (1) Patient\'s expectations Patient\'s preferences (1) Patient requires treatment due to positive Helicobacter pylori serology (3) Patient requires treatment due to positive Helicobacter pylori serology (3) ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- :::

PubMed Central

2024-06-05T03:55:58.363499

2005-5-15

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173099/", "journal": "BMC Gastroenterol. 2005 May 15; 5:15", "authors": [ { "first": "Kerstin", "last": "Knutsson" }, { "first": "Bodil", "last": "Ohlsson" }, { "first": "Margareta", "last": "Troein" } ] }

PMC1173100

Background ========== With the development of molecular techniques to identify genetic polymorphisms, numerous markers and in particular Single Nucleotide Polymorphisms (SNPs) \[[@B1],[@B2]\] are now available within and between genes for establishing their possible role in disease susceptibility. One recurrent question in the literature is the way these markers could be used to test for association with complex diseases \[[@B3],[@B4]\]. Should one focus on one marker at a time and simply compare allele or genotype distributions for the studied markers in cases and controls or should one consider haplotypes formed by several linked markers? Haplotypic methods have been suggested to be more powerful at detecting the role of a given genomic region in disease susceptibility \[[@B4]-[@B7]\]. Indeed, disease susceptibility may be due to the combined effects of variants at different markers. Different methods have been proposed to test for association between markers located on the same reconstructed haplotypes and disease susceptibility \[[@B4],[@B7]-[@B10]\]. In these tests, the number of haplotypes being large, the degrees of freedom to compare cases and controls are also large, decreasing the power to detect the association. Moreover, as some haplotypes would only be carried by one or two individuals, there could be statistical problems owing to small sample sizes making difficult the evaluation of their possible effect on the susceptibility to the disease. A solution to reduce these problems is to group haplotypes in order to decrease the degrees of freedom and to increase the number of individuals in the different haplotype groups. The cladistic method as described by Templeton et al. \[[@B11]\] is a method to carry out such groupings of haplotypes and to determine which haplotype or group of haplotypes is likely to be responsible for the phenotypic variation observed among the population. It consists in using parsimony methods to build a phylogeny of the haplotypes, and to group haplotypes according to the clades defined by the phylogenetic tree. When applied to a case/control data set, the proportions of cases and controls in different clades are compared. If a clade shows a significantly larger number of cases than the others, one will conclude, first, that there is an association between the disease and one haplotype or a group of haplotypes belonging to the clade, and, second, that the mutations defining the clade containing an excess of cases are good candidates to be functional gene polymorphisms involved in the disease susceptibility. The cladistic analysis relies for a large part on the reconstruction of an accurate phylogenetic tree, but recombination is known to bias phylogenetic reconstruction processes \[[@B12]\]. The recent discovery of the haplotype block structure of the human genome \[[@B13]\] now provides an interesting framework for cladistic methods. Indeed, haplotype blocks define regions in which the recombination rate is quite low. Moreover, once a block is found to be associated with a disease, it will not be suitable to use classical methods based on recombination to fine map susceptibility loci. Haplotype phylogenies being based on the history of mutations may then help to identify disease susceptibility loci. The cladistic method was first described in a series of five articles \[[@B11],[@B14]-[@B17]\] and applied in different case-control studies focusing either on quantitative or qualitative data. For example, Haviland et al. used a cladistic analysis to study the association between the apolipoprotein AI-CIII-AIV and plasma lipid, lipoprotein and apolipoprotein levels \[[@B18]\]. Keavney et al. \[[@B19]\] used cladograms to localize the sites that significantly influence Angiotensin-I Converting Enzyme (ACE). Zhu et al. \[[@B20]\] worked on quantitative data simulated for the twelfth Genetic Analysis Workshop (GAW). Using regression analysis and analysis of variance (ANOVA), they were able to find a short region containing the simulated mutation. Wang et al. \[[@B21]\] used a nested linear model to analyze the association between *ApoA4*haplotypes and plasma lipid level in the cladistic framework. Among the studies that consider discrete data is the one by Kittles et al. \[[@B22]\] who tested the association between haplotype variations on the Y chromosome and alcohol dependence and related personality traits. Lobos and Todd applied the same cladistic method to four RFLPs in the tyroxine hydroxylase gene \[[@B23]\], and to five SNPs in the dopamine receptor D2 gene \[[@B24]\]. Darlu and Génin \[[@B25]\] applied the method to the GAW 12 simulated data and were able to identify some of the disease susceptibility sites. For a more detailed review on the subject, see \[[@B26]\]. More recently Seltman et al. \[[@B27]\] developed a method that can be applied on family-based data. They proposed a test, the ET-TDT, that combines the transmission disequilibrium test (TDT) and the use of unrooted phylogenetic trees to group haplotypes. They further extended their method to the analysis of qualitative and quantitative data from case-parent trios or population-based studies \[[@B28]\]. Finally, an alternative method where phylogenetic trees are reconstructed by a distance method was proposed by Durrant et al. \[[@B29]\] and implemented in the software CLADHC. From these different examples, cladistic methods appear to be innovative methods to detect an association between a candidate gene and a disease and to localize disease susceptibility sites. However, the validity and the power of these methods to detect association have only been tested in a very limited number of situations (fixed pattern of linkage disequilibrium, fixed phylogenetic tree) \[[@B27],[@B29]\]. In this paper, we develop a new cladistic test (CT), based on the analysis of a rooted tree, which allows to make hypotheses about functional polymorphisms that are involved in the susceptibility to the disease. Through simulations on ten different tree topologies, we evaluate the conditions in which this cladistic method is efficient and compare it to other methods. Finally, we apply this method to a real data set concerning Crohn disease and the CARD15 gene \[[@B30]\]. Results ======= Association test ---------------- As detailed in the method section, four different tests are performed: a single locus allele-based test referred to as SbST (Site by Site Test), an haplotype-based test referred to as HT (Haplotypic Test), a test based on haplotype phylogenies referred to as CT (Cladistic Test) and a test based on the clustering of haplotypes described by Durrant et al. \[[@B29]\] and referred to as CLADHC. To compare their respective power to detect an association, simulations were performed under three disease models involving one (model 1) or two (model 2 and model 3) susceptibility sites for ten different tree topologies. Results for model 1 are presented in Table [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}. When the susceptibility site is included in the studied SNPs (Table [1](#T1){ref-type="table"}), SbST is the most powerful test for all trees but one (tree 8) (between 99.1% and 99.5% depending on the tree, 99.35% on average), while the other tests have a weaker power (92.9%, 93.7%, and 91.2%, on average, for CT, HT, and CLADHC respectively). CT and CLADHC have almost the same average type I error, slightly but significantly lower than the type I error of HT and SbST. In these simulations, as the type I errors cannot be fixed a priori, powers may be difficult to compare. However, if the type I errors of CT and CLADHC were increased, their power would also be increased and, consequently, the difference in power with HT would be even more pronounced and the difference in power with SbST would be more reduced. In the trees showing comparable type one error (Trees 1, 5, 7), the power of SbST is clearly higher than the power of CT. When the susceptibility site is removed (Table [2](#T2){ref-type="table"}), the performances of the four tests are, on average, similar (about 89%), CT being the most powerful test for four trees, SbST for four other trees, HT and CLADHC for one tree each. However, the type I errors being slightly different, by adjusting them one expects that CT and CLADHC could be on average, slightly more powerful than HT and SbST. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Power to detect an association, one susceptibility site simulated and included in the analysis ::: Tree CT CLADHC HT SbST ------------- --------------------------- --------------- --------------- ---------------------------- 1 96.4 *(3.3)* 88.9 *(3.3)* 91.1 *(5.1)* [99.3]{.underline}*(3.4)* 2 99.0 *(4.4)* 95.7 *(4.1)* 91.3 *(6.0)* [99.5]{.underline}*(5.3)* 3 86.0 *(2.5)* 91.1 *(4.7)* 91.2 *(4.6)* [99.3]{.underline}*(4.4)* 4 96.3 *(3.8)* 95.6 *(2.9)* 91.5 *(4.2)* [99.3]{.underline}*(4.6)* 5 88.8 *(4.1)* 95.3 *(3.3)* 90.4 *(4.7)* [99.1]{.underline}*(4.6)* 6 84.5 *(3.5)* 88.8 *(2.7)* 90.7 *(4.4)* [99.5]{.underline}*(4.4)* 7 97.4 *(4.0)* 96.3 *(3.7)* 91.9 *(4.8)* [99.2]{.underline}*(4.3)* 8 [99.7]{.underline}*(3.2)* 98.5 *(2.4)* 91.2 *(4.9)* 99.4 *(4.7)* 9 88.1 *(3.5)* 94.0 *(4.4)* 91.3 *(4.7)* [99.4]{.underline}*(4.3)* 10 92.5 *(2.6)* 93.0 *(2.9)* 91.3 *(4.7)* [99.5]{.underline}*(3.6)* Average 92.9 *(3.5)* 93.7 *(3.4)* 91.2 *(4.8)* [99.35]{.underline}*(4.4)* Std dev.^a^ 5.63 *(0.62)* 3.24 *(0.76)* 0.41 *(0.49)* 0.13 *(0.54)* Sample size: 100 cases and 100 controls, 1000 replicates, penetrance vector: \[0.03;0.06;0.3\], frequency of the susceptibility allele: 0.205 ± 0.005. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test. ^a^Standard deviation ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Power to detect an association, one susceptibility site simulated and removed from the analysis ::: Tree CT CLADHC HT SbST ------------- --------------------------- --------------------------- --------------------------- ---------------------------- 1 [94.8]{.underline}*(3.8)* 89.1 *(3.5)* 91.2 *(5.6)* 89.5 *(3.9)* 2 94.0 *(3.8)* 92.7 *(4.1)* 87.6 *(5.6)* [98.5]{.underline}*(5.3)* 3 79.3 *(3.2)* 88.9 *(4.7)* 86.2 *(4.7)* [94.2]{.underline}*(4.1)* 4 88.7 *(2.3)* 85.7 *(3.1)* 87.8 *(4.6)* [93.6]{.underline}*(4.4)* 5 85.6 *(4.3)* [94.4]{.underline}*(4.1)* 90.5 *(4.5)* 94.0 *(4.1)* 6 78.6 *(3.0)* 81.6 *(3.3)* [88.0]{.underline}*(4.2)* 67.0 *(4.2)* 7 92.4 *(3.8)* 90.5 *(3.7)* 87.8 *(5.2)* [92.7]{.underline}*(4.0)* 8 [99.7]{.underline}*(3.0)* 98.6 *(2.8)* 91.2 *(4.7)* 99.6 *(4.5)* 9 [91.9]{.underline}*(3.7)* 89.2 *(4.5)* 91.2 *(4.8)* 89.1 *(4.1)* 10 [88.8]{.underline}*(2.9)* 82.8 *(3.0)* 81.3 *(4.1)* 79.1 *(3.5)* Average 89,38 *(3.4)* 89.35 *(3.7)* 88,28 *(4.8)* [89,73]{.underline}*(4.2)* Std dev.^a^ 6.71 *(0.6)* 5.2 *(0.6)* 3.06 *(0.5)* 9.81 *(0.5)* Sample size: 100 cases and 100 controls, 1000 replicates, penetrance vector: \[0.03;0.06;0.3\], frequency of the susceptibility allele: 0.205 ± 0.005. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test. ^a^Standard deviation ::: As expected, the power is higher when the susceptibility site is included than when it is removed. In this latter case, the power depends on the degree of linkage disequilibrium (LD) between the susceptibility site and the other sites as expressed by the maximum value of LD, referred to as *LD*~*max*~. Moreover, since the pattern of LD can be different from one tree to another, one expects more differences in power between trees when the susceptibility site is removed. Indeed the standard deviation of the power is larger in this case. This standard deviation is also particularly large for SbST (*σ*= 9.81). This test, being an allele-based test, is more influenced by LD than the three others tests that are haplotype-based tests, as proved by the high positive correlation observed between the power of SbST to detect association and the *LD*~*max*~(Spearman rank correlation ρ equals 0.851, p-value \< 0.01). For CT and CLADHC, Spearman rank correlations are not significant (p \> 0.1) confirming that LD has a lesser impact on these two tests. Therefore, the high standard deviations observed for these two tests are probably more related to the phylogeny of haplotypes than to the linkage disequilibrium. The results for model 2 and model 3, differing by the values of the penetrance vector, are presented in Table [3](#T3){ref-type="table"} and in Table [4](#T4){ref-type="table"} respectively. There is no real difference in power between the four tests for model 2 (about 97%). For model 3, the power of the four tests is approximately half the power for model 2 owing to the reduced value of the penetrance for the genotype associated to the highest risk in this model. In this case, SbST and CT are, on average, the most and the less powerful test respectively ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Power to detect an association for two susceptibility sites simulated under model 2 ::: Model 2 ------------- --------------------------- ---------------------------- --------------------------- --------------------------- Tree CT CLADHC HT SbST 1 97.8 *(3.7)* [99.3]{.underline}*(3.8)* 97.7 *(5.0)* 99.1 *(6.9)* 2 95.1 *(3.6)* 97.7 *(2.9)* [97.9]{.underline}*(5.6)* 95.8 *(4.3)* 3 94.2 *(4.9)* [99.4]{.underline}*(4.9)* 97.9 *(5.1)* 98.6 *(4.9)* 4 97.6 *(4.6)* 97.8 *(4.6)* 97.8 *(5.5)* [98.6]{.underline}*(4.6)* 5 98.5 *(5.3)* [99.5]{.underline}*(4.5)* 98.9 *(5.7)* 98.2 *(7.4)* 6 [98.9]{.underline}*(4.4)* 97.9 *(3.9)* 98.5 *(5.7)* 96.9 *(6.6)* 7 93.0 *(5.1)* 93.5 *(4.6)* [96.9]{.underline}*(5.6)* 95.5 *(5.2)* 8 95.2 *(6.1)* [99.0]{.underline}*(5.3)* 98.2 *(6.2)* 95.4 *(4.2)* 9 96.7 *(3.6)* 97.3 *(3.1)* 97.6 *(4.1)* [97.7]{.underline}*(4.0)* 10 [99.9]{.underline}*(4.5)* 99.5 *(4.0)* 98.2 *(60.)* 99.8 *(5.8)* Average 96.7 *(4.6)* [98.06]{.underline}*(4.2)* 97.96 *(5.45)* 97.56 *(5.4)* Std dev.^a^ 2.24 *(0.8)* 1.83 *(0.8)* 0.54 *(0.6)* 1.58 *(1.2)* Sample size: 200 cases and 200 controls, 1000 replicates, penetrance: 0.9 for the homozygotes carrying the two susceptibility alleles, 0.3 for the other genotypes. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test ^a^Standard deviation ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Power to detect an association for two susceptibility sites simulated under model 3 ::: Model 3 ------------- --------------------------- --------------------------- --------------------------- --------------------------- Tree CT CLADHC HT SbST 1 40.7 *(4.7)* [54.0]{.underline}*(4.5)* 42.0 *(5.9)* 47.5 *(4.7)* 2 31.7 *(3.4)* 32.7 *(3.9)* 43.0 *(5.9)* [45.8]{.underline}*(5.3)* 3 26.8 *(4.2)* [47.9]{.underline}*(4.4)* 40.7 *(6.4)* 45.2 *(3.5)* 4 34.8 *(4.8)* 39.2 *(4.7)* 38.9 *(7.0)* [47.0]{.underline}*(4.4)* 5 41.5 *(5.6)* [49.7]{.underline}*(4.7)* 44.2 *(6.6)* 37.2 *(4.6)* 6 42.0 *(3.6)* 34.8 *(4.7)* [42.4]{.underline}*(5.6)* 41.3 *(5.3)* 7 28.6 *(5.2)* 32.1 *(4.7)* [41.6]{.underline}*(6.4)* 39.9 *(4.3)* 8 33.2 *(4.3)* [47.2]{.underline}*(4.0)* 42.9 *(6.1)* 37.7 *(4.0)* 9 33.5 *(4.4)* 39.1 *(4.1)* 41.2 *(7.0)* [48.5]{.underline}*(5.3)* 10 [65.6]{.underline}*(6.6)* 48.1 *(4.1)* 42.1 *(7.8)* 56.9 *(4.2)* Average 37.84 *(4.7)* 42.48 *(4.4)* 41.9 *(6.5)* [44.7]{.underline}*(4.6)* Std dev.^a^ 11.07 *(0.9)* 7.84 *(0.3)* 1.45 *(0.7)* 5.94 *(0.6)* Sample size: 200 cases and 200 controls, 1000 replicates, penetrance: 0.6 for the homozygotes carrying the two susceptibility alleles, 0.3 for the other genotypes. The corresponding type I errors are indicated in italics within brackets. For each tree, the method giving the best result is underlined. CT: Cladistic Test, CLADHC: test developed by Durrant et al. \[29\], HT: Haplotypic Test, SbST: Site by Site Test ^a^Standard deviation ::: Localization of the susceptibility site(s) ------------------------------------------ The efficiency of SbST, CT and CLADHC to localize the susceptibility site(s) when it is included among the SNPs is presented in Figure [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}. The efficiency of CT is plotted against the efficiency of SbST or CLADHC. To correctly interpret these figures, it has to be noted that the criteria used to compare the methods are not strictly identical. Indeed, the identified SNP is always unambiguously detected by the SbST method, since this is the site giving the most significant chi-square and two chi-squares are never exactly identical. On the other hand, the susceptibility site may not be the only one detected by CT because two or more sites could have the same values of the criteria (*V*~*i*~, see the Methods section). However, in our simulations, the susceptibility site is the only one detected in from 80% (for 1 tree) to 100% (for 5 trees) of the replicates. When the susceptibility site is not found alone, it is almost always found with only one other site, rarely with two other sites (\<0.5% of the replicates). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Comparison of the efficiency to localize the susceptibility site between CT, and SbST or CLADHC**. The efficiency to localize the susceptibility site is measured by the percentage of replicates in which the simulated susceptibility site is either uniquely detected or detected among other putative sites. black circles: comparison of CT and SbST, crosses: comparison of CT and CLADHC. Each symbol corresponds to one of the ten tree topologies. ::: ![](1471-2156-6-24-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Comparison of the efficiency to localize the two susceptibility sites between CT and SbST**. The efficiency to localize the two susceptibility sites is measured by the percentage of replicates in which the two sites are detected (black circles) or at least one of the two sites (white circles) Each symbol corresponds to one of the ten tree topologies. (A): penetrance 0.9 for homozygotes carrying the two susceptibility alleles and 0.3 for all other genotypes; (B): penetrance 0.6 for homozygotes carrying the two susceptibility alleles and 0.3 for all other genotypes. ::: ![](1471-2156-6-24-2) ::: For model 1 (Figure [1](#F1){ref-type="fig"}), CT is slightly less efficient than SbST (for 9 trees out of 10) and more efficient than CLADHC (for all 10 trees). For SbST, we note a high negative correlation between the efficiency to localize the susceptibility site and the *LD*~*max*~between the susceptibility site and the other sites (Spearman rank correlation *ρ*equal -0.88 with a p-value of 0.008). It shows that the presence of a site in relatively high LD with the susceptibility site disturbs the localization process, increasing the uncertainty of the susceptibility site\'s identification. Interestingly, there is no such statistically significant correlation with CT and CLADHC (p \> 0.08), due to the fact that these last two tests use both the information on the whole haplotypes and on their history through the tree reconstruction. The presence of sites in high LD with the susceptibility site does not prevent these two methods from localizing correctly the susceptibility site. For model 2 and model 3, (Figure [2A](#F2){ref-type="fig"} and [2B](#F2){ref-type="fig"} respectively), CT is more or at least as efficient as SbST to localize the two sites, whatever the penetrance model. For the detection of at least one of the two sites, we can observe that, whatever the tree, CT is always more efficient than SbST under model 2, whereas this is only true for 6 out of the 10 trees under model 3. Application to CARD15/NOD2 polymorphisms and Crohn disease ---------------------------------------------------------- The reconstruction of the phylogenetic tree of the 33 haplotypes leads to a very high number of equally parsimonious trees. One thousand of them were retained and analyzed. In table [5](#T5){ref-type="table"}, for the two rooting methods mfH and consH, the SNPs are ranked from the least to the most significant (based on the *V*~*i*~parameter, see the Methods section). Association is detected with CT (*p*~*mfH*~= 2 × 10^-5^, *p*~*consH*~= 3 × 10^-5^) and SNP 8 is selected as potential susceptibility site. If haplotypes carrying the susceptibility allele at SNP 8 are removed, an association can still be detected (*p*~*mfH*~= 4 × 10^-5^, *p*~*consH*~= 3 × 10^-5^) and SNP 12 is identified as potential susceptibility site. When removing haplotypes with the susceptibility allele at SNP 12, there is still a remaining association (*p*~*mfH*~= 8 × 10^-5^, *p*~*consH*~= 3 × 10^-5^) and depending on the rooting method either SNP 13 (with consH) or SNP 9 (with mfH) are retained. When haplotypes with the susceptibility allele at SNP 13 are removed, the association test is not significant (*p*~*mfH*~= 2.9 × 10^-1^, *p*~*consH*~= 1.1 × 10^-1^). When haplotypes with the susceptibility allele at SNP 9 are removed, the association test is significant (*p*~*mfH*~= 2 × 10^-5^, *p*~*consH*~= 5 × 10^-5^) and the identified site is SNP 13 for both rooting method. In the end, the combination of SNP 8, SNP 12 and SNP 13 is the most parsimonious combination that can explain the susceptibility to the disease. Moreover, we can note that these 3 SNPs are those showing the highest *V*~*i*~on the whole data set. So, we finally have identified three SNPs involved in the susceptibility to Crohn\'s disease. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Cladistic analysis of Crohn\'s data 1000 equiparsimonious trees are studied. SNPs are ranked from the least to the most significant. In bold, the presumed susceptibility sites previously identified by Hugot et al. \[30\]. ::: least significant \-\-\-\-\-\-\--\> most significant ---------- ------------------- ------------------- ------------------ ------- -- ------------ ------------ ---------- mfH^a^ SNP 5 SNP 9 SNP 6 SNP 7 **SNP 13** **SNP 12** **SNP8** consH^b^ SNP 9 SNP 5 SNP 6 SNP 7 **SNP 13** **SNP 12** **SNP8** ^a^Rooting on the most frequent haplotype ^b^Rooting on a consensus sequence ::: Discussion ========== In this paper, we have presented a method to perform phylogeny-based nested haplotype analysis of case/control data and introduced the cladistic test (CT). We have shown by simulations that the CT test is approximately as powerful to detect an association as a test that compares haplotype distributions in cases and controls with no prior grouping. However, a major advantage of CT towards other haplotypic tests is the possibility of making inferences on potential susceptibility sites. Indeed the method of parsimony used to build the phylogeny allows the reconstruction of character state changes on the tree and thus the identification of the most likely susceptibility sites. Finding the polymorphisms responsible for the susceptibility to a disease is an important issue in genetic epidemiology and in the past few years, different methods have been proposed that use linkage disequilibrium to fine-map candidate genes and to localize disease mutations (\[[@B31],[@B32]\] and \[[@B33]\] for example) These methods are based on the fact that a disease variant arises on a particular haplotype, the haplotype and the disease thus being initially associated. Then, over time, this association is broken by recombination and only a small region of the ancestral haplotype is preserved around the disease variant. In such models, recombination is the event of interest. Other parameters such as mutations can be integrated in the models, but they are considered as nuisance parameters. The cladistic test developed here is different in the sense that it is based on the reconstruction of the evolutionary history of the mutations that occur to form the haplotypes. Contrary to the other fine-mapping methods, the event of interest is mutation, and recombination is a nuisance parameter. Through the simulation study, we have shown that this test is a powerful tool to suggest etiological SNPs. This is also illustrated here on Crohn disease since the cladistic test was able to correctly identify the 3 sites within the CARD15 gene that are already reported to be involved in the disease susceptibility \[[@B30]\]. The cladistic test requires the building of a phylogeny of the different haplotypes present in the sample. Different methods are available to build haplotype phylogenies: distance methods, parsimony methods and maximum likelihood (ML) methods. A parsimony method was chosen here because it allows the reconstruction of apomorphies that can be used to define potential sites involved in the disease susceptibility and because it runs faster than ML methods, thus allowing a simulation study. We have compared CT with another clustering method, CLADHC, a distance-based method that also differs from CT in the statistical analysis (CLADHC performs regression analysis, CT is based on chi-square tests). We have shown that the power of these two tests to detect association is quite similar, CT being slightly more powerful when one site is simulated, and CLADHC when two sites are simulated. For the localization of the susceptibility site, the efficiency of the two methods is difficult to compare because CT directly localizes the site on short sequences while CLADHC localizes a window containing the susceptibility site on longer sequences. The two test should thus be used in different situations. Nevertheless CT appears to be more efficient in our simulation conditions In the phylogeny reconstruction, one is faced to different problems. The first one is the choice of the ancestral sequence to root the phylogeny. With real data, the ancestral sequence is usually not known, unless at least two outgroups can be postulated (in the parsimony context). In the absence of ancestral sequences, hypotheses have to be made, and they are numerous, like taking the most frequent haplotype, but this haplotype could be different from one population to an other or from one sampling to another, and this most frequent haplotype may not have a significantly different frequency from the next most frequents. Or one can take the consensus haplotype, but this sequence may not be observed in the sample, or not be among the most frequent sequences. In this study, our strategy was to consider the uncertainty of the ancestral sequence as a nuisance parameter, since our goal was not primarily to focus on this problem and because we must be in the same situation for all the performed simulations. Extensive further simulations will be needed to assess the properties of the different rooting methods. However, several ancestral hypotheses should be considered, discussed, and tested on real data, as we have done for the Crohn data. Second, the presence of recombinations would restrict the use of phylogenetic reconstruction. Indeed, phylogenetic reconstruction is known to be biased by recombinations \[[@B12]\] in a manner depending on how and when the recombination events occur in the tree. This context led us to simulate the history of mutations along the haplotypes without introducing recombinations between haplotypes. However, because in our method the reconstructed tree has not to represent the exact history of the haplotypes, but only gives a way to cluster them in an appropriate way, we believe that our method can tolerate some low degrees of recombination, as it is probably the case for the Crohn data. Further simulations would however be required to better assess this point. Third, the choice of the way characters are optimized on the tree is another important point in cladistic analysis. Two classical options can be used to optimize the character state changes along the tree: deltran, (delayed transformation) that favors convergences or acctran(accelerated transformation) that favors reversions. In this paper, only the results performed with the deltran option are presented. Simulations (one susceptibility site) have also been performed with the acctran option. For the association detection, the results are very similar to those obtained with deltran showing that this parameter does not influence the power of the CT method to detect association in our simulation conditions. For the localization of the susceptibility site, the results are different. The method is more efficient with the deltran option than with the acctran option for eight trees out of ten, it is less efficient for one tree and the results are equivalent for the last tree. It is partly due to to the uncertainties of the character *S*and to the frequency of the disease susceptibility allele. When the frequency of the susceptibility allele is low, as in our simulations, and when *S*is coded \"?\" in some nodes of the tree, one expects the change of *S*from 0 to 1 to occur later in the tree, and the delayed option (deltran) will then be more appropriate. Conversely, when the frequency of the disease is large, the change of *S*from 0 to 1 has more chances to occur earlier in the tree since the *S*mutation can then include more cases, and the accelerated option (acctran) will be more appropriate. However, when analyzing a real data set, it is advisable to compare the results obtained with both options. The variability of the results obtained with the cladistic method is due for a minor part to the LD between the susceptibility site(s) and the other sites and for the major part to parameters specific to the tree. Among these parameters, the position of the mutation in the tree (near the root or the leaves, sample size of the clades near the mutations\...) has already been reported to have an influence on the power of cladistic methods to detect association \[[@B27]\]. Apart from the position of the mutation, the presence of sites that co-mutate with the susceptibility site, the degree of multiple mutations of the susceptibility site in the tree, and the existence of one or more reversions for the susceptibility site influence the power of the cladistic test to detect an association, and they have an even greater impact on the efficiency of the cladistic test to localize the susceptibility site. As all these factors are not independent, the evaluation of their own effects remains problematic. The four methods compared here (CT, HT, SbST and CLADHC) suppose that haplotypic data are available, which is not usually the case. The most likely haplotypes for an individual have to be reconstructed using numerical methods (for a review of these methods, see \[[@B34],[@B35]\]). The availability of familial data may help to determine the phase of the markers and then, to obtain the most likely haplotypes by using the genetic information of the two parents. However, even when these familial data are not available, the numerical methods have been shown to be very efficient to reconstruct haplotypes, except for rare haplotypes that may either be missed or falsely created. In this work, since we choose to focus on the comparison between several methods, the problem of haplotype inference is ignored. Moreover, as the four studied methods are all based on already known or inferred haplotypes carried by the individuals, the uncertainty due to the haplotype inference should not play differently from one method to another. Further extension of the CT method will take this uncertainty into account by using, for example, likelihood-based tests. Conclusion ========== In conclusion, as proved by our simulations and the Crohn example, the use of phylogenies to group haplotypes by clades and the nested analysis of case-control data on the basis of these clades turns out to be an interesting and complementary strategy for the genetic study of multifactorial diseases, especially to pinpoint the sites that are likely to be involved in disease susceptibility among the different markers identified within a gene. Methods ======= Principle of the method ----------------------- The material on which the method is based is a sample of haplotypes composed of a combination of SNPs, each haplotype being labeled either as control or as affected depending on the phenotype of the individual. After building the phylogenetic tree of these different haplotypes, a series of nested homogeneity tests are performed to detect differences in the distribution of cases and controls in the different clades. Once an association is detected, a new character is defined according to the proportion of cases carrying an haplotype. Then, it is optimized on the tree and the sites that significantly co-mutate with this new character are putative susceptibility sites for the disease. Simulation study ---------------- To test the power of this method to detect an association and its efficiency to precisely localize the susceptibility site, we performed the following steps : i) simulation of the haplotypes and choice of the susceptibility site(s) (one or two susceptibility sites are considered) ; ii) attribution of a status (case or control) to these haplotypes taking into account the genetic model of the disease; iii) reconstruction of the phylogeny of haplotypes and inference of the character state changes along the branches of the tree; iv) association test: nested analysis of the case/control ratio in the clades; v) localization of the susceptibility site using the equiparsimonious (ie, all the trees that have the minimum number of character state changes) ; vi) estimation, by simulations, of the power to detect the association between haplotypes and the disease and comparison with other methods; vii) estimation of the efficiency to localize susceptibility sites and comparison with other methods. ### Haplotype simulation 24 different haplotypes are simulated using the TREEVOLVE software (version 1.32) \[[@B36]\]. An ancestral haplotype sequence of 20 sites is randomly produced, a guide tree is generated using the coalescent model without recombination and the evolution of this ancestral sequence along the tree is simulated to obtain a sample of 24 different haplotypes. Since TREEVOLVE can only simulate the evolution of DNA sequences (4 nucleotides) and not the evolution of bi-allelic markers, we set the frequencies for bases *A*and *T*to very low values (*A*= *T*= 0.001 and *G*= *C*= 0.498) in order to obtain an ancestral sequence composed only of *G*and *C*, as an equivalent of SNPs coded by 0/1 character states. The mutation rate is set to 2 × 10^-6^, and the substitution rates are chosen so that only G-C transversions could occur: *A*- *C*= 0.001, *A*- *G*= 0.001, *A*- *T*= 0.001, *C*- *T*= 0.001, *G*- *T*= 0.001, *G*- *C*= 0.995. No heterogeneity in the substitution rate among sites is specified. The population size was set to 100,000 and its growth was stationary with no subdivision. Once the haplotypes are obtained by simulation, the susceptibility sites are chosen among the 20 sites. For the simulations with only one susceptibility site, the program picks up the first site in the sequence having a minor allele frequency (*p*) in the sample of haplotypes within the range *F*\<*p*\<*F*+ 0.01 where *F*is a predefined allele frequency (in our simulations, *F*= 0.20). For two simulated sites, the program picks up the first two simulated sites verifying a user-defined constraint. In our simulations, the two sites are chosen such that *f*(*A*~1~*B*~1~) = *f*(*A*~1~*B*~2~) = *f*(A~2~B~1~) = *f*(*A*~2~*B*~2~) = 0.25, *A*~1~and *A*~2~being the two alleles at the first site and *B*~1~and *B*~2~, the two alleles at the second site. When no site is found satisfying these conditions, a new set of haplotypes is simulated until the required conditions are obtained. This entire procedure is repeated ten times to generate ten different data sets. The pairwise linkage disequilibrium (*r*^2^measure, mean values for the 1000 replicates) and the tree topologies corresponding to these ten data sets are provided as Additional file [1](#S1){ref-type="supplementary-material"} and [2](#S2){ref-type="supplementary-material"}. ### Attribution of disease status To generate the genotypes of each individual, pairs of haplotypes are sampled at random with replacement and the disease status (case or control) is determined by his genotype at the susceptibility site(s) and by the penetrances. In our simulations, three disease models are studied, involving either one susceptibility site (model 1) or two interacting susceptibility sites (model 2 and model 3). The first model is the disease model described in Bourgain et al. \[[@B37]\] that roughly corresponds to the effect of *ApoE4*in Alzheimer\'s disease \[[@B38]\]. We set the penetrance of the three genotypes *DD*, *Dd*and *dd*to respectively *P*~*DD*~= 0.03, *P*~*Dd*~= 0.06, *P*~*dd*~= 0.30, *d*being the disease susceptibility allele. For model 2 and model 3, an at-risk haplotype formed by the combination of the alleles *A*~2~at the first locus and *B*~2~at the second locus is defined. The penetrance associated to the genotype *A*~2~*B*~2~*A*~2~*B*~2~is set to 0.9 for model 2 and to 0.6 for model 3. All the other penetrances are set to 0.3 for both models. The sampling process is carried on until *N*cases and *N*controls are obtained (*N*= 100 or 200 in our simulations, depending on the number of simulated susceptibility sites). This constitutes a replicate. As the haplotypes are sampled randomly from the set of 24 different haplotype, they are almost equi-frequent in the sample of cases and controls, and their frequencies are close to ![](1471-2156-6-24-i1.gif). ### Phylogeny of haplotypes All the different haplotypes are used to build a phylogenetic tree by a parsimony method which consists in choosing the tree that minimizes the number of mutation events \[[@B39]\]. The PAUP software (version 4.0b10 \[[@B40]\]) is used to reconstruct the tree. A heuristic search is performed, using stepwise addition and TBR (tree bisection and reconnection). All sites are equally weighted, and changes from allele 0 to allele 1 and from allele 1 to allele 0 are equally allowed for all sites, including the susceptibility site (unordered option). Among the list of all parsimonious trees, only the first one obtained is used for the association test, for computation time reasons and because we found that identical results are obtained for more than 98% of the equiparsimonious trees (data not shown). For the localization of susceptibility sites, all the equiparsimonious trees (or up to 100 different trees if there are too many equiparsimonious trees) are analyzed. The trees are rooted using the real ancestor obtained with TREEVOLVE. The character states are inferred along the branches by using the classical option deltran (delayed transformation), but acctran (accelerated transformation) was also investigated. ### Association test Starting from the root of the tree, series of nested homogeneity tests comparing the number of cases and controls in different clades are performed. The principle of the method is explained in Figure [3](#F3){ref-type="fig"}. Briefly, at each level of the tree, homogeneity in the distribution of cases and controls is tested among all the *n*clades defined at this level. If the test is significant, an association is detected and the analysis ends. If the test is not significant, one homogeneity test is performed between all the sub-clades descending from the *n*clades. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Description of the nested clade analysis**. (A) shows the homogeneity test performed at level k (between clades *C*~1~and *C*~2~). If it is not significant (B), a test will be performed at the following level (k+1), between all the sub-clades descending from clades *C*~1~and *C*~2~, i.e between clades *C*~1.1~, *C*~1.2~, *C*~2.1~and *C*~2.2~(3 degree of freedom). If it is significant the analysis ends because an association is detected. ::: ![](1471-2156-6-24-3) ::: P-values of the homogeneity tests are calculated by Pearson chi-square tests except when only two clades are compared in which case a Fisher exact test is performed, and except when more than two clades are tested and the sample size is less than four in a category. In this latter case p-values are estimated by permutation tests. The influence of the LD on the power of the different tests is assessed by using the *r*^2^measure of LD. ### Localization of the susceptibility sites For each replicate, several equiparsimonious trees may exist. We will analyze the *T*first equiparsimonious given by PAUP, *T*being limited to 100 for computation time reasons. For a tree *t*, a new character *S*is allocated to each haplotype *h*. The state of *S*is \"0\", \"1\" or \"?\" depending on the proportion (*p*~*h*~) of cases carrying the haplotype *h*compared to the proportion *p*~0~of cases in the whole sample. if ![](1471-2156-6-24-i2.gif), *S*is coded \"0\" (high number of controls); if ![](1471-2156-6-24-i18.gif), *S*is coded \"1\" (high number of cases); else, *S*is coded \"?\" (missing data). with *n*~*h*~being the number of individuals carrying the haplotype *h*. This character *S*is optimized on the *T*equiparsimonious trees using PAUP and the deltran option. For each site *i*, let ![](1471-2156-6-24-i3.gif) and ![](1471-2156-6-24-i4.gif) be the observed number of times each transition (0→1 for ![](1471-2156-6-24-i5.gif) and 1→0 for ![](1471-2156-6-24-i6.gif)) co-mutates with a 0→1 change of the character *S*on tree *t*. Let ![](1471-2156-6-24-i7.gif) and ![](1471-2156-6-24-i8.gif) be the expected number of co-mutations on tree *t*under the hypothesis of a random distribution of the mutations on tree *t*and an equal probability of mutation on each branch. ![](1471-2156-6-24-i9.gif) where ![](1471-2156-6-24-i10.gif) (resp. ![](1471-2156-6-24-i11.gif)) is the number of 0→1 (resp. 1→0) transitions of the site *i*on tree *t*; *s*~*t*~is the number of 0→1 transition of the character *S*on tree *t*; *b*~*t*~is the number of branches of tree *t*. Then, for each site *i*on tree *t*, we measure the correlated evolution of the site *i*and the character *S*, by defining ![](1471-2156-6-24-i12.gif) (resp. ![](1471-2156-6-24-i13.gif)) as follow: ![](1471-2156-6-24-i14.gif) then, for the *T*equiparcimonious trees, we define ![](1471-2156-6-24-i15.gif) (resp. ![](1471-2156-6-24-i16.gif)) as follow: ![](1471-2156-6-24-i17.gif) Finally, *V*~*i*~is defined as the max between ![](1471-2156-6-24-i15.gif) and ![](1471-2156-6-24-i16.gif). The site or the two sites corresponding to the highest *V*~*i*~are selected as putative susceptibility sites ### Estimation of power and efficiency For each of the ten samples of 24 haplotypes simulated along the ten different tree topologies, the process detailed above (attribution of disease status, phylogenetic reconstruction, association test and localization of the susceptibility site) is repeated 1000 times to obtain 1000 replicates, each of them including *N*cases and *N*controls. The power to detect the association is measured by counting the proportion of replicates in which a significant heterogeneity in the distribution of cases and controls is detected between clades. The efficiency of CT to localize the true susceptibility site is simply evaluated as the number of times, among the 1000 simulations, the true susceptibility site is among the detected sites. When two susceptibility sites are simulated, we evaluate the number of times the two sites are detected and the number of times at least one of the two sites is detected. The test using the cladistic method (CT) is compared to three other association tests: i) a haplotypic test (HT) that compares the haplotype distributions in cases and controls: if *h*different types of haplotypes are observed, a chi-square test with *h*- 1 degree of freedom (df) is performed. As with the CT test, when sample sizes are small, a permutation procedure is used to perform the chi-square test. This HT test is only used to test for association and not to localize the susceptibility sites. ii) An allelic test, referred to as Site by Site Test (SbST), that compares the allele distributions in cases and controls at each site by a chi-square: if the number of sites is *s*, then *s*chi-squares, each with one df, are performed. The site or the two sites that give the most significant results are proposed as susceptibility sites. iii) A test based on CLADHC, the program developed by Durrant et al. \[[@B29]\]. In this test, the data set is analyzed by overlapping windows sliding along the haplotypes. In each window, a tree is reconstructed using a distance method, and a regression analysis is performed at each level of the tree (nested analysis). The level of the tree that maximizes the evidence of disease marker association in the likelihood-ratio test is selected. For each window, the program also provides a significant threshold calculated using the Bonferroni correction. In our simulations, a window size of 6 is selected. We consider the association as detected when at least one of the windows is significant. Concerning the localization of the susceptibility site, CLADHC can only find a window containing the susceptibility site, but not the susceptibility site itself. Therefore, we only use CLADHC in the case of one susceptibility site and we consider the localization as correct when the window with the maximum value of the statistic contains the susceptibility site. For CT, SbST and CLADHC, we are confronted to a problem of multiple testing. When a real data set is analyzed, a permutation procedure can be used to estimate p-values of the association tests that are corrected for multiple testing. This procedure is very time-consuming and cannot be used on the simulations we have performed here. As an example, the 100,000 repetitions for the Crohn data run in about 24 hours on a Pentium III, 930 MHz, 512 Mo of RAM. Therefore, for SbST and CT, we estimate the nominal type I error required to obtain a global type-one error of 5% by a simulation under the null hypothesis of no association between the studied sites and the disease (5000 replicates). We found that for the different simulation conditions studied here, these nominal errors are equal to 1% for CT and varies from 0.2 to 0.4% for SbST. For the analysis with CLADHC, the program provides a significant threshold calculated using Bonferroni correction. As the Bonferroni correction is very conservative, to obtain similar type I errors with CLADHC as with CT, nominal type I errors were set to values varying from 15 to 17% for simulations with one susceptibility site and from 20 to 25% for simulations with two susceptibility sites. Analysis of real data on Crohn disease -------------------------------------- The data set is described in a paper of Hugot et al. (2001) \[[@B30]\] and includes 232 nuclear families with two affected children and their parents genotyped for 13 SNPs in the CARD15/NOD2 gene. Haplotypes were reconstructed in these families using GENEHUNTER 2.0b\[[@B41]\]. In each family, one affected child is selected at random to create the case sample (his two haplotypes are labeled cases). The control sample is formed with the parental haplotypes non transmitted to the selected children. The phylogenetic reconstruction assumes that no recombination has occurred between the SNPs. To determine if this was true for the different SNPs tested here or at least for a subset of them, linkage disequilibrium (LD) is estimated between all the SNPs. From the pairwise LD, a group of 7 SNPs (5, 6, 7, 8, 12, 9 and 13) can be identified as forming a block in strong LD among which no or very few recombinations are likely to have occurred (see [Additional file 3](#S3){ref-type="supplementary-material"}). Interestingly, this same block of 7 SNPs was described by Vermeire et al. \[[@B42]\] in the CARD15/NOD2 gene. Only these 7 SNPs that form 33 different haplotypes are used in the cladistic analysis. The phylogenetic trees are reconstructed using the same parameters as in the simulations. However here, contrary to the simulations, the ancestral sequence being not known, two different analysis are performed using as ancestral sequence either the most frequently observed haplotype in our sample (mfH) or a consensus sequence formed by the combination of the most frequent alleles at each SNP (consH). For the association test, p-values are estimated on the first equiparsimonious tree obtained by PAUP, using an algorithm similar to the one described in a paper of Becker and Knapp (2004) \[[@B43]\]. This permutation process corrects for multiple testing in a lesser conservative way than the Bonferroni correction because it takes into account the dependency between the tests. For the localization, among all the equiparsimonious trees, 1000 of them are retained and analyzed with CT. Authors\' contributions ======================= CB perfomed the simulation analyses, contributed to the conception of the study, to the elaboration of the algorithm and wrote the manuscript. VD contributed to the elaboration of the algorithm. J-PH provided the CARD15 data. PD and EG contributed to the design and to the conception of the study and to the manuscript preparation. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 **Average pairwise linkage disequilibrium for the 10 simulated date sets**. The average pairwise LD (over 1000 replicated) is calculated using the *r*^2^measure for each of the 10 tree topologies (T1 to T10). The matrices are plotted with GOLD\[[@B44]\]. ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 2 Tree topologies of the 10 simulated date sets ::: ::: {.caption} ###### Click here for file ::: ::: {.caption} ###### Additional File 3 **Pairwise linkage disequilibrium for the Crohn data**. (A) shows the pairwise linkage disequilibrium (LD) calculated with the *D*\' measure. (B) shows the LD calculated with the *r*^2^measures. The matrices are plotted with GOLD \[[@B44]\]. A high degree of LD can be observed. The differences between the *r*^2^and the D\' values are explained by the difference in the allelic frequency for the different SNPs. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We wish to thank Marie-Claude Babron and two anonymous reviewers for helpful comments and Arnaud Legrand for letting us use one of his perl function. We are also grateful to Habib Zouali and Suzanne Lepage from the fondation Jean Dausset for providing us with Crohn data and to Caroline Durrant for kindly providing us with the software CLADHC. Many thanks to the runtime team from the LaBRI for letting us run a part of our simulations on their cluster.

PubMed Central

2024-06-05T03:55:58.366282

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173100/", "journal": "BMC Genet. 2005 May 18; 6:24", "authors": [ { "first": "Claire", "last": "Bardel" }, { "first": "Vincent", "last": "Danjean" }, { "first": "Jean-Pierre", "last": "Hugot" }, { "first": "Pierre", "last": "Darlu" }, { "first": "Emmanuelle", "last": "Génin" } ] }

PMC1173101

Background ========== N-acetylation polymorphism is one of the earliest discovered and most intensively studied pharmacogenetic traits that underlie interindividual and interethnic differences in response to xenobiotics. In humans, acetylation is a major route of biotransformation for many arylamine and hydrazine drugs, as well as for a number of toxins and known carcinogens present in the diet, cigarette smoke and the environment \[[@B1]-[@B3]\]. Genetically determined differences in N-acetylation capacity have been proved to be important determinants of both the effectiveness of therapeutic response and the development of adverse drug reactions and toxicity during drug treatment \[[@B4]\]. In the last decades, numerous investigations have been made to elucidate the genetic basis of N-acetylation polymorphism in various ethnic groups in order to develop efficient genotyping tests and to adapt therapies to specific patients and populations in accordance with their genetic makeup. Some of the drugs excreted by acetylation are indeed crucial in the treatment of diseases representing a worldwide concern, such as tuberculosis and AIDS-related complex diseases \[[@B5],[@B6]\]. Moreover, a number of epidemiological studies have suggested possible associations between the N-acetylator phenotype and a variety of complex human diseases, the most consistent findings being those regarding urinary bladder cancer and familial Parkinson\'s disease \[[@B7]-[@B10]\]. The gene coding for the arylamine N-acetyltransferase 2 (*NAT2*) enzyme has been established as the site of the classic human acetylation polymorphism \[[@B11]-[@B13]\] and the molecular basis of individual and interethnic variation in acetylation capacity is now well documented \[[@B14],[@B15]\]. All mutations reported to date are found within the 870-bp coding region of the *NAT2*gene. Among the seven single nucleotide polymorphisms (SNPs) that are commonly found in human populations, four result in an amino acid substitution that leads to a significant decrease in acetylation capacity (single base-pair substitutions at positions 191, 341, 590, 857). The other three are either silent mutations (C282T, C481T) or a non-synonymous substitution that does not alter phenotype (A803G). In the consensus gene nomenclature of human *NAT2*that encompasses all currently recognized alleles \[[@B16],[@B17]\], sets of SNPs located throughout the coding region are linked in terms of haplotypes, that is they are organized as they segregate together on one individual\'s chromosome at the *NAT2*locus. Each combination of SNPs identified so far constitutes a distinct haplotype that is treated as an allele of the haplotype system. The consideration of multilocus haplotypes seems more desirable since there is growing evidence that for genes containing multiple SNPs in high linkage disequilibrium (LD) such as *NAT2*\[[@B18]\], haplotype structure rather than individual SNPs can be the principal determinant of phenotypic consequences \[[@B19]-[@B21]\]. A functional polypeptide is indeed the product of a haplotype, covering the entire coding region and coded by a single chromosome. The *NAT2*alleles described so far contain up to four of the acknowledged mutations in various combinations. Each allele is associated with an acetylator phenotype depending on which mutations they contain: for instance, substitutions at positions 191, 341, 590, and 857 are diagnostic for defective *NAT2*function and hence for the slow acetylator phenotype (Table [1](#T1){ref-type="table"}). Three *NAT2*phenotypes have been described: subjects with two low activity alleles are classified as slow acetylators, while those with two functional alleles are considered rapid acetylators. If only one allele is of the slow type, an intermediate phenotype is observed. Many early studies did not distinguish between fast and intermediate acetylators, categorizing both types of subjects as fast acetylators. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### The major human *NAT2*alleles and their associated phenotype^a^. ::: Allele Nucleotide change ^b^ Phenotype ----------------- ----------------------- ----------- ------- --- ------- --- ------- ------- *NAT2\*4* rapid ***NAT2\*5A*** **x** x slow ***NAT2\*5B*** **x** x x slow ***NAT2\*5C*** **x** x slow ***NAT2\*6A*** x **x** slow ***NAT2\*6B*** **x** slow ***NAT2\*7A*** **x** slow ***NAT2\*7B*** x **x** slow *NAT2\*12A* x rapid *NAT2\*12B* x x rapid *NAT2\*13* x rapid ***NAT2\*14A*** **x** slow ***NAT2\*14B*** **x** x slow Nucleotide substitutions shown in bold have a functional consequence on enzyme activity. Bold-faced alleles contain functional polymorphisms and are hence associated with the slow acetylator phenotype. Classification of *NAT2*alleles into different clusters is based on the most functionally significant nucleotide substitution present: the *NAT2\*5*, *NAT2\*6*, *NAT2\*7*and *NAT2\*14*clusters possess signature nucleotide substitutions at positions 341, 590, 857 and 191, respectively and are hence all decreased function alleles (\'slow alleles\'). The others display enzymatic activity comparable to the rapid acetylator allele *NAT2\*4*. There are significant interethnic differences in *NAT2*allele distribution and frequency \[14, 15\]. ^a^Adapted from Hein et al. \[3\]. *NAT2*nomenclature is accessible on the internet at website <http://www.louisville.edu/medschool/pharmacology/NAT.html> ^b^Only changes from the reference sequence (*NAT2\*4*) are indicated. ::: Problems may occur when individual multi-site *NAT2*genotypes have to be assigned correctly to a particular combination of two multilocus haplotypes. Indeed, current routine genotyping and sequencing methods typically do not provide haplotype information in diploid organisms such as humans, and the gametic phase of haplotypes is inherently ambiguous when individuals are heterozygous at more than one locus. As illustrated in Figure [1](#F1){ref-type="fig"}, a subject carrying two inactivating mutations can be either rapid or slow acetylator depending on whether these mutations are located in the same or different chromosome, respectively. It is thus crucial to unequivocally assess mutation linkage patterns, this step being a prerequisite to obtain accurate haplotype frequency estimates in populations and reliable genotype-phenotype predictions. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The ambiguous gametic phase of haplotypes for a given multilocus genotype. To illustrate the relevance of linkage phase ascertainment, let us consider the following case of a four-site heterozygous individual at positions 191, 341, 481 and 803 within the *NAT2*coding sequence. Eight possible combinations of haplotypes can be inferred from this multilocus genotype, two of whom are shown here. Depending on the location of mutations to either DNA strand, the individual\'s *NAT2*genotype composed of two multilocus haplotypes will not be the same. Moreover, an incorrect resolution of mutation linkage patterns may entail an error in individual phenotype prediction: the subject will be classified either as a slow or as a rapid acetylator depending on the haplotypic combination chosen. Symbol (**\***) points at mutations leading to a decrease in *NAT2*enzyme activity, while symbol (**×**) indicates those with no impact on the acetylator phenotype. ::: ![](1471-2156-6-30-1) ::: However, in spite of its high relevance, this issue has not been handled properly by most past studies investigating *NAT2*polymorphisms. Early genotyping studies only screened for the presence of three polymorphisms (C481T, G590A, G857A), and a subject was defined as a slow acetylator if he was homozygous for one, or heterozygous for two (each located on one DNA strand), of this three nucleotide changes. Such a definition assumed that there could be no single allele with two or more of the tested mutations. In other studies that screened a larger number of SNPs within *NAT2*, patterns of LD between point mutations were often assumed, in reference to the haplotypes previously described and which are commonly found in populations of European origin. For instance, the designation of *NAT2*alleles is usually based on the assumption that 481T and 803G are strongly linked to 341T, and 590A and 857A are linked to 282T \[[@B22]\]. However, in rare cases the typical allelic linkage pattern of mutations may be disrupted because of genetic recombination and this may result in misclassification of alleles. Indeed, although such assumed linkage patterns are very strong, other allelic variants carrying either unusual combinations of mutations or mutations in isolation have been described in a few cases \[[@B23]\]. Furthermore, the designation of *NAT2*alleles in such a way that it necessarily conforms to the existing consensus nomenclature of acknowledged haplotypes precludes the disclosure of unexpected combinations of mutations, different from the established allelic variants, and hence, the discovery of new alleles. Such a manner of inferring haplotypes from unphased multilocus genotypes may introduce biases in *NAT2*allele designation and individual phenotype prediction, and these potential biases are of higher magnitude when non-European populations are concerned. Most studies assumed particular patterns of linkage previously described in populations of European origin, but which may not hold in other ethnic groups. Indeed, recent works have shown that patterns of LD can differ markedly among populations with different ethnic and demographic backgrounds. As an example, Loktionov and colleagues \[[@B24]\] pointed out the high occurence of isolated mutations 803G and 282T (defining alleles *NAT2\*12A*and *NAT2\*13*, respectively) in Black South Africans, while these nucleotide changes are almost always tightly linked to other mutations in European populations \[[@B25]\]. Likewise, Dandara *et al.*\[[@B26]\] recently identified a novel mutation linkage pattern (*NAT2\*6E*) that appeared to be common in three African populations and that had not yet been reported in Europeans. As well, Anitha *et al.*\[[@B27]\] revealed a new combination of acknowledged mutations (*NAT2\*5G*) in the Malapandaram tribe of South India that has not been described so far in any other world population. The genotype-phenotype discordance observed in many ethnic groups where mutation linkages have not been extensively proven experimentally might result from such unexpected compound alleles. It is thus necessary to systematically verify the postulated allelic combinations. To avoid such potential problems, many authors designed special analytical techniques to unequivocally determine the allocation of mutations to either DNA strand. Molecular methods using combinations of mutation-specific polymerase chain reaction (PCR) reamplification coupled to restriction mapping of the PCR products have been developed; these allow the separate analysis of each allele in order to obtain a complete map of both genes in every individual. Some studies applied these procedures to all multiply heterozygous subjects \[[@B24],[@B28]-[@B31]\], while others limited their application to particular cases as those where an alternative linkage pattern of mutations would lead to a change in phenotype \[[@B5],[@B27],[@B32]-[@B36]\]. However, these experimental methods of molecular haplotyping are not entirely satisfying because they entail an additional cost and are currently labour-intensive, time-consuming, prone to experimental errors and difficult to automate. Therefore, they do not appear to be good candidates for routine clinical applications and for a generalization at a large scale. A cheap and relatively straightforward alternative for haplotype reconstruction based on genotype data from unrelated individuals is the use of computational algorithms. The most widely used algorithms developed so far are based either on a parsimony, a maximum-likelihood, or a Bayesian approach (see \[[@B37]\] for a review). In the last decade, numerous investigations based on empirical data and extensive simulation studies have demonstrated that such *in silico*haplotype-inference methods could give effective and accurate prediction of haplotype phases, especially in regions with high LD values between polymorphic sites and small probabilities of recombination events \[[@B18],[@B38],[@B39]\]. Therefore, they could be fairly efficient alternatives to molecular-haplotyping methods when applied to *NAT2*gene data. Surprisingly, to our knowledge, only three studies have used computational methods to reconstruct *NAT2*haplotypes and estimate allele frequencies in population samples \[[@B40]-[@B42]\]. One explanation for such limited use may be the lack of evidence documenting the performance of *in silico*approaches when applied to actual *NAT2*data. Indeed, the accuracy of these strategies, compared with molecular methods, needs to be assessed before their applications can be advocated at a large scale. A recent study provided preliminary results on this issue: Xu and colleagues \[[@B18]\] empirically evaluated and compared the accuracy of the Clark\'s algorithm \[[@B43]\], the expectation-maximisation (EM) algorithm and a Bayesian method implemented in the PHASE program \[[@B44]\] in phase inference at *NAT2*, taken as an example of a locus with pronounced LD over a 850-bp region. In this study, *NAT2*haplotypes (consisting of five genotyping SNPs at position 282, 341, 481, 590, and 803 nt) were experimentally determined through cloning and sequencing in 81 individuals of European ancestry. They found that all three computational methods provided remarkably accurate and reliable estimates for *NAT2*haplotype frequencies and individual haplotype phases. The objective of the present study was to extend this investigation to more precisely assess the performance of the computational approach. We conducted an extensive study based on experimental data from a larger number of samples, issued from populations of various ethnic origin, and tested for haplotypes involving the seven major polymorphic loci of *NAT2*. Furthermore, the larger population samples investigated are of greater significance: as the sample size grows, there is more opportunity to observe rare haplotypes that are the most difficult to infer statistically. This comparative study is designed to evaluate the performance of different haplotyping algorithms and to assess the consistency of their estimates. In addition, it provides information on the impact of various data-set characteristics (sample size, haplotype frequency distribution, haplotype frequencies, deviation from Hardy-Weinberg (HW) equilibrium, \...) on estimation accuracy; we then explore the utility of data-based diagnostics for assessing probable accuracy. Results ======= Molecular haplotyping of the *NAT2*locus revealed between eight and twelve distinct haplotypes in each of the five population samples investigated. The theoretical maximum number of haplotypes for a set of seven biallelic variable sites is 128 (2^7^) if there is random association between polymorphic sites, whereas it is only 8 in the absence of recombination, recurrent and back mutation. Hence, the small number of haplotypes observed at *NAT2*suggests strong LD over the short physical distance spanning this gene. Indeed, we observed complete or near complete LD for all pairs of SNPs with sufficiently high frequencies (only alleles with frequencies in the range 0.05--0.95 were included in the analysis because estimates of LD for low-frequency alleles in small samples are not informative): 85% of all pairwise r^2^values were highly significant (Exact *p*-value \<0.0001). Although LD patterns were rather similar among the different population samples, substantial differences in LD levels were observed (Figure [2](#F2){ref-type="fig"}): the Korean sample, and especially the South African sample, displayed much smaller values of average pairwise r^2^(0.27 and 0.20, respectively) than the two European and the Nicaraguan samples (values between 0.39 and 0.57), for which a strong haplotypic structure was observed. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Linkage disequilibrium (r^2^value) between SNP markers in the *NAT2*locus. Graphical representation of the disequilibrium matrices obtained through computation of the r^2^coefficient between each pair of markers, for the Spanish, Korean and Black South African samples. The British and Nicaraguan samples provided patterns and levels of LD comparable to those of the Spanish data. For each marker pair, GOLD \[60\] plotted the color-coded pairwise r^2^statistics at the Cartesian coordinates corresponding to marker location, and the plots were completed by interpolation. These graphs point out the strong level of LD between markers at positions 341, 481, and 803, as well as between SNPs located at 282 and 590: these markers are thus strongly predictive of one another. In Black South Africans, LD patterns are less pronounced and more diffuse across marker pairs. ::: ![](1471-2156-6-30-2) ::: Among the 1608 individuals investigated over the five data sets, 45.5% (732/1608) were either homozygous for all SNP sites or heterozygous at only one SNP site; thus, their haplotype pairs could be assigned directly. Besides, 35.7% (574), 10.0% (160), 0.9% (15), and 7.9% (127) individuals were heterozygous at two, three, four and five SNP sites, respectively. We inferred their haplotype phases with four computational haplotyping methods, and compared the results with those obtained through molecular haplotyping. Since the Hapar program often provided several equally parsimonious solutions for a given multilocus genotype, it could not resolve a relatively large fraction of heterozygous individuals in each sample and hence, we could not deduce frequency estimates for the haplotypes observed. Therefore, we evaluated Hapar only on its ability to identify the set of haplotypes present in a sample. Haplotype identification ------------------------ Hapar found for each sample the smallest set of haplotypes that could explain the genotype data, and PL-EM, Haplotyper and PHASE provided the list of all the haplotypes selected to appear in at least one of the subjects in the \"best\" reconstruction, that is when the most likely pair of haplotypes is selected for each individual. The *I*~*H*~indices of the four programs are displayed for each population sample in Table [3](#T3){ref-type="table"}. For the British and Korean samples, all computational methods identified exactly the same haplotypes as those determined experimentally. In contrast, in the other three samples, the algorithms sometimes inferred one additional haplotype, that was not actually present, and/or missed one haplotype that was shown to be present by means of molecular haplotyping. Nevertheless, these prediction errors always concerned rare haplotypes of frequency \<0.75% (singletons in most cases). The PHASE algorithm performed the best. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Performance of the four computational methods in haplotype identification, as measured by the *I*~*H*~index. ::: Hapar PL-EM Haplotyper PHASE --------------------------------- ------------- ------------- ------------- ------------- 258 Spanish \[45\] 0.933 *(1)* 0.933 *(1)* 0.933 *(1)* 0.933 *(1)* 137 Nicaraguans \[30\] 0.952 *(1)* 0.952 *(1)* 0.909 *(2)* 0.952 *(1)* 112 UK Caucasians \[24\] 1 1 1 1 101 Black South Africans \[24\] 0.917 *(2)* 0.917 *(2)* 0.917 *(2)* 1 1000 Koreans \[31\] \- \* 1 1 1 Numbers in brackets indicate the number of haplotypes for which an error of prediction was made. \* The size of the Korean sample was too large to be correctly handled by the Hapar program. ::: Prediction of individual haplotype phases ----------------------------------------- We also evaluated and compared the effectiveness of the computational methods in reconstructing haplotype pairs for individuals. Table [4](#T4){ref-type="table"} gives, for each data set and for each algorithm, the individual error rate. Whatever the method, the number of incorrectly reconstructed individuals was remarkably low, with error rates always under 4%. The largest number of mistakes were observed for the African sample and, among the three algorithms tested, PHASE yielded the lowest error rates. Furthermore, it is interesting to note that, in all cases of incorrectly predicted phases, there was no impact on phenotype prediction. Thus, despite these errors, the proportions of slow, intermediate and rapid acetylators in each population sample were in 100% agreement with those deduced from molecular haplotyping. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Individual error rate in haplotype reconstruction ::: PL-EM Haplotyper PHASE --------------------------------- ------- ------------ ------- 258 Spanish \[45\] 0.39% 0.39% 0.39% 137 Nicaraguans \[30\] 2.19% 3.65% 2.19% 112 UK Caucasians \[24\] 0.89% 0% 0% 101 Black South Africans \[24\] 3.96% 3.96% 2.97% 1000 Koreans \[31\] 0.30% 0.30% 0.30% The individual error rate is defined as the ratio of erroneous phase calls to the total number of phase calls (see text). ::: Estimation of haplotype frequencies ----------------------------------- A comparison of the haplotype frequencies determined molecularly with those estimated computationally showed very high concordance. Both PL-EM and PHASE methods provided similarity index (*I*~*F*~) values very close to the maximal value of 1 in all investigated data sets (Table [5](#T5){ref-type="table"}). Such high values may be explained by the fact that the *I*~*F*~index gives more weight to common haplotypes whose frequencies are the most accurately estimated by computational algorithms. To investigate the effect of haplotype frequency on estimation accuracy, we plotted the change coefficient (*C*) against the larger of the two haplotype frequencies (Max \[![](1471-2156-6-30-i1.gif), *p*~0*i*~\]), for all possible haplotypes with nonzero frequency estimates determined by either analysis in any of the five population samples. As shown in Figure [3](#F3){ref-type="fig"}, substantial percentage changes \>30% occur only at the lowest haplotype frequencies (\<0.007). Even the moderate changes (range 10%--30%) occur only when the frequency estimates are \<0.03, and any change in percentage value \>5% concerns only haplotype frequencies \<0.035. Two-thirds (67%) of the haplotype frequency estimates showed either no change or a small change (\< 3%). In Table [6](#T6){ref-type="table"}, we compared the relative estimation accuracy of PL-EM and PHASE programs by computing the average change coefficient for three classes of haplotype frequencies. The two methods perform similarly for haplotypes with frequencies \>0.05, whereas PHASE provides more accurate estimates when rarer haplotypes are concerned. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Index of similarity (*I*~*F*~) between haplotype frequencies estimated with and without molecular haplotyping information. ::: PL-EM PHASE --------------------------------- ------- ------- 258 Spanish \[45\] 0.996 0.996 137 Nicaraguans \[30\] 0.986 0.986 112 UK Caucasians \[24\] 0.994 0.998 101 Black South Africans \[24\] 0.981 0.988 1000 Koreans \[31\] 0.997 0.998 Computations of *I*~*F*~indices were based on the haplotype frequency estimates obtained by considering all possible haplotype configurations (with a nonzero probability) inferred for each subject, weighted by their estimated probability. Note that such estimates can be rather different from those obtained by gene counting on the basis of the \"best\" reconstruction (that is, when only the most probable pair of haplotype is selected for each sampled individual). Since Haplotyper only provides a summary of the frequency with which each haplotype occurred in the \"best\" reconstruction, it was excluded from the comparison. ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### The change coefficient (*C*) as a function of haplotype frequency. The change coefficient reflects the discrepancy between haplotype frequencies deduced from phase-known data and those estimated computationally (here with the PHASE program). All haplotypes occurring in any of the five population samples are considered. ::: ![](1471-2156-6-30-3) ::: ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Average change coefficients of PL-EM and PHASE programs computed for three classes of haplotype frequency. ::: Haplotype frequency ------- --------------------- ------ ------ PL-EM 30.6% 8.3% 1.2% PHASE 17.4% 6.8% 1.2% ::: Partially resolved data sets ---------------------------- We also performed similar analyses on six other previously published data sets, in which linkage phase patterns were only partially resolved by molecular haplotyping. These data concerned 844 German \[[@B32]\], 248 Polish \[[@B33]\], 303 Turkish \[[@B34]\], 50 non-caste Dogons from Mali, 52 Gabonese and 60 Caucasians \[[@B5]\]. Haplotype phase information was available for 41%--74% individuals in these six population samples (including phase-resolved genotypes as well as non ambiguous homozygous or simply heterozygous genotypes). The PHASE algorithm was applied on the unphased genotype multilocus data of each of these samples, and a 100% concordance was observed between individual haplotype phase reconstruction through the computational method and the empirically determined linkage patterns, for all investigated data sets (data not shown). This means that, despite the efforts invested, in terms of work, time and money, to resolve mutation linkage phase in a part of each sample, no more information was added by molecular haplotyping than what could be extracted from computational algorithms applied to these data. Discussion ========== This empirical study demonstrates how closely the frequencies computationally estimated from phase-unknown data approximate those from gene-counting estimates based on phase-known data. In the particular case of the *NAT2*gene, where there is near complete LD between SNPs within the coding region, all *in silico*approaches provided highly effective and accurate estimates for haplotype frequencies and individual haplotype phases. Estimated frequencies of common haplotypes were nearly identical to those empirically determined, whereas rare haplotypes were occasionally miscalled when their presence/absence had to be inferred. As already pointed out by Stephens *et al.*\[[@B44]\] and Lin *et al.*\[[@B39]\] and confirmed in this study, lower-frequency variants are less easily estimated statistically; indeed, there is less contextual information about phase for singletons versus nonsingletons. Thus, for those research questions for which the *NAT2*common haplotypes are most important, frequency estimates based on the unphased SNP-typing results from unrelated individuals will be sufficient. However, accurate identification of rare haplotypes may be critical for many researchers, such as population geneticists interested in detecting features of recent demographic history that are population-specific or signatures of selective effects in *NAT2*sequences; as well as for epidemiologists and clinicians concerned with the possibility that rare haplotypes may be important for disease risk or for predicting drug response. In such cases, molecular haplotyping will be necessary to determine linkage phase unambiguously \[[@B57]\]. For a locus such as *NAT2*where a strong haplotypic structure is observed, all algorithms provided highly effective and accurate results for haplotype reconstruction. Thus, such \"ideal\" data for statistical inference did not permit to properly discriminate between the different methods investigated. Nevertheless, despite roughly similar performances, slightly better results were observed with the PHASE program. In particular, PHASE outperformed the other programs when frequencies of rare haplotypes have to be inferred. This is consistent with the results of some previous studies which evaluated and compared the performance of several algorithms on both empirical and simulated data \[[@B44],[@B54],[@B61]\]. PHASE provided the most accurate reconstructions, probably because the true haplotypes conformed more closely to the assumptions of the approximate coalescent prior than to those of the Dirichlet prior. Many factors may influence the estimation accuracy of computational approaches. They can be assessed empirically within a dataset, to be further used as \"diagnostics\" for predicting potential inaccuracies in estimation caused by features in the relevant data set \[[@B38]\]. Sample size did not appear to have a large effect on the haplotype frequency estimates comparing phase-known and phase-unknown results for the five data sets included in this study. Perhaps the low error rate observed in Koreans is partly due to the huge size of this sample (1000 individuals): an improvement in accuracy of the estimation procedure with increased sample size is indeed expected since information redundancy in the form of multiple copies of the same haplotype in the data set is required for the statistical algorithms to work properly \[[@B38],[@B48]\]. On the other hand, computational methods may also perform best in small samples, in which there is little chance to observe rare haplotypes that are the most difficult to infer statistically. Nevertheless, since the number of new haplotypes is not expected to increase linearly with sample size, the analysis of sufficiently large samples should guarantee a good reliability in the resulting estimates. Although most of the tested algorithms assume HW equilibrium, significant departures from HW proportions did not seem to have any impact on the accuracy of their predictions. HW equilibrium holds well for the Nicaraguan, UK Caucasian, and Black South African samples (nonsignificant results), whereas the genotype distribution in the Spanish and Korean samples show significant departures from HW proportions (Table [2](#T2){ref-type="table"}). The data sets investigated are not the most suitable to evaluate the effect of a deviation from HW equilibrium: significant results for tests of HW ratios are quite close to the threshold value (5%), and in the case of the Korean sample, an excess homozygosity is observed, which should not compromise the algorithm performance. Indeed, in such a case, there is a balance between loss of accuracy caused by violation of HW equilibrium and gain of accuracy caused by the decrease in missing phase information through an excess of homozygotes \[[@B38],[@B62]\]. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### The five phase-resolved *NAT2*molecular data sets investigated. ::: **Population sample** **Proportion of phase-unknown multiple heterozygotes^a^** ***NAT2*gene diversity^b^** **Deviation from the Hardy-Weinberg equilibrium (exact p-value)^c^** --------------------------------- ----------------------------------------------------------- ----------------------------- ---------------------------------------------------------------------- 258 Spanish \[45\] 66.7% 0.65 **0.012** 137 Nicaraguans \[30\] 59.1% 0.70 0.072 112 UK Caucasians \[24\] 52.7% 0.69 0.222 101 Black South Africans \[24\] 63.4% 0.86 0.122 1000 Koreans \[31\] 50.0% 0.52 **0.016** All population samples were genotyped for the same seven nucleotide changes (G191A, C282T, T341C, C481T, G590A, A803G, G857A), except Koreans where the C190T mutation was investigated instead of G191A. ^a^Proportion of multiply heterozygous individuals with ambiguous genotype whose phase has been resolved molecularly. ^b^Expected heterozygosities for the *NAT2*haplotyped system were estimated as where n is the number of gene copies in the sample, and pi is the sample frequency of the i-th haplotype. ^c^The significance of deviations from Hardy-Weinberg equilibrium was tested for genotypic data with known gametic phase using the random-permutation procedure implemented in the Arlequin package \[51\]: a Fisher\'s exact test using a Markov chain random walk algorithm was performed for each data set. The resulting p-values were considered significant if inferior to 0.05 (significant p-values are shown in bold). ::: Among the five data sets investigated in this study, Black South Africans displayed the highest error rate in haplotype computational inference. One possible explanation may be the presence in this sample of a large number of different multiple heterozygotes with ambiguous multilocus genotypes, occurring at roughly similar frequencies (this is reflected in the high *NAT2*gene diversity displayed by this population sample (Table [2](#T2){ref-type="table"})). Indeed, both the number of different ambiguous multiple heterozygous genotypes and their relative frequencies have been shown to be of high importance in the assessment of haplotype estimation accuracy \[[@B57]\]: both would be good indicators of the difficulty level of a given data set for haplotyping algorithms. The existence of many different multilocus genotypes uniformly distributed implies that many different haplotypes occur at low frequency, and that, consequently, greatest error and uncertainty occur in the estimation of haplotype frequencies (since no single haplotype is overwelmingly frequent). In contrast, the presence of a small number of multiple heterozygous genotypes at proportionately high frequencies implies that some individual haplotypes exist at high frequencies, and the estimation of those haplotype frequencies will be easier and accomplished with greater accuracy \[[@B38],[@B57]\]. In such cases, molecular haplotyping may add little information for the resolution of haplotype phases. The amount of LD between SNP markers may be another determining factor for the prediction of estimation reliability since when multiple polymorphic sites display little disequilibrium, as was observed in the African sample compared to the others, a large proportion of the chromosomes may occur as uncommon or rare haplotypes, implying a greater difficulty level in the statistical inference of haplotypes. Therefore, we advocate to examine beforehand the unphased *NAT2*genotype data for both the frequency distribution of multiply heterozygous genotypes and the level of LD between polymorphic markers; this will allow to assess the difficulty level displayed by the data set for statistical inference, and hence, to predict the ability and accuracy with which computational algorithms would infer haplotype phases from such data. Of course, statistical methods can be used in conjunction with experimental methods to provide more accurate estimates of individual haplotypes. It has been claimed that the ability of certain computational methods to accurately assess the uncertainty associated with each phase call gives them the substantial practical advantage of allowing experimental effort to be directed at sites and/or individuals whose phases are most difficult to reconstruct statistically or that are critical to the conclusions of the study \[[@B20],[@B44],[@B61]\]. However, in our study, we observed that most erroneous phase calls inferred at the individual level were strongly supported, with a probability close to the maximal value of 1. Thus, these errors could not have been avoided since they would not have been selected for the molecular targeting. This stresses why, in the case of the discovery of a novel *NAT2*allele through computational haplotyping methods, the unusual linkage pattern should always be confirmed by cloning and sequencing the allele under question, as advocated by Cascorbi and Roots \[[@B25]\] for novel allelic combinations detected by molecular techniques. Throughout this study, we assumed that the *NAT2*linkage patterns molecularly determined were the \"true\" ones, and hence, that there was no error in the haplotype assigments based on experimental methods. However, molecular techniques may have experimental error rates as high as the rate of statistical error associated with the computational haplotype determination algorithms \[[@B19]\]. Indeed, molecular haplotyping bears the risk of false positive or false negative allele-specific amplification (because of the nucleotide-dependent specificity of that technique) as well as uncomplete or non-specific digestions with the enzymes used in restriction analyses \[[@B25]\]. In the present study, we have estimated the computational error rates to be of no more than 3--4% for all investigated algorithms. This is not higher than the corresponding error rate from molecular haplotyping techniques, on the order of 2--3% \[[@B20]\]. Therefore, it is difficult to determine whether the discrepancies observed between experimental and computational estimates are actually due to statistical errors from algorithms; they may be due to technical errors during manipulations and the molecular data used as a reference for comparisons might be wrong. The disadvantage of *in silico*approaches is that algorithmic techniques are statistical and require the analysis of a population rather than a single individual. This is not a limitation in clinical trials and epidemiological surveys, which are always performed on a cohort basis. In clinical pharmacy, however, if a specific individual\'s haplotypes are of interest to predict his response to drug treatment, his unphased multilocus genotype must be combined with a standard reference set of haplotypes to infer the phasing \[[@B20]\]. This implies a thorough knowledge of the *NAT2*genotypic distribution in the ethnic population from which this individual was drawn. Conclusion ========== This study demonstrates that computational methods can provide an effective and accurate prediction of haplotype phases, in the particular case of the *NAT2*gene which displays high values of LD between polymorphic sites. The objective of this study is not to advocate the systematic use of computational approaches for *NAT2*haplotype inference at the expense of molecular haplotyping methods. We are convinced that these last ones remain the most reliable and effective way to resolve linkage phase patterns and that they can produce, for a fixed sample size, much more precise estimates of haplotype frequencies than other approaches \[[@B63]\]. However, the considerable effort required to obtain and analyse individual chromosomes make alternative designs preferable; and the *in silico*approach appears to be the most practical one. Thus, for researchers not willing to invest time and money in the preliminary step of *NAT2*haplotype reconstruction, the use of computational algorithms constitutes a safe and effective way to get reliable haplotypic data on which further analyses could be carried on. Once haplotypes are constructed, various statistical methods can be applied on *NAT2*haplotype data to detect allele-disease associations or to classify patients according to their acetylation status. Methods ======= NAT2 molecular data sets ------------------------ To evaluate the performance of *in silico*approaches in *NAT2*haplotype reconstruction, we based our study on data collected from the literature, for which linkage phase was resolved directly through molecular haplotyping. Molecular data from five previously published data sets were analysed: they concerned 258 Spanish from Central Spain \[[@B45]\], 137 Nicaraguans with a Central American Indian-European mixed origin \[[@B30]\], 112 British from the Cambridge area \[[@B24]\], 101 Black South Africans (mostly Tswana-speaking people) \[[@B24]\], and 1000 Koreans \[[@B31]\]. All subjects included in these studies were randomly selected, unrelated healthy volunteers whose ethnic origin had been clearly defined. In each population sample, seven SNPs were typed at *NAT2*for all individuals (no missing data), and mutation linkage phase of all multiply heterozygous individuals was resolved molecularly through allele-specific PCR and restriction mapping. A summary description of the data sets is given in Table [2](#T2){ref-type="table"}. These data provide an opportunity to compare haplotype frequencies estimated by direct gene counting on experimentally haplotyped data with haplotype frequencies estimated by haplotyping algorithms when phase information is ignored. Throughout this report, we will use the term «phase-known» to refer to the individual\'s genetic constitution for the *NAT2*haplotyped system, including the linkage phase of the component SNP alleles. Whereas we will use the term \"phase-unknown\" to refer to an individual\'s multilocus genotype in the absence of phase information. Computational haplotyping methods --------------------------------- We evaluated the ability of four population-based haplotype inference methods to reconstruct *NAT2*haplotypes from the phase-unknown genotype data. ### Hapar The first method is based on maximum parsimony: it searches for a set of minimum number of haplotypes that explain the observed genotype samples. Clark\'s method, the first developed algorithm for haplotype reconstruction \[[@B43]\] which can be viewed as a sort of parsimony approach, requires homozygote or single-site heterozygote in the sample to start its inferential cascade. Wang and Xu \[[@B46]\] overcame this limitation by designing an algorithm with a global optimization goal. This method, recently implemented in the Hapar program \[[@B46]\], was tested at its default settings on the phase-unknown *NAT2*data. ### PL-EM We also applied the EM algorithm \[[@B47]\] to obtain the maximum-likelihood estimates of haplotype frequencies in the samples, given the observed data \[[@B48]-[@B50]\]. This algorithm starts with initial arbitrary values of haplotype frequencies and iteratively updates the frequency estimates, to maximize the log-likelihood function, until convergence is reached. Several EM-based algorithms have been developed. We used three different implementations, Arlequin \[[@B51]\], HAPLO \[[@B49]\] and PL-EM \[[@B52]\], that all gave us identical results on comparable analyses (data not shown). Thus, we presented only the results obtained with the PL-EM program: this software implements an algorithm derived from the standard EM but incorporating the computational strategy of partition-ligation \[[@B53]\] to handle a larger number of loci. We performed 50 independent runs with different initial conditions to minimize chances of local convergence so as to ensure finding the global maximum likelihood estimates. For a given unphased genotype pattern, the probability of each possible haplotype configuration was calculated by using the estimated population haplotype frequencies, and all compatible haplotype phases with nontrivial probabilities were generated. The haplotype pair with the greatest probability was considered to be the haplotype phase for each individual, and population haplotype frequencies were estimated as a function of each inferred haplotype pair, weighted by their estimated probability. ### Haplotyper and PHASE Finally, two Bayesian statistical methods based on Gibbs sampling procedure were applied to the phase-unknown *NAT2*data. Such methods treat the unknown haplotypes as random quantities and combine *prior information*-beliefs about what sorts of patterns of haplotypes are expected to be observed in population samples- with *the likelihood*-the information in the observed data \[[@B54]\]. The conceptual difference between the two investigated Bayesian algorithms lies in the prior information incorporated into the statistical model. The algorithm implemented in the Haplotyper program \[[@B53]\] uses a Dirichlet prior distribution, which assumes that the genetic sequence of a mutant offspring does not depend on the progenitor sequence \[[@B54]\]. Instead, the algorithm implemented in the PHASE program \[[@B44]\] uses a prior approximating the coalescent, which is one of the evolutionary models most commonly used in population genetics (see \[[@B55]\] for a review): it assumes that unresolved haplotypes will tend to be the same as, or similar to, known haplotypes. We employed the lastest version of PHASE (PHASE v 2.1 \[[@B54]\]) to evaluate the performance of this method, using the default parameter values in the Markov chain Monte Carlo simulations. For each data set investigated, we applied the algorithm ten times with different seeds for the random number generator, and checked for consistency of the results across the independent runs in order to verify that the algorithm did not converge to a local, rather than global, mode of the posterior distribution. We chose the results from the run displaying the best average goodness-of-fit of the estimated haplotypes to the underlying coalescent model. Besides, to evaluate the Bayesian algorithm implemented in Haplotyper, we performed 20 independent runs of the program on each sample. This software could not be run directly on the 1000-Korean sample as it can only handle 500 individuals at most per data set. To circumvent this limitation, we randomly generated ten pairs of complementary data sets, each composed of 500 individuals, and we ran Haplotyper on each of them. Results were averaged over the ten complete data sets. Both programs Haplotyper and PHASE provide a list of the most likely pairs of haplotypes for each subject. They also quantify the uncertainty associated with each phase call by outputting an estimate of the probability that each call is correct. This prevents inappropriate overconfidence in statistically reconstructed haplotypes. Measures of estimation accuracy ------------------------------- Computational algorithms of haplotype reconstruction may be used for many different purposes. We focus here on three particular tasks: finding the list of all haplotypes present in a sample, inferring the most likely pair of haplotypes for each sampled individual, and estimating haplotype frequencies in the population. Thus, three different measures of accuracy were used to evaluate the performance of the tested algorithms. ### -haplotype identification To assess accuracy in terms of haplotype identification, we used the *I*~*H*~index introduced by Excoffier and Slatkin \[[@B48]\]. It compares the number of different haplotypes detected experimentally with the number of different haplotypes inferred by the computer programs. We considered that a given haplotype is identified as being present in the true sample if its estimated frequency is above the threshold value of 1/(2n) in a population sample of n individuals. *I*~*H*~is given by: ![](1471-2156-6-30-i3.gif) where *k*~*true*~is the number of haplotypes in the true sample, *k*~*est*~is the number of estimated haplotypes with frequency above the threshold, and *k*~*missed*~is the number of true haplotypes not identified in the sample. Values of *I*~*H*~can vary between 1 (when the computational identified haplotypes are exactly the same as those determined experimentally) to 0 (when none of the true haplotypes are identified computationally). ### - reconstruction of the haplotypes of each sampled individual We specified the haplotype pair for an individual by choosing the most probable haplotype pair consistent with the individual\'s multilocus genotype. We measured performance by the *individual error rate*, which is the proportion of individuals whose haplotype pairs were incorrectly inferred by the program \[[@B53]\]. ### - estimation of sample haplotype frequencies To examine how close the computationally estimated haplotype frequencies are to the observed frequencies in the phase-known data, we used the similarity index *I*~*F*~of Renkonen \[[@B56]\], defined as the proportion of haplotype frequencies in common between the estimated and observed frequency distributions \[[@B18],[@B48]\]. ![](1471-2156-6-30-i4.gif) where ![](1471-2156-6-30-i1.gif) and *p*~0*i*~denote, respectively, the estimated and observed sample frequency of the i-th haplotype. This measure incorporates all *h*haplotype frequencies and thus captures the overall difference between estimated and observed frequencies for a particular data set. It varies between zero, when true haplotypes have estimated frequencies tending to zero, and one, when observed and estimated frequencies are identical. Since this index gives more weight to the high-frequency haplotypes, we used a second criterion to assess the accuracy of computational algorithms in haplotype frequency estimation: the change coefficient *C*, defined in Tishkoff *et al.*\[[@B57]\] as ![](1471-2156-6-30-i5.gif) where Max \[![](1471-2156-6-30-i1.gif), *p*~0*i*~\] indicates the maximum value of ![](1471-2156-6-30-i1.gif) or *p*~0*i*~. This coefficient measures the percentage change in haplotype frequencies across the two information conditions (phase-known versus phase-unknown data). *C*coefficients were computed for each possible haplotype in each population. The value of *C*ranges from 0 to 1, with 0 indicating that the estimated and observed frequency are identical. The maximal value of 1 indicates that molecular haplotyping showed either the presence of a haplotype that was assigned a zero through computational haplotyping, or *vice versa*. Measure of pairwise LD between SNP markers ------------------------------------------ We used the phase-known data to quantify the amount of LD between all pairs of polymorphic sites by computing the correlation coefficient r^2^\[[@B58]\] for each population sample separately. These statistics are expected to be 1 (perfect LD) when the variation is segregating in a population as only two distinct haplotypes. Statistical significance of LD between pairs of sites was assessed by Fisher\'s exact tests. Computations were performed with the software PowerMarker v3.21 \[[@B59]\], and a graphical summary of disequilibrium matrices was displayed by the GOLD program \[[@B60]\]. Abbreviations ============= Single nucleotide polymorphism (SNP) Expectation-Maximisation (EM) algorithm Linkage disequilibrium (LD) N-acetyltransferase 2 (NAT2) Authors\' contributions ======================= AS participated in the conception and design of the study, in collection of data, in performing all the computational analyses, and in drafting the manuscript. PD participated in the design of the study, in interpretation of data and in revising the article critically for important intellectual content. All authors read and approved the final manuscript.

PubMed Central

2024-06-05T03:55:58.370975

2005-6-2

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173101/", "journal": "BMC Genet. 2005 Jun 2; 6:30", "authors": [ { "first": "Audrey", "last": "Sabbagh" }, { "first": "Pierre", "last": "Darlu" } ] }

PMC1173102

Background ========== The accurate replication and repair of genetic material, which is a prerequisite for normal functioning of the eukaryotic genome and the prevention of cancer, relies on coordinated and faithful DNA synthesis. One important mechanism that ensures a high fidelity of DNA replication is a \"cleansing\" of the DNA precursor pool from deoxyribonucleoside triphosphates containing a modified base \[[@B1]-[@B4]\]. Such modified bases may have ambiguous base-pairing properties that will result in a high mutagenic activity after their incorporation into DNA during replication. A classic example of the detoxification mechanism is the elimination of dUTP and 8-oxo-dGTP from the dNTP pool by the *E. coli*dUTPase and MutT proteins, respectively \[[@B1],[@B5]\]. Purine analogs 6-hydroxyaminopurine (HAP) and 2-amino-HAP (AHA) are powerful mutagens in bacteria, yeast, and higher eukaryotes \[[@B6],[@B7]\]. It has been suggested that HAP-deoxyriboside-triphosphate (dHAPTP) is a possible endogenous contaminant of nucleotide pools under peroxyl radical stress \[[@B8]\]. HAP and AHA closely resemble the natural purines, hypoxanthine and xanthine (Fig. [1](#F1){ref-type="fig"}), and therefore, could be exploited to investigate the mechanism preventing mutations that are caused by non-canonical purine nucleotides \[[@B9]-[@B11]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Chemical structures of HAP and AHA and natural purine bases. ::: ![](1471-2156-6-31-1) ::: It was proposed that purine salvage enzymes convert base analogs to the corresponding deoxyribonucleoside triphosphates, which are misincorporated or misreplicated during DNA synthesis, resulting in induction of mutations \[[@B12],[@B13]\]. HAP-induced mutagenesis in yeast is elevated in strains with defects in proofreading activity of replicative DNA polymerases \[[@B14],[@B15]\] and does not depend on excision, mutagenic recombination, and mismatch repair systems \[[@B14]-[@B16]\]. We have described several systems protecting cells from the mutagenic and inhibitory effects of HAP (see review \[[@B16]\]). One is the novel molybdenum cofactor-dependent system in *E. coli*\[[@B17]\]. It has yet to be determined if a similar system exists in higher eukaryotes. Another, versatile HAP-detoxification pathway relies upon the action of triphosphatase, Ham1p, which hydrolyze HAP-containing ribo- and deoxyribo-nucleotides to nucleoside monophosphates, and which prevent incorporation of base analog into DNA and RNA. We initially described the elevated sensitivity to HAP in yeast due to mutations in the *HAM1*gene \[[@B18]\]. When we cloned and sequenced the *HAM1*gene, we found that it has homologs in many organisms, from bacteria to humans \[[@B13]\], and proposed that the gene might code for new triphosphatase \[[@B16]\]. Then, the crystal structure of the Ham1p homologue from a thermostable bacterium (protein Mj0226) was determined \[[@B19]\]. It was found that the Ham1p ternary structure has common features with MutT. Homologs of the yeast Ham1p from other organisms possessed triphosphatase activity on dITP, ITP, XTP, and dHAPTP substrates (Kozmin and Pavlov, unpublished; Burgis and Cunnigham, personal communication; and \[[@B19]-[@B21]\]). There are additional, less thoroughly studied, factors modulating purine base analogues mutagenesis in yeast (see \[[@B16]\] for review). For example, *aah1*mutants are sensitive to HAP, suggesting that adenine deaminase Aah1p may deaminate HAP base to hypoxanthine \[[@B16]\]. In the present study, we carried out a genome-wide search for HAP and AHA sensitive mutants. The release of several complete sets of deletion mutants by the Yeast Deletion Project provides a powerful approach for different types of genome screens in yeast \[[@B22]\]. Haploid and diploid strains have already been used to detect new genes controlling sensitivity to different agents such as UV, ionising radiation, iron, and methyl methane sulfonate (MMS) \[[@B23]-[@B27]\], as well as spontaneous mutability \[[@B28]\]. This approach, when combined with other genomics approaches, helps to establish the biological functions of uncharacterized ORFs in yeast, many of which have human orthologs. This approach also allows us to decipher the network responses to endogenous and environmental stress \[[@B29]\]. The present study is the first systematic, genome-wide search for the mutations conferring sensitivity to mutagenic purine base analogs. Results ======= Development of the screening method ----------------------------------- To develop a useful method for searching the yeast mutants sensitive to base analogs, we calibrated the experimental conditions using the wild-type strain, BY4742, and two previously described HAP-sensitive mutants, *ham1*and *aah1*(see \[[@B6],[@B16]\]), created on BY4742 background. As shown in Fig. [2](#F2){ref-type="fig"} and described in Materials and Methods, yeast were grown in a 96-well microtiter plate and then transferred, using a multiprong replicator device, to YPD plates containing base analogs. An induction of the Can^r^mutants was monitored by replica-plating to the minimal-medium plates containing L-canavanine. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Scheme of the protocol for screening the yeast deletions library for base analog sensitivity and induced mutagenesis. ::: ![](1471-2156-6-31-2) ::: The results are presented in Fig. [3](#F3){ref-type="fig"}. The left panel of Fig. [3](#F3){ref-type="fig"} shows the induction of canavanine-resistant mutants by HAP and AHA; and the right panel represents the survival of the tested strains on YPD plates in the presence of base analogs. In the wild-type strain, as in the *ham1*and *aah1*mutants, 1--3 spontaneous canavanine-resistant colonies per spot arise in the absence of mutagen (Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}, left panel). In our experimental conditions, 1 μg/ml of HAP did not induce Can^r^mutants in the wild-type strain. A moderate induction of Can^r^clones (fewer than ten per spot) was observed at 10 μg/ml of HAP and a very strong HAP-induced mutagenesis was observed in the wild-type strain at 100 μg/ml of HAP (Fig. [3A](#F3){ref-type="fig"}, left panel). For comparison, 100 μg/ml of AHA were only moderately mutagenic (compare Fig. [3A](#F3){ref-type="fig"} and Fig. [3E](#F3){ref-type="fig"}, left panel). Furthermore, both HAP and AHA did not affect the viability of the wild-type strain, even at the maximal concentration of 100 μg/ml (Fig. [3A](#F3){ref-type="fig"} and [3E](#F3){ref-type="fig"}, right panel). In the *ham1*strain, 1 μg/ml of HAP induced Can^r^mutants with the similar frequency that was observed in the wild-type strain at 2 orders in magnitude higher concentration of analog, 100 μg/ml (compare Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}, left panel, phenotype of hypermutability, HM). The hypersensitivity of the *ham1*mutant to the toxic action of HAP was clearly detectable at 100 μg/ml of HAP (Fig. [3B](#F3){ref-type="fig"}, right panel). We will refer later to this phenotype as hypersensitivity, HS. Note that a reduction of the number of canavanine-resistant clones at 10 and 100 μg/ml of HAP, in comparison with 1 μg/ml of HAP observed in the *ham1*mutant (Fig. [3B](#F3){ref-type="fig"}, left panel, another manifestation of hypersensitivity phenotype, HS), is also due to a dramatic decrease of survival. When the *aah1*mutant was tested (Fig. [3B](#F3){ref-type="fig"}, second row, for HAP and Fig. [3F](#F3){ref-type="fig"}, first row, for AHA), the drop of viability was less severe than that for the *ham1*mutant (phenotype of elevated sensitivity, S). HAP-induced mutagenesis was detectable at a low dose of 1 μg/ml, but was much less compared to the *ham1*mutant (phenotype of elevated mutability, M). Mutagenesis was somewhat stronger at the dose 10 μg/ml and was not seen at 100 μg/ml (another manifestation of elevated sensitivity, S). For the *aah1*mutant, AHA-induced mutagenesis was moderate at the dose of 10 μg/ml and very strong at a concentration of 100 μg/ml. The *aah1*mutant exemplified what we expect to observe for mutants moderately sensitive to both HAP and AHA. These data suggested that the procedure we devised for micro-titre plate format is effective for the detection of mutants with altered parameters of HAP and AHA sensitivity and mutability. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Results of the screening of the yeast deletion library for elevated mutagenesis and sensitivity in micro-titer plates**. Left panel -- Mutagenesis on selective plates with canavanine. Right panel -- The estimation of the number of colony-forming units on YPD medium. ::: ![](1471-2156-6-31-3) ::: Screening of the deletion strains library ----------------------------------------- We screened the yeast deletion strains library as described above. After an initial screening of 4,823 deletion strains, 43 mutants strains appeared to affect base analog-induced killing or mutagenesis. However, in subsequent testing of candidate strains in same type of plate tests, we have confirmed HAP-sensitivity of 16 mutants (Fig. [3](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}, column 4, where we refer to phenotypes of mutants according to abbreviations described in previous section). Next, we examined the mutability and survival of these 16 mutants in quantitative tests with HAP (see Materials and Methods). Based on the results of these two types of tests, we categorize HAP-sensitive strains in three groups, as shown in Table [1](#T1){ref-type="table"}. Group I comprises, in addition to *ham1*and *aah1*strains, mutants *ade12*and *ade2*. These strains were hypersensitive to the mutagenic and lethal action of HAP in both types of tests (Fig. [3](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}, columns 4--6). Strains of this group were sensitive to the low doses (1 and/or 10 μg/ml) of HAP, in contrast to the wild-type strain. The *ade12*mutant was almost as mutable as the *ham1*strain, but the hypermutability could only be demonstrated in a quantitative test, due to poor plating efficiency (compare Fig [3B](#F3){ref-type="fig"}, row 3 and Table [1](#T1){ref-type="table"}, column 5, row 3). The *ade2*mutant was as mutable as the *aah1*mutant (Table [1](#T1){ref-type="table"}). Deletion mutants of the first group showed variable degree of sensitivity to HAP-induced killing Table [1](#T1){ref-type="table"}, column 6). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### HAP-sensitivity of the mutants selected in our screening. ::: Class of mutants Gene or ORF name deleted Functional group Response HAP in spot tests^†^ Induced mutant frequency (×10^-7^) by HAP^\#^ Survival in presence of HAP^\#^ -------------------------------------------------------- -------------------------- ------------------ ------------------------------- ----------------------------------------------- --------------------------------- Wild type \- 400 100% Class I: mutants hypersensitive to HAP *HAM1* DNA^1^ HM, HS 13000 17% *ADE12\** DNA HM, HS 10000 30% *AAH1\** DNA M, S 5500 30% *ADE2* DNA M, S 4400 70% II class: Mutants sensitive to mutagenic effect of HAP *VIP1* Cell^2^ M, S 840 76% *VID27* Metabolic^3^ M, S 600^°^ 82% *IPK1* Metabolic M, S 1400 100% *ADE5,7* DNA M 1900 100% *ADE8* DNA M 1500 100% *ADE6* DNA M 1100 100% *RIM101* meiosis M 1100 100% *ADE3* DNA M 940 100% *ADE1* DNA M 860 100% *YGR035c* unknown M 500^°^ 100% *Yjl055w\** Metabolic M 500^°^ 100% Class III: Mutants sensitive to killing *YMl013c-a* Unknown S 230^°^ 60% *SHE4* Cell S 600^°^ 30% *TRP2* Metabolic HS 250^°^ 60% ^†-^HM -- hypermutable, M -- more mutable than wild-type, HS -- hypersensitive, S -- more sensitive than wild-type, see first paragraph od the Results Section for the explanation. ^\#^-- 25 μg/ml ^1^-- \'DNA\' -- includes genes involved in the control of DNA precursor metabolism, purine salvage, DNA repair. ^2^-- \'Cell\' -- includes genes involved in cytoskeleton organization, cell walls and organelles. ^3^-- \'Metabolic\' -- includes genes involved in general metabolic pathways. \* -- These mutants were also sensitive to the mutagenic or toxic effect of AHA (see Fig. 3). ^°^-- small difference from the wild-type strain was reproducible and statistically significant by Wilcoxon-Mann-Whitney test. ::: Eleven mutants fall into Group II. Mutants of this class were sensitive to the mutagenic effect of HAP, but their growth was not severely inhibited on HAP-containing medium. As a result, there is a smaller difference, in comparison to mutants of the group I, or no difference in the number of Can^r^mutants induced by 10 and 100 μg/ml of HAP in spot-tests (Fig. [3C](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}, column 4). These strains produced some HAP-induced Can^r^mutants at 10 μg/ml of HAP, whereas the parental strain was non-mutable at this HAP concentration. We also did not detect any substantial drops of viability after growth in liquid media containing 25 μg/ml of HAP (Table [1](#T1){ref-type="table"}, column 6). Group II was not homogeneous in respect to HAP mutability and sensitivity. The six most sensitive mutants in the group are *vip1, vid27, ade1, ade5,7, ade6, ade8, ipk1*and *rim101*(see Table [1](#T1){ref-type="table"}). These mutants showed a decrease in the number of Can^r^mutants in a qualitative test when concentration of HAP increased ten-fold (up to 100 μg/ml), which is an indication of some cell death at very high doses of HAP (Fig. [3C](#F3){ref-type="fig"}). Mutants *yjl055w*, *ygr035c*, and *ade3*were more resistant to HAP, since the number of Can^r^colonies was similar at 10 and 100 μg/ml of HAP. As could be seen from the results of quantitative assay, mutant classification is quite conditional and there is substantial variation in responses between mutants of the Group II, but all of them were more mutable that the wild-type strain. Finally, Group III includes mutants *she4, trp2*, and *yml013c-a;*which were sensitive to HAP-induced killing, but not to HAP-induced mutagenesis (Fig. [3D](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}, column 6). In quantitative tests the *yml013c-a*and *trp2*mutants were even less sensitive to the mutagenic action of HAP than the wild-type strain (Table [1](#T1){ref-type="table"}). The *she2*mutant was slightly more mutable that the wild-type strain only in quantitative test. The existence of such a type of mutants suggests that the toxic effect of HAP in yeast may be not only due to the induction of lethal mutations, but also due to inhibition of certain metabolic pathways. In the present study, we have also characterized three AHA-sensitive mutants, *aah1*, *ade12*, and *yjl055w*(Fig. [3E](#F3){ref-type="fig"} and [3F](#F3){ref-type="fig"}). Remarkably, all of these mutants were also sensitive to HAP (Fig. [3C](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}). Two of those mutants, *aah1*and *yjl055w*, were AHA-hypermutable; whereas a*de12*strain did not mutate in the presence of AHA, but was sensitive to AHA-induced killing. In a quantitative test, 50 μg/ml of AHA did not inhibit survival of the wild-type strain, *aah1*and *yjl055w*strains; whereas survival of the *ade12*mutant was reduced to 50%. In the *aah1*and *yjl055w*strains, 50 μg/ml of AHA induced canavanine-resistant mutants with the frequencies 120 × 10^-7^and 270 × 10^-7^, respectively, that was 4--9-folds higher than observed in the wild-type and *ade12*strains (30 × 10^-7^, in both strains). Discussion ========== In this study, we elaborated the technology for testing base analog-induced mutability and killing of thousands of yeast strains in microtiter format (see Fig. [2](#F2){ref-type="fig"}). We found that the method is quite sensitive and reliable. Next, we screened the library of haploid yeast strains carrying deletions in all nonessential ORFs for the sensitivity to mutagenic base analogs HAP and AHA. We have found 16 novel HAP-sensitive mutants that fall into several groups, based on the sensitivity profiles (Fig. [3](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}). One group comprises the mutants that are HAP-hypermutable and grow poorly in the presence of HAP. Another class comprises strains with elevated HAP-mutability that grow normally on medium containing HAP. Finally, the third group includes the mutants sensitive to HAP-induced killing, but not to HAP-induced mutagenesis. We have also isolated three AHA-sensitive mutants. All of them were HAP sensitive as well. We summarized the properties of the genes found in our screening in [Additional file 1](#S1){ref-type="supplementary-material"}. One interesting result from our study is that none of the genes involved in the control of base analogs sensitivity except two genes, *YML013c-*a and *ade12*, were found in screenings for genes controlling sensitivity to the other types of mutagens, MMS, UV, and ionizing radiation \[[@B23]-[@B27]\] or for elevated spontaneous mutagenesis \[[@B28]\]. It was previously reported that deletion of the *YML013c-a*and *ADE12*open reading frame specifically enhanced sensitivity to killing (as in case of HAP) induced by γ-radiation and bleomycin, but did not affect sensitivity to MMS, UV, and hydroxyurea \[[@B23],[@B30]\]. It is known that there are overlaps of the sets of genes detected in genome screenings for the MMS-, UV- or ionizing radiation-sensitive strains \[[@B29]\]. Usually, those spectra include genes controlling DNA replication, recombination, and DNA repair. In our study, we did not find any of those genes. This is consistent with our earlier observation that the mutagenic action of HAP in *S. cerevisiae*does not depend on nucleotide excision repair, mutagenic repair, and mismatch repair \[[@B16]\]. A system to repair non-canonical purines and, probably HAP and AHA, in DNA has been recently characterized in *E. coli*\[[@B9],[@B10]\]. It is possible that yeast *S. cerevisiae*lacks such a repair system. Based on our data, we propose that, in yeast, the major base analogs protective mechanism is a control of the quality of DNA precursor pool that prevents incorporation of base analogs into DNA. This mechanism may work at several levels: transport of analogs into cells, detoxification of analogs by metabolic enzymes, maintenance of nucleotide pools, and fidelity control of DNA replication (Fig. [4](#F4){ref-type="fig"}). HAP and AHA are likely transported into the yeast cell by the same permeases, which are involved in transport of natural purines. One candidate is purine-cytosine permease, Fcy2p, a major purine (adenine, guanine, and hypoxanthine) and cytosine transporter in yeast \[[@B31]\]. According to our unpublished data, *fcy2*mutants are resistant to HAP. Thus, the active transport of HAP is the first critical step in the HAP mutagenic pathway. The next step is a conversion of the base analog to the corresponding ribonucleoside monophosphate by enzymes of the purine salvage pathway. Previously we observed that the inactivation of the *APT1*gene, encoding adenine phosphoribosyl transferase, led to a severe decrease of the mutagenic effect of HAP \[[@B16]\], suggesting that this enzyme plays a key role in the biosynthesis of HAP-riboside-5\'-monophosphate (HAPMP). HAPMP then may be converted to the corresponding nucleoside triphosphate, which could be ambiguously incorporated into DNA by DNA polymerases and provoke replication errors in the subsequent replication cycles \[[@B13]\]. The mechanism preventing HAP- and AHA-induced toxicity most likely involves the conversion of base analogs to non-mutagenic metabolites by purine salvage enzymes. The main argument for this hypothesis is that HAP could be utilized by yeast cells as a sole purine source \[[@B13],[@B18],[@B32]\]. The first enzyme in this HAP and AHA detoxification pathway is adenine aminohydrolase, encoded by the *AAH1*gene. Aah1p from several microorganisms has been biochemically characterized (see \[[@B33]\]). Is has a broad substrate specificity and is capable of converting both adenine and its six-substituted analogs into hypoxanthine *in vitro*. According to our data, yeast Aah1p may convert HAP to hypoxanthine \[[@B12]\] and AHA to guanine *in vivo*, since the inactivation of the *AAH1*leads to a defect in this conversion that is readily detected by UV spectroscopy of yeast medium (unpublished observation). In the *aah1*mutant, the base analog intracellular concentration increases. We propose that this causes the elevation of base analog-induced mutagenesis (Fig. [3B](#F3){ref-type="fig"} and [3E](#F3){ref-type="fig"}, Table [1](#T1){ref-type="table"}). Interestingly, in our screening we did not detect the *amd1*mutant, deficient in AMP aminohydrolase, that catalyze deamination of AMP to IMP \[[@B34]\]. We disrupted this gene by the *kanMX*cassette in several yeast strains and did not see any effect on HAP-induced mutagenesis anywhere. Thus, deamination of HAP at the mononucleotide level does not play an important role in HAP detoxification. This can be due to several reasons: the inability of the Amd1p to hydrolyse HAPMP, the minor role of the *AMD1*gene in yeast, or the short life-time of the HAPMP in yeast cells. We have found that inactivation of adenilosuccinate synthase (ASS or Ade12p) encoded by the *ADE12*gene strongly enhanced HAP-induced mutagenesis and AHA-induced killing (see Fig. [3B](#F3){ref-type="fig"} and [3F](#F3){ref-type="fig"}, and Table [1](#T1){ref-type="table"}). The primary function of this enzyme is the conversion of IMP to SAMP in the pathway of AMP biosynthesis *de novo*(Fig. [4](#F4){ref-type="fig"}). We propose that the reason for this hypersensitivity is the dysregulation of purine biosynthesis, as follows. First, the blocking of this step of purine biosynthesis causes accumulation of IMP in the cell. The excess of IMP is probably toxic for the cell, since *ade12*mutants have a slow-growth phenotype that can be rescued by blocks of the earlier steps of the *de novo*purine biosynthesis pathway (Dr. A. M. Zekhnov {St-Petersburg State University}, personal communication). Accumulated IMP can be phosphorylated to ITP by nucleotide phosporylases. Thus, we propose that, in the *ade12*mutant, an excess of ITP may saturate Ham1p triphosphatase, an essential enzyme for the destruction of HAPTP and dHAPTP, which leads to increased HAP-sensitivity (see Fig. [3](#F3){ref-type="fig"}). The data obtained recently in bacteria are consistent with this hypothesis. It was shown that, *in vitro*, *E. coli*Ham1p homologues protein, RdgB, is a triphosphatase that acts to hydrolize non-canonical DNA precursors, dIPT and dXTP. The Ham1p protein was shown to possess a similar activity on dITP, dXTP, and dHAPTP substrates (Kozmin and Pavlov, unpublished; Burgis and Cunnigham, personal communication; and see refs. \[[@B19]-[@B21]\]). In *E. coli*, the *rdgB*mutation is synthetically lethal with *recA*. As proposed, absence of RdgB leads to a dramatic increase of hypoxanthine and xanthine in DNA. Accordingly, base excision repair of such modified bases occurring in opposite strands may generate double strand breaks that require the RecA function to be repaired. However, over-expression of adenilosuccinate synthase PurA (homolog of yeast Ade12p) suppresses this lethality \[[@B9]\]. This suggests a critical role of ASS in the regulation of intracellular concentration of genotoxic hypoxanthine-containing nucleotides. We found that certain mutations affecting IMP biosynthesis *de novo*enhance HAP-induced mutagenesis (Table [1](#T1){ref-type="table"}). Seven of sixteen newly identified genes controlling HAP and AHA sensitivity are involved in AMP biosynthesis (see [Additional file 1](#S1){ref-type="supplementary-material"} and Fig. [3](#F3){ref-type="fig"}). It is known that the regulation of the AMP biosynthesis pathway by adenine is mediated by SAICAR, one of the precursors in adenine biosynthesis *de novo*\[[@B35]\]. The accumulation of certain purine biosynthesis by-products may play a role in the regulation of the nucleotide pool. A defect in endogenous purine biosynthesis probably alters nucleotide pools to favour dHAPTP mis-incorporation into DNA or leads to HAP toxicity. In respect to this hypothesis, it is important that there is a difference in the level of HAP-induced mutagenesis among the strains carrying mutations in the genes of AMP biosynthesis. The less sensitive mutant is *ade3*(Fig. [3](#F3){ref-type="fig"} and Table [1](#T1){ref-type="table"}). The *ADE3*does not directly control any steps of purine biosynthesis. It encodes the C1-tetrahydrofolate synthase that provides C1-tetrahydrofolate, an indispensable precursor for AMP, histidine, thymidylate, and methionine biosynthesis (see Fig. [4](#F4){ref-type="fig"}). In this respect it is of interest that *ade4, ade16*and *ade17*mutants, also defective in IMP biosynthesis, were not found in our screen and were not sensitive to HAP or AHA when constructed *de novo*and tested directly (data not shown). The *ADE4*stands apart because PRPP, a substrate of the product of the *ADE4*gene, serves as a precursor for additional biosynthetic pathways. This prevents by-product accumulation and might be one of the explanations of lack of HAP sensitivity of the *ade4*mutants. The single *ade16*and *ade17*mutants also do not lead to byproduct accumulation because *ADE16*and *ADE17*are isozymes and the inactivation of one gene does not block the pathway. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Purine salvage and purine biosynthesis *de novo*in yeast *Saccharomyces cerevisiae***. Intermediates of the purine biosynthesis *de novo*are designated in blue. The salvage pathway is presented in black. Genes, whose deletions lead to HAP and/or AHA sensitivity are highlighted by red boxes. The proposed conversions of HAP and AHA are represented in brackets below the adenine and guanine metabolites, respectively. Dashed arrows represent hypothetical pathways that were not demonstrated experimentally for a given substrate. Abbreviations:Purine biosynthesis *de novo*: PRPP -- 5-phospho-ribosyl-1α-pyrophosphate, PRA -- 5-phospho-β-D-ribosylamine, GAR -- 5-phosphoribosylglycinamide, FGAR -- 5\'-phosphoribosyl-*N*-formyl glycinamide, FGAM -- 5\'-phosphoribosyl-*N*-formylglycinamidine, AIR -- 5\'-phosphoribosyl-5-aminoimidazole, CAIR -- 5\'-phosphoribosyl-5-aminoimidazole-4-carboxylate, SAICAR -- 5\'-phosphoribosyl-4-(*N*-succinocarboxamide)-5-aminoimidazole, AICAR -- 5\'-phosphoribosyl-4-carboxamide-5-aminoimidazole, FAICAR -- 5\'-phosphoribosyl-4-carboxamide-5-formamidoimidazole, SAMP -- adenylosuccinate, 5,10-methylene-THF -- 5,10-methylenetetrahydrofolate, 10-formyl-THF -- 10-formyltetrahydrofolate. Salvage: Ade -- adenine, AdeRib -- adenosine, Hyp -- hypoxanthine, Gua -- guanine, Xan -- xanthine. ::: ![](1471-2156-6-31-4) ::: We have found that 6 genes detected in our screening, *VIP1, VID27, IPK1, YGR035c, YML013c-a*and *SHE4*, are putatively involved in cell organization or genetically interact with cell-cycle control genes (see [Additional file 1](#S1){ref-type="supplementary-material"}). This observation provides new perspectives on the mechanisms of base analogs-induced mutagenesis. It is possible that there is a specific structural route, mediated by cell cycle and cytoskeleton components, initiated by penetration of the analog inside the cell to its final target. We have also identified several hypothetical genes critical for the HAP and/or AHA resistance. This may be an initial clue to their functional significance. Finally, we would like to mention that out of the 18 genes we found to be involved in HAP and AHA sensitivity control, 11 (60%) have orthologs in all groups of organisms, including mammals. Therefore, the results have relevance to higher eukaryotes and humans as well (see [Additional file 1](#S1){ref-type="supplementary-material"}). Conclusion ========== We identified novel mutants sensitive to mutagenic and toxic effects of purine base analogs. AHA sensitivity was not previously described for three of the mutants identified in this study. The results reveal a complex control of base analogue mutagenesis by genes encoding the components of metabolic pathways and cytoskeleton. Methods ======= Yeast strains and media ----------------------- We have used a set of 4,823 *S. cerevisiae*mutants carrying deletions of all non-essential ORFs created in the haploid strain BY4742 (*MATα his3Δleu2Δlys2Δura3Δ*). The information about the deletion strains set is available from the Yeast Deletion Project site: <ftp://ftp.resgen.com/pub/deletions/mat_alpha_041902.txt>. Deletion strains were constructed by replacement of the ORF\'s with the *kanMX4*cassette, which confers resistance to geneticin \[[@B36]\]. The standard yeast complete media (YPD) and minimal synthetic media (SD) were used \[[@B37]\]. Deletion strains were cultivated on YPD medium, supplemented with 200 μg/ml of geneticin. Sensitivity to HAP and AHA was examined on YPD media containing analogs in concentrations 100 μg/ml, 10 μg/ml and 1 μg/ml. SD medium containing 40 μg/ml of L-canavanine was used for the selection of Can^r^mutants. Base analogs sensitivity tests ------------------------------ HAP was purchased from MP Biomedicals (Irvine, California, USA). AHA was custom- synthesized by Dr. I. Kuchuk at the Department of Chemistry of Indiana University (Bloomington, Indiana, USA) by the method of Janion \[[@B38]\]. Both chemicals were dissolved in DMSO (Fisher, USA) with mild heating. A single colony of each tested strain was inoculated into a well of 96-well microtitre plate containing 200 μl of liquid YPD medium (see Fig. [2](#F2){ref-type="fig"}). Microtiter plates were incubated for 2 days at 30°C with agitation, to reach a stationary phase (approximately 10^8^cells/ml). For mutagenesis assay, cells were then plated by a multiprong replicator device (approximately 5 μl of cell suspension per prong) to the YPD plates containing various concentrations of HAP or AHA, as shown on Fig. [2](#F2){ref-type="fig"}. After one day of incubation, the plates were replica-plated on SD minimal-medium plates containing L-canavanine. The plates were incubated 5 days and inspected for induction of Can^r^mutants. For the survival test, cell cultures from the microplates were diluted in water in series of 96-well microplates, using a multichanel pipette (see Fig. [2](#F2){ref-type="fig"}). Diluted cells suspensions were plated to YPD plates containing HAP or AHA by a multiprong replicator device. After 2--3 days of incubation, the number of colonies was recorded. Strains that produced smaller colonies or a smaller number of colonies on the HAP-containing medium, relatively to wild-type strain, were classified as sensitive to killing. Quantitative assay of the base analog-induced mutagenesis --------------------------------------------------------- For each strain and each concentration of base analogs to be tested, we prepared six independent cultures by inoculating a single colony into 1 ml of liquid YPD medium with or without mutagens. After two days incubation in the roller drum, the mutant frequencies were determined by plating of appropriately diluted cell suspensions on minimal-medium SD plates supplemented with L-canavanine (to determine the number of canavanine-resistant cells per culture), and on YPD plates (to obtain the total number of cells per culture). Then the frequency of mutants was calculated as described \[[@B14]\]. Each experiment was repeated at least three times. We have used several doses of HAP and found that in this type of test the most reproducible results are obtained at dose 25 μg/ml. The statistical significance of differences between variants was estimated by Wilcoxon-Mann-Whitney nonparametric criterion. Authors\' contributions ======================= Youri Pavlov and Elena Stepchenkova designed this study. Elena Stepchenkova performed the experimental work and wrote the initial draft of the manuscript. Vladimir Alenin, Youri Pavlov, Stanislav Kozmin, and Elena Stepchenkova analyzed the data set and wrote the final version of the paper. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Annotation of the genes whose deletion results in HAP and AHA sensitivity ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ We are grateful to Polina Shcherbakova for comments during this study and for critical reading of the manuscript. This work was supported in part, by NE DHHS LB506 grant \#9934 for YIP, and by the CRDF grant provided by The Russian Ministry of Education \# ST-012-0.

PubMed Central

2024-06-05T03:55:58.375806

2005-6-2

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173102/", "journal": "BMC Genet. 2005 Jun 2; 6:31", "authors": [ { "first": "Elena I", "last": "Stepchenkova" }, { "first": "Stanislav G", "last": "Kozmin" }, { "first": "Vladimir V", "last": "Alenin" }, { "first": "Youri I", "last": "Pavlov" } ] }

PMC1173103

Background ========== The antigenic regions of protein recognized by the binding sites of immunoglobulin molecules are called B-cell epitopes \[[@B1]\]. These epitopes can be classified into two categories; i) conformational/discontinuous epitope, where residues are distantly separated in the sequence and brought into physical proximity by protein folding and, ii) linear/continuous epitope, comprised of a single continuous stretch of amino acids within a protein sequence that can react with anti-protein antibodies \[[@B2],[@B3]\]. Most of the B-cell epitopes were thought to be discontinuous. However, in late 1980s it was shown that this conformational restriction is not a necessary condition for the production of protein-reactive anti-peptide antibodies \[[@B4]\]. The designing of the conformational epitopes is difficult and so experimental B-cell epitopes largely include linear epitopes. These linear epitopes can be exploited in the development of synthetic vaccines and disease diagnosis. A number of vaccines based on B-cell epitopes are currently under clinical phase trials against viruses \[[@B5]\], bacteria \[[@B6]\] and cancer \[[@B7]\]. These epitopes are also important for allergy research and in determining cross-reactivity of IgE-type epitopes of allergens \[[@B8]\]. A large number of B-cell epitopes have been reported in the literature in last two decades. There is a need to collect and compile these epitopes to evaluate the performance of existing B-cell epitope prediction methods and to further develop better methods \[[@B9],[@B10]\]. We have observed that the performance of the existing physico-chemical scales is not very high \[[@B11]\]. Recently, Blythe and Flower examined 484 amino acid propensity scales in predicting of B-cell epitopes and found that even the best set of scales and parameters performed only marginally better than random \[[@B12]\]. The evaluation of the existing scales indicates that there is a need to develop better methods by using artificial intelligence techniques. The collection of B-cell epitopes should help in deriving new scales for accurate in silico prediction of linear epitopes. Also it will help the immunologist to understand the complex nature of immunogenic peptides and for the development of vaccines. There are many databases available on T-cell epitopes \[[@B13]-[@B15]\]. In contrast, there are limited number of databases on B-cell epitope for example JenPep \[[@B16],[@B17]\] and HIVDB \[[@B18]\]. Recently, JenPep has been superseded by AntiJen 2.0, which has included peptides bound to MHC ligand, TCR-MHC Complexes, T cell epitope, TAP, B cell epitope molecules and immunological protein-protein interactions. AntiJen also contains peptide library, copy numbers and diffusion coefficient data. Though AntiJen provides information about different types of peptides (\>24000 entries) from a single source but it provides limited information about B-cell epitopes and tools to analyze and retrieve the data. Recently, we have created a comprehensive database, Bcipep, of B-cell epitopes. Latest version of Bcipep contains 3031 entries, where each entry provides detailed description of a B-cell epitope. Currently, we have covered only the continuous B-cell epitopes. The aim of this database is to assist the scientific community working in the areas of synthetic peptide vaccines (based on B-cell epitopes) and allergy research. The database will complement the existing databases such as AntiJen \[[@B16],[@B17]\]. Availability ------------ The database is available at <http://www.imtech.res.in/raghava/bcipep>, <http://bioinformatics.uams.edu/mirror/bcipep/> (Mirror Site) and <http://srs.ebi.ac.uk/srs6bin/cgi-bin/wgetz?-page+LibInfo+-id+1X2XW1JU5_L+-lib+BCIPEP> (SRS version). Database construction ===================== The PostgreSQL relational database management system (RDBMS) has been used for storing, retrieval and managing the data. The scripts, which provide interface between user and database, were written in PERL, CGIPerl and Pgperl. B-cell epitopes were collected from the literature (PubMed, <http://www.ncbi.nlm.nih.gov/pubmed/>; ScienceDirect, <http://www.sciencedirect.com/>). A large number of HIV epitopes were extracted from a book \[[@B18]\]. Database description ==================== The aim of Bcipep database is to provide; i) comprehensive information about B-cell epitopes, ii) tools for extraction and analysis of this information and, iii) hyperlinks to related databases. The overall architecture of Bcipep database is shown in Figure [1](#F1){ref-type="fig"}. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### A schematic representation of Bcipep database. ::: ![](1471-2164-6-79-1) ::: Database information ==================== General ------- This database provides compehensive information about the linear B-cell epitopes which includes; i) amino acid sequence of epitope, ii) source protein from which epitopes were obtained, iii) experimental methods used in accessing the immunogenic potential of epitopes, iv) pathogen group (e.g., bacteria, virus, fungi, protozoa) of source protein and, v) miscellaneous information in the \'*Comment\'*field. Immunogenicity and model organisms ---------------------------------- The immunogenicity of a peptide is a semi-quantitative measure of its immunogenic potency. In Bcipep it is divided into three categories; i) immunodominant, if it increases 2--3 folds anti-peptide antibodies in comparison to reference or control (carrier protein, eg., BSA or KLH), ii) immunogenic, if it enhances anti-peptide antibodies by one-fold in comparison to reference, and iii) null-immunonogenic, where no difference was observed when compared to reference. This information is very important for developing B-cell epitope prediction method. The database also provides information about \'*Model organism\'*used for immunization. Antibodies and neutralization ----------------------------- The database provides full information about monoclonal or polyclonal antibodies produced against an epitope. The information includes, isotypes of immunoglobin and name/number of monoclonal antibodies. The database also contains information about neutralization potential of anti-peptide antibody, which is crucial for considering a peptide for synthetic vaccine design. Links to databases ------------------ The Bcipep provides hyperlinks to various sequence databases in order to provide detailed information about peptides in database. The \'*database reference*\' field consists name/code of protein available in SWISS-PROT \[[@B19]\]. The \'*Antigenic structure*\' field consists of PDB codes \[[@B20]\] of protein structures having matching peptides. These PDB codes are linked to OCA browser <http://pdb.tau.ac.il/> in order to provide detailed structure information of these proteins. It also provides structure information about 242 antibodies, where each antibody is hyperlinked to PDB database through the OCA browser. The \'*Publication reference*\' field provides full information about related publications with link to PUBMED \[[@B21]\]. Bcipep is also linked to MHCBN database \[[@B15]\] in order to identify the peptides that are B-cell as well as T-cell epitopes. Web tools ========= Bcipep has following major web-based tools for retrieval and analysis of information in Bcipep database (Figure [1](#F1){ref-type="fig"}). These web tools have been designed to facilitate the user in retrieving information from database. Keyword search -------------- This option allows users to perform search on all fields of the database (\'*Peptide Sequence\', \'Source Protein\', \'Publication Reference\', \'Database Reference*\'). One can restrict the keyword search on any specific field. It also allows users to select the fields to be displayed. An example of keyword search is shown in Figure [2a](#F2){ref-type="fig"}, where key word \'P26694\' is searched in any filed of database. The output/result of this keyword search is shown in Figure [2b](#F2){ref-type="fig"}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### The typical display of Bcipep database for keyword search; a) input page of keyword search; and b) output of keyword search. ::: ![](1471-2164-6-79-2) ::: Peptide search -------------- The database provides option to search a peptide in Bcipep. The tool will display full information about the peptides included in Bcipep. The server also permits users to search their query sequence in any pathogen group. Search can be restricted on the basis of immunogencity that is immunodominant, immunogenic or null-immunogenic. An example of input and output of peptide search is shown in Figures [3a](#F3){ref-type="fig"} and [3b](#F3){ref-type="fig"} respectively. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### This example illustrate peptide search on Bcipep; a) Peptide search page; and b) result of peptide search. ::: ![](1471-2164-6-79-3) ::: Mapping of T-cell epitopes -------------------------- This server allows searching of peptide in Bcipep against MHCBN database. The MHCBN database provides information about components of cell-mediated immunity like MHC binders/non-binders, T-cell epitopes and TAP binders \[[@B14]\]. The peptides related to cell-mediated immunity can be mapped on resultant B-cell epitopes obtained from keyword/peptide search by clicking on \'*Peptide Sequence\'*field (Figures [2b](#F2){ref-type="fig"} and [3b](#F3){ref-type="fig"}). Thus, the server is useful in identifying the potential B-cell epitopes having T-cell epitopes (or MHC binders). The example of mapping of MHCBN peptides on B-cell epitope (by clicking on peptides sequence field of Figure [3b](#F3){ref-type="fig"}) is shown in Figure [4a](#F4){ref-type="fig"}. The full information of each map peptide can be obtained by clicking on the mapped sequence. One such example is shown in Figure [4a](#F4){ref-type="fig"} and [4b](#F4){ref-type="fig"}. The mapping allows user to detect the regions in B-cell epitope having promiscuous MHC binders (peptides that can bind to large number of MHC alleles) or T-cell epitopes. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Mapping of peptide in MHCBN database; a) mapping of MHCBN peptides on B-cell epitope, and b) full information about a MHCBN peptide. ::: ![](1471-2164-6-79-4) ::: Peptide mapping --------------- The peptides of Bcipep can be mapped on query sequence using this option. The full information about mapped peptide can be obtained by clicking on it. The tool will assist the researchers in gaining knowledge about the known immunogenic or non-immunogenic regions in target protein of interest. The example input and output of peptide mapping is shown in Figures [5a](#F5){ref-type="fig"} and [5b](#F5){ref-type="fig"} respectively. The users can specify the pathogen group and/or immunogenicity level of peptides to be mapped on query sequence. As shown in Figure [5b](#F5){ref-type="fig"}, the graphical mapping of peptides/epitopes on query allows one to easily detect the regions that bind to large number of B-cell peptides. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Mapping of B-cell epitopes on antigen sequence, a) submission page of B-cell epitope mapping, and b) mapping results ::: ![](1471-2164-6-79-5) ::: BLAST search ------------ This tool allows users to search their query protein against antigenic proteins maintained at Bcipep. The sequence of 1070 antigenic proteins has been obtained from SWISS-PROT. The similarity search is performed using the GWBLAST server <http://www.imtech.res.in/raghava/gwblast/>. The GWBLAST also allows users to analyze the BLAST output like multiple alignments, phylogenetic analysis. Online data submission ====================== The database has the facility to submit data to Bcipep on-line via Internet. Users can submit information about their experimentally determined B-cell epitopes. We hope that immunologists will submit their information on B-cell epitope in Bcipep, similar to the sequence data at GenBank and SWISS-PROT. The Bcipep team will add more data from literature to maintain B-cell data up-to-date and cross check the data submitted by users to maintain quality of the data. Potential utility and limitations ================================= One of the major challenges in the field of subunit vaccine design is to identify the antigenic regions (B and T cell epitopes) that can generate antigen specific memory cells. Thus, the identification of regions/stretches on an antigen from the data pool of known epitopes is an important step in vaccine design. The Bcipep database would be very useful as it consists of comprehensive information about experimentally verified linear B-cell epitopes and tools for mapping these epitopes on an antigen sequence. In case query antigen contains known epitopes, this database might aid in the wet experimentation and lower the cost by reducing the overlapping repeats. This strategy is frequently used for the screening of transgenic proteins by searching linear IgE-binding epitopes \[[@B22]\]. Unlike discontinuous epitopes, the linear epitopes are easy to design, as they do not require tertiary structure information. Bcipep also provides information on neutralizing B-cell epitopes where an antibody generated against a B-cell epitope neutralizes the parent antigen. The current version of Bcipep provides neutralizing information on about 1309 such B-cell epitopes. This information is very important for selecting functional B-cell epitopes. This database also provides a link with MHCBN to search for overlapping regions of MHC binders and T-cell epitopes in the B-cell epitopes. Thus, the user can identify both antigenic regions that can activate B-cell and T-cell, which can lead to the development of better vaccine. The epitopes in Bcipep can be used to derive the rules for predicting B-cell epitopes. The aim of designing synthetic linear peptides as epitope-vaccine is to induce neutralizing antibodies against the pathogen \[[@B23]\]. There are many reports that the linear B-cell epitopes were characterized as neutralizing antibodies as in Clostridium botulinum neurotoxin type A (Btx A)\[[@B24]\]. In Bcipep, there are 748 neutralizing anti-peptide antibodies entries. However, in some cases these linear epitope(s) fail to produce neutralizing antibodies and do not give protective immunity. For instance, it has been shown in the past that the antibodies against the synthetic peptides and short recombinant proteins of approximately 100 amino acids of hepatitis E virus (HEV) do not neutralize, suggesting that the HEV neutralization epitope(s) is conformation dependent \[[@B25]\]. The elicitation of a bactericidal and protective immune response to *Borrelia burgdoferi*decorin binding protein requires a properly folded conformation for the production of functional antibodies \[[@B26]\]. Recently, Corcoran *et al*, 2004, observed that B-cell memory is established and maintained against conformational epitopes of Parvovirus VP2 and against linear epitopes of VP1 but not against linear epitope VP2 \[[@B27]\]. Thus, it is not necessary that the linear B-cell epitope will always give rise to memory cells. One should also check the neutralizing information of B-epitopes, as only 748 B-cell epitopes out of 1309 in Bcipep were able to neutralize the parent protein. For an effective use of Bcipep, it is important to understand the limitation of linear B-cell epitopes and data in Bcipep. The few limitations of current version of Bcipep are; i) it does not cover discontinuous epitopes, ii) it has limited number of unique peptides (1590) in 3031 entries and, iii) it contains peptides having only natural amino acids. One should be careful in using linear B-cell epitopes in developing epitope based subunit vaccine. The organism used for immunization (information included in the database) should also be taken into consideration, since immune response is T-helper cell (MHC-II-peptide complex) dependant and B-cell epitope alone may not generate protective antibodies \[[@B28]\]. In some cases, the nature of the adjuvant used and the route of immunization (information not included in the database) might also play important roles in the induction of protective anti-peptide antibody response against the pathogen \[[@B29]-[@B31]\]. Authors\' contributions ======================= SS collected and compiled the data as well as developed the web server. MB helped in designing website and stored data in PostgreSQL. GPSR conceived the idea and supervised the work. Acknowledgements ================ Authors are thankful to Dr Grish C. Varshney for critically reading the manuscript. We are also thankful to Council of Scientific and Industrial Research (CSIR) and Department of Biotechnology (DBT), Govt. of India, for financial assistance.

PubMed Central

2024-06-05T03:55:58.378766

2005-5-29

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173103/", "journal": "BMC Genomics. 2005 May 29; 6:79", "authors": [ { "first": "Sudipto", "last": "Saha" }, { "first": "Manoj", "last": "Bhasin" }, { "first": "Gajendra PS", "last": "Raghava" } ] }

PMC1173104

Background ========== Dinoflagellates play critical roles in marine ecosystems as primary producers and grazers of other bacterial and eukaryotic plankton \[[@B1]\]. Approximately one-half of the ca. 4,000 species of dinoflagellates contain plastids, although many are mixotrophic \[[@B2]\]. Many taxa produce potent toxins and form harmful algal blooms, or \"red tides\", resulting from populations of more than 20 million cells per liter of seawater. The toxins cause a variety of poisonings that affect humans and marine wildlife \[[@B1]\] and have a significant impact on coastal ecosystems throughout the world \[[@B3]\]. Yet, other dinoflagellates, like *Symbodinium*, are central contributors to the health of reef ecosystems as the symbionts of corals \[[@B4]\]. Loss of the dinoflagellate symbiont results in coral bleaching. In addition to their ecological role, dinoflagellates display some fascinating and unique aspects of cell biology. One intriguing character is nuclear biology. The nucleus of dinoflagellates is unlike that of any other eukaryote because the chromosomes are condensed throughout the cell cycle except during DNA replication \[[@B5]\]. The morphologically similar chromosomes are attached to the nuclear envelope and can number in the hundreds \[[@B6]\]. Dinoflagellates also lack nucleosomes \[[@B7]\], instead the nuclear DNA is associated with basic proteins that are moderately similar to bacterial histone-like proteins (HLPs \[[@B8],[@B9]\]). Dinoflagellates were thought to lack histones \[[@B10]\], but in a recent gene expression study, a putative histone H3 was annotated in *Pyrocystis lunula*, although the sequence was not analyzed further \[[@B11]\]. The general lack of nucleosomes raises many questions about transcription and gene regulation in these organisms. Dinoflagellate nuclei also contain vast amounts of DNA compared to other eukaryotes. Estimates range from 3 -- 250 pg·cell^-1^, or approximately 3,000 -- 215,000 megabases (MB) \[[@B12]\]. In comparison, human nuclei contain 3.2 pg·cell^-1^(3,180 MB). The dinoflagellate nucleus contains such a high concentration of DNA that it exists in a liquid crystal state, which is responsible for the unique morphology \[[@B13],[@B14]\]. The DNA to basic protein ratio of dinoflagellate chromosomes has been estimated to be 10:1, which is dramatically higher than the 1:1 ratio observed in most eukaryotes. This indicates that very little basic protein is associated with dinoflagellate chromosomes and that the crystal structure is the primary cause of the unusual morphology. Dinoflagellates are also the only eukaryotes to contain hydroxymethyluracil, a deaminated nucleotide that can be produced by oxidative damage of DNA, which replaces 12 -- 70% of the thymidine \[[@B15]\]. The role of polyploidy or potentially, genome amplification within particular life history stages remains to be clarified for dinoflagellates. It is highly unlikely, however, given their relatively simple morphology that the immense DNA content is explained solely by gene content. The most widespread plastid in dinoflagellates contains the unique photopigment peridinin. The \"peridinin plastid\" is remarkably different from this organelle in other eukaryotes because it lacks a typical genome. Plastids normally contain a circular genome of about 150 kb that encodes 100 -- 200 genes that are necessary for plastid function. In peridinin-containing dinoflagellates, the plastid genome has been broken into minicircles that encode a single, or a few genes per circle. However, only 16 genes have been identified thus far on minicircles \[[@B16],[@B17]\]. Recent studies show that most of the plastid genes have been transferred to the nucleus \[[@B18],[@B19]\] with 15 of these genes found exclusively on the plastid genome in all other photosynthetic eukaryotes \[[@B18]\]. The peridinin dinoflagellates encode therefore the smallest number of plastid genes of any photosynthetic eukaryote, making them a model for understanding organellar gene transfer. Nuclear-encoded plastid proteins are targeted to the plastid using a tripartite N-terminal targeting signal \[[@B20]\]. As in *Euglena*, nuclear-encoded plastid proteins are co-translationally inserted into the endoplasmic reticulum and embedded in this membrane using a stop-transfer sequence in the N-terminus. Through algal endosymbioses, the dinoflagellates have been able to acquire four other types of plastids from distantly related evolutionary lineages including the haptophytes, cryptophytes, diatoms, and prasinophytes \[[@B1],[@B21]\]. This aspect of their evolutionary history highlights the unmatched ability of dinoflagellates to capture and retain foreign plastids. *Alexandrium tamarense*is one of the best-studied dinoflagellates. This species forms toxic blooms and causes paralytic shellfish poisoning through saxitoxin production. It has a peridinin-containing plastid and in North America, *A. tamarense*blooms from Alaska to Southern California in the Pacific and along the Canadian and New England coasts in the Atlantic. There has been a recent increase in blooms of *A. tamarense*and other *Alexandrium*species in other parts of the world making this genus of high importance to the world\'s fisheries. We undertook a gene discovery project with this organism using expressed sequence tag (EST) data to investigate dinoflagellate evolution and to create a genomic resource for scientists working on different aspects of *A. tamarense*and dinoflagellate biology. The EST method was the most reasonable approach in this case because haploid *A. tamarense*cells contain approximately 143 chromosomes and have a genome size of 200 pg/cell (ca. 200,000 Mb \[Erdner and Anderson unpublished data\]). Our EST results comprise the first extensive high-throughput, genome-wide data set for a dinoflagellate. Results and discussion ====================== Clustering and sequence analyses -------------------------------- The collection of 11,171 ESTs comprised of single-pass 3\'-reads (483 from the start library and 10,688 from the normalized library) from *A. tamarense*was assembled into 6,723 clusters. The normalized library showed a high degree of complexity, with a novelty rate of 60.18% and about 52% of the sequenced clones contained inserts that were longer than the single sequence read (ca. 750 bp). Clustering of the total EST set showed that most of the reads were singletons (4,618 sequences) and the largest cluster was comprised of 46 ESTs that are closely related to HLPs (Table [1](#T1){ref-type="table"}). Other highly represented transcripts were those encoding luciferin-binding protein (a protein involved in the regulation of bioluminescence) and several photosynthetic proteins (e.g., Rubisco, ATP synthase C chain, light harvesting proteins). Several large clusters were transcripts that lacked a similarity (e-value \< e^-5^) to known proteins. One of these ESTs has an open reading frame that encodes a protein with a potential plastid-targeting signal (Figure [1A](#F1){ref-type="fig"}). Interestingly, database searches against NCBI\'s nr and dbEST returned hits only to other dinoflagellate ESTs. Another of the largest clusters only had hits to ESTs from other dinoflagellates (Figure [1B](#F1){ref-type="fig"}). These two proteins are therefore candidates for dinoflagellate-specific proteins. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Cluster size and frequency of the *A. tamarense*ESTs. ::: Cluster Size Frequency Cluster Size Frequency Best BLAST hit(s) -------------- ----------- -------------- ----------- ---------------------------------------------------------------- 1 4618 14 7 2 1249 15 1 unknown 3 427 16 1 HSP90 4 176 17 4 peridinin-chl a protein, Cytochrome C6, EF1-alpha, unknown 5 81 18 1 ATP synthase C chain 6 44 19 2 Form II Rubisco, unknown putative dino. specific protein 7 32 21 1 fucoxanthin chlorophyll a/c binding protein like 8 15 22 1 Unknown putative plastid protein 9 21 23 1 Unknown 10 13 24 3 peridinin-chlorophyll a protein, ATP synthase C chain, unknown 11 10 29 1 luciferin-binding protein 12 7 46 1 histone-like protein/basic nuclear protein 13 6 ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Putative dinoflagellate-specific proteins.**Amino acid sequence alignments of putative dinoflagellate-specific proteins. A) putative plastid protein that was highly represented in the *A. tamarense*cDNA library (cluster size = 22). *A. tamarense*sequences 1, 2, and 3 correspond to clones GC1-aba-e-13, GC1-abh-e14, and GC1-abd-o-22, respectively, and are aligned with highly similar ESTs from the dinoflagellates *L. polyedrum*(CD809498) and *P. lunula*(BU582532). The boxed region indicated a possible plastid targeting sequence. B) Putative dinoflagellate specific protein with significant blast hits only to other dinoflagellate ESTs. The *Alexandrium*sequence corresponds to clone UI-D-GC1-abh-f-23-0-UI. ::: ![](1471-2164-6-80-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **GO category assignment of *A. tamarense*ESTs.**Classification of 1,203 *A. tamarense*ESTs into the GO categories. ::: ![](1471-2164-6-80-2) ::: Each cluster was searched against the SwissProt protein database using blastx. A total of 515 hits with an e-value less than 1e-20 were identified that terminated within 10 amino acids of the end of the SwissProt entry. From these hits, we estimated that the 3\'-UTRs ranged in length from 25 -- 620 nt with a mean length of 155 nt. This is shorter than the average length observed for fungi (\~200 nt) and metazoans (300--600 nt) \[[@B22]\]. However, this analysis is likely to be an underestimate of the average 3\'-UTR length because only ESTs that were sequenced into the coding region were included in the analysis. The 3\'-UTRs of *A. tamarense*cDNAs are also interesting because of their apparent lack of a polyA signal. Both simple n-mer searches (e.g. hexamer, pentamer) and the Gibb\'s sampler were used to assay the canonical region from -11 to -30 preceding the polyadenylation site in search of a polyadenylation signal. We were unable to find a single or a related set of hexamers or pentamers that are enriched in the 3\'-UTRs (data not shown). Clearly, polyadenylation of transcripts occurs in *A. tamarense*, however, the mechanism by which this process takes place apparently does not involve a typical polyA signal. These ESTs were also analyzed for GC-content and codon usage. Coding region GC-content was 60.8%, whereas GC-content in the 3\'-UTR was slightly less at 57.6%. The GC-content is reflected in the codon usage (Table [2](#T2){ref-type="table"}), whereby 3^rd^positions are strongly biased towards Gs or Cs. The stop codon TGA is also significantly favoured over TAG and TAA (frequencies of 411, 71, and 25 occurrences, respectively). The accession numbers of SwissProt hits with an e-value of 9e^-10^and below (1,292 sequences) were submitted to the ProToGo server for GO category assignment \[[@B23]\]. A total of 1,203 of the SwissProt accession numbers could be assigned to GO categories. The results are summarized in Figure [2](#F2){ref-type="fig"}. The functional distribution of the *A. tamarense*ESTs that could be placed among GO categories is typical of other eukaryotes. However, the overall small number (i.e., 20%) of significant hits to GenBank is surprising, suggesting that many *A. tamarense*proteins may be either highly diverged and/or encode novel dinoflagellate-specific functions (e.g., Figure [1](#F1){ref-type="fig"}), or the sequence does not extend into the coding region of the transcript. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Codon Usage in the *A. tamarense*ESTs. ::: ------- ------ -------- ------- ------ ------- -------- ------ ------- -------- ------ -------- TTT F 703 23.1% TCT S 482 10.0% TAT Y 372 18.8% TGT C 251 15.6% TTC F 2335 76.9% TCC S 1348 27.9% TAC Y 1612 81.2% TGC C 1356 84.4% TTA L 61 0.9% TCA S 413 8.6% TAA \* 29 5.6% TGA \* 411 79.8% TTG L 1118 15.7% TCG S 926 19.2% TAG \* 75 14.6% TGG W 1051 100.0% CTT L 902 12.7% CCT P 751 17.6% CAT H 464 25.7% CGT R 475 9.6% CTC L 2296 32.3% CCC P 1382 32.4% CAC H 1340 74.1% CGC R 1779 35.8% CTA L 139 2.0% CCA P 829 19.4% CAA Q 433 14.6% CGA R 426 8.6% CTG L 2596 36.5% CCG P 1307 30.6% CAG Q 2535 85.4% CGG R 1128 22.7% ATT I 715 19.1% ACT T 542 13.1% AAT N 508 21.0% AGT S 344 7.1% ATC I 2770 74.1% ACC T 1442 34.9% AAC N 1915 79.0% AGC S 1310 27.2% ATA I 253 6.8% ACA T 638 15.4% AAA K 415 8.5% AGA R 253 5.1% ATG M 2096 100.0% ACG T 1510 36.5% AAG K 4485 91.5% AGG R 910 18.3% GTT V 686 11.2% GCT A 1195 15.2% GAT D 1117 24.9% GGT G 943 13.8% GTC V 2214 37.8% GCC A 2899 36.8% GAC D 3371 75.1% GGC G 3957 58.1% GTA V 268 4.6% GCA A 1559 19.8% GAA E 750 13.8% GGA G 767 11.3% GTG V 2694 46.0% GCG A 2218 28.2% GAG E 4682 86.2% GGG G 1142 16.8% ------- ------ -------- ------- ------ ------- -------- ------ ------- -------- ------ -------- Analysis is of 515 proteins (81,893 codons). Third position nucleotide usage was T = 12.8%, A = 9.3%, C = 40.7%, G = 37.2%. The asterisk (\*) indicates a stop codon. ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Analyses of *A. tamarense*histone H2A.X.**A) Alignment of *A. tamarense*H2A.X with eukaryotic homologs. The alignment is shaded according to the level of conservation. The symbols above the alignment indicate the location of functional residues (T = trypsin cleavage site, \^ = arginines that contact the DNA helix, \* = H2A-H2B interaction sites, U = ubiquitination site). The annotation below the alignment indicates conserved structural features including the α-helices, loops, and the SQ(E/D)Φgotif. B) A ML tree of H2A and H2A.X. The numbers above and below the branches are the results of ML and NJ bootstrap analyses, respectively. The thick branches indicate \> 0.95 posterior probability from Bayesian inference. Only bootstrap values ≥ 50% are shown. Branch lengths are proportional to the number of substitutions per site (see scale bar). ::: ![](1471-2164-6-80-3) ::: Dinoflagellate gene content and gene families --------------------------------------------- Of species with sequenced genomes, the apicomplexan *Plasmodium falciparum*is the most closely related organism to *A. tamarense*. Both of these species are members of the alveolate lineage with dinoflagellates and apicomplexans forming a monophyletic clade that is sister to the ciliates (e.g., \[[@B23]\]). Sequence comparisons using BLAST revealed that 609 of the 6723 *A. tamarense*ESTs had a significant hit (e-value less than 1e^-10^) to *P. falciparum*proteins. The top 20 most significant hits are shown in Table [3](#T3){ref-type="table"}. The most highly conserved proteins between these organisms include many \"housekeeping\" proteins such as α-tubulin and heat shock protein 70. Despite their close evolutionary relationship, there are however likely to be substantial differences between *A. tamarense*and *P. falciparum*with respect to gene content. Due to the apicomplexan intracellular lifestyle, *P. falciparum*has lost most of the genes related to plastid function as well as other metabolic genes. Many of these same proteins appear in the list of the top BLAST hits against the nr database of GenBank (Table [4](#T4){ref-type="table"}). There were 1,349 hits to the nr database that were better than 1e^-10^. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Top 20 *A. tamarense*EST blast hits against the genome of the apicomplexan *P. falciparum*. ::: ***A. tamarense*EST** **E-Value** **GI Number** **Protein Description** ------------------------ ------------- --------------- ----------------------------------------------------------- UI-D-GC1-aao-m-13-0-UI 1.00E-112 23613558 α-tubulin UI-D-GC1-aav-f-09-0-UI 6.00E-86 23508137 flavoprotein subunit of succinate dehydrogenase UI-D-GC1-aad-d-15-0-UI 9.00E-86 23509363 serine/threonine protein phosphatase UI-D-GC0-aae-b-08-0-UI 2.00E-85 23509135 actin UI-D-GC1-aaz-h-12-0-UI 3.00E-85 23507885 26S proteasome regulatory subunit 4 UI-D-GC1-abe-o-23-0-UI 8.00E-85 23510155 bifunctional dihydrofolate reductase-thymidylate synthase UI-D-GC1-abh-e-16-0-UI 1.00E-84 23612827 hsp70 UI-D-GC0-aae-p-02-0-UI 2.00E-84 23613232 adenosylhomocysteinase UI-D-GC1-aay-i-10-0-UI 3.00E-82 16804988 helicase UI-D-GC1-aau-b-16-0-UI 1.00E-80 23509325 eukaryotic translation initiation factor 2 gamma subunit UI-D-GC0-aae-h-03-0-UI 8.00E-78 23509820 glyceraldehyde-3-phosphate dehydrogenase UI-D-GC1-aao-o-20-0-UI 4.00E-77 23508006 ADP ribosylation factor 1 UI-D-GC0-aae-f-01-0-UI 1.00E-76 23509545 calmodulin UI-D-GC1-abb-n-18-0-UI 2.00E-76 23510206 eukaryotic initiation factor UI-D-GC1-abf-g-07-0-UI 4.00E-75 23612467 HSP86 UI-D-GC1-abd-m-07-0-UI 2.00E-74 23612587 40S ribosomal protein S5 UI-D-GC0-aae-b-08-0-UI 3.00E-74 23509345 actin II UI-D-GC1-aab-m-24-0-UI 4.00E-74 23509670 ribosomal protein S2 UI-D-GC1-aar-f-11-0-UI 3.00E-72 23509852 protein serine/threonine phosphatase UI-D-GC1-aao-b-16-0-UI 1.00E-69 23509877 RNA helicase 1 ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Top 20 hits of the *A. tamarense*ESTs to the GenBank nr database. ::: ***A. tamarense*EST** **E-Value** **GI Number** **Protein Description** **Organism** ------------------------ ------------- --------------- ------------------------------------------------------------- -------------------------- UI-D-GC1-abg-i-22-0-UI 1.00E-110 845405 ribulose 1,5-bisphosphate carboxylase *Gonyaulax polyedra* UI-D-GC1-aao-m-13-0-UI 1.00E-109 135433 alpha tubulin *Oxytricha granulifera* UI-D-GC1-abh-e-16-0-UI 2.00E-98 20143982 hsp70 *Crypthecodinium cohnii* UI-D-GC1-abe-o-23-0-UI 1.00E-96 1169423 bifunctional dihydrofolate reductase-thymidylate synthase *Arabidopsis thaliana* UI-D-GC0-aae-p-02-0-UI 1.00E-91 4416330 S-adenosyl-homocysteine hydrolase like protein *Alexandrium fundyense* UI-D-GC0-aae-h-11-0-UI 2.00E-91 21913167 oxygen evolving enhancer 1 precursor *Heterocapsa triquetra* UI-D-GC1-abh-d-23-0-UI 4.00E-91 32307578 glutamate 1-semialdehyde 2,1-aminomutase *Bigelowiella natans* UI-D-GC1-abe-e-15-0-UI 1.00E-89 27450753 proliferating cell nuclear antigen *Pyrocystis lunula* UI-D-GC1-abb-n-18-0-UI 3.00E-88 28277876 Similar to DEAD box polypeptide 48 *Danio rerio* UI-D-GC1-aav-f-09-0-UI 3.00E-87 15240075 succinate dehydrogenase flavoprotein subunit, mitochondrial *Arabidopsis thaliana* UI-D-GC1-abc-o-16-0-UI 8.00E-85 13560096 ALA dehydratase *Gonyaulax polyedra* UI-D-GC1-aao-o-20-0-UI 1.00E-83 7025460 ADP ribosylation factor 1 *Toxoplasma gondii* UI-D-GC0-aae-b-23-0-UI 5.00E-83 1076185 luciferin-binding protein *Gonyaulax polyedra* UI-D-GC1-aay-i-10-0-UI 9.00E-83 18416493 DEAD/DEAH box helicase, putative *Arabidopsis thaliana* UI-D-GC1-aau-b-16-0-UI 1.00E-82 4503507 eukaryotic translation initiation factor 2, subunit 3 gamma *Homo sapiens* UI-D-GC1-aad-d-15-0-UI 5.00E-81 1346753 Serine/threonine protein phosphatase PP1 isozyme 2 *Acetabularia cliftonii* UI-D-GC1-aaz-h-12-0-UI 1.00E-77 23507885 26S proteasome regulatory subunit 4, putative *Plasmodium falciparum* UI-D-GC1-abc-m-19-0-UI 1.00E-77 32307576 geranyl-geranyl reductase *Bigelowiella natans* UI-D-GC1-abj-e-13-0-UI 2.00E-76 4033509 Calmodulin *Tetrahymena pyriformis* UI-D-GC1-abd-m-07-0-UI 3.00E-75 6831665 40S ribosomal protein S5 *Cicer arietinum* ::: As previously mentioned, our bioinformatic analyses identified 6,723 clusters of unique genes. However, this is likely to be a conservative estimate of the number of unique transcripts that were sequenced. A combination of short 3\'-UTRs and highly conserved coding regions caused many related transcripts to be assembled together, even though their 3\'-UTRs contained sequence differences. For example, two large clusters comprise ESTs that correspond to the plastid *atp*H gene that encodes the ATP synthase C chain. This gene is normally plastid encoded in other photosynthetic eukaryotes. These two clusters form closely related, but clearly distinct sets of transcripts. An additional *atp*H-encoding transcript was identified by a single EST. Together, the three clusters contain 43 ESTs, 16 of which are unique. The N-terminal extensions, which encode the tripartite plastid-targeting signals, share an average 74.3% nucleotide and 68.6% amino acid identity, respectively. Similar to many other species, the dinoflagellate transit peptides appear to be under selection to maintain hydrophobicity rather than a conserved amino acid sequence. This may explain why the nucleotide conservation is greater than that of the encoded amino acids. Five hydrophobic amino acids (phenylalanine, leucine, isoleucine, methionine, and valine) are, for example, encoded by codons with a T in the second position. This combined with the high GC-content at third positions results in higher conservation at second and third positions than at first positions. In addition, the high proportion of alanine (28.6%), leucine (10.2%), and valine (11.8%) rather than phenylalanine (2.4%), isoleucine (3.6%), methionine (4.3%, excluding starting methionine), and tyrosine (0.3%) in the N-terminal extensions may reflect the underlying GC-richness, because alanine, leucine, and valine are encoded by GC-rich codons. It is unclear if these amino acids are evolutionarily selected for specifically, or if they are selected for the combination of their hydrophobic character and the GC-content of their codons. In contrast, the conserved core of the protein shared an average 88.4% nucleotide and 98% amino acid identity, respectively, which corresponds to the more typical pattern of third position variation resulting from selection. The 3\'-UTRs of the *atp*H genes show substantial variation and were difficult to align. There are several groups of more closely related 3\'-UTRs that may be the result of recently duplicated genes. In all, there are five alignable groups of UTRs (and one singleton) that may have originated from more closely related genes. Histone and histone-like proteins in dinoflagellates ---------------------------------------------------- A significant finding of this study is the identification of two rare (2/11,171) ESTs that encode a partial histone H2A.X. The longest cDNA isolated from the library using PCR was predicted to encoded a protein of 169 amino acids that shares high sequence identity to eukaryotic histone H2A.X (Figure [3A](#F3){ref-type="fig"}). This clone putatively lacked only the start codon at the N-terminus. The divergent N-terminus of *A. tamarense*H2A.X is somewhat longer than in other homologs but the remainder of the sequence is conserved (in particular the α-helices of the histone fold). Several functional residues from the known crystal structure are also present in *A. tamarense*H2A.X including the lysine at the trypsin cleavage site, the arginines in the loops that interact with the DNA α-helix, and the lysine ubiquitination site \[[@B24]\]. The sites of interaction with histone H2B are also present. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Analysis of dinoflagellate HLPs.**A) HLPs from dinoflagellates (red taxa names) and bacteria (blue) and HU proteins from bacteria (black). B) The predicted secondary structure of HLPs from *A. tamarense*and *B. pertussis*aligned with the known secondary structure of *E. coli*HU. Curled lines indicate α-helices and jagged lines indicate β-strands. The arrow indicates the position of a conserved lysine. The asterisk indicates the proline that intercalates into the DNA in HU proteins. C) An ML tree of HU and HLP proteins from bacteria and eukaryotes. The numbers above and below the branches result from ML and NJ bootstrap analyses, respectively. The thick branches indicate \> 0.95 posterior probability from Bayesian inference. Only bootstrap values ≥ 50% are shown. Branch lengths are proportional to the number of substitutions per site (see scale bar). ::: ![](1471-2164-6-80-4) ::: H2A.X proteins are closely related to the canonical H2A except for the C-terminus which contains the distinctive SQ(E/D)Φ motif (where Φ is a hydrophobic residue). H2A.X plays an important role in the recognition and repair of double-strand DNA breaks by non-homologous end-joining. At the site of double-strand breaks, the serine of the SQ(E/D)Φ motif is rapidly phosphorylated \[[@B25]\]. The phosphorylation signal spreads a large distance down the chromosome around the breaks, signalling the recruitment of the DNA repair proteins Rad50, Rad51, and BRCA1 \[[@B26],[@B27]\]. We also identified histone H2A and H2A.X from the haptophyte *Emilania huxleyi*through high-throughput EST sequencing of this alga (J. D. H. and D. B. unpublished data). Phylogenetic analysis places *A. tamarense*H2A.X in its predicted position (with moderate bootstrap support) as sister to the *E. huxleyi*homolog within a group of chromalveolates that includes haptophytes, stramenopiles, and apicomplexans (Figure [3B](#F3){ref-type="fig"}). H2A.X from *A. tamarense*, *E. huxleyi*, and *Toxoplasma gondii*do not, however, form a monophyletic group suggesting multiple origins within chromalveolates. This is not surprising because H2A.X appears to have arisen independently many times during eukaryotic evolution \[[@B28],[@B29]\]. We tested the strength of these results using the Approximately Unbiased (AU-) statistical test. A 16-taxon ML backbone tree was generated without *A. tamarense*H2A.X and then we made a set of 17-taxon trees by placing this sequence on every possible branch (29 in total). This analysis provides good support for the position shown in Figure [3B](#F3){ref-type="fig"} (P = 0.827), however, many alternative positions were included in the 5% confidence set of trees (i.e., as sister to *Thalassiosira pseudonana*, *Phaeodactylum tricornutum*, *Homo sapiens*, or *Drosophila melanogaster*, and at the base of or sister to either of the land plants). The lack of robust phylogenetic signal for the divergence point of *A. tamarense*H2A.X likely reflects the short length and high conservation of these histones. Dinoflagellate chromosomes do not contain nucleosomes, instead the DNA is associated with HLPs \[[@B10],[@B30],[@B31]\]. The similarity between dinoflagellate HLPs and bacterial HU and HLPs has only recently been noted and these proteins have not yet been subjected to phylogenetic analysis with a broad taxon sampling \[[@B32]\]. In our *A. tamarense*EST data, HLPs were the most highly represented transcripts (45/11,171 ESTs) and encoded 5 closely related proteins. Alignment of the HLPs from *A. tamarense*and other dinoflagellates with HLPs and HU proteins from bacteria and eukaryotes showed moderate sequence similarity (a representative alignment is shown in Figure [4A](#F4){ref-type="fig"}). This alignment was constructed using information from secondary structure predictions (discussed below). One group of proteins (referred to here as bacterial HLPs) is more closely related to dinoflagellate HLPs and includes Bph2 from *Bordetella pertussis*. Bph2 has a role in virulence gene expression and shares limited (likely convergent) sequence similarity with histone H1 \[[@B33]\]. The dinoflagellate and bacterial HLPs also contain an N-terminal extension in comparison to HU proteins. This extension is rich in alanine, lysine, and proline, which is reminiscent of the C-terminus of histone H1. The dinoflagellate HLP N-termini are however, also enriched in methionines. Compared to the bacterial HLPs, this N-terminal region is generally shorter in the dinoflagellates, although there is variability among species in both groups (Figure [4A](#F4){ref-type="fig"}). In contrast to the primary sequence, secondary structure predictions for these three classes of proteins are remarkably similar. The crystal structure of *E. coli*HU has been determined (PBD ID: 1MUL) and the known secondary structure was compared to the predicted secondary structures of *B. pertussis*Bph2 and an *A. tamarense*HLP (Figure [4B](#F4){ref-type="fig"}). Both types of HLPs are predicted to have two α-helices that are identical in size and spacing to the N-terminal helices in *E. coli*HU, followed by two β-strands that are similar in size and position. We conclude from this analysis that dinoflagellate HLPs show structural similarity to HU proteins from bacteria, however, it is unclear if these proteins are functional homologs. It is also apparent that dinoflagellate HLPs are distantly related to bacterial HU proteins. The dinoflagellates have one putatively homologous functional residue corresponding to Lys3 (arrow in Figure [4A](#F4){ref-type="fig"}) of HU proteins, which interacts with the DNA and is involved in wrapping the DNA around the protein \[[@B34]\]. A proline residue (asterisk in Figure [4A](#F4){ref-type="fig"}), which intercalates into the DNA during HU binding, appears to be conserved among HU proteins and bacterial HLPs, but is not present in the dinoflagellate HLPs \[[@B35]\]. However, there are several prolines conserved among dinoflagellates in the C-terminal end of the protein. The C-terminal arms of HU are critical for the interactions that bend the DNA. Given the low level of sequence similarity and the absence of a homologous proline in this region, it is unclear if the dinoflagellate HLPs are able to interact with DNA in the same manner as HU proteins. In our phylogenetic analyses, the proteobacterial HLPs form a well-supported monophyletic group with the dinoflagellates (Figure [4C](#F4){ref-type="fig"}) suggesting an origin of the dinoflagellate gene through lateral transfer (followed by several rounds of gene duplication). It is also noteworthy that dinoflagellates are the only eukaryotes to possess a proteobacterial form II rubisco \[[@B36]\]. The position of the dinoflagellate HLPs is distinct from that of other eukaryotic HU proteins. These latter proteins group with the canonical HU proteins from bacteria and have likely originated through intracellular transfer from the mitochondrial or plastid endosymbiont. Statistical support for the monophyly of the dinoflagellate and proteobacterial HLPs was tested using the AU-test. In these analyses (details not shown), a sister group relationship between the HLPs was the most highly favored topology (P = 0.659) and all other positions for the dinoflagellates (except branching inside the bacterial HLP clade) had significantly lower probabilities (P \< 0.05). Dinoflagellates no longer use the nucleosome as the major DNA packaging protein complex. Chromosomal DNA strands in these taxa are smooth, in contrast to the \"beads on a string\" conformation in other eukaryotes \[[@B12]\]. The chromosome structure is also unique in that they are uniform in size and morphology, remain condensed throughout the cell cycle, and are birefringent, indicating a liquid crystal state \[[@B5],[@B14],[@B37]\]. Transcription is thought to take place in DNA loops that protrude from the condensed chromosome \[[@B38]\]. It appears that dinoflagellates have acquired DNA binding proteins from a proteobacterium possibly to facilitate the compaction of their immense genomes. HU and related proteins from bacteria induce sharp bends in DNA strands and some models suggest that HLPs are responsible for creating DNA bends at the periphery of the chromosomes \[[@B39],[@B40]\]. Immunolocalization shows dinoflagellate HLP to be associated with the periphery of chromosomes \[[@B41]\]. However, the HLP concentration is very low relative to DNA content. Dinoflagellate chromosomes have a 1:10 protein:DNA ratio (in contrast to the 1:1 ratio in other eukaryotes). The HLP concentration may therefore be too low to function in DNA compaction, rather they may act as transcriptional regulators \[[@B41],[@B42]\]. In summary, our discovery of H2A.X in *A. tamarense*shows that, whereas dinoflagellates appear to no longer use nucleosomes for DNA packaging, at least one histone has been retained and is weakly expressed. Interestingly, in a recent paper, histone H3 appears in a table of redox-regulated genes in the dinoflagellate *Pyrocystis lunula*\[[@B11]\]. Until now, only these two histones have been identified in dinoflagellates and it is unclear if all dinoflagellates possess either of these two genes, or others that have not yet been found. If other histones are present (which is likely), they may however also be expressed at a low level (as is the case for *A. tamarense*H2A.X). This would render difficult their identification using the EST-based approach unless comprehensive sequencing of normalized and subtracted cDNA libraries is used. In metazoans, replication-dependant canonical histone (H2A, H2B, H3 and H4) mRNAs are not polyadenylated, raising the possibility that they have been excluded from this poly-A primed cDNA library \[[@B43]\]. However, these histone mRNAs are polyadenylated in plants, apicomplexans, and ciliates, suggesting that if they are present, they may be in dinoflagellates as well \[[@B44]-[@B46]\]. Given the important role that H2A.X plays in DNA repair, we speculate that this gene may have been maintained specifically to perform this function. Consistent with this idea, the core region of *A. tamarense*H2A.X is highly conserved, indicating that it may still be able to interact with DNA in a manner similar to H2A in other species. Conclusion ========== This collection of ESTs is the most extensive genomic resource for a toxic dinoflagellate species to date and provides a useful glimpse into its nuclear genome. These data will be instrumental to future research to understand the unique and complex cell biology of these organisms and for understanding the method of toxin production in these species. We have likely not yet exhausted the gene discovery potential using the EST approach (i.e., note the high discovery rate of our normalized library). In the future, we will use serial subtraction of cDNA libraries to improve/maintain the novelty rate of our cDNA library and create cDNA libraries from *A. tamarense*under various growth conditions and life history stages to get generate a more complete catalog of the gene content of this important organism. Methods ======= Library construction -------------------- Total RNA from a culture of the toxic dinoflagellate *Alexandrium tamarense*(CCMP 1598) was extracted using Trizol (GibcoBRL) and mRNA purified using the Oligotex mRNA Midi Kit (Qiagen). This culture strain was produced by isolating a single cyst, a diploid resting stage that produces haploid vegetative cells by meiosis. However, it is unknown if a single or multiple vegetative cells were isolated after antibiotic treatment to make the culture axenic. If a single vegetative cell was isolated, the culture would be clonal. The culture was grown at 20°C on a 13:11 hour light:dark cycle (80 μEinsteins of light) in L1 media. Start and normalized directionally cloned (3\' NotI-5\'EcoR1) cDNA libraries were constructed as previously described \[[@B47]\]. ESTs were sequenced from the 3\' end to maximize clustering accuracy using the 3\' untranslated region (UTR). All ESTs were processed as previously described \[[@B48]\]. Identification of a total of a non-redundant \"unigene\" set of 6,723 unique clusters from 11,171 sequences was accomplished using using UIcluster v3.0.5 \[[@B49]\]. Phylogenetic analyses --------------------- Data was gathered from GenBank (including the recently released *Karenia brevis*EST data, Frances Van Dolah, unpublished data) using blast searches. Maximum likelihood (ML) analyses were done with PHYLIP using the JTT model of protein evolution with gamma corrected rates (JTT + Γ) with 5 random additions \[[@B50]\]. ML bootstrap analyses (100 replications) were done as described with either 5 (histone H2A) or 1 (HLPs) rounds of random taxon addition. Bayesian analyses were done using MrBayes V3.0b4 \[[@B51]\]. Four chains (1 cold, 3 heated) were run for 1 million generations, sampled every 1000 generations, using the JTT + Γ model. The first 500 trees were discarded as burn-in. Neighbor joining (NJ) bootstrap (500 replicates) analyses were done with PHYLIP using the JTT + Γ model. Minimum evolution (ME) analyses done with PHYLIP using the JTT + Γ model with global rearrangements and 10 rounds of random taxon addition (1 round was used in the bootstrap analysis). The Approximately Unbiased test was done using CONSEL \[[@B52]\]. ML trees without the groups of interest were generated as described above. A pool of trees was then generated by adding the group of interest (*A. tamarense*H2A.X or dinoflagellate HLPs) to every possible branch in the ML tree. For the HLP analyses, a reduced taxon set was used that included *Bordetella*, *Ralstonia*, *Xylella*, *Pasteurella*, *Nostoc*, *Synechocystis*, *Agrobacterium*, *Rikettsia*, *Escherichia*, *Guillardia*, *Cyanidioschyzon*, *Sorghum*, *Toxoplasma*, *Xenopus*, and *Homo*. *A. tamarense*1 and *C. cohnii*HCC2 were added as a monophyletic group to every branch in this reduced ML tree. Secondary structure prediction was done using Jpred \[53, 54\]. The consensus secondary structures were used in the comparison to the know structure of E. coli HU (PDB ID: 1MUL). Authors\' contributions ======================= JDH constructed the cDNA libraries and did the sequence and phylogenetic analyses, the Blast and GO analyses on the EST dataset, the histone and HLP analyses, and drafted the manuscript. TES did many of the other global sequence analyses of the EST dataset. HSY contributed intellectually to the manuscript. Library construction and high-throughput EST sequencing was done in the laboratory of MBS and was supervised by MFB. The bioinformatics group led by TLC did the EST sequence processing and clustering. DB conceived of and supervised this study and contributed to the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ JDH was supported by an Institutional NRSA (T 32 GM98629) from the National Institutes of Health. This work was supported by grants from the National Science Foundation awarded to DB (DEB 01-07754, MCB 02-36631). TES was partially supported by a Career Development Award from Research to Prevent Blindness.

PubMed Central

2024-06-05T03:55:58.380180

2005-5-29

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173104/", "journal": "BMC Genomics. 2005 May 29; 6:80", "authors": [ { "first": "Jeremiah D", "last": "Hackett" }, { "first": "Todd E", "last": "Scheetz" }, { "first": "Hwan Su", "last": "Yoon" }, { "first": "Marcelo B", "last": "Soares" }, { "first": "Maria F", "last": "Bonaldo" }, { "first": "Thomas L", "last": "Casavant" }, { "first": "Debashish", "last": "Bhattacharya" } ] }

PMC1173105

Background ========== Measurement of foot pressure distribution (FPD) is clinically useful because it can identify anatomical foot deformities \[[@B1]\], guide the diagnosis and treatment of gait disorders and falls, as well lead to strategies for preventing pressure ulcers in diabetes. Age-related anatomical and physiological changes in foot bone and ligament structure affect FPD during gait \[[@B1]\]. Gait analysis of healthy elderly people has revealed decreased stride length, reduced step force and increased variability in gait parameters. These findings indicated that unsteadiness during walking is increased in the community-dwelling elderly people, posing a risk for falls \[[@B2]\]. Age was independently associated with lower pressure under the heel, midfoot, and hallux in the multivariate analysis \[[@B3]\]. Foot pressure studies during walking have focused on specific pathology and deformity \[[@B4]-[@B6]\] specific anatomical areas \[[@B7]\], exercise \[[@B8]\] and younger subjects \[[@B9]-[@B11]\]. Knowledge of the plantar FPD map during normal walking in healthy elderly people is lacking. It is not known if distribution of plantar pressure, force, and load across several anatomical regions of the foot during walking is different between young and old. To determine the effects of normal aging on FPD, we evaluated the anatomical distribution of plantar pressure, force, and relative loads in healthy young and old subjects during walking. Methods ======= Subjects -------- The study was performed in the Syncope and Falls in the Elderly (SAFE) Laboratory at the Beth Israel Deaconess Medical Center, and all subjects signed informed consent approved by the Institutional Review Board. Subject characteristics are shown in the Table [1](#T1){ref-type="table"}: 9 young subjects (5 men and 4 women; mean age 30 ± 5.2 years, mean ± SD, range 23--39 years), and 6 elderly subjects (4 men and 2 women, mean age 68.7 ± 4.8, range 63--74 years). All subjects were screened with a detailed medical history, physical activity questionnaire, electrocardiogram, and were not treated for any systemic disease. Inclusion criteria were: age range 20--40 or 60--80 years, ability to walk a flight of stairs, and ability to walk for 12 minutes. All subjects were normotensive by medical history, confirmed by measurements of the sitting and standing heart rate and blood pressure. Height and body mass were measured and used in FPD analysis. Shoe size and foot imprints were taken to determine the insole size. Exclusion criteria were: neurological or musculoskeletal abnormalities affecting gait, peripheral neuropathies, diabetes, past hip/leg/foot trauma or surgery, dementia, epilepsy, alcohol, smoking and other drug abuse, current pregnancy, history or physical evidence for coronary artery disease, abnormal electrocardiogram (ST-T wave changes, old myocardial infarction, arrhythmias, bundle branch block), history of more than one fall or syncope in the past year, orthostatic hypotension, systemic disease requiring continuous medical treatment for \>1 month, malignant neoplasms, hepatic, renal, heart disease or failure, hypertension, and body mass index (BMI) \>35. The arch indexes of the subjects were not measured. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics ::: **Demographics** **Young** **Old** **P** --------------------------------- ------------- ------------- ---------- Age (years) 30 ± 5.2 68.7 ± 4.4 \<0.0001 Men/Women 4/5 2/4 Mass (kg) 72.3 ± 7.5 70.3 ± 13.6 NS Height (cm) 175 ± 9.5 169.9 ± 7.8 NS BMI 23.2 ± 4.6 24.2 ± 3.1 NS NWS (m/s) 1.25 ± 0.25 1.2 ± 0.2 NS Stride (s) 1.1 ± 0.2 1.04 ± 0.1 NS Step (s) 0.6 ± 0.1 0.6 ± 0.1 NS Stance (s) 0.6 ± 0.1 0.6 ± 0.1 NS Swing (s) 0.5 ± 0.1 0.4 ± 0.04 NS Rate Perceived Exertion (RPE) 2.5 ± 1.4 4.2 ± 1.2 0.004 Δ RPE Beginning-End 0.22 ± 0.7 0.1 ± 0.4 NS Physical Activity Questionnaire 4.8 ± 1.1 4.9 ± 1.1 NS Mean ± SD NWS -- normal walking speed BMI -- body mass index ::: Experimental protocol and data acquisition ------------------------------------------ Foot pressure distribution was measured during walking on the treadmill at individual normal walking speed. A treadmill was used to ensure a consistent speed, despite of its artificial milieu, as plantar pressure and force vary at different gait speeds \[[@B6]\]. Self-paced normal walking speed (NWS) was determined with hallway walking for 12 minutes. During walking, the subject was followed by the investigator for safety and timing purposes. Total distance (m) was divided by the walking time to determine individual NWS. Gait characteristics at NWS (step, stride, stance and percentage of the initial and terminal double stance) were not different between the groups (Table [1](#T1){ref-type="table"}). The treadmill walk started at 0.8 mph for all subjects and increased 0.2 mph every 30 seconds until NWS was achieved. All subjects walked at their NWS for 6 minutes. FPD analysis was done on 100 steps (50 left and 50 right) selected from steady data segments after 2 minutes of sustained treadmill walking. At the end of walking, the treadmill decelerated over 1-minute and stopped. Subjects rated perceived exertion using 10-point Borg Rating Perceived Exertion scale \[[@B12]\] before the treadmill walk started, at the end of the speed increase period, and at the end of normal walking. Subjects sat on a chair while heart rate and blood pressure were monitored for six additional minutes during the post-walk period. Foot pressure distribution was measured using the shoe insoles with 99 capacitive sensors, connected to a small portable data acquisition device that sampled pressure for each sensor at 50 Hz (Pedar Mobile, Novel Electronics Inc., GmbH Munich, Germany). The insoles were calibrated regularly using a \'Trublu\' calibration device (Novel GmbH, Munich, Germany). Two insole sizes (size WW = European shoe size 40/41, size XW = European 41/42) were used to account for differences in foot sizes. To control for differences in personal footwear, all subjects were provided with a standard, thin pair of slippers. Data analysis ------------- All data were visually inspected prior to analysis to assure high quality of data acquisition. Figure [1A](#F1){ref-type="fig"} shows distribution of maximum pressure for one step for all sensors. Time-series pressure measurements for all sensors were grouped into nine anatomical masks \[[@B5],[@B13],[@B14]\] (Figure [1B](#F1){ref-type="fig"}). These masks corresponded to the following anatomical areas: medial calcaneus, lateral calcaneus, medial arch, lateral arch, first metatarsal, metatarsals two and three, metatarsals four and five, hallux, and toes. The following 5 variables were calculated for the each mask: maximum pressure, maximum force, mean pressure, mean force, and relative load. All variables were calculated for each step and then averaged over the 50 steps for each foot. Maximum pressure was defined as the greatest pressure any single sensor in each mask measured in a single step, and these values were averaged separately for each mask over 50 steps. Mean pressure was defined as the average of all activated sensors in a mask for a single step. To calculate maximum and mean forces, the pressure time-series data were converted to force by multiplying each pressure value with the cross-sectional area of the corresponding sensor. All sensors in a defined mask were added together for each time frame to give the summed time-series for force, which was the total force for each mask. The maximum force was defined as the greatest force exerted for each mask in a single step. The mean force was defined as the average force exerted in each mask for a single step. Body weight was significantly different between men and women (p \< 0.0001). All variables were normalized by body weight (BW) and the area of each mask, to account for these factors. Relative load was defined as the ratio of the total force in a specific mask to the total force of all masks combined, expressed as a percentage \[[@B5]\]. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Foot pressure distribution. A. Maximum pressure distribution on all sensors during stance for one subject. B. The nine anatomical masks superimposed on the insole (MC = medial calcaneus, LC = lateral calcaneus, MA = medial arch, LA = lateral arch, MT1 = first metatarse, 3 = second and third metatarse, 4 = fourth and fifth metatarse, H = hallux, and T = toes). ::: ![](1471-2318-5-8-1) ::: Statistical analysis -------------------- The maximum and mean pressure and force were compared between the groups for all masks. In addition, we compared the mean and maximum pressure and force in the medial masks (medial calcaneus, medial arch, first metatarsal, and hallux) to the lateral masks (lateral calcaneus, lateral arch, second and third metatarsal, and toes) and between the groups. We also compared the anterior masks (hallux, toes, first metatarsal, second and third metatarsal, and fourth and fifth metatarsals) to the posterior masks (medial arch, lateral arch, medial calcaneus, and lateral calcaneus) and the anterior masks (hallux and first metatarsal) similarly. Within group comparisons between masks were done using nonparametric ANOVA (Friedman test). Comparisons between groups were done using Wilcoxon nonparametric test using JMP 5.0.1 software (SAS Institute 2003). Results ======= Foot pressure distribution was highly significantly different between masks for the young and old groups for all variables (maximum and mean pressures p \< 0.00001, normalized maximum and mean pressures \<0.00001 and maximum and mean force p \< 0.00001 and normalized maximum and mean force p \< 0.00001). Differences in the foot pressure distribution between the young and old groups for the maximum and mean pressures were confined to the calcaneus region and to the medial masks of the foot. Figure [2](#F2){ref-type="fig"} shows differences in maximum pressure distribution (normalized for body weight (BW)) for all 9 anatomical regions. Elderly subjects had lower normalized maximum pressure in the medial (3.2 ± 0.5 vs. 4.6 ± 1.1 %BW, p = 0.001) and lateral (2.8 ± 0.6 vs. 3.4 ± 0.8 %BW, p = 0.036) calcaneal masks, for all medial masks combined (2.6 ± 0.2 vs. 3.3 ± 0.2 %BW/cm^2^, p = 0.019) and marginally reduced in the hallux mask (2.5 ± 1.2 vs. 3.7 ± 1.6%BW/cm^2^, p = 0.07). In the medial calcaneus region, the elderly group also had lower maximum pressure (22.2 ± 6.3 vs. 32.9 ± 11.8 N/cm^2^, p = 0.01), as well as lower mean (6.2 ± 1.8 vs. 8.9 ± 3.1 N/cm^2^, p = 0.01) and normalized mean pressure (0.9 ± 0.2 vs. 1.2 ± 0.2 %BW/cm^2^, p = 0.0006) (Figure [3A](#F3){ref-type="fig"}). The elderly subjects had also lower maximum force (240.9 ± 77.9 vs. 328.4 ± 138.7 N, p = 0.049) and normalized maximum force (34.3 ± 6.2 vs. 45.6 ± 8.5 %BW, p = 0.001) and the mean force (126.3 ± 34.9 vs. 178.8 ± 71.0 N, p = 0.02) and the normalized mean forces (18.3 ± 3.5 vs. 24.6 ± 4.2 %BW/ cm^2^, p = 0.0006) (Figure [3B](#F3){ref-type="fig"}). In the medial region masks, the elderly group had reduced maximum pressure (17.7 ± 1.6 vs. 23.2 ± 1.4 N/cm^2^, p = 0.02), reduced normalized maximum pressure (2.6 ± 0.2 vs.3.3 ± 2.0 %BW, p = 0.019); and reduced normalized mean pressure (0.5 ± 0.1 vs. 0.7 ± 0.1 % BW, p = 0.037) and borderline mean pressure (3.5 ± 0.5 vs. 4.8 ± 0.4 N/cm^2^, p = 0.06). In all anterior region masks, the elderly group exerted lower mean pressure (4.1 ± 2.3 vs. 4.8 ± 2.2 N/cm^2^, p = 0.046), borderline reduced mean force (57.3 ± 36.0 vs. 67.2 ± 33.6 N, p = 0.052) and borderline reduced normalized mean force (8.2 ± 4.5 vs. 9.8 ± 5.0 %BW, p = 0.055). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Pressure distribution by anatomical region. Normalized maximum pressure distribution for the young (white bar) and elderly (black bar) group for each anatomical region (medial calcalneus mask p = 0.0001, lateral calcaneus mask p = 0.03). ::: ![](1471-2318-5-8-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Medial calcaneus mask. A. The average maximum and mean pressures, and normalized mean and maximum pressures for medial calcaneus mask for young and old subjects (\*\*p = 0.01, \*\*\* normalized maximum pressure p = 0.001, \*\*\* normalized mean pressure p = 0.0006). B. The average mean and maximum forces and normalized mean and maximum forces for medial calcaneus for the young and old groups (mean ± SD, maximum force \* p = 0.05 \*\*\* normalized maximumn force p = 0.001, mean force \* p = 0.02, \*\*\* normalized mean force p = 0.0006). ::: ![](1471-2318-5-8-3) ::: In the anterior medial region (hallux and first metatarsal), the elderly group displayed reduced mean pressure (3.6 ± 2.2 vs. 4.7 ± 2.1 N/cm^2^, p = 0.03), reduced normalized mean pressure (0.5 ± 0.03 vs. 0.7 ± 0.3 % BW, p = 0.027). The relative load in the medial masks (11.0 ± 8.8 vs.12.0 ± 7.8%) and the lateral masks (11.2 ± 6.8 vs.10.4 ± 6.2%) were not different. The relative load over the medial and lateral arches was not different between groups. Elderly subjects lower medial pressure values compared to young subjects, indicates that older people had tendency for greater weight bearing on the lateral mask relative to young subjects. The arch height, has not been measured, however, the contact mid-foot area was not different between the groups. Rate perceived exertion was higher in the older subjects at the end of treadmill speed increment (beginning of normal walk) (p \< 0.01), as well as at the end of normal treadmill walking (p \< 0.004). However, the difference in perceived exertion between the beginning and end of normal walk was not significant between groups. There was no significant difference between groups in the physical activity questionnaire on the scale 0--10 (Table [1](#T1){ref-type="table"}). Discussion ========== Our study has shown that elderly people exert less pressure and force under the medial masks of the foot (medial calcaneus, hallux, anterior and posterior medial masks) during heel touch and toe-off phase. This implicates that elderly subjects preferentially bear weight on the lateral foot during normal walking. Lateralization of foot pressure suggested that medial weight bearing from heel-strike to toe-off is limited in older people compared to younger subjects. Well-distributed weight bearing and foot pressure compensate for the forces and heavy loads imposed on the foot during normal walking. Treadmill walking is different than normal walking due to an inability to change speed voluntarily and reduced stride variability. Although an artificial pace and walking environment are imposed by use of a treadmill, it was a tool used to maintain experimental control. Because foot pressure distribution is affected by walking speed and stride variability, \[[@B6]\] it was deemed necessary to control the speed using treadmill walking. In the posterior masks, the older subjects exerted lower maximum pressure and force on the calcaneus region when normalized for body weight, indicating that, along with the results above, maximum pressure at heel strike is also lower in old subjects than in young subjects. These findings may indicate that forces needed to stabilize the ankle during heel touch phase are reduced in older people. In the anterior masks, the elderly subjects also exerted lower normalized mean pressure and lower normalized mean force compared to young subjects. These findings, supported by results in the hallux mask, support the notion that old subjects have lesser ability to push-off in anticipation of the swing phase. Walking may present a challenge to elderly people, and several age-related gait changes have been identified \[[@B2],[@B3]\]. Morag et al. 1997\[[@B3]\] found that age correlated with heel pad stiffness, but to a lesser degree with walking speed, soft tissue characteristics, and height of the medial longitudinal arch. Our study did not confirm an age-related arch-flattening phenomenon to the extent of altered FPD, as forces in the medial arch area were not different between groups. Normal walking speed, stride intervals, timing within the gait cycle and the relative load in the arch area were similar between young and old subjects. These findings rebuke the notion that age-related decline in pressure is due to flattening of the longitudinal arch or that the stride length would be the primary factor underlying the reduced pressure and forces at heel strike. Anatomical foot structure, including soft tissue thickness and arch height, account for 35% of plantar pressure differences during gait \[[@B6]\]. Pressure values under the heel and midfoot are predominantly affected by weight bearing at the heel strike and midstance, whereas pressures in the anterior regions are determined to a greater extent by flexibility, muscle strength, and muscle recruitment \[[@B5]\]. Therefore, age-related soft tissue and bony structure degradation may reduce the capability of the plantar foot to deflect load \[[@B15]\]. FPD pattern in older people was similar to the pattern of experimentally reduced plantar sensation by cooling, emphasizing that decline in proprioception with aging may contribute to these results \[[@B11]\]. Older runners exhibited significantly more knee flexion at heel strike, but the range of motion and peak maximal vertical forces were reduced. The ground impact force and the initial rate of loading at heel strike were greater, indicating loss of shock absorbing capacity in older people \[[@B8]\]. Limitations ----------- The number of subjects in our study was relatively small, but was comparable to other studies of gait and foot pressure \[[@B5],[@B9]\]. However, our study used the large number of steps during steady walking compared to previous studies. Moreover, selection of steps from the middle portion of the walk minimized inconsistencies that may accompany gait initiation and termination \[[@B5]\]. The foot shape (high vs. flat arch) has not been assessed. However, the midfoot contact area and the relative load in the medial and lateral arch masks were not different between groups. Conclusion ========== Healthy aging affects the dynamics of foot pressure distribution during normal walking. The forces and pressures under the medial foot masks were reduced in elderly people, resulting in lower propulsion during the step from the heel-touch to the toe-off phases. Clinically, lateralized foot pressure and lessened propulsion may affect walking ability in elderly people, posing difficulties in balance, forward thrust, and terrain adaptation. List of abbreviations ===================== BMI = body mass index BW = body weight FPD = foot pressure distribution NWS = normal walking speed Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MH -- the recipient of American federation on Aging Scholarship, designed and performed data analysis and wrote the first version of the manuscript. MV -- participated in the experiments, software development and analysis. JL -- participated in the experiments and analysis. KH -- participated in the statistical analysis. LL -- contributed to data interpretation and manuscript preparation. VN -- designed the study, conducted the experiments, participated in the analysis, data interpretation, and manuscript preparation. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2318/5/8/prepub> Acknowledgements ================ This study was sponsored by The Older American Pepper Center Grant 2P60 AG08812-11; Research Resource Cores A and B and the General Clinical Research Center grant MO1-RR01302. The authors wish to thank Shiva Gautam PhD for consultation on statistical methods.

PubMed Central

2024-06-05T03:55:58.384462

2005-5-19

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173105/", "journal": "BMC Geriatr. 2005 May 19; 5:8", "authors": [ { "first": "Mary Josephine", "last": "Hessert" }, { "first": "Mitul", "last": "Vyas" }, { "first": "Jason", "last": "Leach" }, { "first": "Kun", "last": "Hu" }, { "first": "Lewis A", "last": "Lipsitz" }, { "first": "Vera", "last": "Novak" } ] }

PMC1173106

Background ========== Patient choice is a key part of current U.K. National Health Service (NHS) Strategy \[[@B1]\]. At a time when the population has access to 24 hour banking and to supermarkets which are open 24 hours a day, specialist consultations in U.K. hospitals for non-emergency conditions remain largely confined to the hours of 9-5 pm. In this regard it would appear that the NHS has failed to keep pace with a society which generally demands fast, convenient and 24/7 services. The Department of Health\'s 10 year plan has echoed these sentiments and specifically states that the NHS has been too slow to change its ways of working to meet modern expectations in this consumer age. While it recognises that the NHS is under funded, money alone can only be a starting point. Cost-effectiveness, it states, is the key to success and this can only be achieved if available resources are used to achieve real benefits for patients. Clearly, the redundancy of most outpatient clinics after 5 pm and during the weekend is not an efficient use of resources, nor is it a modern way of working. Medical practice nowadays strives to be patient-centred \[[@B2]\]. There is good evidence that involving patients in decision-making and listening to the patient voice improves health outcomes and patient satisfaction. Hence, when we undertook an extensive literature review, we were surprised to find no published work concerning medical patients\' views on out of hours clinics in secondary care. Given the current government proposals we felt it necessary to explore patients\' views about out of hours clinics amongst patients attending medical outpatients\' clinics in a central London teaching hospital. After embarking on our study of medical outpatients we subsequently found two other reports on patients\' opinions on out of hours ophthalmology clinics. The first is a study of patients with ophthalmological conditions attending a hospital in Leeds and Bristol which looked at whether patients found their current appointments and surgery times convenient or inconvenient \[[@B3]\]. They found that 89% of their patients were apparently happy with their current appointment times but when asked about alternatives would find Saturday mornings convenient. The second study was a critique of the first and reported the responses of 100 consecutive patients attending Saturday morning ophthalmology clinics at an outer London hospital \[[@B4]\]. They were asked whether they preferred a Saturday morning clinic, a regular 9-5 appointment, or whether they didn\'t mind either. They found that 50% had no preference, 41% preferred a Saturday morning and only 9% preferred a 9-5 pm weekday appointment. Both these studies have their limitations in that the first essentially asked about convenience not preference and the second may be affected by bias in that verbal responses alone were reported of a select group of patients i.e. only those already attending a Saturday morning clinic. These studies are based in ophthalmology clinics and may not be representative of all patients attending hospital outpatient clinics. Our study is based on respiratory and cardiology outpatient clinics. These two disease areas account for a major part of the burden of medical illness in the UK with 8 million people having lung disease alone. Methods ======= Centre of data collection: Charing Cross Hospital, London Clinics: Respiratory, Monday morning, Wednesday morning and Thursday afternoons; Cardiology, Thursday morning Duration of data collection: 6 months Patient Profile: Outpatients routed through the central booking office (excluding rapid or open access respiratory and cardiology clinics). 300 sequential attendees at Cardiorespiratory clinics were asked to complete a questionnaire \[see [additional file 1](#S1){ref-type="supplementary-material"}\] to determine their opinions regarding access to out of hours (defined as after 5.00 pm) and weekend outpatient clinics. They were then asked to give up to five reasons as to why they might be interested in such clinics. The patients included a mixture of patients attending for the first time and those attending for follow-up appointments. The reception staff were specifically asked to give the questionnaire to all patients who had not previously received them. The items selected for inclusion in the questionnaire had been raised repetitively in another study (not yet published) looking at reasons why patients did not attend clinics (forgetfulness, long waits, seeing a different doctor each time, nowhere to park, time off work, no one available for childcare). Data was analysed using the statistical package SPSS (version 12.0), however as this was an observational study power calculations were not calculated and data was only subjected to descriptive analysis Results ======= 264 completed questionnaires were obtained (response rate 88%) from 124 males and 122 females with a mean age (SD) of 58 years (16.94). 144 of the patients were attending with a respiratory condition (amongst which were 52 with asthma, 18 with sleep related breathing disorders, 13 with unexplained cough) 65 with a cardiac condition (25 with ischaemic heart disease, 12 with arrhythmias, 9 with hypertension), and the remainder did not state their condition or had an ill-defined medical problem. 165 patients (62.5%) wanted either an out of hours clinic or a clinic on a Saturday or Sunday. Of the reasons given, 81 patients (49.1%) said it was the greater flexibility that such clinics offered with 64 (38.8%) specifically stating that it was because they worked during the day. Forty one (24.8%) responded that it was easier for them to get to the hospital in the evening because someone could drive them there. The same number responded that it was easier to get someone to accompany them in the evening, and 22 (13.3%) stated that the reason was because they had family commitments during the day. 18 patients listed other reasons (see Table [1](#T1){ref-type="table"}) which included important practical issues such that parking restrictions in the environs of the hospital would not apply in the evening and at weekends. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### \'Other\' Reasons volunteered by patients for wanting out of hours clinics ::: Reason Number of patients -------------------------------------------------------------------------------------- -------------------- I am a shiftworker or self-employed so out of hours clinics would be more convenient 4 Less traffic and less stressful to get to the hospital 3 Parking restrictions would not apply 3 I would not have to finance child care at these times 3 Waiting times would probably be reduced 2 ::: When the responses were sub-analysed according to whether the respondent was over or under the age of 65 years, 152 patients were seen to be under the age of 65. Of those 152 patients, 110 (72.4%) expressed a preference for out of hours clinics, whereas amongst the 100 patients aged over 65, 49 (49%) expressed such a wish. Not surprisingly amongst those aged under 65, the commonest reason stated for why they were interested in an out of hours clinic was that they worked during the daytime, and this response was given by 58 out of the 110 of the patients (53%) The responses also varied according to the medical condition of the patient. Of the 144 patients attending with a respiratory condition, 97 (67.4%) wanted out of hours clinics, compared to 36 out of the 65 (55.4%) patients attending with a cardiac condition. However, the mean age of respiratory patients was 55 years compared to 65 years in the cardiac group. Of those patients who wanted out of hours clinics, weekend clinics appeared to be consistently more popular than evening clinics and this trend was upheld even when the responses were analysed according to age, gender and medical condition. Discussion ========== The majority of medical outpatients we questioned would like more flexible timing of clinics. The reasons expressed to justify such a preference appear to be entirely reasonable, such as work commitments during the day. An extension of hours for routine consultations may therefore enhance care for those who are in employment. It also represents a logical fuller use of outpatient consultation facilities. However, out of hours clinics involve more than a doctor-patient consultation. Additional reception and nursing staff are likely to be needed and if the consultation is to be truly effective for the patient they would need to be able to have simple investigations like blood tests, ECG and imaging tests at the same time. Further economic evaluation of the cost of making such services available would be necessary before the widespread introduction of out of hours clinics. However, not all outpatients require investigations and it may not be difficult to construct a list of conditions where patient attending for follow-up can be offered out of hours services without other departments needing to be available. The patients surveyed in this study were patients with medical conditions attending an inner city London teaching hospital. In a patient centred health service similar studies should probably be done amongst those with other conditions living in other areas. The response rate for our study was high and reasons for non response are not known but could include the patient having gone straight into see the doctor and having no waiting period in which to complete the form. In a multicultural society some may have had difficulty with a questionnaire in the English language and up to 15% of patients attending some UK out patient clinics may be functionally illiterate \[[@B5]\]. The response rate is unlikely to have altered the validity of the conclusions which are similar to one study in an ophthalmology clinic \[[@B4]\] but dissimilar to another \[[@B3]\]. The differences almost certainly reflect the question which is asked (stating actual preferences versus stating satisfaction with what you have) and how it is asked and by whom. Conclusion ========== We feel that this is an interesting and important area of research with the results suggesting that the hours of opening of traditional medical outpatients clinics should be extended. Each hospital and each speciality should probably survey their own clientele, but to enable comparisons to be made some form of standardised wording should be used and the questionnaire itself should be included in any subsequent report. Competing interests =================== MRP, CF and NJR do not have any significant competing interests (financial or non-financial). Authors\' contributions ======================= MRP developed the study design and protocol. NJR and CF were responsible for implementing the study and analysing the results. All contributed to drafting the report and approved final manuscript. [Additional file 1](#S1){ref-type="supplementary-material"}: Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1472-6963/5/47/prepub> Supplementary Material ====================== ::: {.caption} ###### Additional File 1 A copy of the questionnaire used for this study. ::: ::: {.caption} ###### Click here for file :::

PubMed Central

2024-06-05T03:55:58.386451

2005-6-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173106/", "journal": "BMC Health Serv Res. 2005 Jun 24; 5:47", "authors": [ { "first": "Claire L", "last": "Feeney" }, { "first": "Nicola J", "last": "Roberts" }, { "first": "Martyn R", "last": "Partridge" } ] }

PMC1173107

Background ========== The cancellous bone microarchitecture corresponds to the spatial organization and morphology of the trabecular network. Sugita *et al*. \[[@B1]\] observed that the mechanical behavior of the cancellous bone varied following the testing direction; these variations were interpreted as an anisotropic feature of the bone stress. Trabecular bone anisotropy corresponds to the preferential orientation(s) of trabeculae. Anisotropy is constituted under the influence of prefererential oriented strength applied to bone \[[@B2]\] and permits to establish resistance to these strengths in a given preferential direction. Sugita *et al*. \[[@B1]\] concluded that anisotropy of the cancellous bone should be considered to predict the fracture risk. The determination of collagen and crystal orientation in connective tissues at molecular scale has been studied, for instance by diffraction \[[@B3]-[@B6]\]. Amorphous materials such as reinforced concrete have been characterized by x-rays and the orientation of fibers investigated by the Fourier transform method \[[@B7]\]. Defossez et al. assessed several methods on images of the femoral neck region: co-occurrence and run length matrix, spatial-frequency and fractal techniques to determine the trabecular direction \[[@B8]\]. These methods exhibited a high degree of accuracy, suggesting a possibility of trabecular orientation characterization in clinical practice \[[@B8]\]. Different methods are available to characterize the structural anisotropy on bone radiographs. In 1970, Singh *et al*. \[[@B9]\] developed a semi-quantitative index applied to femoral neck radiographs. This index is based on the existence of several arches of trabeculae in the femoral neck. Some arches preferentially disappear with age and osteoporosis, and the count of these arch systems can be used to determine the Singh index. Aggarwal *et al*. \[[@B10]\] developed a similar index for the calcaneus. Whitehouse *et al*. \[[@B11]\] introduced the Mean Intercept Length (MIL) defined as the average length between two bone/marrow interfaces. The MIL polar diagram generates an ellipse for 2D slice images \[[@B11]\] or projectional images such as radiographs \[[@B12]\], and an ellipsoid for 3D images \[[@B13]\]. A MIL tensor can be calculated \[[@B13]\] by fitting the MIL values and can be used to determine the Degree of Anisotropy (DA) \[[@B14]\]. The MIL method requires a previous binarization of the gray-level images, then it is a suitable method for the calculation of DA from 3D images obtained by Quantitative Computed Tomography (QCT) \[[@B15],[@B16]\] or Magnetic Resonance Imaging \[[@B17]\]. Therefore, the binarization on 2D radiographs remains a critical step due to partial volume effect. The MIL method was used by Luo *et al*. \[[@B18]\] to compare the anisotropy of the 3D structure and of the 2D projected image of one *in vitro*sample and 14 simulation models derived from this sample. The authors reported a very good correlation (r = 0.99) between anisotropy values assessed on the 3D trabecular structure and the 2D projection images. Luo *et al*. \[[@B18]\] insisted on the interest of the estimation of architectural fabric from plain radiographs. Few methods evaluating the trabecular structure have been developed on radiographs. Caldwell *et al*. \[[@B19]\] developed a method characterizing trabecular bone structure on vertebral digitized radiographs. An evaluation of the orientation named the fraction orientation edges is obtained from the histogram of the magnitude of the edge gradient versus the direction of the edges. Geraets *et al*. \[[@B20]\] developed an index called Line Fraction Deviation (LFD) derived from the tresholded radiographs and based on the fraction of bone pixels. A diagram plotting the standard deviation of this fraction in all directions could reflect the orientation of the trabecular bone. This method has been found more sensitive than MIL method on bone radiographs to evaluate anisotropy \[[@B12]\]. Jiang *et al*. \[[@B21]\] developed a method for a non invasive evaluation of bone mechanical properties: textural features, the global Minkowski dimension and trabecular orientation were determined using Minkowski dimension analysis. Jiang *et al*. demonstrated the contribution of normalized Bone Mineral Density (BMD), structural features and patient age to bone mechanical properties. In 1993, Oxnard reported slight visible microarchitectural changes from the Fourier transform on bone radiographic images but no parameter calculation was performed \[[@B22]\]. The same group worked on vertebral bodies T1 to L5 in seven male columns and applied the FFT to the radiographic images of these vertebrae. They characterized the orientation and size of the architectural elements of vertebral cancellous bone, and underline the potential of this technique to define the effects of ageing \[[@B23]\]. Wigderowitz *et al*. considered the properties of Fast Fourier Transform (FFT) to evaluate trabecular bone structure: the directional pattern of frequencies with high-magnitude can identify the orientation of trabeculae. He defined three indices including spectral trabecular index, longitudinal and transversal trabecular indices \[[@B24]\]. These indices approximate the evaluation of the trabecular structure in terms of trabeculae spacing and orientation. Wigderowitz *et al*. concluded that this quantification detects structural changes occurring with age and may be useful in osteoporosis studies \[[@B24]\]. Caligiuri *et al*. \[[@B25]\] also used the FFT to perform a structural analysis of the spine trabecular bone. The hypothesis was that the magnitude and the first moment of the power spectrum could correspond to the coarseness or the fineness of the texture pattern. The two texture measurements were compared to BMD. Caligiuri *et al*. concluded that this texture analysis might lead to a better prediction of the osteoporotic fracture risk \[[@B25]\]. Recently, Gregory et al. have proposed a FFT-based analysis of trabecular bone structure on hip radiographs from patients with and without hip fracture. They worked on FFT profiles parallel and perpendicular to the preferred orientation. Principal conponents analysis was used to generate scores from the profiles and was able to discriminate fracture and control groups better than fractal dimension \[[@B26]\]. Our research group has previously experienced fractal analysis on bone radiographs \[[@B27]-[@B29]\]. The measured parameter (Hurst coefficient) offers a global evaluation of the image irregularity, of its complexity. It is considered as an indicator of roughness. The Maximum Likelihood Estimator is applied first on one direction, then the process is repeated in 36 directions by 10 degrees steps. The directional results can be expressed on a polar diagram that can be itself be fitted to an ellipse; the shape of this polar diagram can be characterized in order to determine the textural anisotropy \[[@B30],[@B31]\]. When comparing radius and calcaneus in the same subjects, Lespessailles *et al*. have shown significant differences \[[@B31]\] but the interpretation is equivocal because this technique characterizes the variations of grey level roughness following directions, and not directly the trabecular network orientation. The osteoporotic fracture risk prediction is important especially in postmenopausal women. BMD is currently measured in clinical practice and textural parameters can be assessed on trabecular bone radiographs \[[@B19]-[@B22],[@B24],[@B25],[@B31]\], but the anisotropy was not clearly quantified in these texture analyses. To address this question, we have worked on a quantitative evaluation of anisotropy on bone radiographs. The development of a structural parameter such as DA from radiographs is very promising; this parameter could be complementary of the BMD and of fractal texture parameters in the explanation of bone strength. We describe here a new quantitative method based on the FFT to obtain anisotropy indices on bone radiographs of the calcaneus and a validation on synthetic images. The intra and inter-observer reproducibility and a pilot clinical study comparing osteoporotic fracture cases to control cases are presented. Methods ======= Population of the pilot study ----------------------------- Postmenopausal women were recruited from a cross-sectional unicenter case control study. The protocol screened 400 women: 349 were enrolled. The inclusion and exclusion criteria have been previously described in details \[[@B28]\]. For this preliminary study, a sample of 22 osteoporotic fracture cases was randomly selected, and was secondarily age-matched with 44 control cases (for each age, we randomly selected two control cases with an equal age ± 5 years). The distribution of fracture cases was 1 hip fracture, 5 wrist fractures, 11 vertebral fractures and 5 metatarsus fractures, all these fractures were considered as low energy fractures. The ages of the control and fracture cases ranged from 45 to 87 years. The mean age was 69.2 ± 11.9 years for fracture cases and the mean age was 69.0 ± 11.7 years for control cases. The BMD was measured by dual energy x-ray absorptiometry (Hologic^®^4500 device) at lumbar spine and femoral neck. The mean lumbar spine BMD was respectively 0.814 ± 0.09 g.cm^-2^and 0.895 ± 0.16 g.cm^-2^(p \< 0.05) for fracture cases and control cases. The mean femoral neck BMD was respectively 0.650 ± 0.09 g.cm^-2^and 0.686 ± 0.12 g.cm^-2^(ns) for the fracture cases and control cases. Study design ------------ ### Image realization This study was performed on trabecular bone radiographic images of the calcaneus. The calcaneus radiographic images were performed following a standardized procedure reported elsewhere \[[@B27]\]. Briefly, the X-ray tube voltage (48 kV), the exposure conditions (18 mAs) and the focal-calcaneus distance (1 meter) were fixed for all patients. Kodak single emulsion Min RG films were used and all developed by the same film processor at fixed developer and fixer temperatures. In order to obtain digitized images of the calcaneus trabecular bone, the radiographic films were digitized with an AGFA Duoscan scanner (AGFA GEVAERT N.V., Morstel, Belgium) with 256 gray-levels. Calcaneus is known to be a heterogeneous site of trabecular bone \[[@B32]\]. For this reason we have selected a large region of interest (ROI) of 2.7 × 2.7 cm^2^(256 × 256 pixels with a pixel size of 105 μm). The ROI of 256 × 256 pixels (Figure [1](#F1){ref-type="fig"}) was defined from anatomic marks \[[@B27]\]. The basis of the ROI was positioned on a line linking the plantar aponeurosis insertion to the superior end of the Achilles tendon insertion, the ROI comprising both compressive and tensile trabecular network \[[@B10]\]. ### Noise filtering The low frequency noise of an image corresponds to the gray-value variations over large distances, due to the radiological artifacts and to the fat tissues projections on the radiograph. In order to remove the low frequency noise and to take into account only trabecular components of the image, we used a convolution filter previously described by Geraets \[[@B33]\]. A kernel box was used; for each pixel, the average gray-value of the box was allocated to the box middle pixel. The new image obtained is the low-frequency image. The window must be small enough to extract the low frequency noise and large enough to prevent the trabecular pattern leaking into the low frequency region. For our images, the optimal size of the box was 25 × 25 pixels. The filtered image was obtained by subtracting the low-frequency image from the original images, this filter also removed cross of high intensity centered on the central pixel due to the finite bondaries of the ROI. ### Fourier transform The Fourier transform represents a signal in spatial frequency space. An image can be considered as a repartition of bright intensities in a (*xOy*) plane and can be expressed as a two-dimensional function *f*(*x*, *y*). The Fourier transform is expressed by a function *F*(μ,υ) with the two variables μ and υ corresponding to spatial frequencies in (μ *Oυ*) plane. The 2D Fourier transform spectrum of an image is expressed by the following formula: ![](1471-2342-5-4-i1.gif) where μ, υ and *x*, *y*are the respective variables of the frequency and the spatial domains and N the size of the image. Any periodic structure in the original spatial-domain image is represented by peaks in the frequency-domain image at a distance corresponding to the period and a direction at right angle of the original orientation. If a mild degree of disorientation is introduced in an oriented periodic structure, the frequencies of the Fast Fourier transform (FFT) spectrum are spread over an angle corresponding to the deviation of the structure in the original image. By analogy, we hypothesized that the periodic structure is represented by trabeculae projection, and the degree of disorientation by anisotropy. The magnitude of the transform corresponds to: ![](1471-2342-5-4-i2.gif) where *Re*= real part of the FFT and *Im*= the imaginary part of FFT. The FFT was calculated on the gray level filtered images of the trabecular bone using the Visilog 5.1 software (Noesis). Then the magnitudes of the frequency images were divided to the total magnitude of the transform to normalize the contrast in the images according to the following formula: ![](1471-2342-5-4-i3.gif) ### Measurements on FFT spectrum Trabecular bone was assimilated to an oriented structure with two main directions, the ROI including longitudinal and transversal trabeculae (Figure [1](#F1){ref-type="fig"}). All the FFT spectra of the calcaneus radiographs involved horizontal and vertical conponents (Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}). The horizontal components were indicative of the longitudinal trabeculae fabric and the vertical components were indicative of the transverse trabeculae fabric. We have to notice that the so-called horizontal and vertebral trabeculae were so described on the ROI, which was slightly rotated by comparison to the original image. The limit between high energy area representing trabeculae and low energy area corresponding to the noise on the FFT spectrum were visually determined by the operator. We measured the spreading angles of the longitudinal and the transversal trabeculae named respectively the Dispersion Longitudinal Index (DLI) and the Dispersion Transverse Index (DTI). In order to obtain an average value, the DTI and DLI parameters were measured in the two symmetrical parts of the FFT spectrum conponents (Figures [2b](#F2){ref-type="fig"} and [3b](#F3){ref-type="fig"}). An index relative to the trabecular fabric or degree of anisotropy was derived from the measured parameters DLI and DTI. The Degree of Anisotropy (DA) was defined as: ![](1471-2342-5-4-i4.gif) For a perfectly isotropic structure, the FFT spectrum has a disc shape and DA is equal to unity. More isotropic trabecular bone structures will have DA values closer to unity. Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"} illustrate two examples of digitized radiographic images and FFT spectra of trabecular bone at calcaneus. Figure [2](#F2){ref-type="fig"} corresponds to a control case and Figure [3](#F3){ref-type="fig"} to a vertebral fracture case. ### Synthetic images To test the reliability of DA assessment, we tested this method on projected volume with known disorientations. We synthesized 6 series of 8 images composed of beam like structures more or less aligned following two directions in order to obtain structures close to trabecular bone of calcaneus. For the less anisotropic structure, horizontal and vertical disorientations varied from 0 to 80° (Figure [4a](#F4){ref-type="fig"}), and for the most anisotropic structure, the disorientation of the beam like structures varied from 0 to 17° (Figure [4c](#F4){ref-type="fig"}). These synthetic images comprised a wide variety of structures, from anisotropic to almost isotropic, a variety much wider than expected in the clinical X-rays. The projections of these volumes were analyzed with the same software as bone radiographs. Data analysis ------------- ### Intra and inter-observer reproducibility of the measurements The two measured parameters DLI and DTI and the derived parameter DA were determined to calculate the intra and inter-observer reproducibility. To determine the intra-observer reproducibility, a single observer performed two sets of measurements on the FFT spectra of 20 subjects with a one day interval between each set. To determine the inter-observer reproducibility, two sets of measurements were performed on the FFT spectra of 20 subjects by two observers. The observers were blinded for each sets of measurements. The intra-observer and inter-observer reproducibilities were calculated for *n*subjects with the root mean square RMS average according to the following formula \[[@B34]\]: ![](1471-2342-5-4-i5.gif) where *SD*~*j*~is the standard deviation for the subject *j*and ![](1471-2342-5-4-i6.gif) is the average of the measurements for the subject *j*. ### Clinical evaluation Results in these two groups were compared using Student\'s *t*-test for comparisons of the means after checking Gaussian distribution. The area under ROC curves was calculated for BMD measurements and DA. Results ======= The results of synthetic images are represented on Table [1](#T1){ref-type="table"}. The DA varied from 1.3 to.3.6, respectively corresponding from more anisotropic (Figures [4a](#F4){ref-type="fig"} and [4b](#F4){ref-type="fig"}) to almost isotropic structures (Figures [4c](#F4){ref-type="fig"} and [4d](#F4){ref-type="fig"}). Table [2](#T2){ref-type="table"} shows the intra-observer and inter-observer RMSCVs. The RMSCVs of DA were respectively 1.5% and 3.1 % Table [3](#T3){ref-type="table"} presents the results of the fabric indices obtained on the 22 osteoporotic fracture cases and 44 control cases. Results of DLI and DTI were statistically significant lower in osteoporotic cases (p \< 0.01, p \< 0.05, respectively) leading to a significantly higher DA. In control cases DA was closer to 1 due to a large spreading of frequencies in the FFT spectra while DA was higher in osteoporotic cases in relation to a narrower frequency spreading. Comparing the DA values from Table [3](#T3){ref-type="table"} to Table [1](#T1){ref-type="table"}, it appears that the controls (DA = 1.7) were in the 0--55° category and the fractures cases (DA = 1.9) were in the 0--45° category. The difference of the mean of DA between cases and controls was 8.7 % and close to the figured variation ![](1471-2342-5-4-i7.gif) corresponding to the 95% confidence interval of the measurements \[[@B35]\]. Differences in DA determined by spectral analysis (p \< 0.01) and lumbar spine BMD (p \< 0.05) were significant (Figure [5](#F5){ref-type="fig"}). Whereas the difference between fracture cases and controls for femoral neck BMD was not significant. From the ROC analysis, the Area Under Curve (AUC) were respectively 0.65, 0.62, 0.64, 0.77 for lumbar spine BMD, femoral neck BMD, total femoral BMD and DA. There is a trend to higher AUC for DA comparatively to BMD measurements, but according to the small size of the population the statistical significance was not reached. Discussion ========== We have developed a new index of trabecular bone anisotropy on radiographs based on a spectral analysis of the gray-level images. In this pilot study, this new index applied to radiographs of the calcaneus has shown a reasonable reproducibility for a technique at its first step of development. As it was demonstrated with synthetic images, the DA of an almost isotropic structure is close to one and the DA of an anisotropic structure is greater than one. We hypothesized that the trabeculae did not randomly disappear, the effect might be a reduction of frequencies on the FFT spectra in preferential directions leading to an increase of anisotropy. The results of this pilot study involving 22 osteoporotic fracture cases and 44 controls suggest that the DA parameter may be potentially useful to distinguish fracture cases from control cases. The main interest of this technique is that it can be applied directly to gray level radiographic images without previous binarization. The radiographic technique provides a projection image. However a good 2D-3D correlation has been described concerning microarchitecture \[[@B29]\] and anisotropy \[[@B18]\] from gray level analysis. The significantly higher DLI and DTI in control cases suggest that there are more various orientations of the elemental structures around the two main orientations (longitudinal and transversal) in control cases than in osteoporosis cases with vertebral fractures. The larger range of orientations in controls corresponds to a lesser anisotropic structure than in osteoporosis. DTI was the best discriminant parameter between fracture patients and control cases but also the less reproducible. There are only few transversal trabeculae comparatively to longitudinal ones and they disappear first with osteoporosis due to a less contribution in bone strength. The difference between fracture cases and controls was close to errors attributed to the intra or inter-observer reproducibility. The intra-observer reproducibility is close to the 1 to 2% of reproducibility found in usual bone densitometry method as dual x-ray absorptiometry \[[@B37]\]. The poor inter-observer reproducibility is due to the difficulty to identify the limit between high energy area of the spectrum (corresponding to trabeculae) and low energy area (corresponding to noise) in relation to the few numbers of trabeculae projections. Since the population size is small, further studies are required with larger groups. We could expect more stringent results with a more homogenous population of fractured patients with vertebral fracture, for instance. The higher anisotropy found in osteoporosis cases is in accordance with the findings of Newitt *et al*. \[[@B38]\] who report that the increase of bone resorption in osteoporosis leads to a loss of thinner trabeculae first, resulting in an increase of anisotropy. The study of Newitt *et al*. was performed with the MIL method on 3D magnetic resonance images at the radius. The MIL method is widely used to determine the trabecular 3D structure anisotropy but some authors \[[@B39],[@B40]\] discuss its reliability since it reflects the boundary orientation rather than the real anisotropy of the structural elements. This concept of transversal and longitudinal systems of trabeculae must be cautiously interpreted in our study. Indeed the ROI on the calcaneus radiographic images is tilted around 45 degrees (Figure [1](#F1){ref-type="fig"}). The trabeculae were named longitudinal and transversal in reference to the radiographic image (Figure [1](#F1){ref-type="fig"}) and not to the ROI orientation (Figures [2a](#F2){ref-type="fig"} and [3a](#F3){ref-type="fig"}). Longitudinal trabeculae correspond to the compressive trabecular network extended from the subtalar joint and the transverse to the tensile trabecular network sweeping backwards and upwards the great tuberosity \[[@B41]\]. The anisotropy of trabecular bone is different according to the skeletal sites: in a study comparing the properties of calcaneus, distal femur, proximal femur and vertebrae on human specimens, Majumdar *et al*. \[[@B17]\] found the highest anisotropy of trabecular bone at the calcaneus followed by distal femur and proximal femur, vertebrae constituting the least anisotropic site. Our results corroborate the studies of Geraets *et al*. \[[@B33]\] on hip and radius, Wigderowitz *et al*. \[[@B24]\] on wrist radiographs, Mosekilde *et al*. \[[@B42]\] on vertebrae and Ciarelli *et al*. \[[@B43]\] on femoral head samples. The Line Fraction Deviation was developed by Geraets on binarized radiographic images, and leads to the determination of the preferential orientations. There was no calculation of an anisotropy index but it was possible to quantify the anisotropy from the deviation of polar diagram from a circle. The lower values of the Line Fraction Deviation index found in osteoporotic subjects were consistent with the early loss of the secondary compressive trabecular group of the hip \[[@B44]\]. Moreover, Geraets *et al*. showed in a study of the distal radius that the Line Fraction Deviation values decreased along the transversal direction with age whereas the orientation along the axial direction remains stable during the entire life \[[@B33]\]. Wigderowitz *et al*. \[[@B24]\] found using the spectral analysis at the distal radius that the transversal trabeculae are preferentially absorbed, thinned and spared with age. Mosekilde *et al*. in vertebrae \[[@B42]\] showed that compressive strength was greater in the vertical direction than in the transversal direction. This anisotropy in vertebral strength increased with age and indicated according to the authors that the transverse trabeculae were selectively removed. Buck et al. have found that the oblique conponents declined in the cranio-caudal direction particularly for age superior to 60 years leading to variation in anisotropy \[[@B23]\]. Ciarelli *et al*. \[[@B43]\] also found that fracture cases have proportionally more trabeculae aligned along the primary load axis (and thus proportionally fewer transverse trabeculae) than control cases; the authors hypothesized that the loss of transverse trabeculae leads to the difference in anisotropy between groups. Zhao *et al*. characterized iliac trabecular bone by micro-QCT and showed that trabeculae thinning led to a more isotropic structure in the first postmenopausal years whereas the structure became more anisotropic in the later years \[[@B45]\]. They hypothesized that, in the later years, the remaining trabeculae would be more widely separated, less connected and some more thickened leading to an increase of anisotropy. It has been well established that the calcaneus structure is heterogeneous \[[@B32],[@B46]\]. Lin *et al*. \[[@B32]\] have studied the calcaneus microarchitecture on MRI images. They analyzed 20 to 25 ROIs 1 × 1 cm^2^in each calcaneus. They found a spatial heterogeneity in the posterior region of 40 %. In our study the ROI was 2.7 × 2.7 cm^2^and represented a much larger area; it was accurately defined by anatomic marks, this point avoiding large variation in fractal analysis \[[@B27]\]. Furthermore our ROI contained both transversal and longitudinal trabeculae, if transversal trabeculae disappeared, it should be possible to calculate DA and a high value will be obtained. As the DLI and DTI measurements were not yet automated, the parameters were measured in the two symmetrical parts of the FFT spectrum in order to obtain an average value. In a near future the automation of our method could be performed, allowing applications to large sets of images. In spite of this lack of automatization the reproducibility of the DA parameter was acceptable and allowed for the accurate characterization of osteoporosis changes in small series. At this first step of development, this clinical study must be considered as a preliminary study evaluating the potential of this anisotropy evaluation. Conclusion ========== This study has shown that the DA can be determined on plain radiographs using spectral analysis. The reproducibility of the DA values may be improved by automating the method. The distinction between fracture cases and control cases is very promising, but further studies are necessary to know if the DA evaluation could improve the osteoporotic fracture risk determination when combined with BMD and other textural parameters such as fractal analysis. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= BBI participated in the design of the study, carried out the measurements and the manuscript preparation, G L carried out measurements of reproducibility and synthetic images, CC performed statistical analysis and participated to the manuscript preparation, R H supervised the application of the Fourier Transform and CLB carried out the design of the study and supervised the manuscript preparation. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2342/5/4/prepub> Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Radiographic image of the calcaneus showing the localization of the Region of Interest (ROI). Inside the ROI, longitudinal trabeculae correspond to the compressive trabecular network extended from the subtalar joint and the transverse to the tensile trabecular network sweeping backwards and upwards the great tuberosity. Arrows show anatomical landmarks corresponding to the bottom of the ROI. ::: ![](1471-2342-5-4-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Control case **a**: ROI of the digitized radiographicimage of calcaneus trabecular bone. **b**: FFT spectrum **A**: corresponds to longitudinal trabeculae **B**: corresponds to transverse trabeculae DLI: Dispersion Longitudinal Index, spreading angle of the longitudinal trabeculae DTI: Dispersion Transverse Index, spreading angle of the transverse trabeculae ::: ![](1471-2342-5-4-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Vertebral fracture case **a**Region of Interest (ROI) of the digitized radiographic image of calcaneus trabecular bone. **b**FFT spectrum DLI : Dispersion Longitudinal Index, spreading angle of the longitudinal trabeculae DTI : Dispersion Transverse Index, spreading angle of the transverse trabeculae ::: ![](1471-2342-5-4-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### Synthetic images **a**: Synthetic image of astructure with orientations between 0 and 80° in 3 dimensional space **b**: Fast Fourier Transform (FFT) spectrum of the image 4a **c**: Synthetic image of a structure with orientations between 0 and 17° in 3 dimensional space **d**: FFT spectrum of the image 4c ::: ![](1471-2342-5-4-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### Box plots of lumbar spine BMD and DA. A T-test was used to compare controls and fractured patients. ::: ![](1471-2342-5-4-5) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### DA assessment on synthetic images composed of cylinders more or less aligned following two directions leading to isotropic and anisotropic structures. ::: Synthetic images ------------------ -------- -------- -------- -------- -------- -------- Angles 0--80° 0--69° 0--52° 0--40° 0--28° 0--17° DA 1.3 1.3 1.8 1.9 2.5 3.8 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Root Mean Square Standard Deviation Coefficient Variation RMSCV(%) for measured indices DLI and DTI and the degree of anisotropy. ::: Intra-observer reproducibility Inter-observer reproducibility ---------------------- -------------------------------- -------------------------------- Measured parameter DLI 2.4 3.4 DTI 2.4 4.1 Calculated parameter DA 1.5 3.1 DLI: Dispersion Longitudinal Index DTI: Dispersion Transversal Index DA: Degree of Anisotropy ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Fabric indices (mean ± SD) from the fabric parameters in osteoporotic fracture cases and control cases. ::: Osteoporotic fractures Controls Statistical significance p ----- ------------------------ ------------- ---------------------------- DLI 55.6 ± 6.2 59.1 ± 6.6 0.04 DTI 41.3 ± 5.9 46.8 ± 6.7 0.002 DA 1.87 ± 0.15 1.72 ± 0.18 0.001 :::

PubMed Central

2024-06-05T03:55:58.387634

2005-5-31

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173107/", "journal": "BMC Med Imaging. 2005 May 31; 5:4", "authors": [ { "first": "Barbara", "last": "Brunet-Imbault" }, { "first": "Gerald", "last": "Lemineur" }, { "first": "Christine", "last": "Chappard" }, { "first": "Rachid", "last": "Harba" }, { "first": "Claude-Laurent", "last": "Benhamou" } ] }

PMC1173108

Background ========== The detection of an unusual, highly conserved sequence motif \"His-phi-His-phi-His-phi-phi\" (phi representing hydrophobic amino acids) in a variety of organisms of all kingdoms led to the definition of a new family of proteins named HIT ([hi]{.underline}stidine triad) \[[@B1]\]. This family has three main branches: the fragile histidine triad (FHit)-related proteins found in animals and fungi, which act as di-adenosine polyphosphate hydrolases and function as tumor suppressors in humans and mice \[[@B2]\] (although the tumor suppressing function is not dependent on ApppA hydrolysis \[[@B3]\]), the GalT (galactose 1 phosphate uridylytransferase) homologues, which have a modified \"His-phi-His-phi-Gln\" motif, which transfers nucleoside monophosphates to phosphorylated secondary substrates rather than hydrolyzing them \[[@B4]\], and the [Hi]{.underline}stidine-[t]{.underline}riad [n]{.underline}ucleotide binding (HinT) homologues, which in eukaryotes are intracellular receptors and hydrolases of purine mononucleotides \[[@B5]\]. Although HinT homologues are found in all kingdoms and this family is the most ancient and widespread branch of the HIT proteins, the cellular function, the substrates and the interaction partners of HinT proteins are largely unknown. In prokaryotes, knowledge of HinT proteins is generally restricted to sequence analyses. In the cell wall-less prokaryote *Mycoplasma hominis*, the cytoplasmic HinT protein interacts with a surface localized membrane complex by binding to the P80 moiety. Interestingly, the genes encoding P80 and P60, the domains of the membrane complex, form an operon with the HinT gene \[[@B6]\]. The identification of a homologous *hit*locus in *M. pulmonis*and access to several sequenced prokaryotic genomes enabled us, in this study, to hypothetically identify interaction partners and thus propose a functional role for HinT in bacteria with small genomes. Results ======= To find out more about the function of prokaryotic HinT proteins we first analyzed the *hit*loci of bacteria with a restricted genome, with the view that they might represent a model for organisms possessing the minimal genetic make-up essential for life as a free-living organism. As polycistronically organized genes often encode proteins that are functionally related (e.g. in a protein-complex formation or as part of a common pathway) we ran a search of known genome sequences for genes closely neighbouring or overlapping the *hit*gene. Species from the *Mollicutes*and the *Chlamydiaceae*fulfill these requirements and were thus analyzed. *Mollicutes*are phenotypically distinguished from other bacteria by their minute size and total lack of a cell wall. They have evolved as a branch of gram-positive bacteria by a process of reductive evolution. The significant genome \"condensation\" (for example the genome of *M. genitalium*is only 580 kbp) was made possible by adopting a parasitic behavior. The primary habitats of human and animal mycoplasmas are the mucous surfaces which they colonize during infection \[[@B7]\]. While the intracellular localization of *Mollicutes*in insect tissues is well established, cell entry of human or animal mycoplasmas seems to be rare and possibly mediated by a site-directed, receptor-mediated event found in chlamydia \[[@B8]\]. However, *Chlamydiaceae*have a number of features in common with mycoplasmas: their small genomes predict a limited metabolic capability, their major target tissue are mucous membranes and they also cause ocular and sexually transmitted diseases. In contrast, the chlamydia are obligate-intracellular bacteria that undergo a unique developmental cycle in which is an alteration in size between the small metabolically inactive infectious elementar body and the relatively large, metabolically active reticular body which is adapted for intracellular growth \[[@B9]\]. Bearing in mind, that in *M. hominis*the HinT-interacting protein is a secreted antigen and thus predicted to play a role in the pathogenicity of mycoplasmas we expand this analysis to include a family of obligate-intracellular organisms known to secrete antigens in the inclusion body in the infected cell. Organization of the *hit*-locus in *Mollicutes* ----------------------------------------------- *Hit*loci with a genomic organization comparable to that of *M. hominis*\[[@B6]\], with the *hit*AB genes encoding two membrane proteins and *hit*L encoding the cytoplasmic HinT, were detected in nearly half of the mollicute genomes analyzed (Figure [1A](#F1){ref-type="fig"}). The highest similarity was found with the *hit*locus of *M. pulmonis*. The MYPU\_0080 encoded protein had 44.3 % identity to the *M. hominis*HinT protein, and the predicted MYPU\_0060 and MYPU\_0070 proteins were 23.8 % and 26.6 % identical to the membrane proteins P80 and P60 of *M. hominis*, which were encoded by *hit*AB. MYPU\_0060 had structural features similar to those of P80, with an amino-terminal signal sequence with a predicted signal peptidase I (SPase I) cleavage site and a predominantly alpha-helical structure \[[@B10]\]. MYPU\_0070 began with an amino-terminal signal sequence of transmembrane helix (from AA 5 to AA 20) and a signal peptidase II recognition site with a lipoprotein attachment site at position 27. Thus, MYPU\_0070 of *M. pulmonis*appears to encode a P60 homologue, a cysteine-anchored lipoprotein. The order of genes within the *hit*loci of *M. mycoides*subsp. *mycoides*SC (MSC), *Mesoplasma florum*(MF) and *M. mobile*(MMOB) was similar with two genes predicted to encode membrane proteins and a downstream *hit*L gene. The similarities between the predicted sequences of MSC\_0500, MF\_235 and MMOB\_910, and P80 of *M. hominis*were quite low and the proteins would be significantly larger than P80. While MMOB\_910 contained a signal peptidase I recognition site, like P80, MSC\_0500 and MF\_235 were pro-lipoproteins with an amino-terminal SPase II recognition sequence. Only MSC\_500 was predicted to have a predominantly alpha-helical structure, MF\_235 being predicted to have a secondary structure of alternate alpha helical and beta sheet regions and MMOB\_910 to consist mainly of beta sheets (data not shown). The proteins encoded by the gene next to *hit*L had little similarity with P60 of *M. hominis*when the whole sequence was compared. However, when the P60 protein region from AA 164 to AA 177 (CS1; ELQKMLLAKLYLLK) was used, identities of 43 % (MMOB\_900) to 57% (MYPU\_0070) were detected and scrutiny of the sequence from AA 226 to AA 234 (CS2; LYLMKYLVE) revealed 60 % (MYPU\_0070) to 70 % identity (MMOB\_900). The corresponding proteins of *Mesoplasma florum*and *M. mycoides*subsp. *mycoides*SC did not contain these conserved sequences. The different HinT homologues had the highest identity with the HinT protein of *M. hominis*, ranging from 44.3 % (MF\_233) to 55.7 % (MMOB\_890). *Hit*loci with a different genomic organization were identified in the mollicutes *Ureaplasma parvum*,*M. pneumoniae*and *M. genitalium*(Figure [1B](#F1){ref-type="fig"}). While UU271 of *U. parvum*, the gene immediately upstream of *hit*L encoded a putative P60 homologue possessing the P60 consensus sequences CS1 (35.7 % identity) and CS2 (30% identity), it differed from all other P60 homologues in possessing a SPase I recognition sequence and two other transmembrane helices (AA 355 to AA 375 and AA 407 to AA 427). The UU273 gene, immediately downstream of *hit*L, encoded a protein with two transmembrane spanning helices, but with no significant similarity to other known proteins. The deduced protein sequence of UU270 did not have any sequence similarity with P80. With six transmembrane segments, it appeared more likely to form a pore than to be surface exposed. The organization of the *hit*locus of *U. parvum*appeared intermediate between the *hit*loci described above and those of *M. pneumoniae*and *M. genitalium*. The sequence identity between the ureaplasma HinT and the HinT proteins of *M. pneumoniae*(42.7 %) and *M. genitalium*(47.4 %) was higher than that between HinT of *U. parvum*and *M. hominis*(39.2 %), and the genes immediately upstream of *hit*L encoded proteins with seven (MPN\_274) or five (MG\_133) transmembrane domains (Figure [1B](#F1){ref-type="fig"}). MPN\_274 is predicted to encode a permease of an ABC transporter \[[@B11]\] suggesting a comparable function in *M. genitalium*and probably also in *U. parvum*. Thus, these *hit*loci of *Mollicutes*analyzed appear to have a *hit*L gene flanked by genes that are predicted to encode membrane-anchored proteins. As in *M. gallisepticum*, the *hit*L gene is not flanked by other genes on the same strand, also exceptions of a polycistronic organization of the *hit*L gene seem to exist within the *Mollicutes*. Organization of the *hit*loci in *Chlamydiaceae* ------------------------------------------------ In the obligately intra cellular *Chlamydiaceae*the order of genes within the *hit*loci and the function of the encoded proteins appeared to be highly conserved, but distinctly different from that of the *Mollicutes*(Figure [1C](#F1){ref-type="fig"}). In all chlamydial species analyzed, the gene upstream of *hit*L encoded a putative cytoplasmic protein with the signature sequence of a metal-dependent protein hydrolase and a large number of metal binding residues (IPR003226/UPF0160), probably indicative of a phosphoesterase function, and an [o]{.underline}ligonucleotide/oligosaccharide-[b]{.underline}inding OB fold (IPR008994). The deduced protein sequence of the gene located downstream of *hit*L contained a 37--47 AA long tandemly repeated ARM repeat fold (IPR008938), which forms a right-handed superhelix and has been implicated in the mediation of protein-protein interactions \[[@B12]\]. Cp267 contained an RGD motif, which plays a role in cell adhesion \[[@B13]\]. However, the presence of RGD in a sequence alone is not sufficient to suggest a biological function for this motif \[[@B14]\]. The presence of two transmembrane helices (AA 314 to AA 334 and AA 428 to AA 448) suggested that it is more likely that Cp267 may interact with the bacterial cell membrane. In all other chlamydial species analyzed, the Cp267 homologues did not contain domains suggestive of membrane attachment. Are the *hit*loci genes co-expressed? ------------------------------------- In bacteria, overlapping genes on the same coding strand may indicate the polycistronic organization of these genes, and co-expressed proteins are often related in function. As the *hit*locus of *M. hominis*is an operon containing three genes \[[@B6]\] the next step was to analyze whether the polycistronic organization of *hit*genes was conserved within the *Mollicutes*and whether this also occurred in *Chlamydiaceae*. Initially, we chose *M. pulmonis*, the species with the greatest similarity to *M. hominis*, for reverse transcription (RT) PCR analysis. To identify the likely boundaries of an mRNA containing MYPU\_0060, MYPU\_0070 and MYPU\_0080, two different primers upstream of the MYPU\_0060/*hit*A gene and three primers downstream of the *hit*L gene were used (Fig. [2A](#F2){ref-type="fig"}). Amplification occurred with primers binding just upstream of the predicted promoter regions 1 or 2 and just downstream of the TAA stop codon of the *hit*L gene (Fig. [2B](#F2){ref-type="fig"}, lanes A and B) indicating a polycistronic mRNA. Amplification was not seen with a primer hybridizing to the sequence downstream of a predicted hairpin loop structure after the TAA stop codon of *hit*L (Figure [2B](#F2){ref-type="fig"}, lanes C1 and C2) suggesting that the mRNA terminates at the predicted hairpin structure. Next, we examined the genes flanking *hit*L in *U. parvum*. Amplicons which spanned the intergenic regions were obtained with primers hybridizing to the center of UU270 and to the 3\'-end of UU272 (Fig. [3C](#F3){ref-type="fig"}), and with primers hybridizing to the 5\'-end of UU271 and to the 3\'-end of UU273 (Fig. [3D](#F3){ref-type="fig"}). No amplification occurred with a primer hybridizing to the 5\'-end of UU270 and the 3\'-end of *hit*L (Fig. [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). These data suggest that HinT is expressed with the flanking genes UU271 and UU273. These findings are in accordance with the predicted termination of transcription by a hairpin loop with a stem energy of -9.1 kcal mole^-1^located 33 nt downstream of the UU273 gene \[[@B15]\]. Of the organisms analyzed so far *U. parvum*was the most demanding organism of the *Mollicutes*in terms of DNA-free, full-length total RNA preparation. Thus the detection of a common RNA from the center of UU270 up to UU273 (Fig. [3D](#F3){ref-type="fig"}) may be due to an unstable RNA or indicatory for UU270 not taking part in the operon structure. To characterize the organization of *hit*loci genes within the *Chlamydiaceae*we analyzed the locus in *Chlamydophila pneumoniae*. The genes Cp266 (encoding HinT), Cp267 and Cp268, which is predicted to encode a solute symporter family protein \[[@B16]\], were predicted to comprise an operon. No amplification occurred when a primer pair that hybridized upstream of Cp265 and downstream of Cp267 was used (data not shown). Amplification was detected using primers binding within each gene (Figure [4A,B,C](#F4){ref-type="fig"}) and with primers binding within Cp265 and Cp266 (Fig. [4D](#F4){ref-type="fig"}) and within Cp266 and Cp267 (Fig. [4E](#F4){ref-type="fig"}). The predicted co-expression of Cp267 and Cp268 downstream of which several putative transcription terminators were detected, was not supported by RT-PCR analysis (Fig. [4F](#F4){ref-type="fig"}), with amplification being obtained when using genomic DNA as a template but not with cDNA. Interactions between the proteins encoded by the *hit*locus of *C. pneumoniae* ------------------------------------------------------------------------------ As the stability of an RNA is increased in regions with pronounced secondary structures, such as transcriptional terminators, and is often decreased in regions lacking such protective structures \[[@B17]\], partial degradation of the chlamydial mRNA may have occurred. The results from RT-PCR analysis of the mRNA of CP265, Cp266 and Cp267 suggest an operon producing a transient mRNA thus suggesting a physical interaction between the products of these genes. Therefore we analyzed the interactions between Cp265, Cp266 and Cp267 using the yeast two-hybrid system \[[@B18]\]. The system is based on the modular organization of the transcription factor GAL4, which has a DNA-binding domain (DB) and an activation domain (AD). When GAL4 binds (via DB) to its cognate binding site, the activation domain AD is brought close to the promoter and activates transcription of the reporter genes *lacZ*and *HIS3*. In the yeast two-hybrid system these two domains, which alone are not able to activate transcription, are separately cloned into pGADT7 (AD) and pGBKT7 (BD) and expressed in fusion with the proteins under investigation. Growth of a histidine dependent yeast strain on histidine deficient media will only occur after transformation with both plasmids if the AD and DB domains are apposed by an interaction between the fusion proteins, which will result in transcription of the reporter genes. The plasmids pGADT7-T and pGBKT7-53, which encode the SV40 large T-antigen and the murine p53 protein, respectively served as a positive control (Fig. [5](#F5){ref-type="fig"} K+; Fig. [6A](#F6){ref-type="fig"}). HinT is known to dimerize \[[@B19],[@B20]\]. We introduced the coding region of *M. hominis*HinT into both pGADT7 and pGBKT7 and transformed the histidine dependent yeast strain AH109 with them. Quantification of transcriptional activation of the reporter gene was achieved using a liquid β-galactosidase assay measuring the hydrolysis of 4-methylumbelliferyl-β, D-galactopyranoside (MUG), yielding the fluorescent molecule 4-methylumbelliferone (4 MU). As expected, dimerization of the fused HinT peptides led to apposition of the AD and DB domains generating a functional transcription factor. The transcription of the reporter genes resulted in growth of the yeast on histidine-deficient agar plates (Fig. [5](#F5){ref-type="fig"}, top row, left plate) and production of β-galactosidase (Fig. [6C](#F6){ref-type="fig"}). Next we analyzed whether the proteins encoded by the *hit*locus of *C. pneumoniae*interacted. The coding regions of Cp265, Cp266 and Cp267 were introduced into pGADT7 and pGBKT7 and expressed in yeast in all possible combinations. As shown in Figure [5](#F5){ref-type="fig"}, large colonies, comparable in size to those of the positive control, were produced by yeast that expressed Cp265 fused to both the binding and the activation domain. This suggested a strong interaction between the molecules of the Cp265 protein. All other combinations of plasmids encoding Cp265, Cp266 and Cp267 only led to the growth of small colonies, a finding that was difficult to interpret. To differentiate between strong and weak interactions, the β-galactosidase assay was used. As shown in Fig. [6](#F6){ref-type="fig"}, the measurement of β-galactoside activity indicated an interaction between Cp267 and Cp265, Cp266 and Cp267 itself, independent of the fusion partner (the AD or DB domains of the GAL4 protein). Surprisingly, dimerization of Cp266 (HinT) was not detected, although Cp265 interacted strongly with itself. Cp267 did not activate transcription by itself (Fig. [6A](#F6){ref-type="fig"}. 267/BD and 267/AD) nor did it interact with the unrelated protein SV40 large T antigen (T/267) or murine tumor suppressor p53 (267/p53) to activate transcription, as in all these cases the β-galactosidase activity was at background levels (Fig. [6A](#F6){ref-type="fig"}). Immune co-precipitation of Cp265, Cp266 and Cp267 ------------------------------------------------- To confirm the findings derived from the two-hybrid analyses we performed protein-binding assays *in vitro*. Protein C-tagged Cp265, Cp266 and Cp267 were expressed in *E. coli*and purified by affinity chromatography (Fig. [7A](#F7){ref-type="fig"}). Each protein was then incubated with equal amounts of \[^35^S\]-labeled Cp265, Cp266 or Cp267 *in vitro*and precipitated with anti-Protein C antibodies. As shown in Fig. [7B](#F7){ref-type="fig"}, the supernatants of the different samples contained comparable amounts of the respective isotope labeled proteins. As expected the \[^35^S\] labeled Cp proteins alone were neither precipitated by unrelated membrane proteins, such as Protein C-tagged OppA of *M. hominis*nor by the anti-Protein C-sepharose itself (Fig. [7C](#F7){ref-type="fig"}). In contrast, an interaction of Cp266 with Cp267 was suggested by co-precipitation of significant amounts of Cp266 \[35*S*\] with Cp267^C^, and Cp267 \[35*S*\] with Cp266^C^(Fig. [7C](#F7){ref-type="fig"}). The other findings from the two-hybrid analyses were also confirmed as both Cp265 and Cp267 were found to dimerize and there was no interaction between Cp265 and Cp266. Interestingly dimerization of Cp266 (HinT) which was not detected by the two-hybrid system, was demonstrated in this co-precipitation assay. Thus, we can conclude that the chlamydial proteins Cp265, Cp266 (HinT) and Cp267 have a tendency to dimerize and Cp267 interacts with both Cp265 and Cp266. Discussion ========== A remarkable feature of HinT proteins is their ubiquity. They are found in all kingdoms ranging from *Mycoplasma genitalium*, the smallest prokaryote, to one of the most complex eukaryotes, the human \[[@B1],[@B19]\]. However the function of this homo-dimer in the cytoplasm of these different organisms has only been closely examined in eukaryotes, where it has been characterized as an intracellular receptor for purine ribonucleotides \[[@B4],[@B5]\], especially for hydrolases of 5\'-monophosphoramide substrates such as AMP-lysine \[[@B21]\], with the histidine triad motif forming the α-phosphate binding site \[[@B20]\]. HinT has been shown to associate with the protein microphthalmia, an important transcription factor that controls growth and function in mast cells and melanocytes \[[@B22]\], to interact with the human cyclin-dependent kinase 7 (Cdk7), a subunit of the RNA polymerase II C-terminal domain kinase Cdk7/Kin28 \[[@B23]\], and to act as a positive regulator of the yeast Cdk7 homologue Kin28 \[[@B21]\]. That it may not be the key regulator of Cdk7 activity was recently suggested by analysis of HinT^-/-^knock-out mice \[[@B24]\]. This led to the hypothesis that eukaryotic HinT homologues are involved in the regulation of transcriptional processes by hydrolyzing related adenylyl-modified proteins \[[@B25]\]. The observation that HinT homologues are ubiquitous and that they are even encoded by the smallest bacterial genomes, suggests that HinT was present at the cellular root of the tree of life and that its preservation is advantageous for the survival of cells \[[@B20]\]. However the function of bacterial HinT homologues seems to be quite different from those of eukaryotes. In *M. hominis*HinT had been demonstrated to interact both physically and genetically with a surface-localized membrane complex by binding to the P80 domain \[[@B6]\]. The data presented here suggest that the co-expression of HinT with membrane proteins is prevalent within the *Mollicutes*. Comparable organization of the *hit*-loci was found in *Mycoplasma hominis, M. pulmonis, M. mycoides*subsp. *mycoides*SC, *M. mobile*and *Mesoplasma florum*. Analysis of further genomes will most likely lengthen this list. A genomic DNA fragment of *M. bovis*encoding the amino-terminal end of a P80 homologue has been sequenced and this homologue is predicted to possess an SPase I cleavage site. In *M. hyorhinis*, a nearly complete *hit*locus, encoding P80, P60 and HinT homologues has been found adjacent to a genomic region encoding the high affinity transport system (personal communication of Michael Calcutt, Columbia, MO, USA) \[[@B26]\]. The organization of the *hit*ABL genes in an operon, as recently demonstrated for *M. hominis*\[[@B6]\] and shown here for *M. pulmonis*, suggests polycistronic expression of the *hit*loci genes. However, it remains to be elucidated whether the encoded proteins have comparable functions to those already shown in *M. hominis*. P80 of *M. hominis*was recently shown to reside in the membrane as a precursor protein and to be secreted into the extra cellular milieu as a 10 kDa smaller antigen. Processing of P80 was suggested to be initiated by SPase I cleavage \[[@B27]\]. While the P80 homologues in *M. mobile*and *M. bovis*also contained signal sequences for SPase I cleavage, those in *M. mycoides*subsp. *mycoides*SC and *Mesoplasma florum*were putative pro-lipoproteins. Secretion of lipoproteins has been observed in *M. hominis*\[[@B6]\] and other bacteria \[[@B28]\], indicating that similar function may be still possible. *M. pneumoniae*and *M. genitalium*each had only one gene adjacent to *hit*L encoding a pore-forming protein with homologies to ABC permeases, suggesting they are part of a distinct phylogenetic branch. Interestingly, in *M. penetrans*the gene upstream of *hit*L also encodes a protein of an ABC transporter, an ATPase \[[@B29]\]. The *hit*locus of *U. parvum*, which contained a P60 homologue and a gene encoding a permease, may have functions that are a hybrid of those of *M. hominis*and the *M. pneumoniae*groups. RT-PCR analyses infer the presence of a mRNA encoding HinT and UU271, a P60-homologue, and a membrane protein of unknown function. Thus, the interaction of HinT with membrane proteins seems to be a common phenomenon in *Mollicutes*. A quite different situation was detected in the *Chlamydiaceae*. Although we were not able to establish definitely the presence of a polycistronic mRNA derived from the three genes comprising *hit*locus, the overlap of the *hit*L gene with the upstream gene, as well as the conserved order of the three genes within the *hit*loci of *Chlamydiaceae*, suggested a relationship between the gene products. Analysis of Cp265, Cp266 (HinT) and Cp267 in the yeast two-hybrid system and immune co-precipitation assays confirmed our hypothesis that these proteins interacted and demonstrated that they form homo-dimers. Cp267, a protein with large areas of ARM repeats, which are known to mediate protein-protein interactions \[[@B30]\], was shown to interact with HinT, a protein known in eukaryotes to influence transcriptional activation, as well as with Cp265, a putative metal-dependent hydrolase with a binding fold for short, single stranded nucleic acids. These findings suggest that in *Chlamydiaceae*, the function of HinT may be more closely related to intracellular processes than to interactions with the extracellular milieu, as suggested by the findings on *Mollicutes*. Conclusion ========== The data presented here demonstrate that HinT proteins of the *Chlamydiaceae*associate with probable cytosolic proteins likely to function in the regulation of cellular processes, such as nucleotide metabolism. This function would be similar to that of homologues in eukaryotes, where HinT has been shown to influence transcription. The finding that HinT proteins of the *Mollicutes*interact both physically and genetically with membrane proteins may reflect a phylogenic shift towards a different function in this group, or may indicate an additional function of bacterial HinT proteins. Our future work will focus on decoding the intra- and extra-cellular processes that bacterial HinT proteins are involved in. Methods ======= DNA manipulations ----------------- All routine DNA manipulation techniques, including plasmid preparation, restriction endonuclease analysis, ligation and transformation of *E. coli*were performed as described by Sambrook *et al*. \[[@B31]\] or according to the manufacturers\' recommendations (Qiagen, Hilden, Germany, and Roche Applied Science, Mannheim, Germany). Bacterial strains and plasmids ------------------------------ The plasmids pXB (Roche Applied Science, Mannheim, Germany), pGADT7 and pGBKT7 (BD Bioscience Clontech, Palo Alto, USA) were used as expression vectors for the heterologous expression of Cp265, Cp266 and Cp267. The pXB plasmids were propagated in *Escherichia coli*DH5α F\'(Life Technologies), pGADT7 and pGBKT7 in yeast AH109 cells (BD Biosciences Clontech, Palo Alto, CA USA). Sequence analysis ----------------- Analysis of the DNA and protein sequences and the design of oligonucleotides were facilitated by the Lasergene software (DNA Star Inc., Madison Wisc.). The genomes listed in table [1](#T1){ref-type="table"} were analyzed using the web site of NCBI <http://www.ncbi.nlm.nih.gov/genomes/MICROBES/Complete.html>. Computer-based prediction of transcription termination was performed using the GeneBee service (online <http://www.genebee.msu.su/services/rna2_reduced.html>). Distant relation-ships between HinT protein sequences of the analyzed *Mollicutes*were determined by using the BLAST method through Molligen 1.4 \[[@B32]\]. Conserved domains of proteins were predicted using the Conserved Domain Database (CDD) or the InterPro service (<http://www.ncbi.nlm.nih.gov/Structure/cdd/cdd.shtml>, <http://www.ebi.ac.uk/interpro/>). RT-PCR ------ RNA was prepared from exponential phase cultures of *M. pulmonis*and *U. parvum*as described by Chomczynski and Sacchi \[[@B33]\]. *Chlamydophila pneumoniae*respiratory isolate GiD was used for all experiments \[[@B34]\]. Chlamydial RNA was prepared from HEp-2 cells 2 days post-infection as described by Mölleken et al. (Mölleken et al., manuscript in preparartion). Before use as a template for RT-PCR, contaminating traces of DNA were digested with DNase I as described previously \[[@B35]\]. Overlapping regions of the *Ureaplasma*and *Chlamydophila*genes were amplified using the Expand Template PCR system (Roche Applied Science, Mannheim, Germany) by standard PCR conditions (initial cycle of 5 min at 95°C; 35 cycles of 1 min at 95°C, 1 min at 50°C (*U. parvum*) or 58°C (*C. pneumoniae*), 2.5 min at 68°C). The *hit*operon of *M. pulmonis*was amplified using the Expand Long Template PCR system by standard PCR conditions (initial cycle of 5 min at 95°C; 10 cycles of 1 min at 95°C, 1 min at 44°C, 4 min at 68°C; addition of 0.5 U polymerase; then 20 cycles of 1 min at 95°C, 30 sec at 44°C, 4 min (+ 20 sec per cycle) at 68 °C). Southern blot analysis was performed as described previously \[[@B6]\]. Construction of plasmids ------------------------ The regions encoding Cp265, Cp266 and Cp267 in *C. pneumoniae*GiD \[[@B34]\] were amplified by PCR with the primers Cp1 to Cp6 (Table [2](#T2){ref-type="table"}). The amplicons were cut at the primer-introduced restriction endonuclease cleavage sites for *Bam*HI and *Sac*I, and ligated in-frame into the expression vector pGADT7. The *Eco*RI/*Sac*I inserts in the pGADT7 constructs were then ligated in-frame into *Eco*RI/*Pst*I-digested pGBKT7. For heterologous expression of the protein-C tagged proteins, CP265^C^, Cp266^C^and Cp267^C^, in *E. coli*, the Cp inserts of the respective pGADT7 plasmids were digested with *Sac*I, the ends bluntened by incubation with 3 units S1 nuclease (Amersham Bioscience, Freiburg, Germany) for 30 min at 37°C and subsequently cut with *Eco*RI. The blunt-end/*Eco*RI digested Cp265, Cp266 and Cp267 fragments were cloned in-frame into the blunt-end/*Eco*RI digested plasmid pXB and propagated in *E. coli*Dh5α F\'. All plasmid constructs were confirmed by DNA sequencing (ABI 373 A machine) \[[@B36]\]. Expression and purification of Cp265^C^, Cp266^C^and Cp267^C^ ------------------------------------------------------------- One liter of LB broth (Gibco BRL, Life Technologies Inc., Gaithersburg, USA) containing ampicillin (100 μg/ml) was inoculated with the respective *E. coli*DH5α F\' clone and incubated for 16 h at 37°C with vigorous shaking. The cells were harvested by centrifugation (15,000 × g, 20 min, 4°C) and frozen at -70°C. After thawing on ice, the cells were resuspended in 40 ml buffer A (120 mM NaCl, 1 mM CaCl~2~, 20 mM Tris-HCl pH 7.5) and disrupted by three repeated freeze-thaw cycles in liquid nitrogen, followed by three bursts of sonication on ice (5 min bursts at 95 W with a 1 min cooling period between each burst). Insoluble material was sedimented (15,000 × g, 20 min, 4°C) and the supernatant was transferred to an anti-Protein C affinity matrix (Roche Applied Science, Mannheim, Germany). Bound proteins were eluted from the anti-Protein C affinity matrix with 2 mM EDTA. SDS-PAGE and immunostaining of proteins --------------------------------------- Proteins were separated on 12 or 15 % polyacrylamide gels \[[@B37]\], transferred to nitrocellulose (Schleicher and Schüll, Dassel, Germany) using a semi-dry blotting apparatus (Phase, Mölln, Germany) \[[@B38]\], and immunostained using anti-Protein C peroxidase (Roche Applied Science, Mannheim, Germany), anti HA-antibody (pGADT7) or anti c-myc antibody (pGBKT7) (BD Bioscience Clontech, Palo Alto, USA). Immune co-precipitation assay ----------------------------- The \[^35^S\]-labeled Cp265, Cp266 and Cp267 proteins were generated from pGADT7 and pGBKT7 fusion vectors by *in vitro*translation using a T7 coupled transcription/translation system in the presence of \[^35^S\]-methionine (Promega, Madison, USA). Two micrograms of purified Protein C-tagged protein were incubated with 10 μl of the \[^35^S\]-labeled protein in buffer B (20 mM Tris/HCl (pH 7.5), 75 mM KCl, 0.1 mM EDTA, 2.5 mM MgCl~2~, 0.05 (v/v) % Igepal, 2 mM DTT, 10 (v/v) % glycerol and 2 mM CaCl~2~) for 1 h \[[@B39]\]. Thereafter, a 5 μl aliquot of a slurry of the anti-Protein C matrix, equilibrated in buffer B containing 5 μg BSA, was added to each sample and the mixture further incubated for 1 h. The Protein C-matrix was washed eight times in buffer B, resuspended in 20 μl of SDS-PAGE sample buffer and heated at 95°C for 5 min. The proteins were then analyzed by 15 % SDS-PAGE and autoradiography using a phosphorimager (Fujifilm FLA 3000). Yeast two-hybrid assays ----------------------- The MATCHMAKER Gal4 two-hybrid system (BD Bioscience Clontech, Palo Alto, USA) was used for interaction assays. AH109 cells were co-transformed with the Cp265, Cp266 or Cp267 expressing pGADT-T7 and pGBKT7 plasmids, grown in small scale cultures and plated onto histidine deficient agar plates, followed by a β-galactosidase colony lift filter assay. All procedures were carried out as outlined in the BD Bioscience Clontech protocol. The β-galactosidase assay was performed after the inoculation of 7.5 ml YPD Medium with 0.5 ml overnight culture in selective medium and incubation for five hours at 30°C until the culture reached an OD~600~of 0.4 to 0.6. The cells were harvested by centrifugation at 14,000 × g for 1 min and re-suspended in 300 μl Buffer Z (Na~2~HPO~4~+ 7 × H~2~O, NaH~2~PO~4~+ 4 × H~2~O, KCl, MgSO~4~× 7 H~2~O, pH7.0). Lysis of the cells was achieved with four cycles of freezing in liquid nitrogen and thawing at 37°C. Soluble and insoluble components were separated by centrifugation at 14000 × g for 2 min. The β-galactosidase activity was measured using a FluorAce™ β-galactosidase Reporter Assay kit (BioRad, Hercules, USA). Authors\' contributions ======================= MH carried out the immunoassays and Yeast TwoHybrid assays. BH carried out the molecular genetic studies. BH and JH participated in the design of the study and drafted the manuscript. All authors have read and approved of the final manuscript. Acknowledgements ================ This work was supported by the Deutsche Forschungsgemeinschaft (DFG-HE 2028/3-2). We thank Michael Calcutt (Columbia, MO, USA) for information about the *hit*loci in *M. bovis*and *M. hyorhinis*, Marzena Wyschkon for excellent technical assistance and Colin MacKenzie for critically reading the manuscript. Katja Mölleken is thanked for preparation of *C. pneumoniae*RNA and genomic DNA. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Schematic of the *hit*loci in *Mollicutes*and *Chlamydiaceae*.**A schematic representation of genes within the *hit*loci of the following species: A) *M. pulmonis*, *M. mycoides*subsp. *mycoides*SC, *Mesoplasma florum*and *M. mobile*containing genes homologous to *hit*ABL of *M. hominis*; B) *U. parvum*, *M. pneumoniae*and *M. genitalium*, each containing a gene upstream of *hit*L predicted to encode an integral pore-forming protein; and C) *C. pneumoniae*, *C. trachomatis*, *C. muridarum*and *C. caviae*containing *hit*L flanking genes predicted to encode, upstream, a protein with the signature sequence of a metal-dependent hydrolase (light blue region), with an OB nucleic acid binding fold (dark blue region) and, downstream, a protein with ARM repeats (red regions), which are known to mediate protein-protein interactions. Triangles indicate signal peptidase recognition sites of SPase I and SPase II. Transmembrane regions are depicted as striped regions. The position of the RGD tri-peptide is marked by a dotted region. ::: ![](1471-2180-5-27-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **RT-PCR analysis of *M. pulmonis*.**A. The regions amplified by RT-PCR are shown below the genomic region encompassing MYPU\_0060, MYPU\_0070 and MYPU\_0080. The primers used and the lengths of the amplicons are indicated. B. Amplicons were separated on a 0.8 % agarose gel and subjected to Southern blot analysis, here shown for the MYPU\_0070 specific probe, using digoxigenin (DIG)-labeled probes hybridizing to each of the three genes, detected using chemiluminescence. Bands of lower length as expected are degradation products. M, DIG-labeled DNA molecular weight marker VII (Roche Biochemicals). ::: ![](1471-2180-5-27-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **RT-PCR analysis of *U. parvum*.**A. The positions of the different amplicons are shown below the schematic of the *hit*locus genes of *U. parvum*. The primers used (Table 2) and the lengths of the amplicons are indicated. B. The PCR products (A -- D) for genomic DNA (g), cDNA (c) and RNA (r) were separated on a 0.6 % agarose gel and stained with ethidium bromide. Southern blot analysis was performed with digoxigenin (DIG)-labeled probes hybridizing to one of the four genes, (here shown for UU272 probing), and detected using chemiluminescence. M, Gene Ruler 1 kb DNA ladder (Fermentas). ::: ![](1471-2180-5-27-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **RT-PCR analysis of *C. pneumoniae.***A. Below the schematic of the *hit*locus of *C. pneumoniae*the positions of the PCR amplicons (A -- F) are shown. Genomic DNA (g), cDNA (c) and RNA without RT reaction (r) are used as templates. B. The PCR products were separated on a 0.8 % agarose gel and subjected to Southern blot analysis with DIG-labeled probes hybridizing to Cp265 (A, D), Cp266 / *hit*L (B, E) Cp267 (C, F) with visualization using chemiluminescence. In E, the signals of lower length as 1.4 kb may be due to primer dimerization. M, DIG-labeled DNA molecular weight marker VII (Roche Biochemicals) ::: ![](1471-2180-5-27-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **The yeast-two hybrid assay.**AH109 were co-transformed with the pGADT7 (AD) and pGBKT7 (BD) constructs containing *hit*L of *M. hominis*, Cp265, Cp266 or Cp267 from *C. pneumoniae*, pGADT7-T, expressing the SV40 large T-antigen and pGBKT7-53, expressing the murine p53 protein (K+), or pGADT7 and pGBKT7 without fusion partners (K-). Yeast cultures were incubated at 30°C on histidine-deficient agar plates for 4 days. ::: ![](1471-2180-5-27-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Quantitation of β-galactosidase activity.**In yeast cells, co-transformed with pGADT7 (AD) and pGBKT7 (BD) constructs as indicated, β-galactosidase activity was quantified using 4-methylumbelliferyl-galactopyranoside (MUG) hydrolysis to the fluorescent molecule, 4-methylumbelliferone (4 MU). The fluorescence of 4 MU was excited at 360 nm and emission measured at 460 nm. ::: ![](1471-2180-5-27-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **Immune co-precipitation.**A. Affinity purified proteins, Cp265 (265^C^), Cp266 (266^C^), Cp267 (267^C^) and OppA^C^, were analyzed on 15 % SDS-polyacrylamide gels by Coomassie staining (C) or on Western blots by immunostaining with an anti-Protein C antibody (W). \[^35^S\]-labeled Cp265, Cp266 and Cp267 (Captives) were incubated at room temperature for 1 h with Cp265^C^, Cp266^C^, Cp267^C^and OppA^C^(Catcher). Immune complexes were precipitated with anti-Protein C-matrix, and the supernatants (B.) and precipitates (C.) analyzed on 15 % SDS-polyacrylamide gels by autoradiography using a phosphorimager. As a negative control, the \[^35^S\]-labeled proteins were subjected to the Protein C-matrix without a catcher (Sepha.^αC^). M, prestained low-molecular weight marker (Biorad) ::: ![](1471-2180-5-27-7) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Organisms used for genome sequence analyses ::: ----------------------------------------------- ------------ *Chlamydophila caviae*GPIC NC\_003361 *Chlamydia muridarum* NC\_002620 *Chlamydophila pneumoniae*AR39 NC\_002179 *Chlamydia trachomatis*D/UW-3/CX NC\_000117 *Mesoplasma florum*L1 NC\_006055 *Mycoplasma gallisepticum*R NC\_004829 *Mycoplasma genitalium*G-37 NC\_000908 *Mycoplasma mobile*163 K NC\_006908 *Mycoplasma mycoides spp.mycoides SC*str. PG1 NC\_005364 *Mycoplasma pneumoniae*M129 NC\_000912 *Mycoplasma penetrans*HF2 NC\_004432 *Mycoplasma pulmonis*UAB CTIP NC\_002771 *Parachlamydia sp.*UWE25 NC\_005861 *Ureaplasma parvum*serovar 3 str. ATCC 700970 NC\_002162 ----------------------------------------------- ------------ ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Primers used in PCR ::: **Primer** **Length** **Sequence (5\'-3\')** ------------- ------------ ---------------------------------------------- MP1 27 ATTTATTTTTGATTACTATTTTTGAAC MP2 23 AATTTGTTTTATTATTTTGTTTA MP8 20 TTAATGTTCATGTCCTTCTT MP9 21 AATCCAGAAAGTTTTTTTATT MP12 24 ATTTGTTATTGCTAAAAGATTGTA UU1 22 ATGATCAATAAAAACAAAAAAT UU2 19 GCGCCGTTAGTGATGTTAG UU6 32 TGAGCTCTTCCATATTTTTTGATTATACCATA UU7 23 AAAAACGAATTGCCTCTATGTAT Cp1 21 CATGATGCGCGTGGAGTAGGT Cp2 25 TCCAAGAAGCATTGGTACTCACGAT Cp3 22 CGGTTTCCATGGCTTCTCTGAC Cp4 29 TTATCGATGGATTGATAGATTGTGAAAAG Cp5 22 GGACGCCCACCTAAAAGATGAA Cp6 21 TTTCGCTTGGTCCCTCTCCTG Cp7 25 GCTCTTTCAATATCTTCACGGCTCA Cp8 28 TACGGTGCTGAGCTTAAAGATAGAAATG Cp\_265 up 27 [ATGGATCCTA]{.underline}ATGGAGGATTGGCTAAG Cp\_265 low 29 [TTGAGCTC]{.underline}TCATATGATCCCTCTATCTTG Cp\_266 up 27 [TTAGGATCCTC]{.underline}ATGCCTACGTGCTTTG Cp\_266 low 29 [TTTGAGCTC]{.underline}ATCAGGCTATAGCACCTAAA Cp\_267 up 23 [ATTGGATCC]{.underline}CCATGTTCGGCTCG Cp\_267 low 30 [TTAGAGCTC]{.underline}ACTTGAAAATAGAGAAAAGAG Cp1 BamHI 27 [ATGGATCC]{.underline}ATGGAGGATTGGCTAAG Cp2 BamHI 27 [TTAGGATCC]{.underline}ATGCCTACGTGCTTTG Cp3 BamHI 23 [ATTGGATCC]{.underline}ATGTTCGGCTCG :::

PubMed Central

2024-06-05T03:55:58.391092

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173108/", "journal": "BMC Microbiol. 2005 May 18; 5:27", "authors": [ { "first": "Miriam", "last": "Hopfe" }, { "first": "Johannes H", "last": "Hegemann" }, { "first": "Birgit", "last": "Henrich" } ] }

PMC1173109

Background ========== The elucidation of the *P. aeruginosa*strain PAO1 genome sequence in 2000 \[[@B1]\] opened the doors for genome- and proteome-scale research with this important opportunistic pathogen. These include the availability of commercial (Affymetrix, Santa Clara, CA) *P. aeruginosa*GeneChips^®^for transcriptome analyses \[[@B2]\], as well as the *P. aeruginosa*PAO1 gene collection for high-throughput expression and purification of recombinant proteins \[[@B3]\]. A comprehensive transposon mutant library was generated containing over 30,000 sequence-defined mutants, corresponding to an average of five insertions per gene \[[@B4]\]. However, despite the development of many genetic tools for *P. aeruginosa*over the past decade \[[@B5],[@B6]\], isolation of defined deletion mutants is still a relatively tedious process, which relies on construction of deletion alleles, most often tagged with an antibiotic resistance gene, on a suicide plasmid, followed by recombination of the plasmid-borne deletions into the chromosome, usually after conjugal transfer of the suicide plasmid (reviewed in \[[@B6]\]). Merodiploids arising from chromosomal integration of the suicide plasmid are resolved by utilization of a counterselectable marker, most often *Bacillus subtilis sacB*(sucrose counterselection) \[[@B7]\] or, less frequently, *rpsL*(streptomycin counterselection)\[[@B8]\]. In cases where the antibiotic selection markers are flanked by site specific recombination sites, e.g., the Flp recombinase target (*FRT*) \[[@B9]\] or Cre recombinase (*loxP*) \[[@B10]\] sites, they can subsequently be deleted from the chromosome, resulting in unmarked deletion mutants. Unfortunately, the previously described method for one-step inactivation of chromosomal genes in *Escherichia coli*using PCR products \[[@B11]\] does not work in *P. aeruginosa*, although we have used a modification \[[@B12]\] of this approach where the target gene is first cloned into a plasmid, followed by λ RED-mediated recombination of a PCR-generated mutated copy of the gene \[[@B13]\]. Here, we describe a more streamlined method for generation of chromosomal deletion mutants. In this method, a \"mutant\" fragment containing the 5\' and 3\' flanking regions of the target gene plus the antibiotic resistance gene is assembled by a modified \[[@B14]\] PCR overlap extension protocol \[[@B15]\] and then cloned into a suicide vector using the Gateway recombinational cloning system, similarly to a previously described method \[[@B16]\]. The plasmid-borne deletion allele is then transferred to *P. aeruginosa*utilizing a rapid transformation procedure and recombined into the chromosome. The transformants are often merodiploids which are then resolved by sucrose counterselection against the undesired wild-type sequences. The antibiotic resistance marker is subsequently deleted from the chromosome by Flp recombinase. Results ======= Construction of a Gateway-compatible suicide vector --------------------------------------------------- We previously described the pEX family of relatively small, fully sequenced suicide plasmids which are now widely used for gene replacement experiments in *P. aeruginosa*and related bacteria \[[@B9]\]. Wolfgang et al. \[[@B16]\] recently described a Gateway-compatible pEX18Gm derivative. Here, we modified pEX18Ap for use as a Gateway destination vector for cloning of mutations marked with a gentamycin resistance (Gm^r^) encoding gene. We prefer using the Gm^r^marker because it is more easily transferable between *P. aeruginosa*chromosomes by transformation with chromosomal DNA fragments when compared to other selection markers \[[@B17]\]. The resulting vector pEX18ApGW (Fig. [1](#F1){ref-type="fig"}) contains the functional sequences for recombination-based cloning (*attR1*and *attR2*) and the *ccdB*counterselection marker, while maintaining the *sacB*counterselection marker for downstream resolution of merodiploids. Generation of a mutant fragment by PCR overlap extension -------------------------------------------------------- The amplification of each mutant fragment was achieved in a two step PCR that required 8 PCR primers, 4 of which were unique for each fragment to be mutated and 4 of which were common primers (Table [1](#T1){ref-type="table"}). The gene-specific portions of the unique primers were designed to yield T~m~\'s of 55--57°C and PCR fragments ranging from \~200 to 600 bp. These parameters enabled utilization of the same PCR conditions for different target genes, while at the same time provided for amplification of sufficient chromosomal DNA to allow for efficient homologous recombination of plasmid-borne sequence with the chromosome. The basic strategy for generation of the mutant fragment by PCR overlap extension is illustrated in Fig. [2](#F2){ref-type="fig"}. In PCR1, the four gene-specific primers are used to amplify the 5\' and 3\' regions of the target gene. Simultaneously, two of the common primers, Gm-F and Gm-R, are used to amplify the Gm^r^cassette flanked by *FRT*sites from a pPS856 template. Since the Gm^r^fragment is used for all constructs, it can be prepared in larger amounts ahead of time and stored until needed. Fig. [3A](#F3){ref-type="fig"} illustrates typical results of PCR1 utilizing the example of *P. aeruginosa*gene *PA1520*, encoding a putative transcriptional regulator of the GntR family. When we adhered to the prescribed PCR conditions, the amplified fragments were very clean. It was especially prudent to follow the conditions for PCR amplification of the Gm^r^*FRT*fragment since the *FRT*sites possess significant secondary structures. The PCR fragments were gel-purified and equal amounts (50 ng) of each were combined for PCR2. PCR2 was allowed to proceed for 3 cycles in the absence of exogenous primers, which for *PA1520*yields the intermittent 1,709 bp \'*attB1*-*PA1520*\'-*FRT*-Gm-*FRT*-\'*PA1520*-*attB2*\' fragment illustrated in Fig [3A](#F3){ref-type="fig"}. The common Gateway primers GW-*attB1*and GW-*attB2*were then added and the PCR continued for another 25 cycles to obtain the final, recombination-proficient DNA fragment, which constituted the predominant PCR product at this stage (Fig. [3B](#F3){ref-type="fig"}). For *PA1520*, the resulting 1,735 bp *attB1*-*PA1520*\'-*FRT*-Gm-*FRT*-\'*PA1520*-*attB2*fragment was gel-purified and used for the BP clonase reaction for construction of an entry clone. BP and LR clonase reactions --------------------------- The BP and LR clonase reactions (Fig. [4](#F4){ref-type="fig"}) were performed according to Invitrogen\'s (Carlsbad, CA) Gateway cloning manual. For the BP clonase reaction pDONR221 was used, but only half of the recommended BP clonase enzyme mix. Gm^r^and kanamycin resistant (Km^r^) DH5α or HPS1 transformants were selected and the presence of the correct plasmids was confirmed by *Xba*I digestion (the *FRT*sites flanking the Gm^r^gene each contain a single *Xba*I site). The insert of a correct pDONR-*Gene*::Gm entry clone was then recombined into the destination vector pEX18ApGW using the LR clonase protocol, again using only half of the recommended amount of LR clonase mix. Recombinants were transformed into DH5α and Ap^r^colonies were selected. We noticed that, at least under the conditions we used, many transformants exhibited a Ap^r^Km^r^phenotype, indicating that they either contained both pDONR-*Gene*::Gm and pEX18ApGW-*Gene*::Gm or, more likely, a cointegrate of both plasmids. The presence of the correct pEX18ApGW-*Gene*::Gm in transformants that were only Ap^r^was confirmed by *Xba*I digestion. Gene replacement ---------------- To expedite gene replacement, an improved electroporation method \[[@B17]\] was used for transfer of the suicide plasmid pEX18ApGW-Gene::Gm into *P. aeruginosa*instead of the traditionally employed conjugation method. Although after transfer into *P. aeruginosa*double cross-over events can occur frequently (\>50%), they can also be rare (\<1--5%). In the latter instance, merodiploids formed via integration of the suicide plasmid by a single cross-over event (Fig. [4](#F4){ref-type="fig"}). The merodiploid state was then resolved via sucrose selection in the presence of gentamycin, resulting in deletion of the wild type gene. For generation of unmarked deletion mutants, the Gm^r^marker was subsequently removed by deletion of a 967 bp fragment using Flp recombinase. The presence of the desired deletion in either the marked or unmarked *PA1520*mutant was verified by colony PCR utilizing the gene-specific up and down primers (Fig. [5](#F5){ref-type="fig"}). Deletion of 25 genes encoding transcriptional regulators of the GntR family --------------------------------------------------------------------------- We are interested in characterizing the activator involved in regulation of expression of the *fabAB*operon, which encodes two essential enzymes of *de novo*fatty acid biosynthesis \[[@B18]\]. Expression of the *fabAB*operon is repressed by exogenously added fatty acids, most effectively by oleic acid, and preliminary results from our laboratory indicate that this may be due to relief of activation by a transcriptional regulator which binds to a 30 nucleotide site located in the *fabAB*promoter region (Choi and Schweizer, unpublished results). This 30 bp region is only found and conserved in *Pseudomonas*spp. Its deletion dramatically reduces transcription of the *fabAB*operon and the residual transcriptional activity is no longer fatty acid responsive (Fig. [6](#F6){ref-type="fig"}). In *E. coli*, transcriptional activation of the unlinked *fabA*and *fabB*genes is achieved by FadR, a regulator belonging to a group of transcriptional regulators with the GntR family signature \[[@B19],[@B20]\]. The PAO1 chromosome encodes more than 30 regulatory proteins with this signature. Of these, 27 have not been assigned function. Using the methodology described above for one of these genes, *PA1520*, we deleted 25 of the 27 genes (we could not delete the presumably essential *PA1285*and *PA2299*) in less than 4 weeks in strain PAO1 with a chromosomally integrated *fabA\'-lacZ*transcriptional fusion and analyzed β-galactosidase expression in the resulting mutant strains (Fig. [6](#F6){ref-type="fig"}). In exponential phase cells, the overall pattern of *fabA*expression was indistinguishable from the parental strain in all 25 mutants, indicating that none of these genes encodes the *fabAB*activator. We would have expected an expression pattern similar to that observed in the PAO1::*fabA*Δ30\'-*lacZ*control strain which contains a deletion of the 30 nucleotide putative activator-binding site. Discussion ========== In our experience, the method described here for generation of unmarked deletion mutants is the fastest available to date. From start to finish, an individual unmarked *P. aeruginosa*mutant can be generated and characterized in \~14 days, which is up to one week faster than using our previously described method \[[@B9]\]. However, the main advantage of the method does not lie in individual mutant generation but in the speed by which collective deletion mutant pools can be generated. For example, we isolated unmarked deletions in 25 of the 27 genes encoding transcriptional regulators of the GntR family in less than 4 weeks. Two genes could not be deleted, presumably because they encode essential proteins. Besides speed, what are some of the other unique features of this method? First, the deletion of gene sequences is not restricted by the natural availability or synthetic generation of restriction sites that are subsequently used to delete intervening sequences. With proper primer location, the PCR-based method allows clean deletion of more-or-less the entire coding sequence without fear of generating truncated genes which potentially produce products with undesired side effects. Second, Flp excision of the antibiotic resistance gene cassette leaves behind a short 85 bp *FRT*scar which does not exhibit any known polar effects on downstream genes. The method is therefore uniquely suited to study genes organized in polycistronic units. Flp excision is a very straightforward procedure with efficiencies approaching 100% \[[@B9]\]. Third, the method is economical, especially in a more high-throughput format. PCR overlap extension requires only four gene specific primers of reasonable length which keeps oligonucleotide costs to a minimum. In comparison to traditional methods, it also requires fewer hands-on steps and gene-specific reagents (e.g., restriction enzymes). Conclusion ========== Although not yet as fast as the PCR fragment-directed gene replacement method described for *E. coli*, the PCR- and recombinational cloning based method described opens the door for high-throughput generation of *P. aeruginosa*deletion mutants on a genome-wide scale. With appropriate modifications, i.e., choice of appropriate selection markers and Gateway-compatible gene replacement vectors, the method should be applicable to the many other bacteria in which the pEX vector system can be applied \[[@B6]\]. Methods ======= Media ----- *Escherichia coli*and *P. aeruginosa*strains were maintained on Luria-Bertani medium (LB; 10 g per liter tryptone, 5 g per liter yeast extract, 10 g per liter NaCl; Becton, Dickinson & Co., Sparks, MD). For plasmid maintenance in *E. coli*, the medium was supplemented with 100 μg per ml ampicillin, 25 μg per ml chloramphenicol, 35 μg per ml kanamycin or 15 μg per ml gentamycin. For marker selection in *P. aeruginosa*, 200 μg per ml carbenicillin and 30 μg per ml of gentamycin was used, as appropriate. β-galactosidase activity assays ------------------------------- Cells were grown at 37°C with shaking to exponential phase (optical density at 540 nm \~0.4--0.8) in LB medium with 0.05% Brij 58 +/- 0.05% oleic acid. β-galactosidase activity was measured in chloroform/sodium dodecyl sulfate-permeabilized cells and its activity calculated as previously described \[[@B21]\]. Standard DNA procedures ----------------------- Routine procedures were employed for manipulation of DNA \[[@B22]\]. Plasmid DNAs were isolated using the QIAprep Mini-spin kit (Qiagen, Valencia, CA) and *P. aeruginosa*chromosomal DNA was obtained using the QIAamp DNA Mini Kit. DNA fragments were purified from agarose gels utilizing the QIAquick gel extraction kit. Construction of the Gateway-compatible gene replacement vector pEX18ApGW ------------------------------------------------------------------------ The previously described gene replacement vector pEX18Ap \[[@B9]\] was modified by cloning of a 1,755 bp *Hin*dIII-*Kpn*I fragment from pUC18-mini-Tn*7*T-Gm-GW (GenBank accession number AY737004), followed by transformation into the *gyrA*strain DB3.1 (Invitrogen). 1^st^round PCRs --------------- ### PCR-amplification of the gentamycin resistance gene cassette A 50 μl PCR reaction contained 5 ng pPS856 \[[@B9]\] template DNA, 1x HiFi Platinum *Taq*buffer, 2 mM MgSO~4~, 200 μM dNTPs, 0.2 μM of Gm-F and Gm-R (Table [1](#T1){ref-type="table"}) and 5 units of HiFi Platinum *Taq*polymerase (Invitrogen). Cycle conditions were 95°C for 2 min, followed by 30 cycles of 94°C for 30 s, 50°C for 30 s, and 68°C for 1 min 30 s, and a final extension at 68°C for 7 min. The resulting 1,053 bp PCR product was purified by agarose gel electrophoresis and its concentration determined spectrophotometrically (A~260\ nm~) in an Eppendorf Biophotometer (Hamburg, Germany) and using the 50--2000 μl Eppendorf UVettes ### PCR-amplification of 5\' and 3\' gene fragments Two 50 μl PCR reactions were prepared. The first reaction contained 20 ng chromosomal template DNA, 1x HiFi Platinum *Taq*buffer, 2 mM MgSO~4~, 5% DMSO, 200 μM dNTPs, 0.8 μM of PA1520-UpF-GWL and PA1520-UpR-Gm (Table [1](#T1){ref-type="table"}) and 5 units of Platinum *Taq*polymerase. The second reaction contained the same components as the first, but 0.8 μM of PA1520-DnF-Gm and PA1520-DnR-GWR (Table [1](#T1){ref-type="table"}). Cycle conditions were 94°C for 5 min, followed by 30 cycles of 94°C for 30 s, 56°C for 30 s, and 68°C for 30 s, and a final extension at 68°C for 10 min. The resulting PCR products were purified by agarose gel electrophoresis and their concentrations determined spectrophotometrically (A~260\ nm~). 2^nd^round PCR -------------- A 50 μl PCR reaction contained 50 ng each of the *PA1520*5\' and 3\' purified template DNAs, and 50 ng of *FRT*-Gm-*FRT*template DNA prepared during 1^st^round PCR. The reaction mix also contained 1x HiFi Platinum *Taq*buffer, 2 mM MgSO~4~, 5% DMSO and 200 μM dNTPs, and 5 units of HiFi Platinum *Taq*polymerase. After an initial denaturation at 94°C for 2 min, 3 cycles of 94°C for 30 s, 55°C for 30 s, and 68°C for 1 min were run without added primers. The third cycle was paused at 30 s of the 68°C extension, primers GW-*attB1*and GW-*attB2*were added to 0.2 μM each, and the cycle was then finished by another 30 s extension at 68°C. The PCR was completed by 25 cycles of 94°C for 30 s, 56°C for 30 s, and 68°C for 5 min, and a final extension at 68°C for 10 min. The resulting major PCR product was purified by agarose gel electrophoresis and its concentrations determined spectrophotometrically (A~260\ nm~). The identity of the PCR fragment was confirmed by *Xba*I digestion (each *FRT*site of the *FRT*-Gm*-FRT*fragment contains a *Xba*I site). ### BP and LR clonase reactions The BP and LR clonase reactions for recombinational transfer of the PCR product into pDONR221 and pEX18ApGW, respectively, were performed as described in Invitrogen\'s Gateway cloning manual, but using only half of the recommended amounts of BP and LR clonase mixes and either *E. coli*DH5α \[[@B23]\] or HPS1 \[[@B24]\] as host strains. The presence of the correct fragments in transformants obtained with DNA from either clonase reaction was verified by digestion with *Xba*I because each *FRT*site flanking the Gm^r^gene contains a *Xba*I site. However, before plasmid isolation from transformants obtained with DNA from the LR clonase reaction, 25--50 transformants were i) patched on LB+Km and LB+Ap plates, and ii) simultaneously purified for single colonies on LB+Ap plates. This was necessary to distinguish between those colonies containing only the desired pEX18ApGW-*Gene*::Gm from those containing this plasmid and the frequently contaminating pDONR-*Gene*::Gm (pEX18Ap-derived plasmids confer Ap^r^and pDONR plasmids confer Km^r^). Plasmids were then isolated from Ap^r^Km^s^transformants and analyzed by *Xba*I digestion. In those extremely rare instances where all patched transformants contained a mixed plasmid population (\<1% of to date performed LR clonase reactions), retransformation with Ap^r^selection was necessary to obtain colonies containing only pEX18ApGW-*Gene*::Gm. ### Transfer of plasmid-borne deletions to the *P. aeruginosa*chromosome A rapid electroporation method described elsewhere \[[@B17]\] was used to transfer the pEX18ApGW-borne deletion mutations to *P. aeruginosa*. Briefly, 6 ml of an overnight culture grown in LB medium was harvested in 4 microcentrifuge tubes by centrifugation (1--2 min, 16,000 × *g*) at room temperature. Each cell pellet was washed twice with 1 ml of room temperature 300 mM sucrose and they were then combined in a total of 100 μl 300 mM sucrose. For electroporation, 300-500 ng of plasmid DNA was mixed with 100 μl of electrocompetent cells and transferred to a 2 mm gap width electroporation cuvette. After applying a pulse (settings: 25 μF; 200 Ohm; 2.5 kV on a Bio-Rad GenePulserXcell™), 1 ml of LB medium was added at once, and the cells were transferred to a 17 × 100 mm glass or polystyrene tube and shaken for 1 h at 37°C. The cells were then harvested in a microcentrifuge tube. 800 μl of the supernatant was discarded and the cell pellet resuspended in the residual medium. The entire mixture was then plated on two LB plates containing 30 μg per ml Gm (LB+Gm30). The plates were incubated at 37°C until colonies appeared (usually within 24 h). Under these conditions, the transformation efficiencies were generally 30--100 transformants per μg of DNA. A few colonies were patched on LB+Gm30 plates and LB+Cb200 plates to differentiate single- from double cross-over events. To ascertain resolution of merodiploids, Gm^r^colonies were struck for single colonies on LB+Gm30 plates containing 5% sucrose. Gm^r^colonies from the LB-Gm-sucrose plates were patched onto LB+Gm30+5% sucrose, as well as LB plates with 200 μg per ml carbenicillin (LB+Cb200). Colonies growing on the LB-Gm-sucrose, but not on the LB-carbenicillin plates were considered putative deletion mutants. The presence of the correct mutations was verified by colony PCR. To do this, a single, large colony (or the equivalent from a cell patch) was picked from a LB-Gm-sucrose plate, transferred to 30 μl H~2~O in a microcentrifuge tube and boiled for 5 min. Cell debris was removed by centrifugation in a microcentrifuge fuge (2 min; 16,000 × *g*), and the supernatant was transferred to a fresh tube which was placed on ice. 5 μl of the supernatant was used as source of template DNA in a 50 μl PCR reaction containing *Taq*buffer, 1.5 mM MgSO~4~, 5% DMSO, 0.6 μM each of PA1520-UpF-GWL and PA1520-DnR-GWR, 200 μM dNTPs and 5 units *Taq*polymerase (Fermentas, Hanover, MD). Cycle conditions were 95°C for 5 min, followed by 30 cycles of 95°C for 45 s, 55°C for 30 s, and 72°C for 2 min, and a final extension at 72°C for 10 min. PCR products were analyzed by agarose gel electrophoresis. ### Flp-mediated marker excision Electrocompetent cells of the newly constructed mutant strain were prepared as described in the preceding paragraph and transformed with 20 ng of pFLP2 \[[@B9]\] DNA as described above. After phenotypic expression at 37°C for 1 h, the cell suspension was diluted 1:1,000 and 1:10,000 with either LB or 0.9% NaCl, and 50 μl aliquots were plated on LB+Cb200 plates and incubated at 37°C until colonies appeared. Transformants were purified for single colonies on LB+Cb200 plates. Ten single colonies were tested for antibiotic-susceptibility on LB ± Gm30 plates and on a LB+Cb200 plate. Two Gm^s^Cb^r^isolates were struck for single colonies onto a LB+5% sucrose plate and incubated at 37°C until sucrose-resistant colonies appeared. Ten sucrose-resistant colonies were retested on a LB+5% sucrose (master) plate and a LB+Cb200 plate. Finally, two sucrose-resistant and Cb^s^colonies were struck on LB plates without antibiotics, and their Cb^s^and Gm^s^phenotypes confirmed by patching on LB ± Cb200 and LB ± Gm30 plates. Deletion of the Gm^r^marker was assessed by colony PCR utilizing the conditions and primers described in the preceding paragraph Authors\' contributions ======================= KHC, performance of experiments, writing of manuscript. HPS, design and coordination of the study, data evaluation, direct supervision of experimental work, writing of manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Table 2 -- Sequences of PCR primers used to amplify genes encoding transcriptional regulators of the GntR family ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ This work was supported by NIH grant AI058141 to HPS. The authors wish to thank Dr. Donald Moir (Microbiotix, Inc.) for helpful advice regarding use of the PCR overlap extension technology with GC-rich DNA. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Map of the Gateway-compatible pEX18ApGW**. This vector was derived by cloning a Gateway conversion fragment (grey) into the multiple cloning site of pEX18Ap (black). Only selected restriction enzyme cleavage sites are shown. Abbreviations: *attR1*and *attR2*, bacteriophage λ recombination sites; *bla*, β-lactamase-encoding gene; *cat*, chloramphenicol acetyl transferase-encoding gene; *ccdB*, gene encoding gyrase-modifying enzyme (CcdB poisons host DNA gyrase by forming a covalent complex with the DNA gyrase A subunit and thus serves as a counter-selectable marker in *gyrA*^+^cloning hosts \[25\]); *ori*, ColE1-derived replication origin; *oriT*, origin of conjugal transfer; *sacB*, *Bacillus subtilis*levansucrase-encoding gene. The sequence of this plasmid was deposited in GenBank and assigned accession number AY928469. ::: ![](1471-2180-5-30-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Schematic illustration of mutant fragment generation by overlap extension PCR**. During first round PCR (PCR1), the 5\' and 3\' ends of the target genes, as well as the gentamycin (Gm) resistance cassette are amplified using four gene-specific primers (G-UpF-GWL, G-UpR-Gm, G-DnF-Gm and G-DnR-GWR) and the common Gm-specific primers (Gm-F and Gm-R). This generates three fragments with partial overlaps either to each other (indicated by the blue boxes signifying Gm overlap) or the *attB1*and *attB2*λ recombination sites (indicated by the green and pink boxes). These purified fragments are then assembled *in vitro*by overlap extension during second round PCR (PCR2) using common primers GW-*attB*1 and GW-*attB*2, resulting in a recombination-proficient mutant PCR fragment. ::: ![](1471-2180-5-30-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **PCR amplification of the Gm *FRT*cassette, *PA1520*gene fragments and the overlap extension product**. **A.**First round PCR fragments. The left panel illustrates amplification of the 1,053-bp Gm^r^fragment from pPS856 which contains 24 bp (right) and 25 bp (left) overlaps with the *PA1520*\' and \'*PA1520*fragments (blue boxes). The right panel illustrates amplification of the *PA1520*\' and \'*PA1520*fragments. The 5\' fragment contains 388 bp of chromosomal DNA, 25 bp overlapping the left side of the Gm^r^fragment and 16 bp overlapping the GW-*attB1*primer. Similarly, the 3\' fragment contains 236 bp of chromosomal DNA, 24 bp overlapping the right side of the Gm^r^fragment and 16 bp overlapping the GW-*attB2*primer. The sequences of the gene-specific and common primers are listed in Table 1. **B.**Second round PCR. The purified fragments shown in panel **A**were used in the second round PCR illustrated in Fig. 2 to derive the indicated *attB1*-*PA1520*\'-*FRT*-Gm-*FRT*-\'*PA1520*-*attB2*fragment. The entire 50 μl second round PCR reaction was subjected to agarose gel electrophoresis. The desired DNA fragment constituting the major product marked by the arrow was excised from the gel, purified and then used for the BP clonase reaction. Lanes labeled M in both panels contained Hi-Lo molecular size markers from Minnesota Molecular (Minneapolis, MN). ::: ![](1471-2180-5-30-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Gateway-recombinational cloning and return of the plasmid-borne deletion allele to the *P. aeruginosa*chromosome**. The mutant DNA fragment generated by overlap extension PCR is first cloned into pDONR221 via the BP clonase reaction to create the entry clone pDONR221-*Gene*::Gm, which then serves as the substrate for LR clonase-mediated recombination into the destination vector pEX18ApGW. The resulting suicide vector pEX18ApGW-*Gene*::Gm is then transferred to *P. aeruginosa*and the plasmid-borne deletion mutation is exchanged with the chromosome to generate the desired deletion mutant. Please note that, as discussed in the text, gene replacement by double-crossover can occur quite frequently, but it can also be a rare event in which case allele exchange happens in two steps involving homologous recombination. First, the suicide plasmid is integrated via a single-crossover event resulting in generation of a merodiploid containing the wild-type and mutant allele. Second, the merodiploid state is resolved by *sacB*-mediated sucrose counterselection in the presence of gentamycin, resulting in generation of the illustrated chromosomal deletion mutant. An unmarked mutant is then obtained after Flp recombinase-mediated excision of the Gm marker. ::: ![](1471-2180-5-30-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **PCR analysis of marked and unmarked *P. aeruginosaPA1520*deletion strains**. Colony PCR was performed on either wild-type PAO1 and its *PA1520*mutant derivatives, either containing a marked (*PA1520*::*FRT*-Gm-*FRT*) or unmarked (*PA1520*::*FRT*) *PA1520*deletion. The sizes of the expected PCR fragments are indicated. Note that the short deletion removes 41 bp of the *PA1520*coding sequence, corresponding to codons 131 to 145, but replaces these sequences with a 85 bp *FRT*scar. Lane M contained Hi-Lo molecular size markers of the indicated sizes from Minnesota Molecular. ::: ![](1471-2180-5-30-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Effect of deletions of *P. aeruginosa*GntR homologs on *fabA*expression**. The indicated genes were deleted in strain PAO1 containing a chromosomally integrated *fabA\'-lacZ*fusion (labelled PAO1). The control was PAO1 with a *fabA*Δ30\'-*lacZ*fusion; this strain harbours the same *fabA\'-lacZ*fusion but contains a deletion of the putative 30 nucleotide activator-binding site in the *fabA*promoter region. Strains were grown to mid-log phase in LB medium containing 0.05% Brij 58 +/- 0.05% oleic acid and β-galactosidase expression was monitored in triplicate samples. The dotted line marks expression levels observed in the putative activator binding-site mutant and similar levels would be expected in an activator mutant. ::: ![](1471-2180-5-30-6) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### PCR oligonucleotides ::: **Name** **Sequence (5\'→3\')^1^** ----------------------- ----------------------------------------------- Gene-specific primers G-UpF-GWL^2^ TACAAAAAAGCAGGCTccgatcagttgcaacacatc G-UpR-Gm^2^ TCAGAGCGCTTTTGAAGCTAATTCGttgttcctgcaccaccatct G-DnF-Gm^2^ AGGAACTTCAAGATCCCCAATTCGccggcgagttccacctgaa G-DnR-GWR^2^ TACAAGAAAGCTGGGTggttgagcttgctgtcgata Common primers Gm-F CGAATTAGCTTCAAAAGCGCTCTGA Gm-R CGAATTGGGGATCTTGAAGTTCCT GW-*attB*1 GGGGACAAGTTTGTACAAAAAAGCAGGCT GW-*attB*2 GGGGACCACTTTGTACAAGAAAGCTGGGT ^1^Sequences in capital letters are common for all genes amplified and overlap with the Gm or *attB*primer sequences. Lower-case letters indicate gene-specific sequences, here *PA1520*. PCR oligonucleotide sequences for all 27 genes amplified in this study are given in Table 2 (see [Additional file 1](#S1){ref-type="supplementary-material"}). ^2^G, gene specific sequence, here *PA1520*. :::

PubMed Central

2024-06-05T03:55:58.394580

2005-5-23

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173109/", "journal": "BMC Microbiol. 2005 May 23; 5:30", "authors": [ { "first": "Kyoung-Hee", "last": "Choi" }, { "first": "Herbert P", "last": "Schweizer" } ] }

PMC1173110

Background ========== The evolutionary pressures that have driven Hepatitis B virus (HBV) variation remain incompletely understood. Using whole HBV genotype sequencing, this variability can usefully be classified into at least eight families (genotypes) with a characteristic geographic distribution (reviewed in \[[@B1]\]). Alternatively, HBV strains can be classified serologically on the basis of antibody to surface antigen (subtypes). These two classifications broadly correlate, although some subtypes appear in more than one genotype. The extent of genetic diversity reflects the evolutionary history of the virus and the rate of genomic mutation, as well as gene specific selection forces. Several models of HBV evolution have been proposed (reviewed in \[[@B2],[@B3]\]) but fundamental parameters, such as the rate of interspecies transmission or the rate of nucleotide mutation (the molecular clock) remain unresolved \[[@B3]\]. Nevertheless, it is generally assumed that the emergence of HBV families may reflect adaptation to the genotype of the prevalent human host population \[[@B4]\]. The clinical course of HBV infection is very variable. Acute infections in adults are usually effectively controlled, but occasionally lead to fulminant hepatitis and death. In a proportion of individuals however, infection leads to chronic viral replication, which can lead to severe liver damage or hepatocellular carcinoma. Host factors including immune status clearly play a major role in determining clinical outcome. For example, perinatal transmission leads to up to 90% chronic carriership, while the figure is less than 10% for adults. However, pathogenicity has also been linked to virus genotype and several different mechanisms have been proposed for this observation \[[@B5],[@B6]\]. Sequence changes occurring during the course of infection (longitudinal diversity) have also been extensively documented. One common example is the introduction of a stop codon in the precore region which results in downregulation of secretion of a soluble form of HBV core protein (HBVe) whose function remains unclear \[[@B7],[@B8]\]. Interestingly, the downregulation of HBVe secretion is often associated with the appearance of anti-HBVe antibodies in serum, suggesting the protein itself may induce some form of immunological tolerance \[[@B7],[@B9]\]. The role of adaptive immunity both in determining the course of HBV infection and in driving HBV evolution is of special interest. Although pre-existing antibody to HBV surface protein (HBVs) (for example in vaccinated individuals) clearly provides strong protection, antibody to this antigen in natural infection is a late event, usually subsequent to effective control for viremia. In contrast, antibody to HBV core (HBVc), although this protein is internal to the virion, occurs early in infection in almost all infected individuals, irrespective of their ability to control viral replication \[[@B10]\]. T helper and cytotoxic responses to several proteins of HBV have also been detected, and the presence of a higher frequency of HBV specific CTL in liver is associated with the ability to control viremia \[[@B11]\]. As might be predicted CTL responses are not limited to structural proteins, but recognise several non-structural viral proteins. Virus specific T cell immune responses are most readily detected in individuals who effectively control viral replication. In chronically infected individuals, these responses are often much more difficult to detect, suggesting that the chronic state is associated with the establishment of some form of immunological tolerance \[[@B12]\]. HBVc antigen is a small protein, whose three-dimensional structure has been determined by X-ray crystallography \[[@B13]\], and whose immunogenicity in terms of both antibody and T cell responses has been studied rather extensively in both mouse and man. It thus represents an ideal starting point for studies aimed at relating antigen structure to immune response. Indeed longitudinal studies on small groups of HBV infected individuals have suggested that variation is more common in B cell and T helper cell epitopes, suggesting a possible immune driven escape mechanism \[[@B14]\]. As a basis for further functional exploration we have documented HBVc variation in detail. In this study we have collected several hundred protein sequences of HBVc from public databases, and have re-examined variation in relation to structure, immunogenicity and genotype. Results and discussion ====================== 742 protein HBVc sequences were retrieved from the NCBI protein database after some manual curation of very short sequence fragments. The mature HBVc protein sequences 1--185 (not including the precore region) were aligned and the number of polymorphisms found at each position (defined as amino acids occurring in more than 1% of the sequences at each site) was determined (see [Additional file 1](#S1){ref-type="supplementary-material"}). A plot of the number of variants at each position is shown in fig [1a](#F1){ref-type="fig"} and the major structural features, including the four alpha helices, the proline rich loop and the C-terminal arm \[[@B13]\] of the protein are shown in fig [1b](#F1){ref-type="fig"}. Over three quarters (141/185) of positions are completely invariant. Variation clustered within certain regions of the protein, with 18 of the 45 variable amino acids found within positions 59-100. Sequence variation in core protein may reflect the overall genotypic variation among HBV strains driven by drift or other unknown factors (\"hitchhiking\"). Alternatively, specific selection pressures may operate on HBVc driving diversity independently. In order to approach this question, the relationship between HBVc variation and HBV genotype was explored. Using text querying of the data base, a subset of 402 sequences were selected which had been assigned genotype A-D (these were the most frequent assigned genotypes within this set) by analysis of whole genome sequence or sequences outside the core region. These 402 core protein amino acid sequences were aligned and compared to each other and to the overall consensus sequence using Protdist, a Phylip program using the Jones-Taylor-Thornton model \[[@B15]\]. The distribution of distances between each sequence and the consensus was then plotted for each viral genotype as given in the database record (fig [2](#F2){ref-type="fig"}). The distance distribution profiles of the different genotypes were largely overlapping, suggesting that there was no correlation between distance from consensus and genotype. In an alternative approach, the relationship between all the different core sequences was determined using Fitch-Margoliash least squares analysis, and plotted as an unrooted tree (fig [3](#F3){ref-type="fig"}). The different genotypes of each virus are colour coded. Although some broad clustering of genotypes was evident, a significant amount of \"mixing\" can be observed. Similar qualitative results were obtained when phylogeny was determined by Neighbour Joining analysis (using Phylip program Neighbor) or by parsimony (using Phylip program ProtPars). This phenomenon was analysed in more detail in a subset of 40 full length core protein sequences selected from various sections of the tree illustrated in fig [3](#F3){ref-type="fig"} (see Table [1](#T1){ref-type="table"}). The phylogenetic relationship between these forty sequences was analysed again, using Neighbour-Joining (a less computation intense method) incorporating a 1000 fold bootstrap replicate in order to validate the tree topology. The results of this analysis are shown in fig [4a](#F4){ref-type="fig"}. In this smaller subset, genotypes A and D are reasonably well resolved, but genotypes B and C are extensively jumbled. One instance of this is illustrated by two identical sequences (1 and 40 in fig [4](#F4){ref-type="fig"}) derived from genotypes B and C respectively. The DNA sequences corresponding to each of these forty protein sequences were then obtained from GenBank, and analysed using the same bootstrapped Neighbour-Joining procedure. The tree obtained is shown in fig [4b](#F4){ref-type="fig"}. As expected, the bootstrap values are in general higher for the DNA tree (since three times as much sequence information is being analysed). The DNA sequences are somewhat more efficient at classifying B and C as separate grouping. In particular the two identical protein sequences (1 and 40 in figure) are well separated and correctly classified by the DNA phylogeny. Overall, however, significant misclassification remains, reflecting either incorrect genotyping, recombination between viral strains \[[@B16],[@B17]\], or simply insufficient discrimination based on these relatively short viral sequences. Taken together, this data suggested the forces driving the evolution of core were partially independent of the evolutionary forces driving diversification of overall genotype. The evolutionary pressures on sequence diversity were analysed in more detail by analysing synonymous (S) /nonsynonymous (N) nucleotide variation within the 40 viral sequences shown in Table [1](#T1){ref-type="table"}. The overall distribution pattern of synonymous /nonsynonymous mutation rates is shown in fig [5a](#F5){ref-type="fig"}. The protein shows evidence of strong purifying selection throughout the sequence, with dN/dS ratios (using a sliding window of 36 base pairs) mostly below 0.1, The apparent dS rate was not homogenous along the length of the gene. The sharp decrease in substitution rate from around position 110 most likely reflects the start of the overlapping open reading frame of the polymerase, and this area was therefore excluded from further analysis. The identification of site-specific positive or neutral selection operating within an area of overall purifying selection has received considerable attention \[[@B18]-[@B21]\], and a number of methodologies have been adopted. We applied maximum-likelyhood analysis as implemented using Bayes Empirical Bayes method in the software package PAML \[[@B22]\] and compared a number of models of site-specific selection distribution \[[@B18]\]. The best likelihood value was obtained using model 8 (see Methods) which includes a positive selection subset. The posterior probabilities for the positive (ω = 1.38) classes are plotted in fig [5b](#F5){ref-type="fig"}. Interestingly, only one position showed a probability of \>90 of being positively selected, with two more approaching 50% probability. In contrast 133 out of 140 codons showed a probability of \>90% of negative purifying selection. Thus there appears to be only very limited amount of positive selection within the HBV core sequence. The sequence information in [Additional file 1](#S1){ref-type="supplementary-material"}/fig [1](#F1){ref-type="fig"} was rationalised on the three dimensional structure of the capsid obtained from the Brookhaven Data Base. As evident in fig [6](#F6){ref-type="fig"} the variation in sequence is concentrated on the spikes at the outer surface of the capsids. All the amino acids with three or more variants are found on the outer surface of the capsid (fig [6b](#F6){ref-type="fig"} upper panel) while the internal surface facing the capsid lumen are relatively conserved (fig [6b](#F6){ref-type="fig"}, lower panel). A detailed alanine mutagenesis study has been carried out mapping those amino acids critical to proper capsid formation, and/or required for envelopment and virion formation \[[@B23]\]. Mutation of 24 amino acids was found to block capsid formation, virion formation or both. The position of these mutations on the 3D structure is shown in fig [7](#F7){ref-type="fig"}. The position of this set of mutations is quite widely distributed over the structure, suggesting multiple essential interactions are absolutely required either for proper capsid assembly, or for envelope and virion assembly. However, interestingly, all but one of these 24 amino acids were found to be invariant in the data set analysed in this study. The only exception observed was at position 129, where changing proline to alanine was found to block both capsid and virion formation. Both glutamine and threonine are found in a proportion of virus sequences at this position, and further mutagenesis will be required to clarify the constraints imposed by the requirements of virion formation on the sequence at this particular position. Thus the structural requirements of virion assembly seem to impose a significant restraint on HBVc diversity. However, several positions which were found to tolerate alanine mutagenesis in terms of capsid/virion assembly were nevertheless invariant in the set of sequences examined here. The overall high degree of conservation in HBVc therefore probably reflects the multiple functions required from this protein, including control of intracellular targeting \[[@B5]\], pregenome/DNA polymerase packaging, capsid disassembly and viral maturation. Pressure by the host immune system is one obvious candidate driving variation in the protein sequence of HBVc. CD8 cytotoxic cells are believed to play a key role in controlling virus replication during HBV infection \[[@B24],[@B25]\]. Several CD8 T cell epitopes have been characterised in detail (e.g. \[[@B26]-[@B28]\]) by the use of T cell clones and lines or by elution from HLA \[[@B29]\]. Some of the sequences are shown in fig [7](#F7){ref-type="fig"}, together with the known variant sequences in the dataset analysed in this study. Although several of the epitopes lie within the more conserved inner region of the capsid, two of the epitopes show substantial polymorphism (fig [8](#F8){ref-type="fig"}). Interestingly, a previous paper did not find any evidence for emergence of mutations within the major CD8 epitopes during chronic HBV infection \[[@B30]\]. The evidence implicating CD4 T cells in HBV control remains much less clear. Furthermore, although regions containing CD4 epitopes have been described, many of the epitopes are not very well characterised. One putative \"immunodominant\" CD4 T cell epitope (amino acids 50-69) does contain a number of variable amino acids (fig [9a](#F9){ref-type="fig"}), and indeed changes in its sequence have been related directly to changes in T cell response in vivo \[[@B31]\]. A second epitope (core region 147-156) identified as a major target in HLADR13 individuals has also been examined in some detail \[[@B32]\]. Interestingly this epitope also shows considerable variation (fig [9a](#F9){ref-type="fig"}) including a mutation at position 151 shown to be essential for T cell recognition. The epitope also spans the region of a two amino acid insertion which is found exclusively in viruses of genotype A. Further detailed mapping of CD4 T cell recognition sites, in relation to natural sequence variation would be seem to be an area of great interest. Finally, considerable information has been accumulated on the interaction of antibodies to HBV capsids. As shown in fig [9b](#F9){ref-type="fig"}\[[@B10]\] the major defined antibody specificities lie on the outside of the capsid structure, particularly at the tip of the spikes and at the junctions between adjacent spikes. These regions do indeed contain the majority of the HBVc sequence variation, although the most variable amino acids themselves have not been identified as known antibody contacts \[[@B33]\]. However, the contribution of anti-HBVc antibody to protection remains unclear, particularly since the capsid in intact virions is presumably largely shielded from antibody by the HBV envelope. Conclusion ========== This study makes use of the large number of HBVc sequences now available in public databases to characterise sequence variation in HBVc. One limitation of such an approach is that detailed clinical information associated with infection is not available, and in particular, it is not possible to examine variability in the context of the longitudinal course of an HBV infection. This is likely to be an important factor since mutations are often found to arise late in infection, associated with a variety of clinical outcomes (e.g. \[[@B34],[@B35]\]). In addition few of the available sequences have been checked for their ability to make competent infectious virus, and some sequences may therefore represent non-functional proteins. However, despite these limitations, the data available does allow some general conclusions. Overall, HBVc contains a large proportion of invariant amino acids, and a strong over representation of synonymous versus non-synonymous mutations at almost every codon. Both features suggest the presence of strong constraining forces on sequence diversity. Virion assembly is likely to provide one major constraining force \[[@B23]\]. As reported previously (e.g. \[[@B14]\]) sequence diversity appears to be clustered, and mapped predominantly to the spike and external surface of the capsid. These positions may allow greater flexibility in terms of virion assembly. One significant consequence of the strong purifying selection is that protein sequence is a poor predictor of genotype for this gene. DNA sequence which reflects predominantly synonymous mutations, is a better discriminator, particularly in resolving genotypes B and C. Longer sequence analysis, however, is clearly necessary to obtain reliable genotyping information. The putative role of immune selection in driving HBV core diversity is much more unclear. Direct evidence for positive selection, at least using the analysis presented here, identifies only a single amino acid (position 74 at the tip of the viral spike) as showing evidence for positive selection of diversity. Nevertheless, it is clear that several polymorphic positions lie within T or B cell epitopes. Hence, while the overall effect of immune selection on HBV sequence diversity may be small, sequence diversity may have a significant effect on the immune response at an individual level. The data analysis given here will help inform further analysis of the HBV-specific immune response. The combination of T cell and antibody recognition studies with directed mutagenesis of HBVc should determine more precisely the relationship between structure, immunity and pathogenicity. Methods ======= Human HBVc sequences were retrieved from the NCBI protein sequence database \[[@B36]\], limiting the search to organism = Hepatitis B virus and searching for core protein. Additional classification into genotypes A-D was done using the text search tool to look for \"genotype X\". A proportion of hits were verified by manual inspection. The initial 780 hits were aligned using the EMMA program on EMBOSS (a version of Clustal)(see\[[@B37]\] using the BLOSUM 62 similarity matrix. The alignment was further refined by manual inspection using the sequence editor Bioedit, and very short or badly aligned sequences removed. The frequency of amino acids at each position was determined using the EMBOSS program PROPHECY. The matrix observed was converted into polymorphism frequency by setting a cut-off of 1% frequency at each position. A phylogenetic tree of the HBVc sequences was created using the Phylip program Kitsch \[[@B38]\]), which uses a Fitch-Margoliash criterion based on a distance matrix obtained using the Phylip program Protdist. The tree was displayed using Treeview \[[@B39]\], copied in Adobe Illustrator and color coded according to genotype. Because this method is extremely processor intensive (the best tree is analysed at each iteration) it was not possible to bootstrap. Analyses of the same data were also done using nearest neighbor analysis (using the Phylip program Neighbor) and parsimony using the EMBOSS (loc cit) program EPROTPARS. Although the fine details of the trees varied between methods the overall qualitative conclusion were the same. In order to further validate the conclusions of the phylogeny, 40 sequences (shown in Table [1](#T1){ref-type="table"}) chosen manually to cover the major branches of the tree shown in fig [3](#F3){ref-type="fig"} were reanalysed using the Nearest Neighbour with bootstrap option (1000 bootstraps) of ClustalW \[[@B40]\]. The consensus tree set was plotted in Treeview and coloured in Illustrator. Similar analysis was carried out on an alignment of the DNA sequences corresponding to each protein sequence. The analysis of synonymous/nonsynonymous mutations rates was carried out initially using the program DNASP3.0 \[[@B41]\] (using the Nei/ Gojobori algorithm) with a 36 base pair sliding window shifted by 9 nucleotides. A more detailed analysis at individual codons was carried out using the program PAML version 3.14 (using maximum likelihood Bayes Empirical Bayes inference, as described in \[[@B20]\]). Analysis was carried out using a variety of selection models, but the output represented in fig [5](#F5){ref-type="fig"} used model 8 \[[@B22]\]. The models assume that the selection pressure (measured as ω) operating at each codon can fall within a range of different classes. Model 8 assumes a distribution of negative selection values (for all of which ω \<1), or a positive selection class with ω = 1.38. This model was found to give the best likelihood estimate. The program then calculates the posterior probability (p value) that each codon within a sequence falls within a particular class. The crystal structure of HBVc was retrieved as a pdb file from the Brookhaven database, and displayed and coloured using RasMol software (version 2.7.1.1 for Windows). Figures [6](#F6){ref-type="fig"} and [7](#F7){ref-type="fig"} show the structure of four identical monomers, to illustrate the spikes and their interaction, while figs [8](#F8){ref-type="fig"} and [9](#F9){ref-type="fig"} show a single monomer for clarity. Abbreviations ============= Hepatitis B virus HBV; Hepatitis B core protein HBVc; Hepatitis B surface protein HBVs Authors\' contributions ======================= RM provided bioinformatics support and expertise, and carried out the phylogenetic analysis shown in fig 4. BMC initiated the project and was responsible for the text, and the data and analysis shown in the other figures. Supplementary Material ====================== ::: {.caption} ###### Additional File 1 Amino acid polymorphisms along the HBVc sequence. Table containing all the polymorphic amino acid residues found at each position along the 742 HBV sequences analysed. ::: ::: {.caption} ###### Click here for file ::: Acknowledgements ================ I am grateful for much useful discussion with many colleagues at UCL, in particular to Paul Kellam for help with the bioinformatics, to Ziheng Yang for initial help setting up PAML and to Richard Tedder, Antonio Bertoletti, Mala Maini and Nikolai Naoumov for advice on HBV. I am grateful to Dr. Volker Bruss (University of Goettingen) for his help and advice. I am also very grateful to the staff at the UK HGMP Resource Centre for a lot of help and patience with running the various Bioinformatic programs. Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Sequence variation along the HBVc sequence.a)**The number of polymorphisms (variants occurring in more than 1 % sequences examined) found at each position of the HBVc sequence. See [additional file 1](#S1){ref-type="supplementary-material"} for full data. **b)**Major features of secondary structure along HBVc \[13\]. Blue : α-helix Green : Proline rich loop; Red : C-terminal arm. ::: ![](1471-2180-5-33-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Relationship between HBVc sequence and genotype.**The 403 sequences from viruses of known genotype were aligned and compared to the consensus. The distance of each sequence from the consensus was calculated using maximum likely-hood model (Jones-Taylor-Thornton). The frequency distribution of distance from consensus was then plotted for each genotype A-D. ::: ![](1471-2180-5-33-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Analysis of the relationships between HBVc sequences.**All the sequences were classified and plotted as an unrooted tree using the Fitsch Margoliash criterion (Kitsch, see Materials and Methods). Where known the genotype is indicated by colour coding (Yellow = A, Green = B, Blue = C, Red = D). Sequences of unknown genotype are in black. ::: ![](1471-2180-5-33-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Analysis of 40 protein and DNA HBVc sequences.**The 40 sequences detailed in Table 1 were analysed using Neighbor Joining analysis of both protein (panel a) and DNA (panel b) and plotted using TreeView. The bootstrap values (as percentage) are shown in smaller font. The accession number corresponding to each numbered sequence is shown in Table 1. The genotype is indicated by colour coding (Yellow = A, Green = B, Blue = C, Red = D). ::: ![](1471-2180-5-33-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Synonymous/nonsynonymous mutation rates and selection in HBVc.**Panel a shows the distribution of synonymous (dS) and non-synonymous (dN) substitutions (the number of such substations per site), and the ratio dN/dS along the sequence of HBVc. Gaps indicate dS = 0. Panel b shows the posterior probability p^+^that dN/dS\>1 for each position of HBVc, as calculated by PAML. The probability that dN/dS\<1 (purifying selection) is not plotted for clarity but is given by (1- p^+^). ::: ![](1471-2180-5-33-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Sequence variation in the context of capsid structure.**The crystal structure formed by four HBVc subunits was displayed and coloured using RasMol software. Positions containing 2 polymorphisms are coloured blue, 3 polymorphisms are coloured orange, and 4 or above are coloured red. Grey amino acids are invariant. a) The structure is displayed in ribbon form, showing a vertical section though the capsid, with two spikes projecting upwards, and the internal face of the capsid shown at the bottom of the picture. b) The structure is displayed in space fill. Upper Panel The structure is displayed so that the spikes and outward surface of the capsid are shown towards the viewer, and only the outer surface of the capsid is visible. Lower Panel The structure is rotated by 180 degrees so only the lower (inner) face of the capsid is visible. ::: ![](1471-2180-5-33-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **Structural constraints and HBV diversity.**The crystal structure formed by four HBVc subunits was displayed and coloured using RasMol software. The position of the amino acid identified as essential for virion formation (violet) or for both capsid and virion assembly (green) \[23\] are shown. The rest of the colour coding is as shown in fig 6. ::: ![](1471-2180-5-33-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### **Variation within CD8 T cell epitopes of HBVc.**Five of the best defined class I T cell epitopes from HBVc (for references see text) are shown together with alternative amino acids found at each position. The position of the epitopes (green) within the HBVc three dimensional structure is shown in the right hand panel. The rest of the colour coding is as shown in fig 6. ::: ![](1471-2180-5-33-8) ::: ::: {#F9 .fig} Figure 9 ::: {.caption} ###### **Variation with CD4 and antibody epitopes of HBVc.**a) Two of the best defined CD4 epitopes within HBVc (taken from \[31,32\]) are shown together with alternative amino acids found at each position. The position of the epitope 50\#150;69 is shown in the context of the HBVc three dimensional structure above(in green). The rest of the colour coding is as shown in fig 6. Epitope 147\#150;156 is not shown since this region of HBVc was excluded in the crystallographic study. b) The best defined antibody epitopes (taken from \[33\]) are shown (in green) in the context of the HBVc three dimensional structure above. The rest of the colour coding is as shown in fig 6. ::: ![](1471-2180-5-33-9) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Accession numbers and genotype for HBV sequences analysed in fig 4 ::: Number Accession number Genotype -------- ------------------ ---------- --- Protein DNA 1. AAK81691 AY040627 C 2. AAO41301 AY167091 C 3. AAP06648 AY206392 C 4. AAL31832 AF324119 C 5. AAP06544 AY206375 B 6. AAP06549 AY206376 C 7. AAL31776 AF323466 B 8. AAL31802 AF324088 B 9. AAP06642 AY206391 B 10. AAP79869 AY233296 D 11. AAP31176 AY269059 D 12. AAR19326 AY230115 D 13. BAC57440 AB090269 D 14. AAL31837 AF324124 A 15. AAF24670 AF121240 D 16. BAD02318 AB126581 D 17. AAL31826 AF324112 D 18. AAP31178 AY269061 D 19. BAA85376 AB033559 D 20. AAP31169 AY269052 D 21. BAC92696 AB106885 B 22. BAC92693 AB106884 B 23. AAR19322 AY230111 A 24. AAL31839 AF324126 A 25. AAP79778 AY233283 A 26. AAL31785 AF324071 A 27. AAP79724 AY233275 A 28. AAP79715 AY233274 A 29. AAP79729 AY233276 A 30. AAP79736 AY233277 A 31. AAL31856 AF324143 A 32. AAG49707 AF223961 C 33. AAP06636 AY206390 B 34. AAF24738 AF121250 B 35. BAA32843 AB014363 C 36. AAP31199 AY269084 C 37. AAL31815 AF324101 C 38. AAL31819 AF324105 C 39. AAL31781 AF324066 B 40. AAF24705 AF121245 B :::

PubMed Central

2024-06-05T03:55:58.396937

2005-5-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173110/", "journal": "BMC Microbiol. 2005 May 27; 5:33", "authors": [ { "first": "Benjamin M", "last": "Chain" }, { "first": "Richard", "last": "Myers" } ] }

PMC1173111

Background ========== The vast majority of fast synaptic transmission in the mammalian central nervous system is glutamatergic. Proper expression, localization, and regulation of glutamate receptors are critical for brain development and plasticity. Not surprisingly, the molecular mechanisms controlling glutamate receptor expression, localization, and stability are of great interest. The most common recent approach to understanding these mechanisms has been biochemical: Proteins are identified based on a biochemical interaction with a glutamate receptor subunit. This approach has identified a number of important candidates, some of which have been subsequently shown to play important roles in glutamate receptor trafficking and stability at the PSD (for review see \[[@B1],[@B2]\]). An alternative approach is forward genetics: Mutant animals are screened for alterations in glutamate receptor localization. Perhaps the most well-known glutamate receptor trafficking protein identified via genetics is stargazin, which is disrupted in stargazer mutant mice \[[@B3]\]. The identification of stargazin was serendipitous; mammalian genetic screens are unfortunately still relatively time-consuming and expensive (though this may change as RNAi techniques advance). Most forward genetic screens glutamate receptor mutants have used *C. elegans*. Studies using *C. elegans*have highlighted the importance of ubiquitination, CamKII, and PDZ proteins in controlling glutamate receptor number, and identified a novel protein required for nematode glutamate receptor function \[[@B4]-[@B7]\]. Unfortunately, glutamatergic synapses in *C. elegans*are accessible for electrophysiology only with great difficulty. The subunit composition of *C. elegans*receptors in vivo has also not yet been determined, which hinders the study of subunit-specific trafficking mechanisms. *Drosophila*are not only amenable to powerful genetics, but also contain glutamatergic neuromuscular junctions (NMJs) that are individually identifiable and accessible throughout development to high-resolution electrophysiological and microscopic techniques. These techniques have revealed a significant amount of information concerning this popular model synapse, including the exact subunit composition of glutamate receptors in vivo. *Drosophila*NMJs contain two subtypes of postsynaptic glutamate receptor, which are molecularly, pharmacologically, and spatially distinct \[[@B8]-[@B11]\]. The \'A\' receptor subtype contains the subunit GluRIIA, in combination with subunits GluRIIC, GluRIID, and GluRIIE. The \'B\' receptor subtype contains the subunit GluRIIB, in combination with the subunits GluRIIC, GluRIID, and GluRIIE. Both receptor subtypes are most similar in sequence to mammalian kainate receptors. The *Drosophila*genome also encodes receptor subunits with high similarity to mammalian AMPA, delta, and NMDA receptor subunits, but these proteins are not found postsynaptically in the NMJ. Virtually nothing is known about the molecular mechanisms that control *Drosophila*glutamate receptor expression, localization, and stability. To determine the molecular mechanisms controlling *Drosophila*glutamate receptor expression, localization, and stability, we are screening transposon insertion mutants generated by the Berkeley *Drosophila*Gene Disruption Project \[[@B12]\]. This mutant collection contains insertions in over 40% of the entire genome, and the insertion site for almost every mutant has already been identified by inverse PCR. With the addition of transposon insertions from other collections, it is likely that Drosophilists will have access to insertion mutants of almost every gene in the fly genome within a few years. Theoretically, one could quickly and efficiently identify the complete list of genes required for any particular process simply by examining a non-redundant set of mutants for a phenotype of interest. We are testing this idea and using this approach to define the broad categories of genes required for glutamate receptor cluster formation. We identified 57 lethal mutants with qualitatively abnormal glutamate receptor clusters. Here, we present this list and discuss the types of genes that are, and are not, represented. Because of the high prevalence of background mutations even in transposon mutant collections, the role of individual genes must be extensively confirmed before a role for any particular protein is assumed. However, this problem may be circumvented, and perhaps more insight gained, by focusing on the types of genes identified rather than the identity of individual genes. We do so here. Results ======= We sought to identify *Drosophila*mutants with abnormal glutamate receptor cluster development. Elimination of *Drosophila*NMJ glutamate receptors results in paralysis and embryonic lethality \[[@B9]\]. Mutations that reduce (but do not eliminate) NMJ glutamate receptors allow hatching, but typically cause larval or pupal lethality \[[@B8]-[@B10]\]. Any effective screen for glutamate receptor cluster formation mutants must therefore include examination of homozygous lethal mutants. However, proper examination of glutamate receptor clusters in embryos and larvae requires technically challenging dissection techniques and time-consuming confocal microscopy. Therefore, we sought to minimize examination of mutants that do not have glutamate receptor cluster defects. To do this, we made the assumption, based on the studies cited above, that mutants with severe defects in glutamate receptor cluster formation are more likely to be homozygous lethal. Thus, the first step in our screen for mutants with reduced or eliminated glutamate receptors was identification of recessive lethal transposon mutants (e.g. insertions lines with no homozygous viable adults). We concentrated on the collection of *GT1*and *SuPor-P*insertion mutants, since these transposons were engineered for maximal gene disruption \[[@B13],[@B14]\]. Of 2185 mutants (representing insertions in approximately 16% of the entire genome), 220 insertion lines contained lethal mutations. Because a prerequisite for NMJ glutamate cluster formation is development of neuromusculature, we examined dechorionated embryos from each of the 220 stocks to ensure that homozygous mutants developed into \'morphologically mature\' late stage embryos typical of wildtype animals 16--17 hr after egg laying, AEL, which is when NMJs begin to form \[[@B15]-[@B18]\]. Morphological maturity was based on the presence of characteristics typical of late stage 17: clear segmentation, mouthhooks, condensed CNS, malphigian tubules, and visible trachea. 205 of 220 (93%) of the homozygous lethal mutants formed morphologically mature embryos. 202 of these mutants were rebalanced using a GFP-tagged balancer chromosome for unambiguous identification of homozygous mutant animals. We were unable to rebalance 3 mutant lines using chromosome-appropriate GFP-tagged balancers, possibly because the insertion mutant chromosome and balancer chromosome both contained lethal mutations that fail to complement. We then verified that all 202 GFP-balanced stocks did indeed carry recessive lethal mutations and determined the latest stage to which homozygous (non GFP) embryos and larvae lived. For each of the 202 balanced lines, several (typically 5--10) homozygous mutant animals, along with control heterozygous siblings, were manually dissected at the latest viable stage to expose NMJs on ventral longitudinal muscles, then fixed and stained using anti-HRP antibodies that recognize all neuronal membrane (including peripheral axons and NMJs) and anti-GluRIIA antibodies. After dissection, fixation, and staining, each of the 202 rebalanced mutant lines was examined for changes in NMJ morphology and glutamate receptor expression using laser-scanning confocal fluorescent microscopy. 57 of the 202 rebalanced lethal mutants (28%) displayed consistent (most NMJs in several animals) defects in glutamate receptor expression without severe presynaptic morphological abnormalities (e.g. 57 mutants were identified in which NMJs formed, but glutamate receptor clusters were altered). Examples of some of these phenotypes, from both embryonic and larval NMJs, are shown in Fig. [1](#F1){ref-type="fig"}. All phenotypes could be classified into one of two broad categories: 1) loss of glutamate receptors (fewer glutamate receptor clusters or smaller individual clusters), or 2) gain of glutamate receptors (more clusters or larger clusters). We relied on cluster size and number because cluster size and number measurements (compared to fluorescence intensity measurements) avoid a requirement for fluorescence intensity calibration or problems associated with potential differences in background immunofluorescence between genotypes. Assuming constant receptor density, cluster size should be directly proportional to the number of clustered receptors. In support of this assumption, immunoreactive cluster sizes correlate well with high-resolution patch-clamp electrophysiological measurements between genotypes \[[@B11],[@B19],[@B20]\] and throughout embryonic/larval NMJ development (Featherstone, unpublished observations). Most (84%) of the mutants with disruptions in glutamate receptor clusters showed a qualitative loss in receptors. Severe loss of GluRIIA was always associated with embryonic/L1 lethality. One example is *P{SUPor-P}KG00333*, which shows a complete loss of A-type receptors despite the presence of morphologically normal presynaptic terminals (Fig. [1A](#F1){ref-type="fig"}, left). Some mutants with loss of receptors survived until pupation. An example of a third instar viable mutant, *P{SUPor-P} Chro*^*KG03258*^, with reduced GluRIIA is shown in Fig. [1B](#F1){ref-type="fig"} (middle column). A minority (16%) of third instar viable mutants showed a qualitative increase in GluRIIA immunoreactivity. An example of one of these mutants, *P{SUPor-P} vri*^*KG01220*^, is shown in Fig. [1B](#F1){ref-type="fig"} (right column). Except for *P{SUPor-P}KG00212*, all of the mutants that displayed an increase in GluRIIA immunoreactivity were viable as third instar larvae and also displayed an increase in the number of presynaptic boutons. This presynaptic phenotype is consistent with previous *Drosophila*studies showing that overexpression of postsynaptic GluRIIA causes presynaptic overgrowth \[[@B21]\]. BG and KG transposons are designed to generate loss of function mutations. Therefore, the isolation of mutants with increased receptor cluster size suggests that receptor insertion and clustering per se are not rate-limiting. There are molecular mechanisms, revealed by our screen, that actively restrain the number of synaptic glutamate receptors. For all of the mutants identified in our screen, the Gene Disruption Project has precisely determined the genomic insertion site of the transposon using inverse PCR. In the vast majority of cases, inverse PCR results were consistent with a unique insertion, and in most cases the flanking genomic sequence revealed that the insertion was in an annotated gene. For each of the mutants identified in our screen, we used the BDGDP *P*-screen database <http://flypush.imgen.bcm.tmc.edu/pscreen/> and/or BLAST searches with flanking genomic sequence from iPCR results to identify which gene was mutated by the inserted *P*-element. Putative functions were assigned to each of these genes based on previous publications, FlyBase annotations, and/or Genbank BLAST searches. We binned each of the insertion site genes into one of the following functional categories: 1) extracellular matrix proteins (mucins), 2) cell adhesion proteins 3) cytoskeleton proteins, cytoskeletal regulators, and adaptor proteins, 4) kinases and phosphatases, 5) ubiqitination proteins (ubiquitin ligases), 6) transporters/pumps, 7) proteins involved in gene expression and protein translation, 8) enzymes, and 9) proteins of unknown/novel function. These categories, and the relative number of proteins in each category, are illustrated by Figure [2](#F2){ref-type="fig"}. All categories shown in Figure [2](#F2){ref-type="fig"} were represented by at least two mutant genes. Table [1](#T1){ref-type="table"} contains the complete list of all mutants identified in our screen, listed by functional category. For each mutant, Table [1](#T1){ref-type="table"} indicates whether the phenotype was loss (down arrow) or increase (up arrow) in postsynaptic glutamate receptors. Table [1](#T1){ref-type="table"} also lists the *Drosophila*gene in which each transposon is inserted, and the mouse homolog of each of those genes (identified by BLASTP against the mouse genome refseq protein database; <http://www.ncbi.nlm.nih.gov/genome/seq/MmBlast.html>). A definitive test of whether any particular gene is required for glutamate receptor cluster development requires precise excision of each P-element insertion, or transgenic rescue, followed by re-examination of the NMJ at the same developmental stage -- a task which is not practical on a genomic scale. However, we performed two broad types of checks to identify potential caveats in our results: expression analysis and complementation tests for lethality. *Drosophila*glutamate receptors are expressed in neurons and muscles. Thus, any gene required for expression and/or localization of glutamate receptors should be expressed in neurons and/or muscles. The *Drosophila*gene expression database (accessed at: <http://www.fruitfly.org/cgi-bin/ex/insitu.pl>) describes expression patterns of approximately 3300 different *Drosophila*genes (\~24% of the genome), as determined by embryonic in situ cDNA hybridization \[[@B22]\]. Of these 3300 genes, which presumably represent a random \'sampling\' of the genome, 2350 (71%) are expressed in the nervous system and/or body wall muscles. The gene expression database contains data for 28 (49%) of the genes identified in our study. 24 (86%) of these genes identified by our screen are expressed in the nervous or muscular systems. 4 (14%) of the genes identified in our screen are annotated as not detectably expressed in the nervous or muscular systems. Thus, most (86%) of the genes identified in our screen have expression patterns consistent with the conclusion that these genes are important for glutamate receptor cluster development, but we did not identify neuronal and muscle genes significantly above the level expected by a random sampling of the genome -- possibly because the fraction of genes expressed in neurons and muscles (71%) is quite high to begin with. There may also be cell non autonomous roles for some genes with regard to receptor expression and/or localization. Although all of the mutants identified in our screens displayed a receptor phenotype, and all of the mutants contain a unique P-element insertion, an important genetic caveat is that the receptor phenotype may not actually be caused by the transposon insertion. For example, the *P*-element mobilizations that generated each of the *P*-element alleles for the gene disruption project may have also created second-site mutations in an unknown locus. Spontaneous mutations can also be easily stabilized and propagated in a stock carrying a balanced lethal mutant chromosome such as those screened here. To estimate the frequency of background mutations in this transposon mutant collection, we performed complementation tests to determine whether the *P*-element insertions failed to complement lethal alleles of the same insertion locus. Of the 57 mutants identified in our screen, 20 were tested for complementation to deficiencies that remove the insertion site. Ten (50%) of the complementation tests failed, suggesting that lethality in half of the mutants may be caused by a mutation other than the *P*-element insertion. These results are consistent with those from a related screen for presynaptic morphology mutants (Liebl FLW, Werner KM, McCabe BD, Featherstone DE: **A Genome-Wide P-element screen for Drosophila synaptogenesis mutants**. M*anuscript in preparation*), which included complementation tests to over 80 genes. Complementation did not appear to be biased for any particular type of gene (in this or the related Liebl et al study). We did not systematically test for complementation of the receptor cluster phenotypes. Discussion ========== In an effort to identify new genes required for glutamate receptor cluster development, we screened lethal transposon insertion mutants for alterations in postsynaptic glutamate receptor clusters. 202 lethal insertion lines, representing insertions in \~1.4% of the genome, were manually dissected and glutamate receptor clusters were examined using immunocytochemistry and confocal microscopy. This screen identified 57 mutations in 56 different loci. Transposon mutageneses are becoming increasingly popular, because a transposon insertion simultaneously mutates and \'tags\' a gene. An assumption of all transposon mutagenesis (and subsequent screens) is that the insertion locus is responsible for any observed phenotype. Expression analysis showed that the expression pattern of insertion loci identified in our screen is broadly consistent with that expected from genes involved in glutamate receptor cluster development, but also not compellingly different (86% vs. 71%) from that expected from a random sampling of the genome. Furthermore, complementation tests suggest that approximately one-half of the lethality observed in our screen was not due to disruption of the insertion locus. Thus, it is impossible to definitively say which particular genes are important without extensive validation of each candidate. And even with extensive validation sufficient to confirm that a particular locus is involved, it is difficult to know whether a gene\'s role in receptor cluster formation is indirect with regard to glutamate receptor cluster formation (i.e. whether a gene product regulates another protein which in turn clusters receptors). Nevertheless, one cannot discount our results without also casting aside the fact that decades of forward genetic screens have successfully identified many genes critical for many different processes \[[@B23]\]. But how does one decide which screen \'hits\' to trust, and is there any way to usefully interpret results of forward genetic screens? Should we exclude from consideration all mutants in which lethality was complemented (and conversely accept all mutants in which lethality failed to be complemented)? This depends on the degree of coupling between two different phenotypes: lethality and abnormal glutamate receptor cluster development. To enrich for mutants with receptor cluster defects, we examined only homozygous lethal mutants, on the assumption that GluRIIA cluster defects and lethality are strongly linked. Indeed, all of the mutants showing severe loss of GluRIIA were embryonic/first instar lethal. Nevertheless, this assumption probably does not hold for all mutants. At the time this screen was initiated, it was assumed that all *Drosophila*NMJ glutamate receptors contained the subunit GluRIIA, and that the viability of GluRIIA null mutants was due to substitution by the alternate subunit GluRIIB \[[@B24]\]. Thus, the presence and localization of GluRIIA was considered an accurate marker for all NMJ receptors. Subsequently, we and others have determined that the *Drosophila*NMJ contains two independently assembled and localized subtypes of postsynaptic glutamate receptors: A-type receptors, that contain the subunit GluRIIA (plus GluRIIC, GluRIID, and GluRIIE, but not GluRIIB), and B-type receptors, which contain the subunit GluRIIB (plus GluRIIC, GluRIID, and GluRIIE, but not GluRIIA) \[[@B8]-[@B10]\]. GluRIIA therefore serves as a tag for only one-half to two-thirds (depending on developmental stage; Featherstone, unpublished) of fly NMJ glutamate receptors. Complete elimination of A-type receptors does not result in lethality \[[@B24],[@B25]\]. Conversely, lethality can obviously be caused by many defects other than loss of NMJ glutamate receptors. Since viability and glutamate receptor cluster development may be only loosely coupled, complementation tests for lethality indicate relatively little about the reliability with which our screen identified genes required for proper glutamate receptor cluster formation (although these complementation results do give important insights into the properties of this mutant collection, and transposon mutageneses in general). Similarly, the large fraction of genes expressed in the neuromusculature makes it difficult to judge, based on expression, whether any particular gene has simply been randomly selected. Thus, we are left with uncertainty regarding any particular gene until that gene\'s role in synaptogenesis can be extensively validated -- a prospect that is not practical on a genomic scale. However, the sequencing of several metazoan genomes, along with the realization that the function of homologous proteins tends to be conserved across genomes, opens the possibility of another approach, as demonstrated in our study: \'functional category analysis.\' Functional category analysis involves screening a non-redundant collection of mutants for the phenotype of interest, assigning putative functions bioinformatically, and categorizing the \'hits\' by function. The goal of this approach is not to generate a definitive list of individual proteins involved in a process, but to gain insight into the types and relative numbers of proteins required. All of the categories shown in Figure [2](#F2){ref-type="fig"}, for example, are represented by several mutants. Therefore, even with a 50% accuracy rate (which would be a worst case estimate; the real accuracy is probably substantially higher -- see below), the gene function *categories*are likely to be correctly identified. Functional category analysis recognizes that not all hits will be valid, that not all genes will play roles specific to a particular process, and that many proteins are only indirectly required. Functional category analysis reveals unexplored areas of relevant biology and provides a broad roadmap for further study. For example: instead of studying the individual genes identified in Table [1](#T1){ref-type="table"} (which is arguably a hit-and-miss endeavor unlikely to shed much global insight into the process), we plan to focus research toward understanding how RNA regulation and regulated translation are involved in glutamate receptor cluster formation. Similarly, we are directing efforts toward understanding the interactions of receptors with the actin and microtubule-based cytoskeletons. Preliminary results suggest that this approach is insightful and effective \[[@B20]\]; Liebl, Karr, & Featherstone, unpublished observations). What percentage of the genome is involved in glutamate receptor cluster development? Our results allow this question to be addressed (at least with regard to the *Drosophila*NMJ). If glutamate receptor cluster development and lethality are completely uncoupled, then our results imply that approximately 28% (57/202) of the entire *Drosophila*genome is required for glutamate receptor cluster development. Depending on the percentage of insertions that were too hypomorphic to show qualitatively detectable phenotypes via immunocytochemistry, that percentage could be higher. Glutamate receptor cluster formation and lethality, however, are not unrelated phenotypes. Therefore, 28% is likely to be a gross overestimate. Complete loss of *Drosophila*NMJ glutamate receptors results in paralysis and embryonic lethality \[[@B9]\], and mutations that reduce (but do not eliminate) NMJ glutamate receptors allow hatching, but typically cause larval or pupal lethality \[[@B8]-[@B10]\]. In the present study, all severe loss of function phenotypes were associated with embryonic/early larval lethality. If all GluRIIA cluster mutants are lethal, then mutants with normal glutamate receptor clusters would have been eliminated in the selection for lethality early in the screen (indeed, this was the rationale for this step). In this case, our results suggest that approximately 2.6% (57/2185) of the genome is required for glutamate receptor cluster formation. This estimate, however, is also likely to be flawed; although severe disruptions in glutamate receptor cluster formation cause lethality, not all lethality is likely due to disruptions in glutamate receptor cluster formation. Mutations that trigger loss of GluRIIA are not necessarily lethal \[[@B24],[@B25]\]. Lethality and GluRIIA cluster phenotypes are not absolutely coupled, and therefore 2.6% is likely too low of an estimate. Given the uncertainties above, we can say only that somewhere between 2.6 and 28% of the fly genome (360--3900 genes) is required for NMJ glutamate receptor cluster formation. This is a wide range, which in any case represents a surprisingly large number of genes. Is this reasonable? Do all of these genes represent specific machinery for glutamate receptor cluster expression and clustering? To answer these questions, it is helpful to consider the types of genes identified in the screen. As noted earlier, the role of each individual gene needs to be validated before placing too much emphasis on any particular protein\'s role. But a general discussion of the genes implicated is helpful for evaluation of the general results. In support of the idea that our screen correctly identified genes required for glutamate receptor cluster formation, some of the types of genes identified in our screen have been previously identified as important for postsynaptic development. Polo, for example, was identified in our screen, and mammalian polo-like kinases are receiving increasing attention as important players in synapse development \[[@B26]\]. We also identified several other kinases and phosphatases. Activation of the mitogen activated protein kinase (MAPK) pathway facilitates AMPAR surface expression \[[@B27],[@B28]\], and interaction of AMPAR GluR2 subunits with GRIP and PICK1 is dependent upon the phosphorylation of the GluR2 subunit \[[@B29],[@B30]\]. Our screen also isolated mutants in two different ubiquitin ligases. Consistent with this, ubiquitination is known to regulate glutamate receptor number and synapse development \[[@B6],[@B31]\]. Our screen identified a fly neuroligin family member, and mammalian neuroligins were recently implicated in postsynaptic development \[[@B32]\]. One of the largest groups of genes identified in our screen encodes proteins that comprise or regulate the actin and/or microtubule cytoskeletons. Glutamate receptors, like many other proteins, are thought to be transported along microtubules, and anchored to the synaptic actin cytoskeleton \[[@B33]-[@B35]\]. In support of this, we\'ve recently found that GluRIIA-containing receptors are specifically linked to the actin cytoskeleton via the 4.1 homolog coracle, which interacts directly with GluRIIA \[[@B20]\]. We\'ve also found that regulation of synaptic microtubules affects fly NMJ glutamate receptor cluster development (see below). The cytoskeleton might not only be important for receptor protein localization; trafficking and localization of synapse-specific mRNAs (see below) probably also relies on the cytoskeleton. Consistent with this, untranslated regions of GluRIIA appear to be required for proper synaptic receptor localization in the fly NMJ (Karr & Featherstone, unpublished). Other types of genes identified by the screen are consistent with what one might expect. For example, many of the genes identified in our screen are involved in gene expression or protein translation. Mutation of a transcription factor or component of the translation machinery would be expected to disrupt many downstream things, including production of the cellular machinery required for postsynaptic development. Since our screen specifically excluded mutants which did not develop to the later stages of embryogenesis, and which did not form neuromuscular junctions, our screen may have highlighted components of pathways specific for a subset of cell differentiation steps that includes glutamate receptor cluster formation. Some of this machinery is possibly localized to the synapse. *Drosophila*GluRIIA mRNA is localized to the NMJ and locally translated \[[@B36]\]. Mammalian glutamate receptors may also rely on local translation and editing for surface expression \[[@B37]-[@B39]\]. Preliminary results also suggest that *Drosophila*NMJ glutamate receptor cluster formation depends on a burst of receptor subunit transcription that follows contact between pre and postsynaptice cells (Karr & Featherstone, unpublished). Our screen also revealed some important surprises. For example, we identified insertions in two different putative mucin-encoding genes. Mammalian mucins are secreted glycoproteins that are widely implicated in tumor cell adhesion but have no previously identified role in the nervous system \[[@B40]\]. Are these false positives or important insights into postsynaptic development? Interestingly, the \'mucin\' genes identified in our screen are also similar to the *C. elegans*polycystin gene *lov-1*, which is localized to the ciliated sensory endings of dendrites required for male mating behavior, where it may be critical for regulating localization of other transmembrane proteins \[[@B41],[@B42]\]. Our screen also identified the exocyst protein sec-8. Sec-8 has recently been implicated in NMDA receptor trafficking but has not been shown to regulate non-NMDA receptor localization, and the mechanism by which it (and other sec proteins) functions remains unclear \[[@B43],[@B44]\]. We are particularly interested in validating and studying the most surprising candidates from out screen, and have done so for sec 8. This work (Liebl FLW, Chen K, Karr J, Sheng Q, Featherstone DE: **Altered Synapse Development in *Drosophila*Sec 8 Mutants**. M*anuscript in preparation*, and Liebl & Featherstone, unpublished) reveals that *Drosophila*sec 8 regulates the synaptic microtubule cytoskeleton to facilitate transmembrane protein localization (Thus the inclusion of sec 8 as a \'cytoskeletal regulator\' in Table [1](#T1){ref-type="table"}). Thus, even the surprises identified by our screen are, so far, apparently reasonable. But note that even validated candidates, such as sec 8, may not work directly or specifically on glutamate receptors. Approximately 12% of the genes identified in our screen encode proteins with unknown or novel function. A better understanding of these genes is important for understanding both those protein families and synaptogenesis -- assuming these proteins are really regulators of receptor cluster formation. Even with a 50% success rate in identifying genes involved in receptor cluster formation -- and the arguments and data above suggest the success rate was much higher -- it is clear that novel genes encode a large fraction of proteins required for glutamate receptor cluster development in the fly NMJ. Therefore, major unexplored areas apparently may exist with regard to postsynaptic development. Interestingly, our screen did not isolate any PDZ-domain or MAGUK proteins, which are widely regarded as essential trafficking and scaffold proteins at mammalian glutamatergic synapses \[[@B45]\]. However, this may not reflect a complete lack of importance for these proteins in fly NMJ glutamate receptor localization. *Drosophila*discs-large (DLG) is the sole fly representative of the mammalian DLG/SAP 97/SAP102/PSD-95 protein family; DLG is important for formation of fly NMJ glutamate receptor clusters that contain the subunit GluRIIB, but not those containing GluRIIA \[[@B11]\]. Because we screened only for alterations in GluRIIA immunoreactivity, we would not have isolated DLG mutants. However, our results do support the idea that PDZ proteins are not predominant components of the glutamate receptor localization/stabilization machinery in *Drosophila*NMJs. Many of the genes implicated by our screen probably do not work directly or specifically. Kinases only have to regulate something that in turn regulates receptor cluster formation. Transcription and translation factors probably control expression of multiple proteins required for PSD formation. Given the fact that many proteins definitively required for glutamate receptor cluster formation may not work directly, it is reasonable that a large percentage of the genome appears to be required. In other words, it is not difficult to identify proteins required for receptor cluster development, and we feel that the demonstration of such by itself does not give real insight into receptor cluster formation. An alternative approach, which we call \'functional category analysis\' and introduce here, is to focus on the *types*of genes identified rather than the identities of individual genes, in an effort to gain larger insights into the entire process. Subsequent work can then be directed at investigating the processes represented by these gene types, rather than validation of individual genes that may or may not work directly or be applicable to other synapses and organisms. For example, many microtubule and actin regulators were identified in our screen. The target for these proteins, obviously, is likely to be cytoskeletal proteins instead of receptors. But their identification tells which types of cytoskeleton might be important for glutamate receptor cluster formation. Using these clues, we subsequently determined that microtubules are important for fly glutamate receptor trafficking (Liebl FLW, Chen K, Karr J, Sheng Q, Featherstone DE: **Altered Synapse Development in *Drosophila*Sec 8 Mutants**. M*anuscript in preparation*, and Liebl & Featherstone, unpublished), and that A-type receptors are anchored via the 4.1 protein coracle to postsynaptic actin \[[@B20]\]. Conclusion ========== We identified 57 transposon mutants with qualitative differences in glutamate receptor expression and localization. Mutant gene identities need to be validated despite the fact that mutant genes are tagged. Focus on the types of genes identified (\'functional category analysis\') may provide more useful insight into the process of glutamate receptor cluster formation, compared to focus on individual genes. Our results suggest that glutamate receptor cluster formation involves cell adhesion and signaling, extensive and relatively specific regulation of gene expression and RNA regulation, the actin and microtubule cytoskeletons, and many novel/unexplored processes such as those involving mucin/polycystin-like proteins and proteins of unknown function. Methods ======= NMJ staining and microscopy was performed as previously described \[[@B9],[@B19]\]. Briefly, animals were dissected and fixed for 30--60 min in Bouin\'s fixative. Late stage embryos were dechorionated in bleach and then manually devitellinated and dissected. Mouse monoclonal anti-GluRIIA (\'8B4D2\' Iowa Developmental Studies Hybridoma Bank, Iowa City, IA) was used at 1:100. Fluorescently conjugated anti-HRP (Jackson Immunoresearch Labs, West Grove, PA) was used at 1:100. Goat anti-rabbit or goat anti-mouse fluorescent (FITC or TRITC) secondary antibodies (Jackson Immunoresearch Labs, West Grove, PA) were used at 1:400. Confocal images were obtained using an Olympus FV500 laser-scanning confocal microscope. Image analysis and quantification was performed using ImageJ software. Complementation analysis was performed by crossing GFP-balanced *P*-element mutant stocks to a balanced stock containing a deficiency that removes the insertion site. The F1 generation of each cross was examined for the presence or absence of adult flies carrying neither balancer chromosome. Thus, *P*-element insertion chromosomes were tested for their capacity to complement the viability of the lethal mutations in the gene carrying the *P*-element insertion. All *P*-element stocks were obtained from Bloomington Stock Center <flystocks.bio.indiana.edu>. \'Control\' genotypes in all experiments are *w*^*1118*^. Authors\' contributions ======================= FLWL performed most of the immunocytochemistry and microscopy. Analysis was jointly performed by FLWL and DEF. Acknowledgements ================ Thanks to Qi Sheng and Pei-San Ng for technical assistance. This work was supported by NIH/NINDS R01 NS045628 to D.F Figures and Tables ================== ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Example phenotypes identified in the screen.**A: NMJs on embryonic ventral longitudinal muscles 7, 6, and 13, visualized using anti-HRP antibodies (which stain all neuronal membrane) and anti-GluRIIA antibodies (which stain postsynaptic glutamate receptors). HRP immunoreactivity is red; GluRIIA immunoreactivity is green. In wildtype NMJs (A, left column), GluRIIA immunoreactivity appears as puncta, representing clusters of postsynaptic glutamate receptors. In homozygous *P{SUPor-P}KG00333*mutants (A, right column), GluRIIA immunoreactivity is eliminated. B: NMJs on third instar larval longitudinal muscles 7 and 6, visualized using anti-HRP and anti-GluRIIA antibodies. In wildtype NMJs (B. left column), GluRIIA immunoreactivity appears as large blobs, representing developmentally merged puncta, at sites of innervation. In homozygous *P{SUPor-P}Chro \[KG03258\]*mutants (B, middle column), synaptic GluRIIA immunoreactivity is dramatically decreased, although extrasynaptic receptors remain prominent. In contrast, synaptic GluRIIA immunoreactivity is dramatically increased in homozygous *P{SUPor-P}vri \[KG01220\]*mutants (B, right column). Scale bars: 10 micrometers. ::: ![](1471-2202-6-44-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Pie chart showing the different types of proteins encoded by the insertion loci identified in the screen.**The size of the pie wedge represents the proportion of genes of each type identified. Functional categories were based on previously published studies of the *Drosophila*gene and/or sequence similarity to functionally annotated mouse genes. Table 1 contains a complete list of the genes represented in this chart. ::: ![](1471-2202-6-44-2) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Genes affected in glutamate receptor expression/localization mutants. Complete list of the mutants identified in the screen. Table 1 lists all of the mutants identified in the screen, organized by putative function of the protein encoded by the gene in which the transposon is inserted. The qualitative glutamate receptor phenotype in the *Drosophila*mutants is indicated by an arrow pointing upwards (for an increase in immunocytochemically detectable GluRIIA) or downwards (for a decrease in immunocytochemically detectable GluRIIA). The third column lists the *Drosophila*gene that is mutated by the transposon insertion; superscript letters represent complementation test results: \'f\' = transposon insertion failed to complement a deficiency covering the region, and \'c\' = insertion complemented a deficiency. Note that the confirmed presence of a background lethal mutation (denoted by \'c\') does not indicate whether the NMJ phenotype is also complemented. The fourth column lists the mouse homolog, as determined by best BLAST match to an annotated gene in the *mus musculus*refseq database. ::: **Allele** **Phenotype** **Mutant gene** **Mouse protein** ----------------------------------------------------------------- --------------- ----------------- ------------------------------- **Extracellular matrix** *P{SUPor-P}CG18819 \[KG01657\]* ↓ *CG32814* mucin *P{SUPor-P}CG14713 \[KG05924\]* ↓ *CG14713* mucin **Cell adhesion** *P{SUPor-P}mys \[KG02930\]* ↓ *mys* beta integrin *P{SUPor-P}KG00212* ↑ *Nrt* neuroligin **Cytoskeleton, cytoskeletal regulators, and adaptor proteins** *P{SUPor-P}CG10540 \[KG02261\]* ↓ *CG10540*^*c*^ actin capping protein *P{SUPor-P}KG01009* ↓ *Tkr* actinfilin *P{SUPor-P}pnut \[KG00478\]* ↑ *pnut*^*f*^ septin *P{SUPor-P}didum \[KG04384\]* ↓ *didum*^*c*^ myosin V *P{SUPor-P}neb \[KG05913\]* ↓ *neb*^*f*^ kinesin *P{SUPor-P}KG05547* ↓ *Cdic* dynein intermediate chain *P{SUPor-P}CG2095 \[KG02723\]* ↓ *CG2095*^*f*^ sec 8 *P{SUPor-P}lap \[KG06751\]* ↓ *lap* synaptosomal protein 91 *P{SUPor-P}noodle \[KG03815\]* ↓ *CG3210* dynamin **Kinases & Phosphatases** *P{SUPor-P}KG06341* ↓ *KP78a*^*f*^ MAP/MT affin.-reg. kinase *P{SUPor-P}SNF4Aγ \[KG00325\]* ↓ *SNF4γ* AMP-activated kinase *P{SUPor-P}trbl \[KG02308\]* ↓ *trbl* AMP-activated kinase *P{SUPor-P}KG04591* ↑ *CG15072*^*c*^ SNF-like kinase *P{SUPor-P}KG02006* ↓ *Rheb* RAS-homolog *P{SUPor-P}polo*\[16-1\] ↓ *polo* polo-like kinase *P{SUPor-P}KG00853* ↓ *CG10082*^*c*^ inositol hexaphosphate kin. *P{SUPor-P}CG32666 \[KG03058\]* ↓ *CG32666* serine-threonine kinase *P{SUPor-P}KG00564* ↓ *ia2* rec. tyrosine phosphatase *P{SUPor-P}l(1)G0003 \[KG02485\]* ↓ *l(1)G003*^*f*^ Rab coupling protein **Ubiquitination** *P{SUPor-P}Cbl \[KG03080\]* ↓ *Cbl* E3 ubiquitin ligase *P{GT1}sina \[BG02648\]* ↓ *sina* sina (ubiquitin ligase) **Transporters** *P{SUPor-P}CG5802 \[KG01634\]* ↑ *CG5802*^*f*^ UDP-galactose translocator *P{SUPor-P}KG02272* ↓ *CG8029*^*f*^ lysosomal H+ ATPase *P{SUPor-P}Vha44 \[KG00915\]* ↓ *Vha44* vacuolar H+ ATPase **Unknown** *P{SUPor-P}oaf \[KG03408\]* ↑ *oaf*^*c*^ D130038B21 RIKEN cDNA *P{SUPor-P}KG03591* ↓ *CG32345* hyp. Protein XP\_355833 *P{SUPor-P}cmp44E \[KG03925\]* ↓ *cmp44E*^*c*^ *\'*similar to KIAA0953\' *P{SUPor-P}CG2185 \[KG02712\]* ↑ *CG2185* calcium binding P22 *P{SUPor-P}KG00333* ↓ *CG15358* cd209e antigen *P{SUPor-P}CG31694 \[KG04350\]* ↑ *CG31694*^*f*^ interferon development reg. *P{GT1}l(2)35Di \[BG02008\]* ↓ *l(2)35Di*^*c*^ \[none\] *P{SUPor-P}KG06339* ↓ *\[none\]* **Enzymes** *P{SUPor-P}CG4825 \[KG06018\]* ↓ *CG4825* phosphatidylserine synth. *P{SUPor-P}TppII \[KG03294\]* ↓ *TppII*^*c*^ tripeptidyl peptidase II **Gene expression & Translation** *P{SUPor-P}KG02920* ↓ *hsrω*^*f*^ ladybird homeobox-1-like *P{SUPor-P}KG02514* ↓ *hbn* aristaless-rel. homeobox *P{SUPor-P}KG03852* ↓ *Ntf-2* nuclear transport factor-2 *P{SUPor-P}Trn-SR \[KG04870\]* ↓ *CG2848* transportin 3 *P{GT1}CG10689 \[BG01776\]* ↓ *CG10689* DEAH box polypeptide *P{SUPor-P}CG11107 \[KG02727\]* ↓ *CG11107* DEAH box polypeptide *P{SUPor-P}pnt \[KG04968\]* ↓ *pnt* avian leukemia oncogene *P{SUPor-P}KG00625* ↓ *CG18591* snrpE *P{SUPor-P}RpL3 \[KG05440\]* ↓ *RpL3*^*f*^ ribosomal protein L3 *P{SUPor-P}KG03101* ↓ *Ef2b* translation elong. factor 2 *P{SUPor-P}crp \[KG00953\]* ↓ *crp* transcription factor AP-4 *P{SUPor-P}E2f \[KG03332\]* ↑ *E2f* E2f transcription factor *P{SUPor-P}armi \[KG04664\]* ↓ *armi*^*c*^ Moloney leukemia virus 10 *P{SUPor-P}vri \[KG01220\]* ↑ *vri* interleukin3 reg. nuc. factor *P{SUPor-P}Kr-h1 \[KG00354\]* ↓ *Kr-h* zinc finger protein *P{SUPor-P}Bgb \[KG03779\]* ↓ *Bgb* core binding factor beta *P{SUPor-P}KG06256* ↓ *Chro*^*c*^ Domino (histone deactyl.) *P{SUPor-P}Chro \[KG03258\]* ↓ *Chro* Domino (histone deactyl.) *P{SUPor-P}EcR \[KG04522\]* ↓ *EcR* nuclear receptor 1, group H :::

PubMed Central

2024-06-05T03:55:58.400207

2005-6-28

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173111/", "journal": "BMC Neurosci. 2005 Jun 28; 6:44", "authors": [ { "first": "Faith LW", "last": "Liebl" }, { "first": "David E", "last": "Featherstone" } ] }

PMC1173112

Background ========== There has been an apparent increase in the incidence of coeliac disease (coeliac disease) over the past 30 years \[[@B1]\]. For instance, the number of reported cases in the Netherlands increased from 0.18 per 1000 live births from 1975--1990, to 0.54 per 1000 live births 1993--1994 \[[@B1]\]. There is also increasing recognition that symptomatic coeliac disease may be the tip of the iceberg, with many more asymptomatic cases in the community \[[@B2]-[@B4]\]. It is unclear whether the changing prevalence of reported cases of coeliac disease is due to a real increase in the number of cases, enhanced awareness of disease or more reliable serological testing. Dietary changes, such as earlier introduction of gluten, have been suggested as contributing to the changing presentation of coeliac disease in some populations \[[@B5],[@B6]\]. Breast feeding, in particular breast feeding after the introduction of gluten, is protective against the development of coeliac disease \[[@B7]\]. There are few data on the clinical features, incidence and prevalence of coeliac disease in Australian populations. In a population based screening study from a rural community of Western Australia, 10/3001 were positive for anti-endomysial antibodies. This community was of predominantly Anglo-Celtic origin \[[@B4]\]. Australians come from diverse ethnic backgrounds, thus the prevalence reported in this community cannot be generalised to the Australian population. The primary aim of this study was to describe the clinical features of children diagnosed with coeliac disease at one tertiary paediatric centre in Australia. Our secondary aim was to determine the presentation patterns of coeliac disease in the local community at different ages. Methods ======= A retrospective cross sectional study was conducted at Sydney Children\'s Hospital, one of two tertiary referral paediatric hospitals in Sydney, Australia. The study population consisted of infants and children who had been referred to the Gastroenterology Department at Sydney Children\'s Hospital for investigation of possible coeliac disease between 1997 and 2002. Subjects were identified through records kept by the Departments of Nutrition and Dietetics, Gastroenterology, and Anatomical Pathology. Ethics approval was granted from the South East Sydney Area Health Service Research Ethics Committee-Eastern Section. Cases and clinical features --------------------------- Cases were defined as those with histological confirmation of coeliac disease \[[@B8]\]. A single pathologist reviewed all slides. Where the histological diagnosis of coeliac disease was uncertain, clinical and laboratory features, including response to gluten free diet, were included. The clinical details of each patient were extracted from the medical records retrospectively. Features of interest included presenting symptoms as well as the symptoms and signs elicited during consultation with a paediatric gastroenterologist. Weight and height measurements at diagnosis were documented. SD (standard deviation) scores were calculated using reference data \[[@B9]\]. The ethnic mix for the reference population of children admitted to Sydney Children\'s Hospital was described using child\'s country of birth. Laboratory investigations ------------------------- The results of haemoglobin, iron studies, serum IgA, anti-gliadin (AGA) and anti-endomysial antibodies (EMA) were obtained from each patient\'s medical records. AGA and EMA tests had been performed at SEALS (South Eastern Area Laboratory Service, Kogarah, NSW). The SEALS laboratories measure EMA using indirect immunoflurescence, anti-human FITC conjugate IgA against monkey distal oesophagus. The slides are prepared by MeDICa (Encinitas, CA 92024). A value of 10 is used for as positive cut off. AGA were measured by commercial immunoassay kits (Analytica Ltd, Castle Hill NSW Australia). The cut off values used for children were IgA\>25 and IgG \>46. Low haemoglobin values for age were defined as those less than the reference range for age used by the hospital laboratory (SEALS). The low end of the normal range for children 6 months to 2 years was 104 g/L; 2--4 years 107 g/L; 4--8 years 110 g/L; 8--12 y 113 g/L; \>12 y 130 g/L. Patients were considered iron deficient on the basis of a microcytic blood film, low serum iron, low serum ferritin, and elevated iron binding capacity. Statistical analysis -------------------- Data were analysed by simple descriptive analysis performed manually and using Excel (Microsoft Office 97). Fisher\'s Exact tests were used to compare independent data using STATA \[[@B10]\]. Where data was not available, summary statistics were performed using only the data available. Results ======= Subjects -------- A total of 119 patients were identified over the six year period. Forty-three patients were excluded as there was another medical reason to explain their histological features or reason for dietetic referral. The medical records were unable to be located for two patients, leaving 74 subjects with coeliac disease available for analysis. The male: female ratio of patients was 1:2. The age at diagnosis ranged from 11 months to 14 years with median of 5.5 years. Seven patients (9% of total group) were less than 2 years of age (Figure [1](#F1){ref-type="fig"}). The ethnic mix (based on country of birth) of all admissions to SCH consists of approximately 94% Australian, 2% Asian, 1% UK and Europe, 0.05% Africa, and 0.02% Pacific Islands (data compiled by MS using admission data for Sydney Children\'s Hospital for 2002). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Age at Presentation of Coeliac disease in the cohort of 74 subjects**. This figure illustrates the age distribution of the 74 subjects with coeliac disease in this study. Infants less than 2 years of age represented only 12% of the population. 27% of subjects were aged between 4 and 5 years, 36% were aged between 5 and 10 years, 26% were more than 10 years. ::: ![](1471-2431-5-11-1) ::: Clinical features ----------------- Sixty nine of the seventy four children had presented with symptoms (Table [1](#T1){ref-type="table"}). Full details of the presenting symptoms were available in 65 of the patients. Twenty nine of these children were aged less than five (preschool) and forty were aged five years or greater. The most common presenting symptoms in younger children (those \<5 years) were diarrhoea (59%), irritability (34%) and weight loss (38%). In older children (≥ 5 years), the most common presenting feature was abdominal pain (55%), followed by diarrhoea (26%). Twenty-three percent of the symptomatic children had \> 3 presenting symptoms. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Clinical features of 69 symptomatic children with Coeliac disease ::: **Age \< 5 years n = 29 (% of group)** **Age ≥ 5 years n = 40 (% of group)** --------------------------------------------- ---------------------------------------- --------------------------------------- Abdominal pain 6 (20) 25 (62.5)\* Diarrhoea 17 (59) 12 (30)\* Tiredness/lassitude 9 (31) 9 (22.5) Abdominal distension 16 (55) 9 (22.5)\* Constipation 4 (14) 5 (12.5) Irritability 10 (34) 6 (15)\* Mouth Ulcers 2 (7) 6 (15) Anaemia 5 (17) 10 (25) Weight loss or lack of expected weight gain 11 (35) 6 (15)\* Anorexia and vomiting 7 (24) 0\* (\*p \< 0.05) This table displays the number of subjects with each symptom. The numbers in brackets describe the percentage of the age group with that symptom. ::: Five patients had no gastrointestinal symptoms at diagnosis. Four of these patients had positive coeliac serology during routine complication screening for type 1 diabetes (T1DM). The fifth patient had coeliac serology performed as part of investigation into recurrent infections. Overall, ten children were noted to have other medical problems potentially related to the development of coeliac disease. Five patients had T1DM, two patients had Trisomy 21, two patients had juvenile rheumatoid arthritis, and one patient had Raynaud\'s phenomenon. Fourteen of the 79 (18%) patients had a family history of coeliac disease in a first degree relative. The mean weights and heights of the subjects with coeliac disease were not significantly different to the reference sex and age matched population (mean weight SD -0.45 ± 0.14; mean height SD -0.4 ± 0.17). However, 5% of the children with coeliac disease had a weight SD score of more than 2 SD below the mean and 12% of children had a height SD score of more than 2 SD below the mean. Laboratory data --------------- Seventeen of 64 (26%) patients with haemoglobin levels available had haemoglobin levels below the reference range for age (mean haemoglobin 118 g/L ± 0.2). Thirty two of 37 (86%) patients who had iron studies available were deemed iron deficient (low serum iron and ferritin). Of those who were biochemically iron deficient, 8 (25%) had normal haemoglobin values for their age. IgG AGA was elevated in 97% of cases. IgA AGA was elevated in 94% of cases. There were no patients with IgA deficiency. EMA were elevated in 97% of this group of children. There were two patients with positive EMA antibodies who had negative AGA, on the other hand there was one patient (aged 9 years) with negative EMA who had positive AGA. Discussion ========== This study describes the clinical features at presentation in a group of Australian children diagnosed with coeliac disease during a time of increasing awareness of this disease and widespread availability of reliable laboratory screening tests. The children included in this retrospective review presented with a wide variety of clinical features. The most common clinical features of coeliac disease in younger children were the so-called classical symptoms of diarrhoea, irritability and weight loss. However, older children more commonly presented with abdominal pain. There were few cases without gastrointestinal symptoms in our cohort. The clinical features of the children in this study were similar to the clinical features described from populations in New Zealand \[[@B11]\], South Yorkshire \[[@B12]\] and Sweden \[[@B13]\]. Although abdominal pain is a common symptom of older children with coeliac disease, there is no association between classical recurrent abdominal pain and coeliac disease \[[@B14]\]. Anaemia was identified in approximately one quarter of the cases. Furthermore, many children had laboratory evidence of iron deficiency but were not anaemic at the time of diagnosis. Unfortunately only 86% of subjects had a full blood count taken, and 50% had blood taken for iron studies. Thus the true prevalence of anaemia and iron deficiency is uncertain. Cross sectional studies have identified a high prevalence of coeliac disease in patients with iron or folate deficiency \[[@B15],[@B16]\]. The incidence of coeliac disease in patients in the community presenting only with iron or folate deficiency has been reported at 4.7% \[[@B15]\]. Anaemia was reported in only 5% of children with coeliac disease from the Netherlands \[[@B1]\]. In the United Kingdom, iron deficiency anaemia was reported in approximately 20% of adults with coeliac disease. Although the current data may have underestimated the true incidence of anaemia and iron deficiency, it is likely higher than that reported in these previous cohorts. Growth failure in height and weight was not as common in this cohort as was described in coeliac children from the Netherlands, where 49.7% of children had poor growth \[[@B1]\]. In our study, 5% of children had a weight more than 2 SD below the mean, and 12% of children had a height more than 2 SD below the mean. Growth failure is more commonly seen in children diagnosed at a younger age than in our study \[[@B17]\]. The discrepancy in reported growth failure may also be due to different definitions of growth failure. In addition, because this was a retrospective study, it was not possible to obtain adequate information about growth velocity or of the children\'s genetic growth potential. The children in our study were older at the time of diagnosis than previous surveys of children with coeliac disease. In a cohort of patients diagnosed between 1950 and 1969, 70 of 91 were less than 2 years at the time of diagnosis and only 3 were more than 6 years. In a smaller cohort of children diagnosed between 1972 and 1975, 32 of 42 were less than 2 years at the time of diagnosis and 6 were greater than 6 years \[[@B18]\]. In the Netherlands, the proportion of children with coeliac disease aged \< 2 yrs has remained at about 60% from 1975 to 1994 \[[@B1]\]. An epidemic in coeliac disease in Swedish Children in the 1980\'s paralleled the change in feeding practices at the time, in particular increased exposure to wheat, rye and barley and decreased duration of breast feeding \[[@B5]\]. A prospective study of coeliac disease in infants showed that a longer duration of breast feeding, smaller amounts of gluten in the diet in infants less than 12 months \[[@B17]\] and breast feeding after gluten is introduced to the diet reduces the risk of coeliac disease in infants less than 2 years \[[@B7]\]. Thus, the later age of acquiring coeliac disease in the Australian population may be due to the delayed introduction of solids in the diet of Australian children. The current recommendations in Australia are to introduce solids at 6 months (NHMRC Dietary Guidelines for Children and Adolescents in Australia 1993). In this study, there were relatively few cases of coeliac disease detected by the screening of at risk groups. This contrasts with the worldwide trend of an increased prevalence of coeliac disease detected by screening \[[@B2],[@B19]-[@B22]\]. A large multi-centered study of children and adults in the United States of America showed that the prevalence of celiac disease was 1:22 in first degree relatives, 1:56 in symptomatic individuals or a disorder known to be associated with celiac disease (type 1 diabetes, down syndrome, anemia, arthritis, osteoporosis, infertility and short stature), and 1:133 in the not-at-risk population \[[@B23]\]. In our study, the best screening test in children aged over 2 years with symptoms of, or who are at risk of, coeliac disease was IgA EMA. Other studies have shown that this test has a sensitivity of 95--98%, specificity of 94--95%, positive predictive value of 91--95% and negative predictive value of 96--98% \[[@B24]\]. For children less than 2 years of age, IgA AGA may be more reliable \[[@B24]\]. Increasingly over the last few years, reports suggest that tissue-transglutaminase (tTG) antibody may be even more reliable than EMA as a screening test \[[@B25],[@B26]\], and in addition is more reproducible and reliable. This test was not available during the period of this review. Although the utility of these newer tests is improved compared to AGA antibodies, they are not yet accepted as diagnostic tests. Any child with symptoms suspicious of coeliac disease, even if negative serology, warrants further review by a paediatric gastroenterologist, as a small bowel biopsy remains the gold standard test for diagnosis of coeliac disease. Conclusion ========== Coeliac disease in children does not always present with classical clinical features, especially in older children. The possibility of coeliac disease should be considered in any child presenting with diarrhoea, irritability and weight loss, and particularly in older children with abdominal pain. With the widespread availability of minimally invasive screening tests for coeliac disease, identifying cases in at risk groups by screening should also be considered. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= MS conceived the study, performed the data collection and statistical analysis and drafted the manuscript; KW for assistance in collecting the data and reviewing the manuscript; VT performed the histological examination of the biopsy specimens; AD and TB were involved in the study design, writing and reviewing of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2431/5/11/prepub> Acknowledgements ================ Dr Phillip Emder for his support and review of the manuscript.

PubMed Central

2024-06-05T03:55:58.403675

2005-5-21

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173112/", "journal": "BMC Pediatr. 2005 May 21; 5:11", "authors": [ { "first": "Monique L", "last": "Stone" }, { "first": "Timothy D", "last": "Bohane" }, { "first": "Kylie E", "last": "Whitten" }, { "first": "Vivienne H", "last": "Tobias" }, { "first": "Andrew S", "last": "Day" } ] }

PMC1173113

Background ========== The concern for health disparities extends to residents of rural areas, who are recognized to face distance barriers when seeking to access health services. Concerns about disparities apply to mental as well as physical health. Studies addressing rural residence as a risk factor for poor mental health are important to verify the existence of a disparity and to clarify its nature. Singh and Siahpush studied rural residence as a risk factor for suicide. They reported that suicide risk increases for males with increasing levels of rurality and that the rural-urban differentials are increasing over time. Rural suicide rates also were higher for women but the rural-urban differential was found to be decreasing over time \[[@B1]\]. Recently, national experts have emphasized the recognition of mental health as a public health problem \[[@B2]-[@B4]\]. Suicide, though important, may not be the most important focus for a public health approach to mental health. Community health programs typically are targeted at moderately healthy populations, not the seriously ill, because the goal is prevention so as to avoid the need for treatment. Less serious mental health concerns affect a large fraction of the population. The more common mental health problems often are preventable and may be amenable to health promotion programs involving self-help strategies. Population surveys such as the one reported here naturally are more relevant to the less severe, more common mental health issues. Rural health advocates frequently make the case that rural residents have worse health than residents of urban areas, implying that residing in a rural area is a risk factor for poor health \[[@B5]\]. However, it is also possible that the causes of poor health are the same regardless of where one lives, and thus programs should be targeted not at rural locations but at populations that are afflicted with the genuine risk factors (e.g., poverty, low educational attainment, and age). The purpose of the study reported here was to investigate the relationship between rural residence and frequent mental distress, which can be described as poor self-rated mental health. This study differs from some earlier studies in that it includes a variable representing rural or urban residence in an analysis of data drawn from the Behavioral Risk Factor Surveillance System (BRFSS). The BRFSS does not ordinarily contain a variable representing rural or urban residence of the respondent. The Texas Department of State Health Services recently created such a variable, permitting comparisons that previously were not possible. Methods ======= Data were obtained from the 2003 Texas Behavioral Risk Factor Surveillance (BRFSS) survey. The Texas Behavioral Risk Factor Surveillance System (BRFSS), sponsored by the Texas Department of State Health Services in partnership with the Centers for Disease Control and Prevention (CDC), is an ongoing random-digit dialed telephone survey that collects information from the non-institutionalized, civilian adult population. BRFSS surveys adhere to the highest scientific standards for telephone researchand pose questions relating to health status, personal health habits, and use of preventive health services. The 2003 survey data were collected over a 12-month period frominterviews with 6,035 Texas residents. The overall cooperation rate, or proportion of individuals actually contacted on the phone who completed the survey, was 70%. The more conservative response rate favored by the Council of American Survey Research Organizations (CASRO), which adjusts for the estimated number of eligible respondents in the samplewho could not be contacted due to technical or other barriers, was 41%. Missing values were excluded from the analyses if they constituted less than 5% of cases for an independent variable. Missing dummy variables were created for independent variables for which 5% or more of the cases had missing values (household income and body mass index). The data set contained 5,757 observations. Measures -------- The dependent variable was frequent mental distress, which was assessed with the following question: \"Now, thinking about your mental health, which includes stress, depression and problems with emotions, for how many days during the past 30 days was your mental health not good?\" Frequent mental distress was defined as having 14 or more days of \"not good\" mental health. The primary independent variable of interest was whether the individual lived in a metropolitan, suburban, or non-metropolitan/rural area. Many prior studies have categorized the area of residence as metropolitan vs. non-metropolitan. However, a dichotomous classification may mask important differences between residents of highly urban and suburban areas. Thus, we chose a more detailed classification system of metropolitan-central city, metropolitan-suburban, and non-metropolitan areas based on respondents\' reports of their county of residence. Metropolitan-central city was defined as an urban core county containing 50,000 or more persons. Metropolitan-suburban was defined as a metropolitan county adjacent to a core metropolitan-central city area. All other counties were considered non-metropolitan or rural. Additional independent factors included demographics and body mass index. Demographic variables were age category (18--24, 25--34, 35--44, 45--54, 55--64, and 65+), race/ethnicity (non-Hispanic white, Hispanic, black, and other), gender, marital status (single/never married/divorced/widowed vs married/member of a couple), educational status (less than a high school degree, a high school degree, some college, and college graduate), and household income (\<\$25,000, \$25,000 to \$75,000, ≥\$75,000, and income missing). Body mass index (BMI) categories were defined as obese (BMI ≥ 30), overweight (BMI 25--29.99), normal weight (BMI \<25 and ≥ 18.5), underweight (BMI \< 18.5), and a BMI missing category. Analyses -------- The percentage of respondents with frequent mental distress (FMD) was calculated by urban/rural residence and all other independent variables. Multivariate logistic regression analyses were conducted to estimate adjusted odds ratios (AORs) for selected risk factors. Data were weighted to reflect the demographic distributions of the sample. Because we were interested in testing whether the predictors of FMD differ between residents of metropolitan-central city, metropolitan-suburban, and non-metropolitan areas, we then stratified the analyses by metropolitan status. STATA 8.0 was used to conduct descriptive and multivariate analyses to account for the complex sampling scheme and population weights \[[@B6]\]. Results ======= Frequent mental distress was reported by 9.79% of the respondents. The first column of Table [1](#T1){ref-type="table"} describes respondents\' demographic, social, economic, and body mass characteristics. Approximately 62% and 21% of respondents lived in metropolitan-central city and metropolitan-suburban areas, respectively, whereas about 17% resided in non-metropolitan counties. The second column describes the percentage of respondents with frequent mental distress by each selected risk or independent variable. Residents of metropolitan-suburban areas had the highest rate of FMD (11.92%). Those residing in metropolitan-central city and non-metropolitan areas had similar rates of FMD (9.34 and 8.77%). After adjustment for other variables, persons in metropolitan-suburban areas had a significantly higher odds of FMD compared to metropolitan-central city residents (AOR = 1.37). Non-metropolitan residents did not have a significantly different adjusted odds of FMD as compared to individuals residing in metropolitan-central city areas. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Prevalence of frequent mental distress (FMD) and adjusted odds ratios (AOR) for selected risk factors, State of Texas, 2003, (N=5,757) ::: Risk variable Overall % \% with FMD (95% CI) AOR\* (95% CI) *N*= 5,757 *P*value ---------------------------- ----------- ---------------------- --------------------------- ---------- Residence Metropolitan-central city 61.50 9.34 (8.24--10.44) Reference Metropolitan-suburb 21.23 11.92 (9.77--14.07) 1.37 (1.06--1.77) 0.016 Non-metropolitan 17.27 8.77 (6.59--10.96) 0.86 (0.63--1.18) 0.350 Age group 18--24 14.66 11.92 (8.80--15.05) Reference 25--34 20.23 6.79 (5.04--8.54) 0.92 (0.62--1.34) 0.635 35--44 21.09 9.26 (6.91--11.60) 1.17 (0.78--1.69) 0.488 45--54 18.08 11.40 (9.32--13.49) 1.26 (0.86--1.85) 0.227 55--64 11.88 10.39 (8.29--12.50) 0.91 (0.59--1.41) 0.675 65+ 14.07 8.57 (6.84--10.32) 0.51 (0.33--0.78) 0.002 Race/ethnicity non-Hispanic white 58.94 9.54 (8.61--11.68) Reference Hispanic 28.91 9.22 (7.51--10.95) 0.56 (0.41--0.75) \<0.001 Non-Hispanic Black 8.20 13.58 (9.47--17.68) 0.90 (0.62--1.31) 0.587 Other 3.95 9.72 (5.20--14.24) 1.15 (0.67--0.98) 0.607 Sex Female 50.91 12.06 (10.80--13.31) Reference Male 49.09 7.44 (6.16--8.73) 0.60 (0.48--0.76) \<0.001 Marital status Single/never married 38.65 13.28 (11.58--14.98) Reference Married 61.35 7.61 (6.61--8.61) 0.66 (0.54--0.82) \<0.001 Educational status \< high school degree 17.52 13.61 (10.96--16.26) Reference high school degree 27.44 12.84 (10.77--14.91) 0.80 (0.59--1.10) 0.172 some college 26.51 8.72 (7.20--10.24) 0.53 (0.37--0.74) \<0.001 College graduate 28.53 5.50 (4.41--6.60) 0.38 (0.26--0.55) \<0.001 Household income \<\$25,000 30.01 15.79 (13.68--17.90) Reference \$25,000 to \$75,000, 39.27 7.50 (6.28--8.72) 0.52 (0.40--0.67) \<0.001 ≥\$75,000 18.85 6.40 (4.75--8.06) 0.49 (0.33--0.71) \<0.001 Missing 11.87 7.58 (5.41--9.75) 0.50 (0.35--0.71) \<0.001 Body mass index Obese 23.02 12.66 (10.58--14.75) 1.56 (1.19--2.05) 0.001 Overweight 34.61 8.76 (7.31--10.21) 1.22 (0.93--1.60) 0.159 Normal 34.07 8.18 (6.75--9.61) Reference Underweight 2.08 24.38 (15.00--33.76) 3.15 (1.82--5.45) \<0.001 BMI missing 6.22 8.85 (5.42--12.27) 0.88 (0.53--1.44) 0.602 CI = confidence interval. \*Adjusted for all other variables in Table 1. *N*= 5,757 prior to weighting. ::: Age differences in FMD were revealed as those 65 years and older had a significantly lower adjusted odds of FMD than individuals between 18 and 24 years of age (AOR = 0.51). In regard to race/ethnicity, a higher percentage of Blacks/African Americans (non-Hispanic) reported FMD than non-Hispanic whites, Hispanics, or other non-Hispanic racial groups. However, when adjusting for other factors, only Hispanics\' odds of FMD differed from non-Hispanic whites\' (AOR = 0.56). Males exhibited a significantly lower adjusted odds of FMD (AOR = 0.60) as compared to females. Married or living-together persons had a significantly lower adjusted odds of FMD than single persons (AOR = 0.66). Educational attainment and household income, two indicators of socioeconomic status, were also associated with FMD. Compared with those with less than a high school degree, individuals with some college education as well as those who were college graduates had lower adjusted odds of FMD (AOR = 0.53 and 0.38, respectively). Similarly, individuals who had higher household incomes experienced a lower adjusted odds of FMD. Body mass index categories of obesity and underweight were associated with substantially higher odds (AORs = 1.56 and 3.15, respectively) of FMD as compared to normal weight persons. We also performed analyses of the predictors of FMD by metropolitan-central city, metropolitan-suburban, and non-metropolitan status (Table [2](#T2){ref-type="table"}). The findings for each model were similar with a few notable exceptions. Among those residing in metropolitan-central city areas, Hispanics had a significantly lower odds (AOR = 0.53, 95% CI = 0.37--0.77) of FMD than non-Hispanic whites. However, among residents of metropolitan-suburban and non-metropolitan areas, the odds of FMD did not differ between Hispanics and non-Hispanic whites. Similarly, male gender was significant in the model for metropolitan-central city residents (AOR = 0.50, 95% CI = 0.37--0.68) but insignificant in the remaining two models. Married or living-together persons were less likely to have FMD among metro-central city respondents, but marital status was not significant elsewhere in the state. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Adjusted odds ratios (AOR) of frequent mental distress (FMD), stratified by metropolitan status ::: Risk variable Metropolitan-Central City *N*= 3,505 *P*value Metropolitan-Suburban *N*= 1,251 *P*value Non-Metropolitan *N*= 1,001 *P*value ------------------------ -------------------------------------- ---------- ---------------------------------- ---------- ----------------------------- ---------- AOR \* (95% CI) AOR \* (95% CI) AOR \* (95% CI) Age group 18--24 Reference Reference Reference 25--34 1.25 (0.77--2.03) 0.366 0.40 (0.19--0.85) 0.016 0.94 (0.32--2.72) 0.904 35--44 1.40 (0.85--2.30) 0.186 0.63 (0.28--1.40) 0.256 1.73 (0.64--4.67) 0.283 45--54 1.42 (0.87--2.30) 0.157 0.96 (0.46--1.98) 0.904 1.49 (0.50--4.46) 0.476 55--64 1.16 (0.65--2.09) 0.614 0.59 (0.25--1.37) 0.218 0.93 (0.31--2.77) 0.892 65+ 0.60 (0.33--1.08) 0.091 0.26 (0.11--0.59) 0.001 0.77 (0.29--2.02) 0.586 Race/ethnicity Non-Hispanic white Reference Reference Reference Hispanic 0.53 (0.37--0.77) 0.001 0.49 (0.23--1.03) 0.058 0.88 (0.41--1.88) 0.738 Non-Hispanic Black 0.90 (0.59--1.39) 0.648 0.88 (0.31--2.46) 0.804 0.90 (0.33--2.50) 0.845 Other 1.23 (0.62--2.43) 0.547 0.81 (0.25--2.63) 0.725 1.57 (0.37--6.65) 0.540 Sex Female Reference Reference Reference Male 0.50 (0.37--0.68) \<0.001 0.73 (0.47--1.14) 0.171 0.77 (0.44--1.36) 0.368 Marital status Single/never married Reference Reference Reference Married 0.56 (0.43--0.74) \<0.001 0.81 (0.50--1.30) 0.384 0.98 (0.56--1.70) 0.931 Educational status \< high school degree Reference Reference Reference high school degree 0.75 (0.51--1.12) 0.165 0.85 (0.44--1.61) 0.623 0.90 (0.42--1.96) 0.800 some college 0.59 (0.38--0.92) 0.018 0.36 (0.17--0.77) 0.008 0.71 (0.31--1.62) 0.420 College graduate 0.42 (0.26--0.67) \<0.001 0.41 (0.20--0.83) 0.013 0.15 (0.05--0.50) 0.002 Household income \<\$25,000 Reference Reference Reference \$25,000--\$75,000 0.53 (0.38--0.74) \<0.001 0.56 (0.32--0.97) 0.040 0.34 (0.17--0.71) 0.004 ≥\$75,000 0.51 (0.33--0.80) 0.003 0.36 (0.16--0.82) 0.016 0.94 (0.38--2.42) 0.905 Income missing 0.45 (0.28--0.71) 0.001 0.62 (0.27--1.39) 0.245 0.41 (0.18--0.96) 0.040 Body mass index Obese 1.96 (1.37--2.80) \<0.001 1.04 (0.61--1.79) 0.881 1.28 (0.58--2.87) 0.542 Overweight 1.19 (0.82--1.72) 0.357 1.16 (0.70--1.92) 0.562 1.69 (0.78--3.63) 0.181 Normal Reference Reference Reference Underweight 2.98 (1.54--5.78) 0.001 1.26 (0.37--4.29) 0.710 11.94 (3.66--38.97) \<0.001 BMI missing 0.22 (0.67--2.23) 0.513 0.35 (0.11--1.16) 0.087 0.74 (0.19--2.83) 0.656 CI = confidence interval. \* Adjusted for all other variables in Table 1. ::: Discussion ========== The theory that residing in a rural location rather than an urban area is a risk factor for frequent mental distress is not supported by our data. Instead, persons living in suburbia were found to be at risk for FMD after adjusting for other risk factors. We suspect that some stressors may be at work in the suburbs that are not found in rural areas or central cities. This warrants further investigation. In our data, lifestyle variables (obesity and underweight), Hispanic ethnicity, female gender, being unmarried, and having less education are more important risk factors for FMD than rural residence. Because obesity and underweight are related to frequent mental distress, public health programs designed to achieve normal body weight might improve mental health \[[@B7]\]. The potential for population-level health promotion strategies to reduce the prevalence of mental health problems is important, since the medical care system has been chastised for its insensitivity to underlying mental health problems among patients who present with physical symptoms \[[@B8]-[@B11]\]. Efforts directed at improving the quality of mental health services delivered in primary care settings have had mixed results \[[@B9]\]. Clearly, this is another case where population-based preventive strategies by public health agencies might be more effective than individualized medical treatment of mental health problems, at least for less severe or incipient cases \[[@B7],[@B11]\]. The widespread and increasing \[[@B12]-[@B14]\] prevalence of frequent mental distress in the population (8.6 percent of adults nationally a decade ago and ten percent in 2001 \[[@B13]\]) necessitates a public health response. In addition, the psychological consequences of any future terrorist attacks or natural disasters will require population-based responses that can reach large numbers of people in a short time \[[@B15]\]. Conclusion ========== This study examined how rural residence was related to frequent mental distress in the general population in a single state. Our findings are a direct result of the modeling strategy employed. Accordingly, replication of these results by other investigators is important. Limitations of the study include a modest participation rate, the cross-sectional design, and the loss of some cases due to missing data. The cross-sectional nature of the study makes causal conclusions impossible. Furthermore, the use of a self-reported single item to measure frequent mental distress does not equate to providing diagnostic information. Another important limitation is that most seriously mentally ill persons are likely to be missed by a survey such as this because they at increased risk of being homeless or otherwise unavailable for telephone interviews. Nevertheless, the single item has been accepted by public health researchers for use in population surveys \[[@B11]-[@B14],[@B16]-[@B18]\]. Health status in rural populations is believed to be lower than in urban areas, partly due to higher rates of poverty \[[@B5]\]. This assertion specifically relates to rates of chronic disease, infant mortality, injury rates, trauma mortality, and overall mortality. However, most government sponsored surveys do not include a variable that reflects urban or rural residence. Our study is the first to use a Behavioral Risk Factor Surveillance System (BRFSS) survey with a variable reflecting rural residence. Our results do not support the concern about rural residence being a risk factor for poor mental health. The risk factors for poor mental health are found in urban as well as rural areas, although the delivery of services may be more difficult in rural areas. When Singh and Siahpush reported that suicide rates were higher in rural areas, they attributed this finding partly to social isolation \[[@B1]\]. If this theory was correct, then we would expect frequent mental distress also to be more prevalent among rural respondents in our data. However, we did not find rural respondents to be at greater risk of frequent mental distress than urban respondents. We question the generalization that rural people experience greater levels of social isolation, since rural communities sometimes may be more integrated and supportive than urban neighborhoods. Community health programs that are supported by strong evidence will, for the most part, focus on three strategies: policy changes, mass communication, and targeted health education \[[@B19]\]. Applied to risk factors for frequent mental distress, examples of policy changes might be zoning practices that promote physical activity, labeling of unhealthy menu items in restaurants, and increases in liquor and cigarette taxes. Media campaigns could be employed to raise awareness about anxiety, depression, stress, and healthy coping strategies. The third approach to community mental health is to target health education programs at persons who are at high risk. In Texas, this would be the elderly, the poor, and persons who are undereducated. Since none of these population-based strategies involve personal health services or require substantial investments in physical plant, they could each be carried out without regard to whether the targeted populations reside in rural or urban areas. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= TFB carried out the statistical analysis, wrote the methods section, and wrote the results section. JB wrote the section on sampling and the BRFSS in Texas. JER conceived of the study, planned the design, and wrote the background, discussion and conclusions. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/46/prepub>

PubMed Central

2024-06-05T03:55:58.405448

2005-5-16

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173113/", "journal": "BMC Public Health. 2005 May 16; 5:46", "authors": [ { "first": "James E", "last": "Rohrer" }, { "first": "Tyrone F", "last": "Borders" }, { "first": "Jimmy", "last": "Blanton" } ] }

PMC1173114

Background ========== Dietary supplements are defined in the United States Dietary Supplement Health and Education Act of 1994 (United States DSHEA) as any product (other than tobacco) intended to supplement the diet that bears or contains one or more of the following ingredients: a vitamin, a mineral, an herb or other botanical, an amino acid, a supplement used by man to supplement the diet by increasing the total dietary intake, or a concentrate, metabolite, constituent, extract, or combination of any ingredient described above \[[@B1]\]. Hankin \[[@B2]\] and Radimer *et al.*\[[@B3]\] used the term \'nonvitamin, nonmineral (NVNM) supplements\' to differentiate a new class of dietary supplements. NVNM supplements include products such as chrondroitin, sulfate, kava kava, ginseng, echinacea, ginkgo biloba, garlic herbals, botanicals, protein and amino acids, as well as Brewer\'s yeast and shark cartilage \[[@B3]-[@B5]\]. Although use of NVNM supplements is increasing in popularity, patterns of use for these supplements are not well known, and what is available is not always consistent, clear, or easily accessible \[[@B3]\]. People have reported a variety of reasons for taking dietary supplements, including decreasing their susceptibility to health problems such as stress, colds, heart attacks, osteoporosis, neural tube defects, dental caries and cancer, as well as to increase energy \[[@B6],[@B7]\]. The supplements typically can be considered as falling into three distinct categories: supplements that are believed to add nutrients to a system that owing to inadequate dietary practices might otherwise be lacking \[[@B8]\]; supplements that allege rapid weight loss and maintenance of the loss \[[@B9]\], or conversely, supplements that are believed to result in weight gain or muscle development \[[@B10]\]. Furthermore, the rise in popularity of freely available substances purported to enhance performance, such as androsterone (a steroid precursor), and creatine (an amino acid derivative), has also been documented \[[@B11],[@B12]\]. To date, however, the taking of nutritional preparations has not consistently indicated a competitive advantage other than a possible placebo effect, or one resulting from the treatment of concurrent nutritional deficiencies \[[@B13]\]. Consequently, according to the United States Commission on Dietary Supplement Labels, it is important for health and nutrition professionals to become more knowledgeable about all types of dietary supplements in order to help consumers make appropriate choices \[[@B14]\]. Variables that have been found to have a relationship with supplement use are race or ethnicity, age, education, income and lifestyle variables such as drinking, smoking, and exercising \[[@B15],[@B16]\]. There is documented evidence that the use of NVNM supplements in western society is high, and its use is increasing worldwide \[[@B17]-[@B20]\]. In some countries, such as Germany, many botanical products are classified as drugs, and thus are subject to quality, safety, and efficacy regulations \[[@B21]\]. In the United States, although it is acknowledged that some dietary supplements, particularly botanicals, are used to prevent and treat diseases, DSHEA distinguishes these products from drugs and food additives and thus they are not subject to the same regulations \[[@B1],[@B4],[@B5]\]. Use of NVNM supplements increased substantially with the passage of the DSHEA \[[@B1]\], which gave manufacturers greater freedom to market more products as dietary supplements and to provide information about their purported benefits in package labeling and advertising. Although the United States Food and Drug Administration (FDA) regulates additives and drugs, premarket review of dietary supplements is minimal \[[@B22]\]. The use of NVNM supplements has increased recently in many countries, a fact also reflected in our country. However, while the worldwide rate of consumption of these supplements is near to 50% \[[@B17]-[@B19]\], it is still rather low in developing countries. This may be put down to indications that those with higher education and income levels seem to be comparatively more likely to use supplements \[[@B15]\]. In Turkey, there is no regulation of NVNM supplements in that they are widely available over the counter from such places as pharmacies and markets without a prescription. There is no regulation or health policy on the use of NVNM supplements, and furthermore, data on the prevalence and use of NVNM supplements are limited or non-existent \[[@B23]\]. Finally, there is a need to understand the frequency with which people use supplements and what variables are associated with particular frequencies of use. In general, although we consider Turkish NVNM supplement consumption to fit into the prevalence patterns for developing countries, to our knowledge, no data is available either concerning use in Turkish students or the population as a whole. For this reason, the objective of this survey was to quantify the prevalence of NVNMS usage among university students; to identify supplements consumed and rationale for usage; to identify sources of supplement information; and to relate usage to selected demographic characteristics. Methods ======= Sampling -------- The survey was conducted between September 2004 and December 2004 at Osmangazi University, a public university located in the west of Turkey. All students from the schools of medicine, engineering and architecture, science and literature, economics, education, and the college of health services were invited to participate. Participation was voluntary and anonymous, and the Director of the Institution approved the survey. Subjects -------- Subjects were randomly selected with a stratified sampling method. Osmangazi University is an urban, mid-sized university with six schools, all of which were included in the survey. At least one class from the first, second, third and fourth years in each department were randomly selected to participate in the survey. The sample size for the survey was determined by multiplying at least 10% by the number of the students: the total numbers of the students in each department and in our survey, respectively, were 1097 and 203 for the school of medicine; 4212 and 561 for the school of engineering and architecture; 3556 and 403 for the school of science and literature; 1394 and 178 for the school of economics; 858 and 177 for the school of education; and 302 and 212 for the college of health services. Students were excluded from the survey due to: unwillingness to participate (n = 186), handing in incomplete questionnaires (n = 87), and non-attendance of class (n = 109). Thus, the response rate for sample respondents was 83% (1871/2253) of 2253 subjects, with ages ranging from 17 to 28, leaving a total of 1871 students (962 women, 909 men). Questionnaire and interview schedules ------------------------------------- Completion of the questionnaire was self-reporting on the part of the students. The dates on which the study would be conducted were determined in cooperation with class teachers in the schools concerned. The students completed questionnaires in the presence of a member of the research team. ### The first section of questionnaire This section identified consumption of NVNM supplements during the survey or the past year, previous use of dietary supplements, and the probability of students considering the use of NVNM supplements in the near future. If participants reported having used a NVNM supplements during the survey period or the year prior to the study, they were further questioned on a number of topics related to consumption: the number used; the name of the product(s), 73 of which were written on the questionnaire; frequency of use of each product; total product consumption; length of product use; the reasons for using NVNM supplements, revised from previous studies \[[@B24],[@B25]\]; and general demographic information such as age, sex, cigarette smoking, body height and weight, affiliation to which school of the university, class, marital status, parents\' education level, and parents\' total income level. ### The second section of questionnaire This section included the items related to nutritional beliefs about NVNM supplements. Subjects\' views about NVNM supplement use and the health benefits associated with it were used to assess the motivational factors related to supplement use in this population. Items were based on those of previous studies \[[@B24]-[@B28]\], and also contained the most frequently consumed food items in the Turkish population. Twenty two opinions were expressed, consisting of such items as \'vitamins and minerals can provide energy\', \'maintain health\', increase longevity\', \'reduce stress\' \'prevent colds\', \'improve mental function\', \'reduce the risk of chronic diseases\', or \'aid in recovery from fatigue\', all of which were assessed to be commonly-held assumptions connected with use of the supplements. For this section, a five-point Likert scale was used, and responses were scored from one (disagree very strongly) to five (agree very strongly). Thus, the higher score showed a stronger belief on the students\' part that NVNM supplements could provide specific health benefits. Applying it to a sample of 59 randomly selected students from same schools assessed the reproductivity of the questionnaire. The students who participated in the test/retest procedure were not included in the final analyses. Statistical analysis -------------------- Frequency, mean, and standard deviation (SD) were calculated, and the *t*test compared scores on nutritional beliefs between users and non-users, or women and men. The chi-square test (x^2^) examined the relation between general characteristics and NVNM supplements use. The percentages of the reasons cited for using NVNM supplements arranged according to sex, were calculated by first extracting out of the total figure male and female students\' numbers for NVNM supplements usage, and the significance between men and women was evaluated using cross tables. The reliability of the internal consistency of data concerning nutritional beliefs collected from the pretest was calculated by Cronbach\'s alpha and was found to be 0.96, indicating a high internal consistency across nutritional-belief statements. Consequently, belief statement use was used for the final survey. Statistical analyses were performed using SPSS for Windows (Version 10.0, SPSS Inc., Chicago, Illinois). The standard used for statistical significance was p ≤ 0.05. The data were analyzed using descriptive statistics including means, standard deviations, and frequency distributions. Results ======= Of the participants, 909 (48.6%) were men and 962 (51.4%) women. The average age of the participants was 19.9 years (range = 17 to 28 years). Nearly half of the students (49.9%) were between the ages of 19--20. Most (74.8%) were of normal weight. More than 50% of the students (53.9%) were from the schools of engineering and architecture, and science and literature. Most students (72.4%) were in their freshman or sophomore years. Almost 100% (97.9%) were single or divorced. The proportion of students whose mothers had an education level of secondary school and lower was 61.0%, with the figure of 35.0% reported for students\' fathers. Most students\' level of family income (47.0%) was low. The majority of the students reported living in villages or cities (85.1%), and 26.0% were current smokers. The percentage of those participating in sport was 15.4%, and 18.0% reported exercising more than 3 times per month. Most (74.3%) rated their health as \'good/excellent\'. There were significant differences between men and women according to all the descriptive information. The more detailed general characteristics of users and non-users of NVNM supplements are shown in Table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### General characteristics of users and non-users of nonvitamin/nonmineral supplements ::: --------------------------------------------------------------------------------------------------- **Non-users**\ **Users**\ **Total**\ **n(%)**\ **n(%)**\ **n(%)**\ **1563(83.5)** **308(16.5)** **1871(100.0)** --------------------------------------- ------------------------- --------------- ----------------- **Sex** x^2^= 5.594; p = 0.018 Men 758(83.4) 151(16.6) 909(48.6) Women 805(83.7) 157(16.3) 962(51.4) **Ages** x^2^= 35.732; p = 0.000 ≤ 18 358(86.7) 55(13.3) 413(22.1) 19--20 789(84.6) 144(15.4) 933(49.9) 21--22 292(85.1) 51(14.9) 343(18.3) ≥ 23 124(68.1) 58(31.9) 182(9.7) **BMI values** x^2^= 73.392; p = 0.000 Underweight (\<18.5) 199(67.7) 95(32.3) 294(15.7) Normal weight (≥ 18.5 -- ≤ 24.99) 1212(86.6) 187(13.4) 1399(74.8) Overweight (≥ 25 -- ≤ 29.9) 121(90.3) 13(9.7) 134(7.2) Obese (≥ 30 -- ≤ 39.9) 31(70.5) 13(29.5) 44(2.4) **Type of school** x^2^= 23.573; p = 0.000 Medicine 164(79.2) 43(20.8) 207(11.1) Engineering and architecture 536(88.4) 70(11.6) 606(32.4) Science and literature 332(82.4) 71(17.6) 403(21.5) Economics 204(79.7) 52(16.9) 256(13.7) Education 144(77.0) 43(23.0) 187(10.0) College of health services. 183(86.3) 29(13.7) 212(11.3) **Year in school** ns Freshman 570(85.6) 96(14.4) 666(35.6) Sophomore 556(80.8) 132(19.2) 688(36.8) Junior 168(86.2) 27(13.8) 195(10.4) Senior 269(83.5) 53(16.5) 322(17.2) **Marital status** ns Single 1520(83.7) 296(16.3) 1816(97.1) Married 33(82.5) 7(17.5) 40(2.1) Divorced/widow(er)/separated 10(66.7) 5(33.3) 15(0.8) **Mother\'s education level** x^2^= 11.940; p = 0.008 Primary school and lower 848(84.8) 152(15.2) 1000(53.4) Secondary school 106(74.1) 37(25.9) 143(7.6) High school 395(84.9) 70(15.1) 465(24.9) College or university 214(81.4) 49(18.6) 263(14.1) **Father\'s education level** x^2^= 7.807; p = 0.050 Primary school and lower 416(87.6) 59(12.4) 475(25.4) Secondary school 149(83.2) 30(16.8) 179(9.6) High school 537(81.7) 120(18.3) 657(35.1) College or university 461(82.3) 99(17.7) 560(29.9) **Family income** x^2^= 22.971; p = 0.000 Low 763(86.7) 117(13.3) 880(47.0) Moderate 566(83.4) 113(16.6) 679(36.3) High 234(75.0) 78(25.3) 312(16.7) **Most permanent place of residence** x^2^= 13.642; p = 0.001 Village/town 141(79.7) 36(20.3) 177(9.5) City 1207(85.3) 208(14.7) 1415(75.6) Metropolis 215(77.1) 64(22.9) 279(14.9) **Smoking status** x^2^= 15.435; p = 0.000 Nonsmoker 1093(85.7) 183(14.3) 1276(68.2) Current smoker 390(80.1) 97(19.9) 487(26.0) Former smoker 80(74.1) 26(25.9) 108(5.8) **Participation in sport** x^2^= 9.193; p = 0.010 No 724(82.5) 154(17.5) 878(46.9) Former 580(82.4) 124(17.6) 704(37.6) Yes 259(89.6) 30(10.4) 289(15.4) **Exercise** x^2^= 12.205; p = 0.002 Never 440(83.5) 87(16.5) 527(28.2) Fewer than 3 times per month 862(85.6) 145(14.4) 1007(53.8) More than 3 times per month 261(77.4) 76(22.6) 337(18.0) **Self-reported health** ns Fair/poor 400(83.2) 81(16.8) 481(25.7) Good-excellent 1163(83.7) 227(16.3) 1390(74.3) --------------------------------------------------------------------------------------------------- ::: The duration of supplement use was cited as \'between 1 and 6 months\' by 65.3%, followed by \'between 6 and 12 months\' and \'more than one year\' (23.1% and 11.7%, respectively). More than 50% of supplement users (53.6%) were taking just one supplement at the time of this survey, followed by those using 2 supplements (38.6%) and 3 or more (7.8%). The average number of supplement taken was 1.6 (standard deviation = 0.6). Most students (68.2%) rated the frequency of supplement use as \'once/twice a day\' (37.0%) or \'once every other day\' (31.2%), followed by \'once/twice a week\' (20.5%) and \'once/twice a month\' (11.4%). No difference was revealed between men and women by supplement use (Unshown data). Table [2](#T2){ref-type="table"} presents reasons cited for using NVNM supplements. The four most commonly given reasons for use were \'improvement of energy and vitality\' (78.6%), \'promotion of weight loss\' (71.1%), followed by \'enhancement of athletic performance\' (64.3%) and \'retardation the onset of aging\' (63.3%). Women were more likely to be using supplements to promote weight loss, burn-up fat, prevent colds, improve memory, and relieve stress; whereas men were more likely to use supplements to enhance athletic performance, retard the onset of aging, build muscle, and improve sexual function. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Reasons cited for using nonvitamin/nonmineral supplements by sex ::: --------------------------------------------------------------------------------------------------------------------------------------------------------------- **Men**\ **Women**\ **Total**\ **x^2^; *P*values** **n(%)**\ **n(%)**\ **n(%)**\ **151(49.0)** **157(51.0)** **308(100.0)** ---------------------------------------------------------------------------------------- --------------- --------------- ---------------- --------------------- Improve energy and vitality 128(49.6) 114(50.4) 242(78.6) ns Promote weight loss 92(42.1) 127(57.9) 219(71.1) 5.59; 0.018 Enhance athletic performance 117(59.1) 81(40.9) 198(64.3) 6.54; 0.011 Retard aging 124(63.6) 71(46.4) 195(63.3) 14.41; 0.000 Burn-up fat 55(35.1) 102(64.9) 157(50.9) 14.07; 0.000 Prevent(threat) colds(sore throat) 41(35.4) 75(64.6) 116(37.6) 9.97; 0.002 Promote skin(hair) health 41(41.8) 57(58.2) 98(31.8) ns Build muscle 71(74.3) 24(25.7) 95(30.8) 23.25; 0.000 Supplement inadequate diet due to nutrition deficiency 48(53.9) 41(46.1) 89(28.9) ns Improve memory 33(39.3) 51(60.7) 84(27.3) 3.86; 0.05 Relieve stress (improve mood) 29(37.7) 48(62.3) 77(25.0) 4.67; 0.03 Improve sexual function 55(79.7) 14(20.3) 69(22.4) 24.36; 0.000 Enhance sleep 22(52.4) 20(47.6) 42(13.6) ns Reduce dangers of cigarette smoking 18(52.9) 16(47.1) 34(11.1) ns Improve circulation 18(58.1) 13(41.9) 31(10.1) ns Gain weight 18(62.1) 11(37.9) 29 (9.4) ns Prevent illnesses such as cancer, osteoporosis, menopause, hypertension, kidney stones 9(42.9) 12(57.1) 21 (6.8) ns --------------------------------------------------------------------------------------------------------------------------------------------------------------- Note: Subjects could list more than one reason for nonvitamin, nonmineral supplement usage ::: In general, when the scores obtained from all the nutritional beliefs between men and women were taken into consideration, women agreed on the effects obtained from taking NVNM supplements more strongly than men in the categories: \'important in maintaining health\', \'stress reduction\', and \'enhancement sleep\', whereas men agreed more strongly on \'improvement sexual function\' than women (Unshown data). Twenty-six of the 308 NVNM users (26/308, 8.4%) reported having had an adverse reaction to an NVNM supplement. Of 26, the two most adverse effects were nausea and vomiting (61.5%, 30.7%, respectively), followed by gastrointestinal disturbances (15.4%), flashing (11.5%) and liver insufficiency (7.7%), Subjects could list more than one complication cited for NVNM usage (Unshown data). The most frequent used sources for obtaining information about supplements were television (76.3%), magazines/newspapers (41.5%) and internet websites (37.3%); followed by family (20.1%), friends (19.8%), school and teachers (16.5%), books (13.3), radio (5.8%), health food stores (0.5%) and agricultural engineers (0.3%). The rates of those citing dietetic professionals, and doctor and nurse as their sources were only 4.2% and 6.8%, respectively (Unshown data). Of 1871, 308 students (16.5%) reported using NVNM supplements during the past year, 23.0% students indicated that they were considering NVNM use in the near future. About 10% (9.7%) reported past usage. The data revealed no difference between men and women according to supplement usage. There were also no differences between men and women in the categories \'use of NVNM during the past year\', \'considering NVNM supplement use in the future\', and \'past usage of NVNM supplements\' (p \> 0.05 for each one) (Unshown data). Of the 1690 students reporting no use of NVNM supplements in the past, 44.3% indicated that they had not considered using supplements, 34.7% reported that they had only thought about using supplements, and 11.6% reported that they had been unaware of the presence of supplements. There was a 9.4% non-response rate (Unshown data). Table [3](#T3){ref-type="table"} summarizes information regarding the prevalence of NVNM supplement use. The three most commonly used supplements were reported as echinacea, ginseng, and gingko biloba (38.6%, 36.4%, and 32.8%, respectively). Since all of the 73 most commonly purchased supplements in Turkey and the world were listed on the questionnaire, the students reported using no supplements other than those listed. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Prevalence of use of nonvitamin, nonmineral supplements ::: ---------------------------------------------------- **Total**\ **n(%)**\ **308(100.0)** ----------------------------------- ---------------- Echinacea 119(38.6) Ginseng 112(36.4) Gingko biloba 101(32.8) Protein powder and or Amino acids 89(28.9) Fish oil 84(27.3) Bee pollen 78(25.3) Garlic 64(20.8) Green Tea 61(19.8) Bee pollen 57(18.5) St. John\'s wort 51(16.5) Ginger 47(15.2) Creatine 45(14.6) Aloe 27 (8.7) Lecithin 25 (8.1) Flax 21 (6.8) Chestnut seed 19 (6.2) Valerian 16 (5.2) Burdock root 15 (4.9) Coenzyme Q 15 (4.9) Goldenseal 12 (3.9) Cranberry 11 (3.6) Guarana 9 (2.9) Chamomile tea 8 (2.6) Sam-E 7 (2.3) Kava kava 7 (2.3) Melatonin 6 (1.9) Dong Quai 4 (1.3) Angelica 3 (0.9) Cayenne 2 (0.6) Borage 2 (0.6) Astragalus 1 (0.3) Evening primrose 1 (0.3) ---------------------------------------------------- The percentages total to more than 100% because some respondents reported taking more than one NVNM supplement ::: Table [4](#T4){ref-type="table"} shows that NVNM supplement use correlated to nutritional beliefs. Of all the beliefs, the links between \'supplement use and recovery from fatigue\' and \'supplement use and prevention/threat colds/sore throat\' were the ones with which users and non-users most agreed. The lower degree of perception of any \'relationship between supplement use and improved mental health\' was similar between users and non-users. In general, when all the nutritional beliefs between users and non-users were taken into consideration, users agreed on the effects obtained from taking NVNM supplements more strongly than non-users except for the categories of \'prevention/threat of colds/sore throat\', \'prevention of mental illnesses\', \'burning-up fat accumulating in the body\', \'promotion of more weight loss\', and \'reduction of the harmful effects of cigarette smoking\'. Of the 22 beliefs, the score from only one item, \'burning-up fat accumulating in the body\' was found to be lower in users than in the non-users. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Nutritional beliefs about nonvitamin/nonmineral supplements of supplement users and non-users\* ::: ----------------------------------------------------------------------------------------------------------- **Belief statement** **Non-users†**\ **Users†**\ ***P*values‡** **n(%)**\ **n(%)**\ **1563(83.5)** **308(16.5)** -------------------------------------------------------- ----------------- --------------- ---------------- Improve energy and vitality 2.8 ± 0.8 3.2 ± 0.8 0.000 Important in maintaining health 2.9 ± 0.8 3.3 ± 0.8 0.000 Provide living longer 2.7 ± 0.8 2.8 ± 0.7 0.000 Help recovery from fatigue 3.0 ± 0.8 3.3 ± 0.7 0.000 Reduce stress 2.8 ± 0.7 3.0 ± 0.8 0.012 Prevent/threat colds/sore throat 3.1 ± 0.8 3.2 ± 0.7 ns Prevent mental illnesses 2.7 ± 0.8 2.8 ± 0.8 ns Improve mental health 2.6 ± 0.8 2.7 ± 0.8 0.016 Reduces risk of chronic illnesses 2.8 ± 0.7 3.0 ± 0.8 0.000 Retard aging 2.7 ± 0.9 2.9 ± 0.8 0.000 Burn fat accumulating in the body 2.9 ± 0.8 2.8 ± 0.8 ns Promote weight loss more 2.8 ± 0.7 2.9 ± 0.8 ns Build muscle more 2.9 ± 0.8 3.1 ± 0.7 0.000 Enhance athletic performance 2.9 ± 0.8 3.1 ± 0.8 0.001 Enhance sleep 2.8 ± 0.8 3.0 ± 0.8 0.000 Improve sexual function 2.8 ± 0.8 3.0 ± 0.8 0.000 Improve circulation 2.9 ± 0.8 3.2 ± 0.7 0.000 Supplement inadequate diet due to nutrition deficiency 2.9 ± 0.8 3.2 ± 0.8 0.000 Improve memory 2.8 ± 0.8 3.0 ± 0.8 0.001 Promote skin/hair health 2.9 ± 0.8 3.1 ± 0.8 0.000 Gain weight 3.0 ± 0.8 3.1 ± 0.8 0.009 Reduce dangers of cigarette smoking 2.7 ± 0.9 2.8 ± 0.9 ns ----------------------------------------------------------------------------------------------------------- \*Values are shown as mean ± standard deviation † A five-point Likert scale was used; disagree very strongly (one point) to agree very strongly (five points) ‡ *t*test ::: Discussion ========== To our knowledge, this is the first survey on NVNM supplements use conducted in Turkey. This survey allowed the assessment of the prevalence of NVNM supplement use in a sample of university students from different departments. Our results indicate that about one in six students (16.5%) consumed one or more NVNM supplement during the past year. While our finding is compatible with some other study results, it is not with others: Radimer *et al*. \[[@B3]\] found that between 3.3 and 12% of respondents to their survey reported using NVNM supplements in 1994 and 1995, respectively. In a survey on a 1000 non-patient student population by Perkin et al. \[[@B25]\], a NVNM supplement usage rate of 26.3% was observed. Eisenberg et al. \[[@B29]\] reported this proportion as 12.1%. Furthermore, in a survey relevant to NVNM supplement use over a 12-month period by adult members of a large health maintenance organization, an estimated 32.7% of adults used at least one NVNM supplement \[[@B30]\]. Newberry *et al.*\[[@B31]\] reported that out of 272 college students canvassed, 48.5% had taken an NVNM supplement during the previous year and in a survey in the United States, it was found that 14.5% of the adults reported having used a NVNM supplement during the previous year \[[@B32]\]. The aforementioned studies show that the rate of NVNM supplement usage ranges from 3.3 to 48.5%. One explanation for these differences in reported usage rates could be the variation in how questions were asked regarding time frame and the types of supplements used. A further possibility could be relevant to individuals\' sociodemographic characteristics, such as race or ethnicity, age, education, income and lifestyle variables such as drinking, smoking, and exercising \[[@B15],[@B16]\]. As in other populations, echinacea, ginseng, gingko biloba, and protein powder and or amino acids were reported by this population to be the most popular supplements \[[@B3],[@B33]-[@B35]\]. Wingate \[[@B33]\] found that in 1997 the three most reportedly used supplements were garlic, ginseng, and ginkgo biloba. Similarly, the three most commonly consumed NVNM supplements reported in the Slone Survey \[[@B36]\], conducted from 1998 to 1999, were ginseng, ginkgo biloba extract, and allium sativum. This high use of supplements in Turkey and the rest of the world may be due to the fact that those products are prevalent in the market. In this survey, most students reported that they were healthy. This was reflected in the rationale offered by the students for supplement usage, such as increasing energy and enhancing athletic performance. In this survey, a positive relationship was seen to exist between exercising and NVNM supplement usage. Both behaviors may be perceived as congruent with disease prevention. NHANES III \[[@B3]\] also found a positive relationship between exercise and NVNM usage. The current survey found that former smokers used NVNM supplements most, which is in line with the NHANES III survey \[[@B3]\]. One explanation for this is that usage could be viewed as offsetting a negative behavior \[[@B25]\]. In this survey, 9.7% of the students reported past usage, 16.5% currently usage and 23.0% potential for use in the future. That these findings show increasing trends in the use of supplements with a corresponding increase in time may be due to media reports or advertisements in local newspapers about provision of energy and vitality, an increase in athletic performance, the retarding of aging, and the reduction of the risk of chronic diseases such as hypertension, colon cancer, breast cancer, kidney stones, and osteoporosis \[[@B37]-[@B39]\]. A further explanation may be that the use of NVNM supplements increased substantially with the passage of the United States Dietary Supplement Health and Education Act of 1994 (DSHEA), which gave manufacturers greater freedom to market more products through purported benefits reported in package labeling and adverting \[[@B1]\]. Twenty-six of the 308 NVNM users (26/308, 8.4%) reported having experienced an adverse reaction to an NVNM supplement. This rather high rate is in line with results published by Newberry *et al.*\[[@B31]\], which reported a reaction rate of 6.9%. This may be due to the fact that the majority of the students did not seek nutrition information from reliable sources such as doctors, nurses or dietetic professionals. In this survey, only 11.0% of supplement users cited doctors, nurses, and dietetic professionals as sources of information. However, it is important that students get nutrition information from reliable sources and receive proper nutrition education. To put this another way, that 76% of the respondents obtained information about supplementation from television may indicate that television has enormous implications for educational and public policy initiatives, and that television should be used to get more health information about supplementation. The nutritional beliefs that subjects questioned had concerning supplements were an important factor for their use of supplements. Supplement users agreed more strongly than non-users on the health benefits of supplement use for the majority of items. Others reported similar findings \[[@B9],[@B26]\]. In this survey, NVNM supplement use was higher in men than in women, in line with Radimer *et al.*\' survey \[[@B3]\]. However, these findings are not concordant with data reported from Malaysia (40), where the prevalence of supplement use was higher in women (14.9% vs. 9.7%, respectively). No apparent reasons were disclosed for such differences; it is possible that NVNM supplement use might be related to a desire to increase athletic performance in men \[[@B41]\], but further studies are needed to better assess this result. In this survey, nutritional knowledge average scores were low in both sexes (63.3 ± 12.4 for a maximum score of 110), probably because of the absence of nutritional education in primary, secondary, high schools or university in our country. NVNM supplement use by the students was different according to sex, age, Body Mass Index (BMI) values, school type, mother and fathers\' education levels, family income, most permanent place of residence, smoking status, participating in sport, and exercise. This finding shows that the rationale for supplements use changes according to subjects\' general characteristics. Of further interest is the trend for more female students in the current sample to indicate use of significantly more supplements than males. Parallel to our survey, this finding highlights the fact that female students are more likely to be dieting to lose weight, whereas male students are more likely to be controlling their diet to gain weight and build muscle, which is compatible with some other studies \[[@B42],[@B43]\]. We are well aware of the limitations of the present survey. Firstly, it was performed in a single institution, and therefore the sample may not be representative of all Turkish students. Another limitation on the data was the use of the self-reporting system employed in this survey. This may have given rise to bias. Conclusion ========== This descriptive analysis helps to provide a clearer picture of NVNM supplement use. Not only does it indicate that students are indeed using or considering using supplements as a viable means to achieve their short-term fitness and or appearance goals, but also which products are most likely to be used, and indeed which individuals are most in need of education and information concerning the use and abuse of certain dietary aids. Further research, therefore, needs to focus on how NVNM nutritional supplements use information can be applied to specific user groups. Monitoring the use of NVNM supplements within the community or university students should be of interest because these substances may affect health or interact with other medications \[[@B44]\]. The respondents in this survey reported 85 cases of interactions between St John\'s wort and prescription medications. Dietetic professionals and physicians need to consider evaluating NVNM supplement use when assessing overall nutrient and medication use by clients or students. In summary, our results in this sample of Turkish students indicate that the prevalence of NVNM supplement use is still relatively modest, and neither associated with a healthier lifestyle nor related to a better nutritional knowledge. The need for further research on the relationships between dietary supplements and health has also been highlighted in the 1997 report by the United States Commission on Dietary Supplement Labels \[[@B45]\]. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AU conceived of the study, performed its design and coordination, sequence alignment, collected the data, performed the statistical analyses and drafted the manuscript, SN and SO collected the data and entered the data to SPSS packet program, All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/47/prepub> Acknowledgements ================ We thank all the students from the University of Osmangazi, Eskisehir, who participated in this survey, and the lecturers of the schools, who helped in the collection of data.

PubMed Central

2024-06-05T03:55:58.409569

2005-5-16

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173114/", "journal": "BMC Public Health. 2005 May 16; 5:47", "authors": [ { "first": "Unal", "last": "Ayranci" }, { "first": "Nazan", "last": "Son" }, { "first": "Osman", "last": "Son" } ] }

PMC1173115

Background ========== Tobacco is one of the leading causes of the global burden of disease, \[[@B1]\] demanding effective intervention strategies \[[@B2]\]. Smoking cessation treatment is generally considered to be among the most cost-effective life saving interventions in the health system \[[@B3]\] and telephone helplines for smoking cessation (quitlines) have proved to be both effective and cost-effective \[[@B4]-[@B6]\]. In evaluation of smoking cessation programs including surveys and clinical trials the tradition has been to treat non-responders as smokers \[[@B7]-[@B12]\]. However, bias due to failure to respond to follow-up is seldom assessed \[[@B13]\] for the different types of interventions and presently no reliable data exists supporting the pre-judgement that non-responders are more likely than responders to be present smokers. The existing empirical data on difference in smoking between responders and non-responders is often based on public health surveys \[[@B10]\]. However, the possibility exists that non-responders in smoking cessation programs using a telephone quitline may differ from non-responders in general health surveys. Thus, to categorically classify non-responders as smokers may underestimate the true effectiveness of smoking cessation treatment \[[@B9]\]. To our knowledge no previous studies investigating telephone treatment for smoking cessation have assessed the effect of non-response on reported abstinence rates. In the present study we compare abstinence rates and population characteristics in responders and non-responders included in an evaluation of the Swedish quitline and assess reasons for not responding to the original follow-up questionnaire. Methods ======= Study population ---------------- The original study base included 1606 individuals in different stages of change who 12 months earlier had contacted the Swedish quitline for help with smoking cessation. All had accepted to receive a follow-up questionnaire 12 months after first contact with the quitline and participated in the original follow-up. First contacts were made from February 2000 to November 2001 and the follow-up questionnaires were mailed out February 2001 to November 2002. After two reminders answers were collected from 1131 individuals (70%) leaving 475 non-responders. A sample of 84 (18 % of the total non-response) was included in the present assessment of non-responders (Fig [1](#F1){ref-type="fig"}). In the beginning of November 2002 we identified all 84 patients who had not responded to the follow-up questionnaire during the previous four months. The relative narrow time window was to minimise recall bias. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Graphic description of subjects in the study base. ::: ![](1471-2458-5-52-1) ::: Data collection --------------- The non-responders were contacted by a telephone interview during the autumn 2002. Each interview took approximately five minutes to perform and was conducted by only one person, to minimise interviewer bias. Attempts were made to phone first three times during daytime and then twice during evenings/weekends. Since all subjects had completed a questionnaire earlier (following their first call to the quitline) all background information, such as gender, age, language, tobacco habits and telephone number, was already available. In the original follow-up questionnaire abstinence was defined as \"not a single puff of smoke during the previous seven days\", which is a commonly used definition of point prevalence abstinence in larger surveys \[[@B12],[@B14],[@B15]\]. Two questions were used to assess present smoking behaviour: Have you smoked at least one single puff during the previous seven days? Has your smoking behaviour changed since you received the follow-up questionnaire? An additional retrospective question was asked to assess smoking behaviour at the time of the original 12 month follow-up: Have you smoked (one puff or more) in the week previous to receiving the follow-up questionnaire? The date when the original questionnaire was sent out was specified for all subjects. The interview was completed with a question on why he/she had not returned the original questionnaire. Ethical approval was obtained by Karolinska Institutet (Dnr 00-367). Statistical methods and presentation of data -------------------------------------------- We compared the prevalence of abstinence in the original study population (responders) with the 46 non-responders participating in the present study. In the table we start by comparing abstinence in the two groups at the time of the original follow-up and then at the time of the telephone interview. Reasons stated by the subjects for not returning the original follow-up questionnaire are presented in a table as proportion of those 46 participating in the telephone interview. Reasons identified by the interviewer for not participating in the telephone interview are presented in the text as proportion of those 38 not participating. Non-responders (at 12 month follow-up) participating in the telephone interview were compared with responders in the original study population in terms of population characteristics (sex, age, smoke-free at first call, and using nicotine at first call). These data are presented in table form. In the table we also present background data on all 84 non-responders selected for the study and the 38 non-responders not participating in the telephone interview. In the table we dichotomised age as a two category variable (≤40, ≥41), nicotine use at baseline as \"yes\" or \"no\" and stages of change as \"still smoking at the time of first contact\" and \"smoke free at the time of first contact\". For statistical comparison between groups we calculated Fisher\'s exact test, odds ratios (OR) and 95% confidence intervals (CI) on proportions and OR. In table [1](#T1){ref-type="table"} we used one- sided CI on the proportions since our main focus was on the lower limits. Statistical analyses were carried out using the Statistical Package for the Social Sciences (SPSS 11.5). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Percentage and proportions of abstinence in the original study population (responders) and the present study population (non-responders) at 12 months follow-up, and at the time of the telephone interview. ::: **Original study population** **The present study population participating in the telephone interview** ----------- ------------------------------- --------------------------------------------------------------------------- ------------------------------------------------- **Abstinent at 12 months** **Abstinent at 12 months** **Abstinent at the time of the tel. Interview** **% (n/N) One-sided 95%CI** **% (n/N) One-sided 95%CI** **% (n/N) One-sided 95%CI** Men 30 (69/226) 63 (5/8) 38 (3/8) ≥25 ≥29 ≥11 Women 31(285/905) 34 (13/38) 26 (10/38) ≥29 ≥22 ≥15 **Total** 31 (354/1131) 39 (18/46) 28 (13/46) ≥29 ≥27 ≥18 ::: Results ======= Of the 84 subjects not responding to the original questionnaire at 12 month follow-up (non-responders) recruited for the study base 55% (46/84) participated. Of the 38 subjects not participating 61% (23/38) could not be reached, 29% (11/38) declined, and 10% (4/38) were either sick or dead (not in table). Abstinence ---------- Of the 46 subjects participating in the present study 39% reported to have been smoke-free at the time when they received the original follow-up questionnaire (abstinent at 12 months) compared with 31% of responders in the original study population (Table [1](#T1){ref-type="table"}). No significant difference in abstinence was noted between the present study population and the original study population (Table [1](#T1){ref-type="table"}). However, men in the present study population were somewhat more likely to report being abstinent at 12 months compared with the men in the original study population (Table [1](#T1){ref-type="table"}). The reported higher level of 12 months abstinence in men did not persist at the time of the telephone interview (Table [1](#T1){ref-type="table"}). One woman did not remember whether or not she was abstinent at twelve months and was treated as a smoker. Reasons for not returning the postal questionnaire -------------------------------------------------- The most common reason stated for not having returned the original questionnaire was claiming that they had returned it (Table [2](#T2){ref-type="table"}). Approximately one in ten stated that they had believed that abstinence was a prerequisite for answering and therefore had not returned the questionnaire since they were smoking at the time (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Stated reasons among 46 participants for not returning the postal questionnaire ::: **%** **(n)** ----------------------------------------------------- ------- --------- Claimed to have returned the questionnaire 35 (16) Had not received the questionnaire/moved 20 (9) Thought abstinence was a prerequisite for answering 13 (6) Do not know 13 (6) Forgot to return it 9 (4) Uninterested 6 (3) Had lost the questionnaire 4 (2) **Total** 100 (46) ::: Comparison with responders -------------------------- The non-responders comprising the study base in the present study were somewhat younger than the responders in the original study population (Table [3](#T3){ref-type="table"}). The mean ages being 47 for the responders and 42 for the non-responders (data not in table). Men and women were equally represented both among responders and non-responders. Non-responders tended to a higher degree to have been smoke-free when they first called the quitline (Table [3](#T3){ref-type="table"}). They were also significantly more likely to have been totally nicotine free at first call compared with the responders (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Population characteristics of responding and non-responding subjects. Comparing 46 non-responders participating in the non-response analysis with the 1131 responders in the original study population. ::: **Total** **Total sample of non-responders** **Non-responders not participating in the telephone interview** **Non-responders participating in the telephone interview** **Responders in the original study** **Comparison ¶** ------------------------------------- ------------------------------------ ----------------------------------------------------------------- ------------------------------------------------------------- -------------------------------------- ------------------ ------------ **Sex** **Male**(Ref) 20 (17) 24 (9) 17 (8) 20 (226) Female 80 (67) 76 (29) 83 (38) 80 (905) 1.2 0.5 -- 2.6 **Age distribution**: ≥ **41**(Ref) 58 (49) 61 (23) 57 (26) 67 (755) ≤ 40 42 (35) 39 (15) 43 (20) 33 (376) 1.5 0.9 -- 2.8 **Smoke-free at first call:** **No**(Ref) 73 (61) 76 (29) 70 (32) 77 (875) Yes 27 (23) 24 (9) 30 (14) 23 (256) 1.5 0.8 -- 2.8 **Using nicotine\* at first call:** **Yes**(Ref) 82 (69) 87 (33) 78 (36) 89 (1010) No 18 (15) 13 (5) 22 (10) 11 (121) 2.3 1.1 -- 4.8 \*Total consumption of nicotine, including smoked and smoke-free tobacco and NRT. ¶Comparing non-responders participating in the telephone interview to responders ::: Discussion ========== Our results indicate that non-responders in follow-ups of large cohorts of smokers trying to quit with the aid of telephone quitlines, should not by definition be considered as treatment failures. If anything, the non-responders in the present study reported higher abstinence rates at the time when they were supposed to return the original follow-up questionnaire. The observed trend of overrepresentation of non-smokers among non-responders may be explained by the fact that they were more likely to be nicotine free at first call to the quitline. In the original follow-up we presented data separately for different stages of change. Those who were in the contemplation stage at first call reported 12-month abstinence in 19% of the cases compared with 22% for those in preparation and 53% for people who were in the action stage at first call \[[@B14]\]. Thus, being nicotine free at first call is a significant predictor for abstinence at 12 months. Our results indicate that classifying non-responders as smokers may underestimate the true treatment effect of quitlines in line with a recent Dutch study \[[@B17]\]. There is a tendency to view non-responders as a homogenous group with common characteristics but studies have not confirmed this to be the case \[[@B16]\]. A previous Swedish study comparing prevalence of smokers amongst non-responders to a general health survey with responders, showed a tendency for an overrepresentation of smokers amongst non-responders, especially in low income groups \[[@B10]\]. Contrary, a study from Holland did not find such differences \[[@B18]\]. However, caution is needed when comparing these studies to the present study since it is possible that non-responders in general health surveys may differ from non-responders in studies assessing abstinence rates after smoking cessation treatment. Based on a response rate of approximately 70% in the original study, our results suggest that non-responders in the assessment of quitlines are probably not more likely than responders to be smokers at the time of follow-up. However, the possibility exists that non-responders in studies with lower response rate may differ from the present study population and our results may only apply for studies with a similar or higher response rate. The present results are in line with a study based on a one-year follow-up of participants from a national \"Quit and Win\" contest where bias in smoking prevalence because of non-response was studied \[[@B19]\]. Approximately one in three (29%) of those who were reached by telephone refused to participate in the interviews. There is a high probability that those individuals may be present smokers. Methodological considerations include the relative low number of subjects compromising statistical power. The data indicate that non-responders are not more likely than responders to be smokers. However, owing to small numbers we are not able to conclude that the non-responders may be more likely to be smoke-free as the data indicates. Another problem is the retrospective assessment of smoking behaviour at the time the non-responders were supposed to have turned in the questionnaire. Obviously recall bias may have affected the answers. A more conservative way to interpret the data is to compare the responders in the original study with non-responders smoking behaviour at the time of the telephone interview (Table [1](#T1){ref-type="table"}, columns one and three). This more conservative comparison does not change the main results that using methods with no visual contact between counsellor and patient, non-responders may well fail to respond due to other reasons than active smoking. As in all questionnaire surveys on smoking cessation the possibility of underreporting of smoking may exist e.g. due to the impact of social desirability (the desire to appear good). However, since this would most probably affect all subjects equally it is not a major concern in the present study. Also, a self-reported smoking status appears to be a reliable indicator of actual smoking-status \[[@B20]\] and the effect of social desirability in smoking cessation studies is probably less than previously suggested \[[@B21]\]. Further, in a recent study assessing if subjects who decline cotinine tests are lying about their smoking behaviour it was found that failure to comply may result from external factors such as demands or random factors such as being too busy at the time \[[@B22]\]. In the present study, thirteen percent stated that they thought abstinence was a prerequisite for returning the questionnaire (Table [2](#T2){ref-type="table"}). This is a pedagogical problem that needs to be taken seriously since this may be a potential source of bias although not a major problem in the present study. Conclusion ========== The present study, one of the first of its kind, indicates that routinely treating non-responders as smokers in smoking cessation research may underestimate the true effect of cessation treatment. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Tanja Tomson have made substantial contributions to conception and design, analysis and interpretation of data. She has been writing all drafts of this paper. Catrine Björnström have contributed in the planning, design and analysis of this paper and have given final approval of the version to be published. Hans Gilljam participated in the planning of the study design and assisted in the writing of the paper. He have given final approval of the version to be published Asgeir Helgason have made substantial contributions to conception and design, analysis and interpretation of data. He has been involved in drafting the paper and have given final approval of the version to be published. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/52/prepub> Acknowledgements ================ The authors would like to thank Mats Toftgård (Tobacco Prevention) for statistical advice. All staff members at the Swedish quitline are also acknowledged. The Swedish quitline is funded by the Swedish Government through, The National Institute of Public Health, and the Swedish Cancer Society.

PubMed Central

2024-06-05T03:55:58.414220

2005-5-23

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173115/", "journal": "BMC Public Health. 2005 May 23; 5:52", "authors": [ { "first": "Tanja", "last": "Tomson" }, { "first": "Catrine", "last": "Björnström" }, { "first": "Hans", "last": "Gilljam" }, { "first": "Asgeir", "last": "Helgason" } ] }

PMC1173116

Background ========== In the early evening of October 4^th^, 1992, an El Al Boeing 747-F cargo aircraft lost two of its engines just after take off from Schiphol Airport and crashed into two apartment buildings in the Bijlmermeer, a densely populated suburb of Amsterdam (the Netherlands) \[[@B1]\]. The air disaster killed 43 people, and destroyed 266 apartments \[[@B1],[@B2]\]. Fire-fighters and police officers were called to the scene to extinguish fires, to search and rescue people, to assist in the identification of human remains and personal belongings, to secure the surroundings and to clean-up the devastated area. Many of them were faced with bewildered residents and extensive destruction, and some witnessed dead or injured victims. Within a few days the wreckage of the aircraft was transported to a hangar at Schiphol Airport, where employees (i.e. \'hangar workers\') sorted and inspected the wreckage. In the extensive aftermath of the disaster, rumors and questions arose about the cause of the accident, the contents of the cargo, potential exposure to hazardous materials, and health consequences \[[@B2],[@B3]\]. Every now and then the media highlighted stories of individual victims, as well as uncertainties about potential exposures during the disaster \[[@B4]\]. One of the major topics concerned exposure to depleted uranium from the aircraft\'s balance weights, particularly because some of the depleted uranium has never been recovered from the rubble \[[@B1]\]. However, the authors of a retrospective risk analysis \"considered it improbable that the missing uranium had indeed led to the reported health complaints\" \[[@B5]\]. Nonetheless, it appeared that a growing number of exposed workers and affected residents reported health complaints, which some of them attributed to the disaster \[[@B6]\]. Public and political unrest thus waxed and waned in the aftermath of the disaster \[[@B2],[@B3]\]. Eventually, a parliamentary inquiry, that was held in 1998, recommended an epidemiological study on the health effects of the disaster \[[@B1]\]. About the same time, in 1998, the employers of professional fire-fighters and police officers in Amsterdam decided to start an independent assessment of the health status of professional workers involved in the disaster. The mayor of Amsterdam assigned their occupational health service, the KLM Health Services, to organize this assessment. The employer of the hangar workers at Schiphol Airport joined this initiative, as did government representatives of the affected inhabitants and volunteer workers. It was decided to offer a medical examination to all people involved in the air disaster, residents as well as assistance workers, and that an epidemiological study would be performed simultaneously by the Institute for Research in Extramural Medicine (EMGO Institute). In this paper we report on the design of the epidemiological study among professional assistance workers: the Epidemiological Study Air Disaster in Amsterdam (ESADA). Unfortunately, the epidemiological study among residents had to be cancelled, due to low response rates. The ESADA is the first epidemiological study that has ever been conducted after a major technological disaster in the Netherlands. The aim of this study is to assess the long-term psychological and physical health effects of occupational exposure to the air disaster in Amsterdam on professional assistance workers, i.e. fire-fighters, police officers and hangar workers. Based on the scientific literature on the health effects of disasters, the main hypotheses of the ESADA concern unexplained physical symptoms \[[@B7]-[@B12]\], and post-traumatic stress symptoms and associated psychological symptoms \[[@B13]-[@B15]\]. Due to the fact that the ESADA originated partly from societal concerns, we considered it necessary to also include some additional outcomes that will answer questions for some of the affected people, which, in turn, might help to reassure them. These societal questions relate to depleted uranium, Mycoplasma species and carnitine levels in plasma. The first of these questions stems from the concerns about the depleted uranium from the aircraft\'s balance weights, described above. The other two questions are primarily based on an alleged resemblance between the symptoms of some of the people affected by the air disaster in Amsterdam and the symptoms of patients with chronic fatigue syndrome (CFS) and Gulf War (I) Syndrome (GWS). Although some authors may have suggested a link between these syndromes and Mycoplasma species \[[@B16]-[@B21]\] or carnitine deficiency \[[@B22]-[@B26]\], others have rejected the existence of such links \[[@B27]-[@B29]\]. In this paper we describe and discuss the design of the ESADA. More details on the (organization of the) ESADA can be found on its website \[[@B30]\]. Methods / Design ================ Design ------ The ESADA is designed as a historical cohort study, in which the health status of the professional fire-fighters, police officers and hangar workers who were occupationally exposed to the air disaster in Amsterdam is compared with the health status of reference groups of workers with the same jobs and employers at the time of the disaster, but who were not occupationally exposed to this disaster. Study population ---------------- The ESADA study population consisted of professional fire-fighters, police officers and hangar workers. Eligible subjects had to (1) sign informed consent; (2) have sufficient mastery of the Dutch language to fill in the questionnaires; and (3) belong to one of the following three occupational groups: 1\) All professional fire-fighters who were, according to company records, employed in the Amsterdam fire department at the time of the disaster. Additional professional fire-fighters who started working in this fire department after the disaster were also invited to participate in the study, as almost the entire fire department had been exposed to the disaster. 2\) All police officers (*i.e.*constables, warrant officers, sergeants and their supervisors) who were, according to company records, employed in the Amsterdam-Amstelland regional police force on the date of the disaster (October 4^th^, 1992), and were still employed there on the 1^st^of January 2000. 3\) All the hangar workers registered as working for one of the departments involved in the transport, security and sorting of the wreckage on the date of the disaster (October 4^th^, 1992), and who reported to have been involved in these activities; as well as a random sample, matched with their colleagues for age, sex, department and job title, who were also registered as working for these departments on 30^th^November 1992, but who did not report to have been involved in any disaster-related activities. Procedures and data-collection ------------------------------ The study design was approved by the two independent Medical Ethics Committees of the medical facilities involved in this project: the VU University Medical Center (VUmc) and the \'Onze Lieve Vrouwe Gasthuis\' (OLVG) in Amsterdam. Potential participants were initially informed about the study via announcements in staff magazines, after which they were approached via personal letters, and eventually by telephone. All participants signed informed consent and participated voluntarily. Data were collected at the Prinsengracht out-patient clinic of the OLVG from January 2000 to March 2002, i.e. on average 8.5 years after the disaster. In addition, data on about half of the hangar workers were collected at Schiphol Airport for logistic reasons. Trained medical research assistants checked that the questionnaires had been completed, measured body height and weight, drew blood samples, and assisted with the collection of urine and saliva samples. A team of administrative employees carried out the data-entry of the questionnaires. Data of each participant were entered twice by two of these employees independently, after which inconsistencies were reviewed and any mistakes rectified. All remaining problems in the interpretation of data, such as dubious handwriting, were consistently resolved by one of the authors (AH, PS or AW). Blood, saliva and urine samples were dealt with according to standard procedures for collection, transportation, storage and laboratory analysis. Laboratory technicians could have been aware that the samples were from the ESADA, but they were blinded for exposure and health status. The laboratories were all certified according to accredited Dutch standards. Occupational exposure to the disaster ------------------------------------- All participants were asked to fill in a questionnaire on occupational exposure to the air disaster. This questionnaire addressed several specific disaster-related tasks, and also the total time spent on these tasks and the location in which they were performed (e.g. on or near the disaster site, in the hangar where the wreckage was temporarily placed, or elsewhere). They were also asked to describe any other disaster-related task(s) that they had performed. Answers to the latter question were categorized (by PS and AW). The questionnaire also covered disaster-related psychosocial events in a number of items on personal experiences during the disaster (e.g. \"were you in life-threatening danger?\", \"did you see the disaster scene during the first hours after the crash?\", and \"were any of your family members injured?\"). These personal records of occupational exposure to the disaster were used to define \'exposed\' workers, i.e. those who reported at least one disaster-related task, and \'non-exposed\' workers, i.e. those who did not report any disaster-related tasks. In addition to comparing exposed and non-exposed workers, we examined exposure-response relationships among exposed workers, in which level of exposure is characterized by the type of tasks and psychosocial events and the duration of exposure. As an additional dimension of level of exposure, we took into account the differences in potential psychotraumatic impact of exposure items, based on criterion A1 of the diagnostic criteria for Post Traumatic Stress Disorder (PTSD; American Psychiatric Association \[APA\]; Diagnostic and Statistical Manual of Mental Disorders-IV-Text Revision \[DSM-IV-TR, 2000\]) \[[@B31]\]. This criterion states that \"the person has experienced, witnessed, or been confronted with an event or events that involve actual or threatened death or serious injury, or a threat to the physical integrity of oneself or others\". Five experts on PTSD from different universities and psychiatric hospitals independently rated the likelihood of potentially psychotraumatic disaster-related tasks and events to meet this criterion on a 4-point Likert Scale ranging from 1 = \'very unlikely\' to 4 = \'very likely\'. Subsequently, we assumed that items with a mean item score of three or higher met the A1 criterion for PTSD (i.e. A1 tasks and events), as opposed to items with a lower mean score (i.e. non-A1 tasks and events). Table [1](#T1){ref-type="table"} lists the disaster-related tasks and the psychosocial events according to their potential psychotraumatic impact. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Disaster-related tasks and psychosocial events according to their potential psychotraumatic impact ::: ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **A1\* (traumatic)** **Non-A1\* (non-traumatic)** ------------------------- ------------------------------------------------------------------------------------------------------------------ ---------------------------------------------------------------------------------------------------- **Tasks** 1\. Identification or recovery of victims from the rubble/transport or search for human remains\ 1\. Fire-extinguishing\ 2. Rescue people 2. Clean up of destructed area\ 3. Transport of injured victims\ 4. Provide first aid/support injured victims or workers\ 5. Security tasks (surveillance, prevent burglary, keep disaster area free of bystanders)\ 6. Other tasks (e.g. traffic management)\ 7. Sort wreckage in hangar (at Schiphol Airport)\ 8. Other tasks in hangar in the presence of the wreckage\ 9. Transport of wreckage\ 10. Burning of contaminated soil remnants (from disaster site) **Psychosocial events** 1\. Having been in life-threatening danger during disaster\ 1\. Saw the aircraft crash / saw or heard the aircraft when it crashed\ 2. Personal injuries due to disaster\ 2. Felt or heard the impact of the crash\ 3. Witnessed dead or injured victims\ 3. Saw the fire\ 4. Having been in or near one of the destroyed buildings at the time of the disaster\ 4. Saw the disaster site during the first hours after the crash/when the wreckage was still there\ 5. Immediate family members (partner, children) died / in life-threatening danger / injured due to the disaster\ 5. Other family members in life-threatening danger or injured due to the disaster\ 6. Other family members died due to the disaster 6. Friends or acquaintances died, injured or in life-threatening danger due to the disaster\ 7. Apartment of other family members, friends, or acquaintances damaged due to the disaster\ 8. Lived in the affected suburb of Amsterdam (Bijlmermeer) at the time of the disaster\ 9. Visited the hangar where the wreckage was kept ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- \*A1 and non-A1 = items with a mean score of ≥ 3 or \<3, respectively, on a 4 point Likert Scale indicating the likelihood for an item to meet criterion A1 for post-traumatic stress disorder (from \'very unlikely\' \[=1\] to \'very likely\' \[=4\]) (see Methods). ::: Main health outcomes -------------------- ### Self-reported health measures • *Post-traumatic stress symptoms*: (a) The Dutch 22-item Self-Rating Inventory for PTSD (SRIP) \[[@B32]-[@B34]\] and, among exposed subjects only, (b) The 15-item Dutch version of the Impact of Event Scale (IES), which addressed post-traumatic stress symptoms with explicit reference to the air disaster in Amsterdam \[[@B35]-[@B37]\]. • *General mental health*: (a) The 90-item Symptom Checklist (SCL-90) \[[@B38],[@B39]\]; (b) The 20-item General Health Questionnaire (GHQ-12) \[[@B40]\]. • *Fatigue and associated symptoms*: The 20-item Checklist Individual Strength (CIS) \[[@B41],[@B42]\]. • *Health-related quality of Life*: The MOS 36-item Short-Form Health Survey (SF-36) \[[@B43],[@B44]\]. • *Chronic conditions*: One questionnaire assessed the current presence and history of the following chronic conditions, which are considered to have a significant impact on well-being: diabetes; stroke, brain hemorrhage or infarction; heart attack; other heart problems (such as heart failure, or angina pectoris); cancer; chronic osteoarthritis (wear) of the hip or knee joints; hypertension; asthma, chronic bronchitis or lung emphysema (Chronic Obstructive Pulmonary Disease \[COPD\]); serious or persistent intestinal disorders (longer than 3 months); chronic stomach disorders, stomach or duodenal ulcers; serious or persistent back complaints (including hernias); chronic inflammation of the joints (chronic rheumatism, rheumatoid arthritis). Workers with these chronic conditions were subsequently asked in what year the onset was, to determine whether this was before the disaster took place. • *Physical symptoms*: Multiple questionnaires were used to assess the current presence of various physical symptoms, such as a number of respiratory, musculoskeletal, and skin symptoms. • *Attribution of current problems to the air disaster in Amsterdam and its aftermath*. Another questionnaire assessed the extent to which exposed workers related any of their current physical, psychological or practical/financial problems to the air disaster and its aftermath. Those who attributed physical symptoms to the disaster and its aftermath were asked to specify these symptoms. ### Laboratory outcomes General laboratory tests \[[@B1]\]: • *Hematological and blood chemical outcomes*: hemoglobin, leukocyte count, differential count, platelet count and mean corpuscular volume (Sysmex SE 9000, TOA medical electronics Co. ltd); potassium (Roche Modular ISE900, Roche Diagnostics); creatinine, alkaline phosphatase, gamma-glutamyl transferase, alanine aminotransferase, creatine kinase and C-reactive protein (Roche Modular P800, Roche Diagnostics); ferritin and thyroid stimulating hormone (Centauer, Bayer Diagnostics); β2-microglobuline (IMx Abbott). • *Autoantibodies*: nuclear antigen antibodies, anti-double stranded DNA antibodies \[[@B46]\], Immunoglobulin (IgM) rheumatoid factor \[[@B47]\], antineutrophil cytoplasmic antibodies \[[@B48],[@B49]\], and cardiolipin antibodies \[[@B50],[@B51]\]. • *Urine outcomes*: creatinine (Hitachi 747, Roche Diagnostics GmbH, Mannheim, Germany); micro-albumin (Beckman Array 360 system); and β2-microglobuline (IMx Abbott); screening for protein, glucose, pH, blood and leukocytes (teststrip Boehringer Mannheim B.V.), followed by microscopic evaluation of the urinary sediment if indicated. • *Saliva outcome*: cortisol concentration (Wizard 1470, Perkin Elmer). Additional laboratory tests with respect to the societal questions: • *Uranium 238*: concentration in urine (Inductively Coupled Plasma-Mass Spectrometry \[ICP-MS\] analyser, Finnigan Mat Element) and, at concentrations above 50 ng/l or above 50 ng/g creatinine, also the ratio of uranium 235/238 isotopes \[[@B52]\]. • *Total and free carnitine*: concentration in blood plasma (Mira Plus, Roche Diagnostics) \[[@B53],[@B54]\]. • *DNA of any Mycoplasma species*: presence in peripheral blood mononuclear cells (DNA-isolation, Magna Pure, Roche Diagnostics; real time PCR, Taqman, Applied Biosystems); positive samples were subsequently evaluated for the presence of DNA of Mycoplasma fermentans \[[@B55],[@B56]\]. Self-reported socio-demographic characteristics ----------------------------------------------- • *Age*: at time of assessment in years. • *Sex:*male or female. • *Ethnicity:*categorized into those who considered themselves as European (i.e. Dutch, British, Dutch/Irish, Dutch/Chinese, Dutch/Indonesian, Portuguese, Spanish, Dutch/ Spanish and European), and others (*e.g.*Moroccan, Turkish, Surinam). • *Level of education*: highest level of education completed, categorized as: high (higher vocational education, university); medium (intermediate vocational education, higher general secondary education, or pre-university education); and low (no education, elementary school, lower vocational education, or lower general secondary education). • *Current executive function*: yes (i.e. supervising one or more people) or no. • *Level of physical activity*: the total number of hours spent each week on physical activities such as physical exercise, gardening and housekeeping, classified into high, medium and low according to the 33^rd^and 66^th^percentiles. • *Alcohol consumption*: Usual and exceptional consumption of alcoholic beverages, classified into: none; light-moderate; and (extremely) excessive, i.e. consumption of (a) six or more glasses on 9--20 days a month and on 3--4 days in the last week, (b) four or more glasses on at least 21 days a month and on at least 5 days in the last week, and/or (c) more than six glasses a day, on a weekly basis. • *Cigarette-smoking:*categorized as: never, former smoker, and current smoker. • *Negative life events:*the number of reported negative life events, based on a questionnaire which specified 13 such events and also included two open-ended questions in which other events could be described. Subjects were asked to indicate whether any of these events happened to them before or after the disaster. Role of funding sources ----------------------- The study was funded by the Dutch Ministry of Health, Welfare and Sports; the City of Amsterdam; the Amsterdam-Amstelland regional police force; and KLM Royal Dutch Airlines. The funding sources had no role in the collection, analysis, or interpretation of the data, or in the decision to submit a manuscript for publication. Discussion ========== In recent years there has been increasing scientific and societal interest in the health consequences of man-made, technological disasters, i.e. a collective stressful experience with a sudden onset due to technological failure. Technological disasters have had psychiatric consequences \[[@B13]-[@B15],[@B57]\], such as PTSD, as well as medical consequences, in particular those of toxic exposures \[[@B58]-[@B61]\]. In addition to direct toxic health effects, the mere suspicion and fear of exposure to hazardous materials can also take its toll on the quality of physical, psychological and social well-being in the community \[[@B62]-[@B64]\]. Technological disasters strike unexpectedly and suddenly, which puts time-pressure on researchers to develop study protocols, gather exposure data, call in multidisciplinary experts, and obtain financial resources for immediate epidemiological research. Disaster researchers may also have to deal with complicated socio-political and legal aspects. In addition, they have to face a number of methodological problems. These difficulties include: (a) defining the entire potentially \'affected\' population and appropriate reference groups; (b) contacting potential participants, particularly in the case of evacuation and hospitalization; low response rates; usually without data on non-respondents \[[@B65]\]; (c) collecting exposure data immediately after the event, which is actually also needed for long-term epidemiological studies. Probably due to these difficulties, evidence from large-scale epidemiological studies that have been carried out after technological disasters is rather scarce \[[@B66],[@B67]\]. Furthermore, before-after comparisons are rare and only possible by chance in ongoing research projects, due to the unexpected nature of technological disasters \[[@B68]-[@B70]\]. Most of the studies that have been conducted so far have relied on \'convenience samples\', which were mainly composed of those who were directly affected, such as victims and residents; were based on non-epidemiological study designs; and used group-level or retrospective, self-reported exposure data, which can be affected by recall and reporting bias \[[@B71]-[@B74]\]. ESADA approach -------------- The purpose of the ESADA is to assess long-term health effects of occupational exposure to the air disaster in Amsterdam on professional assistance workers. In view of the above-mentioned difficulties in epidemiological research on disasters, the ESADA has some strong methodological points. With respect to the study population, we have been able to identify the complete cohort of exposed and non-exposed workers accurately, based on company records of employment at the time of the disaster. Another strong point of the ESADA is that we included reference groups of colleagues, who had the same jobs and employers, but who were not occupationally exposed to the disaster. Hence, we are able to draw group-level conclusions on associations between health status and occupational exposure to the disaster. With respect to exposure assessment, we were able to collect individual data on occupational exposure to the air disaster. Moreover, this consisted of multiple aspects of self-reported occupational exposure, including the duration and location of various disaster-related tasks and the experience of potentially stressful events during these tasks. Finally, we also included various assessments of long-term health, such as laboratory tests and self-reported symptoms and health-related quality of life, to obtain an integral evaluation of health status. Notwithstanding these strong methodological qualities, some limitations of the ESADA design should also be mentioned. Firstly, although company records of employment were available, we still had to resolve a few difficulties regarding the definition of the study population. For the fire-fighters, this was due to the fact that almost the entire fire department of Amsterdam had been exposed to the disaster. Therefore, in order to achieve an adequate reference group, we decided to also include fire-fighters who joined this fire department after the disaster took place. With respect to the police officers, we were unable to trace those who had left the Amsterdam-Amstelland regional police force in the years after the disaster, due to administrative difficulties. Hence, it was necessary to restrict this group to those who were still working for this police force in 2000. A second methodological issue concerns the self-report nature of occupational exposure status, and the average time-lag of 8.5 years between the disaster and the assessment. Due to administrative deficiencies in the historic registration of the exposure status, we used our detailed questionnaire data to define exposure status for all workers. Strictly speaking, the ESADA is therefore not a historic cohort study, but a cross-sectional one. The time-lag between the disaster and the exposure assessment may have led to recall bias, especially concerning certain details of exposure to the disaster, such as the duration of activities. However, it seems reasonable to assume that the workers did recollect whether or not they performed any as opposed to no disaster-related tasks, which was used to define occupational exposure status. It is therefore very unlikely that recall bias has resulted in (non-)differential misclassification of exposed and non-exposed workers. Nevertheless, recall bias should be kept in mind with respect to exposure-response relationships. We included multiple aspects of level of exposure, such as the duration and the potential psycho-traumatic character of disaster-related tasks, as it is unknown which aspect of occupational exposure to disasters is relevant for long-term health. However, we may still have missed other potentially relevant aspects, such as exposure to disaster-related media reports in the aftermath of the disaster \[[@B4],[@B75]\]. Thirdly, we acknowledge the fact that, with the exception of the laboratory variables, we rely on self-reported health outcomes. However, most of the health questionnaires that we used have been validated and widely accepted, except for those used to assess the physical symptoms. Differential misclassification in self-report health measures could occur if exposed workers are more likely than non-exposed workers to interpret and report bodily sensations as symptoms. On the other hand, hypervigilance and hypochondria themselves could well be adverse health effects of (toxicological) disasters \[[@B76],[@B77]\]. In conclusion, to increase our knowledge of potential health consequences of (technological) disasters, it is important to be prepared for epidemiological disaster research. Incorporating basic multidisciplinary, epidemiological research protocols into disaster management plans will stimulate scientifically sound research on the health effects of disasters. The ESADA will provide additional scientific knowledge on the long-term health effects of technological disasters on professional workers. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= All authors participated in the multidisciplinary ESADA project team of the EMGO Institute, provided comments on the draft versions and approved the final manuscript. PS drafted the manuscript and performed statistical analyses. AH coordinated the acquisition of data and supervised the first part of the ESADA study. JT contributed to the design and supervised the statistical analyses. AW coordinated the rating of the potential psychotraumatic impact of exposure items and performed statistical analyses. HP and IB participated in the conception and design of the ESADA with respect to psychological outcomes, and IB rated the potential psychotraumatic impact of exposure items. NS supervised the second part of the ESADA. As Medical Director JB coordinated the data-collection. LB contributed to the design of the ESADA. WM contributed to the design and is Vice-President of the ESADA project team. TS conceived the study, and participated in its design and coordination as President of the ESADA project team. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/54/prepub> Acknowledgements ================ We wish to thank everyone who contributed to the organization and data-collection as well as the funding sources of the ESADA.

PubMed Central

2024-06-05T03:55:58.416574

2005-5-30

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173116/", "journal": "BMC Public Health. 2005 May 30; 5:54", "authors": [ { "first": "Pauline", "last": "Slottje" }, { "first": "Anja C", "last": "Huizink" }, { "first": "Jos WR", "last": "Twisk" }, { "first": "Anke B", "last": "Witteveen" }, { "first": "Henk M", "last": "van der Ploeg" }, { "first": "Inge", "last": "Bramsen" }, { "first": "Nynke", "last": "Smidt" }, { "first": "Joost A", "last": "Bijlsma" }, { "first": "Lex M", "last": "Bouter" }, { "first": "Willem", "last": "van Mechelen" }, { "first": "Tjabe", "last": "Smid" } ] }

PMC1173117

Background ========== In many developing countries and countries with economies in transition, health care injections are overused and are frequently administered in an unsafe manner \[[@B1],[@B2]\]. The World Health Organization (WHO) estimates that 20 million new hepatitis B virus (HBV) infections, 2 million new hepatitis C virus (HCV) infections, and 260,000 new HIV infections are associated with unsafe injections each year worldwide \[[@B3]\]. Concerns have been expressed over the use of injections within the Kingdom of Cambodia (2003 United Nations population estimate 14.1 million). Anecdotal reports suggest that therapeutic injections are often unnecessary and administered in an unsafe manner, and healthcare waste is inappropriately disposed \[[@B4]-[@B6]\]. A high prevalence of HIV infection has been documented in Cambodia, with 2.6% of the general population infected \[[@B7]\]. Cambodia also has high levels of HBV and HCV infection endemicity. A 1991 community-based study indicated that 8.0% of the population was hepatitis B surface antigen positive and 6.5% was anti-HCV positive \[[@B8]\]. The proportion of bloodborne pathogen infections in Cambodia attributable to unsafe injections is not known. To estimate the magnitude and patterns of unsafe injection practices in Cambodia, a rapid assessment of injection practices was conducted. The objectives of the assessment were to describe healthcare injection practices, including measures of injection frequency and injection safety, and determine knowledge, attitudes and practices related to medical injections among the general population and healthcare providers. This assessment focused mainly on the unregulated private healthcare sector, in which 70% of Cambodians seek medical care \[[@B9]\]. Methods ======= Overview -------- The assessment was conducted using methods established by WHO for the rapid assessment of injection practices \[[@B10]\]. As detailed in the following sections, we interviewed members of the general population, healthcare workers responsible for prescribing injections (i.e., \"prescribers\"), and healthcare workers responsible for administering injections (i.e., \"injection providers\"). All data were collected during face-to-face interviews using standardized questionnaires and forms adapted from the WHO assessment guide \[[@B10]\]. The general population sample was drawn from Takeo province (1998 population census 790,168), a rural province located in southeastern Cambodia near the Vietnamese border. Prescribers and injection providers were sampled in Takeo province and in Phnom Penh (1998 population census 999,809), Cambodia\'s largest city \[[@B11]\]. General population ------------------ We surveyed 500 Takeo province residents in November and December 2002 using standard cluster sampling methodology. To obtain a total of 500 participants, we surveyed 25 participants each in the catchment area of 20 district health centers. The 20 district health centers were selected by probabilities proportional to the population size of their catchment areas \[[@B12]\]. For each health center selected, one village was chosen at random to be surveyed. Households were sampled at random until 25 participants were included. Verbal informed consent was obtained from all household members \>16 years old to participate in the survey. For persons ≤ 16 years old, informed consent was obtained from a parent or guardian. For each participant, basic demographic data were collected, as well as the number of injections received during the 6 month period beginning on April 13, 2002 (New Year\'s Day in Cambodia). Injections were categorized as therapeutic injections, immunizations, and intravenous infusions. Participants \>16 years of age were also interviewed regarding their knowledge and attitudes about injections. Prescribers and injection providers ----------------------------------- We surveyed prescribers and injection providers in Takeo province and Phnom Penh in November and December 2002. In Takeo province, a convenience sample of 30 prescribers and 30 injection providers was accessed through the province\'s five public hospitals. In Phnom Penh, 30 prescribers were randomly selected from a list of registered private outpatient clinics and 30 injection providers were selected from 11 of the city\'s 13 private hospitals. For all prescribers, regardless of where they were identified, the survey focused on their medical practices in their private outpatient clinics. We ascertained their knowledge, attitudes and practices regarding injections. Injection frequency was determined by asking prescribers to estimate the total number of prescriptions they wrote per week and the number that included an injection or infusion. Injection providers were interviewed regarding their knowledge of disease transmission risks through unsafe injections, their hepatitis B vaccination status and the frequency of needle stick injuries during the previous 12 months. In addition, we observed each injection provider administer one injection. During all hospital visits, information was recorded regarding the use of incinerators and the presence of used needles and syringes and other medical equipment on the grounds of the facility. Data analysis ------------- Proportions, means, and confidence intervals were calculated using EpiInfo 6.04d (CDC, Atlanta, Georgia) with adjustment for the cluster sampling design effect. Univariate and multivariate analyses of variables associated with receiving injections were performed using SAS software version 8 (SAS Institute, Cary, North Carolina). Results ======= General population ------------------ The overall injection rate was 5.9 injections per person-year (95% CI: 5.3 -- 6.7), with 40% of the participants reporting receipt of one or more injections during the previous 6 months (Table [1](#T1){ref-type="table"}). Therapeutic injections accounted for 74% of the total number of injections reported, followed by intravenous infusions (16%), and immunization (10%). Among individuals who had any exposure to injections, nearly one-half reported receiving ≥ 5 injections. This group represented only 18% of all participants but accounted for 69% of the total injections reported. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Characteristics of study participants, Takeo province, Cambodia, 2002 ::: ------------------------------------------------------------ --------------------- -------------------------- --------------------------- ------------------------ **Total (n = 500)** **Children\* (n = 212)** **Adult women (n = 189)** **Adult men (n = 99)** **Age:** median 22 9 38 37 range (years) 0--80 0--16 17--79 17--80 ***[Injection experiences (6-month period)]{.underline}*** Number of injections 1483 557 713 213 Overall injection rate (per person-year) 5.9 5.3 7.5 4.3 Proportions of subjects who received: ≥ 1 vaccine injection 15% 35% 0% 0% ≥ 1 therapeutic injection 32% 31% 39% 20% ≥ 1 intravenous infusion 16% 16% 18% 14% ≥ 1 injection (any type) 40% 46% 42% 24% ≥ 5 injections (any type) 18% 18% 22% 9% ***[Knowledge and attitudes\^]{.underline}*** Trusted practitioners if no injections prescribed 90% ND 90% 89% Preferred injection for treatment of fever 32% ND 35% 26% Believed injections more powerful than oral medication 47% ND 50% 40% Aware that dirty syringes can transmit HIV 95% ND 96% 93% Aware that dirty syringes can transmit hepatitis 59% ND 60% 57% ------------------------------------------------------------ --------------------- -------------------------- --------------------------- ------------------------ ND = not determined \* participants aged ≤ 16 years \^ adults only (288 total respondents) ::: The pattern of injection use differed by age and sex. Among adults, women were more likely than men to report receiving any (42% vs. 28%, p \< 0.001) or frequent injections (i.e., ≥ 5 injections during the previous 6 months; 22% vs. 9%, p \< 0.001) (Table [1](#T1){ref-type="table"}). Women were also more likely than men to report having received ≥ 1 therapeutic injection (39% vs. 20%, p = 0.001). Similar proportions of men, women and children reported receiving intravenous infusions, ranging from 14 -- 18%. Immunizations were limited to children, with 35% of children reporting having received ≥ 1 immunization. Ninety-six adult participants (93% of those reporting one or more injection in the previous 6 months) could recall the details of their last injection. All reported the injection was administered with a newly opened disposable syringe and needle. The majority (85%) of these injections was administered by health care workers; at least 13% were administered by lay persons (Figure [1](#F1){ref-type="fig"}). Injections were most commonly administered in the patient\'s home (65%) or in a private clinic (20%). Only 13% of injections were administered at public hospitals or public health centers. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Distribution of injections by provider type and setting\*, Takeo Province, Cambodia, 2002 (n = 96) \*Data refer to the most recent injection received by adults reporting ≥ 1 injection in the previous 6 months ::: ![](1471-2458-5-56-1) ::: Of the 288 subjects interviewed, 90% reported they would trust a medical practitioner who did not prescribe an injection (Table [1](#T1){ref-type="table"}). However, 47% (134/288) believed that therapeutic injections are more powerful than oral medications, with 75% (100/134) of these persons attributing a faster healing effect to injections. In addition, 32% of the respondents reported they preferred injections over oral medications for the treatment of illnesses characterized by fever. Knowledge of the HIV transmission risk associated with \"dirty\" syringes and needles was very high (95%), though only 59% of respondents were aware that hepatitis could be transmitted in this manner. None of the knowledge or attitude variables or other characteristics (besides gender) were associated with the receipt of injections. This was true even when the analyses were restricted to comparisons of persons reporting no injections versus frequent injection use (i.e., ≥ 5 injections during the previous 6 months). Prescribers ----------- The 60 prescribers interviewed were registered medical doctors or medical assistants. Prescribers estimated that nearly half (47%) of the total prescriptions they wrote included either a therapeutic injection (34%) or an intravenous infusion (14%) (Table [2](#T2){ref-type="table"}). The overall prescription rates and distribution by type was similar in Phnom Penh and Takeo Province. The main reasons cited for prescribing injections were illness severity (44%) and perceived patient preferences for injectable medications (40%). Seventy-seven percent of prescribers believed that injection prescriptions resulted in greater reimbursement. Only one prescriber perceived himself as over-prescribing injections. Most (92%) prescribers were aware that HIV, HBV and HCV could be transmitted through unsafe injections (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Characteristics of injection prescribers and providers, Takeo province, Cambodia, 2002 ::: ------------------------------------------------------------------------------------------------ -------------------- ----------------------------- ------------------------- **Total (n = 60)** **Takeo Province (n = 30)** **Phnom Penh (n = 30)** ***[Characteristics of prescribers^1^]{.underline}*** Medication prescription rate (prescriptions/week) average 20 21 20 median 20 21 20 range 4--140 4--140 7--140 Prescriptions including an injection^2^ 47% 48% 45% therapeutic injection 34% 32% 35% intravenous infusion 14% 18% 10% Main reason for prescribing injections illness severity 44% 50% 37% patient preference 40% 40% 40% more effective than oral medications 12% 7% 17% reimbursement 9% 7% 10% Preferred injectable med for treatment of febrile illness 64% 47% 80% Believed patient trust requires injection prescription 42% 53% 30% Believed reimbursement is higher for patient visits that result in injection prescription 77% 87% 66% Perceived themselves as over-prescribing injections 2% 3% 0% Knew HIV, HBV and HCV can be transmitted through unsafe injections 92% 87% 97% ***[Characteristics of injection providers]{.underline}*** Completed hepatitis B vaccination series 20% 7% 33% Needlestick injury in last 12 months 53% 50% 57% Average number (and range) of needlesticks in past 12 months among those reporting one or more 1.7 (1--10) 1.4 (1--6) 2.0 (1--10) Use of single use needles and syringes^3^ 98% 97% 100% Safety box (i.e., sharp container) present in injection administration area^3^ 25% 37% 13% Reported having sufficient number of sharps boxes 85% 77% 93% Practiced two hand recapping of used needles^3^ 58% 53% 60% Left used sharps in preparation area^3^ 13% 23% 3% Knew HIV, HBV and HCV can be transmitted through unsafe injections 90% 87% 93% ------------------------------------------------------------------------------------------------ -------------------- ----------------------------- ------------------------- ^1^Prescribers\' responses pertain to their private outpatient practices ^2^Denominator is the total number of weekly prescriptions ^3^Based on observation of provider by the investigators ::: Injection providers ------------------- All 60 injection providers interviewed were registered nurses, only 20% of whom had received the complete hepatitis B vaccination series. During our observations, 59 (98%) providers used new single use syringes (i.e., traditional plastic disposable syringes) and needles and one administered the injection with sterilized reusable equipment. Over half (58%) of the providers were observed practicing two-handed recapping of used injection equipment and 53% reported they had sustained a needlestick injury during the past 12 months. For only 25% of providers, a safety box (i.e. sharps container) was observed in close proximity to the injection area, yet 85% reported having a sufficient number of safety boxes. We observed that 13% of providers left used sharps in the injection preparation area (23% in Takeo province and 3% in Phnom Penh city). Ninety percent of injection providers stated they were aware that HIV, HBV and HCV could be transmitted through unsafe injections (Table [2](#T2){ref-type="table"}). Waste disposal -------------- In the general population survey in Takeo province, 32% (23/72) of participants who reported receiving their last injection in their home reported that the used injection equipment was left behind by the injection provider. Used sharps were observed on the grounds outside three of the five public hospitals in Takeo province, but not on the grounds of any of the 11 private hospitals in Phnom Penh. Incinerators and sharps pits in which used sharps waste could be burned or buried were present at all of the Takeo hospitals. In Phnom Penh, sharps pits were available in two hospitals. In the remaining nine hospitals, sharps waste was collected along with non-medical waste for disposal at the city landfill. Discussion ========== The rapid assessment of injection practices described in this paper suggests that injections are prescribed and administered in Cambodia at excessive rates. Intravenous infusions were common among the general population surveyed and represented approximately one-third of all injections prescribed in private outpatient clinics. Using WHO standardized measures as a basis for comparison, our assessment indicates that Cambodia has one of the highest rates of overall therapeutic injection usage ever reported worldwide \[[@B1],[@B2]\]. Studies conducted in other countries in the region have documented rates of 2--4 injections/person/year; approximately one third lower than in Cambodia \[[@B2]\]. This assessment documented potentially harmful injection practices including inadequate handling and disposal of used injection equipment. Given the country\'s high prevalence of HIV, HBV, and HCV, the overuse and misuse of injections carry substantial risks. While reuse of injection equipment was uncommon in this survey, the high prevalences of bloodborne infections may reflect past practices of needle and syringe reuse. We did observe breaks in aseptic technique (e.g., used sharps left in the injection preparation area) that may facilitate cross-contamination during injection preparation \[[@B13],[@B14]\]. Other infection control breaks such as mishandling of sterile materials and improper use of multidose vials have been frequently implicated in the transmission of bloodborne viral and bacterial infections, but were not studied in this survey \[[@B13],[@B15]\]. Anecdotal reports suggest that used needles and syringes might be repackaged and resold as new. While there was no evidence to suggest this practice was occurring in Cambodia, we did not evaluate the sterility of needles and syringes. In addition, many of the injection providers in our survey engaged in high risk behaviors (e.g., two-handed recapping needle), highlighting the need for training and other interventions aimed at reducing occupational exposures and infection risks. The patterns of attitudes, knowledge and practices regarding injections among the general population and health professionals observed in Cambodia were similar to those in many countries with high frequencies of injections and comparable healthcare systems such as Pakistan, India, and Indonesia \[[@B16]-[@B18]\]. In these countries as in Cambodia, patients and prescribers frequently believe that injections are more effective than oral treatments, and few health professionals perceive themselves as over-prescribing injections. This study and others have also shown that many healthcare providers believed that patients\' trust hinged on injection prescriptions, whereas only a minority of patients actually stated that they prefer injections for treatment of conditions that could be treated with oral medications. Both misconceptions may facilitate injection overuse. In this study, almost half of the prescriptions made by the private providers we surveyed contained at least one injection, compared with a previous finding that only 2% of prescriptions written in public sector outpatient clinics included injections \[[@B19]\]. Financial interests might lead some private health professionals to over-prescribe injectable medications \[[@B16]\]. Since oral and intravenous medications are readily available directly from pharmacies without prescriptions, it is possible that private healthcare providers provide and promote injections and infusions as a way to attract patients \[[@B20]\]. These factors might partly explain the popularity of the private sector compared with the public healthcare sector where the cost for clinic visits is lower and medications are available free of charge, but are dispensed according to standardized treatment protocols which generally specify oral formulations. High levels of awareness of infection risks associated with syringe and needle reuse, particularly regarding HIV were documented in this study. This is consistent with results of a recent national Demographic Health Survey \[[@B9]\] and likely reflects Cambodia\'s efforts during the late 1990\'s to educate the general public regarding risk factors for HIV infection. These education campaigns included information on the risk of transmission of HIV from re-used syringes. Following the campaigns, disposable needles and syringes became widely available in Cambodia at a relatively low cost. It is possible that the increased awareness of HIV infection risks associated with injections led to increased utilization of disposable needles and syringes rather than a shift toward alternative forms of treatment. This assessment had several limitations. First, the survey did not capture the medical indications that were associated with either prescription or receipt of injections. Second, resource constraints required the general population surveys to focus on one province instead of being conducted nationwide. Takeo province was selected because it is the third most populated province, its economic status is in the middle of the national spectrum, and like the general Cambodian population it is mainly rural \[[@B9]\]. Third, because only formal health professionals were surveyed, we could not assess practices in the informal health sector. We found that approximately one-in-seven injections in Takeo province were administered by untrained paramedical and lay health care workers and their practices may differ from health care workers in the formal sector. There is growing consensus that improvements in injection safety and overuse in developing countries can be achieved \[[@B21]\]. Such efforts require multidisciplinary engagement and behavioral changes on the parts of both patients and health professionals, including those from the private healthcare sector \[[@B22]\]. Although large-scale interventions to improve the safety of therapeutic injections are limited, several demonstration projects have proven to be effective and relatively long lasting. Studies in Indonesia demonstrated a sustained 19% reduction in injection rates following \"interactional group discussions,\" an intervention that brings health care providers and patients together to discuss their preferences regarding injections and oral medications \[[@B18],[@B23]\]. In Burkina Faso, interventions targeted at providing sufficient quantities of single-use syringes and needles resulted in \>90% reductions in unsafe injections that have been sustained for at least 5 years \[[@B24]\]. Comprehensive interventions aimed at reducing overuse of injections, promoting rational use of oral and injectable medications, and reinforcing safe injection and waste disposal practices are likely to be cost-effective in Cambodia \[[@B25],[@B26]\]. While initial efforts in injection safety have been largely focused on infant immunization, Cambodia is now expanding this initiative to include curative care, with an emphasis on the private sector. A national plan of action has been developed to decrease injection use and promote injection safety in accordance with WHO guidelines. The plan aims to involve multidisciplinary partners and improve coordination among relevant departments in the Ministry of Health. This plan also includes demonstration projects that address specific issues related to injection use in the private sector. It is hoped that Cambodia\'s efforts to address its injection safety and overuse problems will provide an example for other countries that are beginning to confront this important public health issue. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= SV, SS, SS and JT adapted the WHO assessment design to Cambodia. SV, SS, and SS conducted the survey, supervised different aspects of its implementation and collected data. SV and JP synthesized analyses and led the writing. SG helped interpret findings and review drafts of the manuscript. YH conceptualized the assessment and reviewed drafts of the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/56/prepub> Acknowledgements ================ We are grateful to Drs. Hok Phalla and Thou Chourn of the Cambodian National Institute for Public Health, and Mrs. Mam Boravann of the Cambodian Department of Drugs and Foods for their collaboration in conducting the survey.

PubMed Central

2024-06-05T03:55:58.419656

2005-6-2

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173117/", "journal": "BMC Public Health. 2005 Jun 2; 5:56", "authors": [ { "first": "Sirenda", "last": "Vong" }, { "first": "Joseph F", "last": "Perz" }, { "first": "Srun", "last": "Sok" }, { "first": "Seiharath", "last": "Som" }, { "first": "Susan", "last": "Goldstein" }, { "first": "Yvan", "last": "Hutin" }, { "first": "James", "last": "Tulloch" } ] }

PMC1173118

Background ========== For years Israel has been considered to be endemic for viral hepatitis, with an incidence of reported cases five to ten times that observed in the United States \[[@B1]\]. Most cases of viral hepatitis in Israel are caused by the hepatitis A virus (HAV), and in the 1970\'s the Israeli Ministry of Health instituted a wide-scale program of post-exposure prophylaxis with immune serum globulin (ISG). In 1999 universal immunization with inactivated virus (Havrix; Smith Klein Beecham) was instituted for infants, the first dose being given at 18 months and the second at 24 -- 30 months of age \[[@B2]\]. The Israel Defense Forces (IDF) have instituted vaccination programs as well, first with ISG and, more recently, with the inactivated virus \[[@B3]\]. Vaccination programs, as well socioeconomic changes, have led to a significant reduction in the prevalence of HAV infection in Israel today \[[@B4],[@B5]\]. In the United States, the U.S. Advisory Committee on Immunization Practices has given clear recommendations for vaccinating communities with high rates of infection (\>20 cases per 100,000) or populations at risk for liver failure \[[@B6]\]. However, no guidelines exist for routine testing for HAV serology prior to vaccination, a policy which could potentially prevent unnecessary immunizations (both active and passive) in seropositive patients as well as target those who are seronegative for active immunization. The following study examined the current prevalence of anti-HAV serology in a selected Israeli population (patients presenting to a primary care clinic), evaluating the benefits of routine testing for populations at risk. Methods ======= The records of teenage and adult patients (aged 16--99 years; mean 33.9) from two primary care clinics in Jerusalem, Israel, were retrospectively searched for anti-HAV IgG serology. The clinics are affiliated with the Maccabi Health Care Services, one of four government-funded services available to all Israeli citizens. Patients had undergone general blood tests for a number of reasons, either for follow-up of existing illness (none of which were hepatic-related) or for routine health screening, including anti-HAV IgG serology. A total of 1,017 patient files were found which contained serological results of anti-HAV antibodies from blood analyzed in the health fund\'s central lab, which used a microparticle enzymatic assay (MEA; AxSYM System, Abbott Diagnostics). Patients who tested seropositive for HAV were then contacted (either in person or by phone) and asked regarding the following: demographic data (country of birth); whether or not they remembered being sick with hepatitis (i.e. clinical jaundice, with or without known serological evidence of active HAV infection), and when; and whether or not they had been immunized, either passively (ISG) or actively (inactivated virus), and when. Recall of past infection was based on patient reporting alone and not through review of medical or laboratory records. Data was compiled and analyzed using the Microsoft Excel program. Results ======= Prevalence of HAV seropositivity -------------------------------- As mentioned above, a total of 1,017 patient records (503 males and 514 females) were found to contain results of anti-HAV antibodies (from March 1998 to September 2003). The mean age of the study group was 33.9 years; 37.8 among males (range: 16--99) and 31.2 among females (range: 16--87). Among the recorded results, 692 (68.0%) were seropositive -- 354 of the male patients (70.4%) and 338 of the female patients (65.8%; p = 0.113). The rates of seropositivity increased with age (see Figure [1](#F1){ref-type="fig"}; p \< 0.005), and remained constant throughout the years of the study (p = 0.451). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Prevalence of positive HAV serology according to age (%) ::: ![](1471-2458-5-60-1) ::: Patients who were born in Middle Eastern countries other than Israel had the highest seropositive rate (91.3%), while those born in South America the lowest (44.1%; p \< 0.005. See Figure [2](#F2){ref-type="fig"}). Immigrants from the Eastern Bloc countries (former Soviet Bloc and Eastern Europe) had a significantly higher prevalence of anti-HAV antibodies than native Israelis (70.3% vs. 58.7%; p = 0.016). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Prevalence of positive HAV serology according to country of birth(%)**. \* **Eastern Bloc**= former USSR and Eastern European countries\*\* **Western Bloc**= North America and Western Europe ::: ![](1471-2458-5-60-2) ::: Recall of past infection and immunization ----------------------------------------- Of the 658 patients with anti-HAV antibodies, 456 were interviewed and asked regarding past infection and vaccinations (231 males and 225 females). Only 91 had any recollection of prior illness (20.0%) -- 51 of the males interviewed (22.1%) and 40 of the females (17.8%; p = 0.249), all of them at least 5 years prior to undergoing the serological testing. A total of 103 respondents remembered receiving post-exposure prophylaxis (passive immunization) with ISG. Those who were unaware of past infection were more likely to have been vaccinated with ISG than those who were aware (26.3% vs. 7.7%; p \< 0.005). The reasons given for receiving ISG are listed in Table [1](#T1){ref-type="table"}. Eight patients (2 male, 6 female) reported receiving the inactivated virus vaccine, of which none were aware of prior illness. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Reasons for Passive Immunization\* in Seropositive Patients ::: **[Reason]{.underline}** **[Male (n)]{.underline}** **[Female]{.underline}** **[Total]{.underline}** ------------------------------- ---------------------------- -------------------------- ------------------------- **Post-exposure prophylaxis** 25 25 50 **Travel to endemic country** 10 3 13 **Army Service** 14 3 17 **[Other]{.underline}** [17]{.underline} [6]{.underline} [23]{.underline} **Total** 66 37 103 **\* passive immunization**= immune serum globulin ::: Discussion ========== Although Israel is considered endemic for HAV infection, the prevalence of anti -HAV antibodies in our study population (68%) is much higher than that found in other studies, especially among the under-20 age group (62.8%). In a 1996 study of teenagers undergoing pre-draft evaluation, only 38.4% were seropositive \[[@B5]\], while serum samples from the Israeli Center for Disease Control\'s national serum bank in 1989 were seropositive for HAV in only 27% of 16 year-olds \[[@B4]\]. The relatively high prevalence of anti-HAV antibodies found in our study may be due to the fact that the study was conducted in a primary care clinic, with higher morbidity rates than the general population, though not liver-related. Another factor is the fact that both clinics are located in Jerusalem, one of Israel\'s poorest and most densely populated cities. Many of the patients were ultra-orthodox Yeshiva students who are at greater risk for HAV infection \[[@B7]\] and are not drafted into the IDF. At the same time, the higher rate of seropositivity among immigrants from Eastern bloc countries is consistent with the findings of an earlier study of this population \[[@B8]\], though another study of pre-draft teenagers did not find a significant difference between these immigrants and native Israelis \[[@B9]\]. Most of the seropositive patients in our study (80%) had no prior recollection of HAV infection. For 70% of children under the age of 6 (in communities with high rates of hepatitis A, 30--40% of children acquire infection before the age of 5 6) HAV infection is a self-limited, subclinical disease \[[@B10]\]. Among older children and adults, infection is usually symptomatic, with jaundice occurring in more than 70% of patients \[[@B11]\]. Signs and symptoms usually last less than 2 months, although 10--15% of symptomatic persons have prolonged or relapsing disease lasting up to 6 months \[[@B12]\]. Case fatality rates also increase with age, rising from 1.5/1000 among children less than 5 years old to 27/1000 after age 50 \[[@B13]\], making vaccination of unimmunized adults that much more important. Both ISG and inactivated HAV vaccines are safe, though not without potential side effects both locally and systemically \[[@B6],[@B14]\]. Aside from having to undergo two injections within a span of 6 months to a year, adult patients in Israel are required to purchase the active vaccine themselves. Seropositive patients in our study who were unaware of their immune status were more likely (by a factor of 3.4) to have unnecessarily received ISG (unnecessarily since they most likely have lifetime immunity) than those who were aware of past infection. This is probably an underestimate, since both post-exposure and pre-induction vaccinations have been implemented in Israel for nearly thirty years, and it is likely that many have forgotten receiving immunizations in the distant past. For the eight patients who received the inactivated virus without prior serological testing, it is possible (and even probable) that they too were immunized unnecessarily. Now it is too late to know whether their current seropositive status is a result of past occult infection (with lifetime immunity) or active immunization. These patients will now have to receive periodic booster vaccinations, which may have been unnecessary if prior immune status had been checked. Testing for HAV serology prior to active immunization has been shown to be both cost-effective and economically valid for Israeli travelers, assuming a cost of \$130 for vaccination (which is subsidized but not free for the patient) and \$30 for the IgG test (free for the patient) \[[@B15]\]. This may be true for the general population as well. Conclusion ========== The recent adoption of a nationwide infant HAV immunization policy in Israel has also been found to be both medically and economically justifiable \[[@B16]\]. However, this program is aimed at the infant population, for the purpose of preventing early childhood infection. Screening the adult population in endemic countries like Israel would enable seronegative adults to receive the inactivated virus, thus preventing morbidity and mortality resulting from infection. For the seropositive adult population, unnecessary immunizations could be avoided. Asking patients whether or not they need to be vaccinated is, at best, unreliable, since most cases of HAV infection occur in childhood and are asymptomatic, and even in cases where jaundice is present the cause may not necessarily be HAV. The health establishment in Israel needs to investigate the benefits of routing testing for anti-HAV serology, especially among the high-risk adult population who were not actively immunized during childhood or military service. Such routine testing is feasible, especially in an era where it is accepted practice to conduct routine screening tests for a number of preventive measures. Competing interests =================== The author(s) declare that they have no competing interest. Authors\' contributions ======================= The design and implementation of the study, as well as statistical analysis of the data, were all done by the author (NS). Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/60/prepub>

PubMed Central

2024-06-05T03:55:58.422936

2005-6-6

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173118/", "journal": "BMC Public Health. 2005 Jun 6; 5:60", "authors": [ { "first": "Noah", "last": "Samuels" } ] }

PMC1173119

Background ========== Ethiopia stands among the world\'s top 22 tuberculosis (TB) high-burden countries, with an estimated annual incidence of 250 TB cases/10^5^population \[[@B1],[@B2]\]. In the Southern Ethiopia Regional State (SNNPRS), TB is among the leading causes for sickness and death \[[@B3]\]. As in many other resource-constrained settings, treatment outcomes for tuberculosis have not been satisfactory \[[@B4]\], mainly due to poor treatment compliance and low coverage of short course chemotherapy (SCC). Delays in the diagnosis and treatment initiation, the devastating HIV/AIDS epidemic and the potential threat of anti-tuberculosis drug resistance represent serious threats to the TB control effort in the region. The HIV co-infection among TB patients in the region is estimated at 19% \[[@B5]\]. Directly Observed Treatment, Short-course (DOTS) was introduced in the region in 1996. However, the impact of DOTS and its effectiveness in the regional context has not been assessed yet. Global reviews \[[@B6]\] and reports from Asian, African and East European countries \[[@B7]-[@B9]\] have favourable implications for the DOTS strategy. On the contrary, some randomised controlled studies have failed to establish the superiority of the most notable component of this strategy, direct observation of treatment, over the conventional non-observed treatment in improving treatment outcomes \[[@B10],[@B11]\]. Nevertheless, DOTS is widely accepted and practised, with some TB high-burden countries achieving almost full coverage of their populations \[[@B2]\]. DOTS is equally effective in curing TB patients with HIV co-infection \[[@B9],[@B12]\] and hence, its importance in the era of the HIV/AIDS epidemic. With this study, we aimed to assess the results and trends in the expansion of DOTS in Hadiya Zone, Southern Ethiopia. Methods ======= Study setting ------------- We did the study in Hadiya zone, Southern Ethiopia (see map in Figure [1](#F1){ref-type="fig"}), with a population of 1.2 million \[[@B13],[@B14]\]. In addition to fifteen diagnostic centres, five health centres and 21 health stations provide treatment for TB patients referred or transferred from the diagnostic centres, thus making the number of treatment centres 41. All treatment units have standard Unit Registers from the National Tuberculosis and Leprosy Control Programme (TLCP). TB drugs used in a combination of two or more in the region included isoniazid (H), rifampicin (R), pyrazinamide (Z), ethambutol (E), streptomycin (S) and thioacetazone (T). The TB and Leprosy control programs were run as two separate vertical programs funded and managed mainly by the All African Leprosy and Rehabilitation Training Centre (ALERT) until the end of 1996 when the National and Regional TLCP took over the responsibility. The zonal TLCP gets administrative and technical support from the regional TLCP, and monitors the activities in the diagnostic and treatment centres through its seven woreda (district) offices. The woreda health offices have standard district TB registers from the National TLCP, and are responsible for the distribution of drugs and reagents, supervision of day to day activities in the treatment units, collection of slides for quality control and record keeping. Quarterly reports on cases detected and treatment outcomes from the woredas are gathered at the zonal level and sent to the regional TLCP. In 1994, only Hossana Hospital provided unobserved SCC for critically sick smear-positive pulmonary TB (PTB+), miliary TB and tuberculosis meningitis cases. DOTS was initiated in 1996 in two health facilities and gradually expanded, first to the health centres in all woredas and then to the health stations. Details of treatment regimens for various categories of patients under DOTS are given in the NTLCP manual \[[@B15]\]. In brief, all new patients were treated with SCC (two months on RHZ +/- E or S, followed by 6 months on EH or RH), and were required to take their medications under direct supervision by the health workers at least during the intensive phase of treatment. During this phase, only critically ill patients were hospitalised, while the rest received treatment on ambulatory basis. Re-treatment cases were treated with SERHZ for 2 months, ERHZ for 1 month and ERH for 5 months. In non-DOTS areas treatment with SCC was limited to critically sick PTB+, TB meningitis and miliary TB cases while others received LCC (two months on EH +/-S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). Design and data collection -------------------------- This is a retrospective trend analysis. The Unit Registers reviewed contain basic information such as patient\'s age, sex, address, category, TB type, drug regimen, date treatment started, treatment follow-up, follow-up sputum result and treatment outcomes. We visited all 41 TB treatment centres during July-November 2002. We checked when DOTS was initiated in each health institution and looked for the availability of TB drugs and reagents during the time of visit. We then reviewed the Unit Registers and entered the data to the computer. At three peripheral treatment centres where the unit registers could not be found, the respective district TB registers were reviewed. For each TB case, we copied the data from the registers to a computer program, SPSS for windows \[[@B16]\], according to the standard definitions of the National TLCP \[[@B15]\] on disease classification, patient categories, treatment regimens and treatment outcomes. To ensure the quality of data entered into the computer database, two people independently cross-checked each entry. Data analysis and Statistics ---------------------------- The data were analysed using SPSS for Windows version 11.0 \[[@B16]\]. For categorical data, we used proportions with 95% confidence intervals, Odds ratio and Chi-square test to compare different groups. Multivariate analysis using logistic regression model was used to analyse the association between treatment outcomes and potential predictor variables. We set the level of statistical significance at 5%. Treatment outcomes were analysed for the years 1994--2000 because a considerable proportion of patients registered during 2001 were still on treatment during the time of data collection. (The Ethiopian fiscal year goes from July to June. Fiscal year 2001, for example, goes from July 2001 through June 2002, and our study was commenced in July 2002). Patients with unrecorded treatment outcome were analysed as defaulters. Results ======= Patient registration and case notification ------------------------------------------ A total of 19971 tuberculosis patients, 11138 (55.8%) males and 8819 (44.2%) females were registered between 1994 and 2001 with the mean (SD) age of 25.6 (13.2) years. Forty-six percent (n = 9232) of the patients were PTB+ cases. Overall, 18687 (93.6%) patients were registered as new cases, 558 (2.8%) as transferred in, 273 (1.4%) as return after default, 142 (0.7%) as failure and 139 (0.7%) as relapse cases whereas patient category was not recorded for 172 (0.9%) cases. Table [1](#T1){ref-type="table"} shows the general characteristics of the patients. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### General Characteristics of the study subjects (n = 19971), 1994--2001 ::: **Characteristics** **Number** **Percent** --------------------------------------- ------------ ------------- ***Age group (years)*** 0--14 3356 16.8 15--24 6262 31.4 25--34 5366 26.9 35--44 2706 13.5 45--54 1248 6.2 55--64 501 2.5 ≥ 65 220 1.1 Unknown 312 1.6 ***Sex*** Male 11138 55.8 Female 8819 44.2 Not mentioned 14 0.1 ***Patient Category*** New 18687 93.6 Transferred-in 558 2.8 Return after default 273 1.4 Failure 142 0.7 Relapse 139 0.6 Unknown 172 0.9 ***TB Classification*** Pulmonary positive 9232 46.2 Pulmonary negative 4225 21.2 Extra-pulmonary 6453 32.3 Unknown 61 0.3 ***Treatment Centre*** Hossana Hospital 5362 26.8 Lemmo district health facilities 4011 20.1 Shashogo district health facilities 1208 6.0 Misha district health facilities 2660 13.3 Gibe district health facilities 1984 9.9 Soro district health facilities 2313 11.6 Duna district health facilities 216 1.1 Badewacho district health facilities 2217 11.1 ::: The proportion of women among TB patients registered for treatment remained at the range of 40--45% across the years and at the range of 43--47% across the age groups below 45. However, among patients older than 45 years, the proportion of women was significantly lower (31%; 95%CI 29--33) compared to those below 45 (46%; 95%CI 45--47). Among new cases, 46% (n = 8557) were PTB+ patients. Except for the years 1997 (38%; 95%CI 36--40) and 2000 (56%; 95%CI 54--58), this proportion showed little variations in the range of 40--50%. Table [2](#T2){ref-type="table"} shows a trend in the case notification of PTB+ over the study period. The proportion of expected PTB+ incident cases notified increased from 45% in 1994 to 116% in 1999, and declined to 67% in 2001. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Trends in case notification of smear-positive pulmonary TB, Hadiya Zone, 1994--2001 ::: Year Zonal Population\* New PTB+ cases Reported (n = 8558) Case Notification/ 10^5^persons/year Proportion of estimated PTB+ incident cases notified, %\*\* --------- -------------------- ------------------------------------ -------------------------------------- ------------------------------------------------------------- 1994 1004000 512 49 45.2 1995 1070160 1025 96 87.9 1996 1101195 1162 106 96.8 1997 1133129 946 84 76.6 1998 1165990 1277 110 100.5 1999 1199804 1510 126 115.5 2000 1234598 1267 103 94.2 2001^§^ 1174118 858 73 67.0 \*Projected from the 1994 Population and Housing Census, CSA \[13\] \*\*Estimated Incidence of PTB+ = 109/10^5^persons/year based on the WHO estimates of the Global Burden of Tuberculosis for 1997 \[1\] ^§^Population adjusted as an area with about 70--80 thousand population was taken away from Hadiya to the neighbouring Silti zone between years 2000 and 2001 PTB+: smear-positive pulmonary tuberculosis ::: Treatment given --------------- Forty percent (n = 7919) of patients received SCC. Treatment regimen was not recorded for 981 (5%) cases. Table [3](#T3){ref-type="table"} presents trend in treatment regimens across the years by TB classification. The proportion of patients treated with SCC increased from 7% in 1994 to 58% in 1999, and 97% in 2001 (χ^2^~trend~, p \< 0.001). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Trend in treatment regimens for the different categories of patients (n = 19970).\* ::: **Category** **1994 n (%)** **1995 n (%)** **1996 n (%)** **1997 n (%)** **1998 n (%)** **1999 n (%)** **2000 n (%)** **2001 n (%)** -------------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- **All TB cases** 1106 2201 2895 2742 3388 3253 2435 1950 SCC 73 (7) 311 (14) 288(10) 471 (17) 927 (27) 1881 (58) 2085 (86) 1883 (97) LCC 1011 (91) 1684 (77) 2360 (82) 2097 (77) 2292 (68) 1302 (40) 304 (12) 20 (1) No record 22 (2) 206 (9) 247 (8) 174 (6) 169 (5) 70 (2) 46 (2) 47 (2) **New PTB+** 512 1025 1162 946 1277 1510 1267 858 SCC 9 (2) 111 (11) 94 (8) 232 (24) 537 (42) 1170 (78) 1187 (94) 837 (97) LCC 496 (97) 780 (76) 938 (81) 611 (65) 660 (52) 321 (21) 68 (5) 7 (1) No record 7 (1) 134 (13) 130 (11) 103 (11) 80 (6) 19(1) 12 (1) 14 (2) **New PTB-** 258 385 402 382 807 772 487 393 SCC 12 (5) 66 (17) 90 (22) 109 (28) 225 (28) 395 (51) 402 (83) 379 (96) LCC 242 (94) 286 (74) 248 (62) 251 (66) 543 (67) 362 (47) 73 (15) \- No record 4 (1) 33 (9) 64 (16) 22 (6) 39 (5) 15 (2) 12 (2) 14 (4) **New EPTB** 257 668 1177 1176 1124 721 528 570 SCC 19 (8) 73 (11) 57 (5) 70 (6) 97 (9) 197 (27) 385 (73) 559 (98) LCC 232 (90) 565 (85) 1081 (92) 1067 (91) 993 (88) 513 (71) 138 (26) 6 (1) No record 6 (2) 30 (4) 39 (3) 39 (3) 34 (3) 11 (2) 5 (1) 5 (1) **Relapse** 8 13 21 29 15 19 19 15 SCC 2 (25) 10 (77) 12 (57) 21 (72) 13 (87) 17 (90) 19 (100) 14 (93) LCC 4 (50) 2 (15) 9 (43) 6 (21) 2 (13) 2 (10) \- \- No record 2 (2) 1 (8) \- 2 (7) \- \- \- 1 (7) **After default** 35 56 27 45 33 25 21 31 SCC 5 (14) 14 (25) 15 (56) 20 (45) 15 (46) 21 (84) 17 (81) 30 (97) LCC 30 (86) 39 (70) 11(41) 23 (51) 18 (54) 4 (16) 4 (19) \- No record \- 3 (5) 1 (3) 2 (4) \- \- \- 1 (3) **After failure** 28 38 17 17 13 19 8 2 SCC 26 (93) 35 (92) 15 (88) 13 (77) 11 (85) 14 (74) 8 (100) 2 (100) LCC 2 (7) 3 (8) 2 (12) 4 (23) 2 (15) 5 (26) \- \- No record \- \- \- \- \- \- \- \- **Transferred in** 6 3 65 131 89 120 86 57 SCC \- \- 3 (5) 3 (2) 22 (25) 43 (36) 53 (62) 38 (67) LCC 5 (83) 2 (67) 55 (84) 124 (95) 58 (65) 57 (47) 19 (22) 7 (12) No record 1 (17) 1 (33) 7 (11) 4 (3) 9 (10) 20 (17) 14 (16) 12 (21) **Missing info** 2 13 17 10 20 46 13 29 SCC \- 2 (15) 1 (5) 3 (30) 3 (15) 11 (24) 9 (69) 29 (100) LCC \- 7 (54) 12 (71) 6 (60) 11 (55) 30 (65) 2 (16) \- No record 2 4 (31) 4 (24) 1 (10) 6 (30) 5 (11) 2 (15) \- \* Percentages are out of the total cases registered in the same category during the year. PTB+ = smear-positive pulmonary tuberculosis; PTB- = smear-negative pulmonary tuberculosis; EPTB = extra-pulmonary tuberculosis SCC = short course chemotherapy (2 months on RHZ+/- E or S, followed by 6 months on EH or RH) LCC = long course chemotherapy (2 months on EH +/- S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). R = Rifampicin,; H = Isoniazid; Z = Pyrazinamide; E = Ethambutol; S = Streptomycin; T = Thioacetazone. ::: During 1994--1997, more PTB- patents received SCC compared to PTB+ patients (19% vs. 12%; p \< 0.05) and EPTB patients (19% vs. 7%; p \< 0.05) among new cases of TB (Table [3](#T3){ref-type="table"}). This trend was reversed after 1997 when more PTB+ patients (76%) were put on SCC, followed by PTB- (57%) and EPTB (42%). There was no significant difference in the treatment regimens across the age groups and between the two sexes. Follow-up --------- Of the 8557 new PTB+ cases, sputum examination was repeated at the end of two months treatment for 60% (n = 5112), and 7.8% (n = 401) remained positive for AFB. At the end of five months treatment, 2747 (32%) patients had their sputum examined for AFB, and 78 (2.8%) remained smear-positive. 1823 (21%) patients had their sputum examined for AFB at (a month prior to) treatment completion, and 1.4% (n = 26) remained smear-positive. There has been a continuous and significant decline over time in the sputum positivity at two months irrespective of treatment regimen (Table [4](#T4){ref-type="table"}). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Trends in the follow-up smear results for new smear positive pulmonary tuberculosis ::: **2^nd^month** **5^th^month** End of treatment --------------------- ---------------- ---------------- ------------------ ---------- ---------- ---------- **Cases on SCC** 1994 6 (67) 0 5 (56) 0 3 (33) 0 1995 79 (71) 18 (23%) 53 (48) 4 (8%) 33 (30) 0 1996 68 (72) 15 (22%) 38 (40) 2 (5%) 25 (27) 0 1997 155 (67) 14 (9%) 94 (41) 0 74 (32) 2 (3%) 1998 342 (64) 24 (7%) 183 (34) 0 140 (26) 0 1999 824 (70) 52 (6%) 411 (35) 7 (2%) 365 (31) 2 (0.5%) 2000 817 (69) 43 (5%) 439 (37) 9(2%) 316 (27) 2 (1%) 2001^§^ 527 (63) 13 (3%) 171 (20) 1 (0.6%) 121 (14) 0 Cases on LCC 1994 330 (66) 45 (14%) 154 (31) 19 (12%) 86 (18) 5 (6%) 1995 508 (65) 75 (15%) 354 (45) 15 (4%) 179 (23) 4 (2%) 1996 564 (60) 59 (11%) 385 (41) 7 (2%) 223 (24) 4 (2%) 1997 302 (49) 15 (5%) 188 (31) 5 (3%) 86 (14) 3 (4%) 1998 263 (40) 19 (7%) 127 (19) 7 (6%) 80 (12) 2 (3%) 1999 182 (57) 7 (4%) 86 (27) 1 (1%) 68 (21) 2 (3%) 2000 46 (68) 1 (2%) 9 (13) 0 4 (6) 0 2001^§^ 2 (29) 1 2 (29) 0 1 (14) 0 **Unknown regimen** 1994 0 0 2 (29) 1 2 (29) 0 1995 9 (7) 0 4 (3) 0 3 (2) 0 1996 25 (19) 0 9 (7) 0 3 (2) 0 1997 23 (22) 0 11 (11) 0 3 (3) 0 1998 27 (34) 0 19 (24) 0 5 (6) 0 1999 4 (21) 0 1 (5) 0 1 (5) 0 2000 4 (33) 0 1 (8) 0 0 0 2001^§^ 5 (36) 0 17) 0 2 (14) 0 sm+ = sputum smear positive for acid-fast bacilli out of examined. SCC = short course chemotherapy; LCC = long course chemotherapy; AFB = acid-fast bacilli. ^§^Follow-up smears for the 5^th^month and end of treatment were not complete for the year 2001 as a considerable number of patients were still on treatment during the time of data collection. ::: Sputum positivity at the completion of fifth month treatment remained at the range of 0--2% for those on SCC, with the exception of 1995 and 1996 that exhibited 8% and 5% respectively. For those on LCC, smear positivity at 5^th^month decreased from 12% in 1994 to 1% in 1999. Overall, patients on LCC contributed more to the 5^th^month and end of treatment smear-positivity (treatment failure), than those on SCC (p \< 0.05; table [4](#T4){ref-type="table"}). Treatment results ----------------- We analysed treatment outcomes for the years 1994--2000 and, as a result, we evaluated 16943 (85% of the total) cases. Of these, 8268 (49%) successfully completed treatment, 3151 (18.6%) defaulted, 110(0.6%) had treatment failure, 446 (2.6%) died and 2000 (10%) were transferred out, while 2968 (17.5%) had no record of treatment outcomes. Assuming that patients with unrecorded treatment outcome had all defaulted from treatment, a total of 6119 (36.1%) patients defaulted. Among new PTB+ (and all TB) patients, treatment success was higher among patients on SCC, while default and failure rates were higher among those on LCC (Table [5](#T5){ref-type="table"}). Death in all categories was higher among those on SCC. Among new cases on SCC, PTB+ cases had 1.8 (65% vs.37%) and 1.4 (65% vs. 48%) times higher rate of treatment success compared to PTB- and EPTB cases, respectively (p \< 0.001). Female PTB+ patients had significantly higher treatment success (58% vs. 54%; p = 0.001) and lower defaulter rate (26% vs. 30%; p \< 0.001) than males. ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Treatment outcomes for different categories of patients on short and long course regimens (n = 16943) ::: **New cases** --------------------- --------------- ---------- ----------- --------- --------- --------- ---------- **Patients on SCC** Treatment success 2166 (65) 475 (37) 436 (48) 62 (66) 49 (46) 65 (53) 79 (64) Default\* 726 (22) 483 (37) 320 (36) 13 (14) 45 (42) 31 (25) 33 (26) Failure 25 (0.7) \- \- 2 (2) 1 (1) 6 (5) 3 (2) Death 117 (3.5) 72 (5) 24 (3) 11 (12) 5 (5) 2 (1) 2 (2) Transferred out 306 (9) 269 (21) 118 (13) 6 (6) 7 (7) 18 (15) 7 (6) **Patients on LCC** Treatment success 1986 (51) 772 (39) 1932 (42) 14 (56) 32 (25) 11 (61) 189 (59) Default\* 1365 (35) 972 (48) 1922 (42) 8 (33) 79 (61) 2 (11) 24 (8) Failure 66 (2) \- \- 1 (4) 2 (2) 3 (17) 1 (0.3) Death 107 (3) 57 (3) 37 (1) 2 (8) 2 (2) 1 (5) 7 (2) Transferred out 350 (9) 204 (10) 697 (15) \- 14 (11) 1 (6) 3 (1) \* Patients with missing treatment outcome information were assumed to have defaulted and analysed as defaulters. PTB+ = smear positive pulmonary tuberculosis; PTB- = smear negative pulmonary tuberculosis; EPTB = extra-pulmonary tuberculosis SCC = short course chemotherapy (2 months on RHZ+/- E or S, followed by 6 months on EH or RH) LCC = long course chemotherapy (2 months on EH +/-S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). R = Rifampicin,; H = Isoniazid; Z = Pyrazinamide; E = Ethambutol; S = Streptomycin; T = Thioacetazone. ::: Patients registered as \"return after default\", had significantly lower treatment success compared to new cases, both for all TB (34% vs. 49%; p \< 0.001), and for PTB+ (34% vs. 58%; p \< 0.001). Furthermore, return after default cases were more likely to default again compared to new cases among PTB+ patients (53% vs. 29%; p \< 0.001) and all TB patients (53% vs. 36%; p \< 0.001). Treatment failure was five times higher among cases that received re-treatment for previous failure compared to new cases of PTB+ (6.4% vs. 1.2%; p \< 0.001); however, both groups had comparable treatment success, mainly due to a lower defaulter rate among failure cases (24% vs. 29%). Patients for whom follow-up sputum smear remained positive at the second month during treatment had significantly lower treatment success than those with negative smear result (53% vs. 74%; p \< 0.001), mainly as a result of higher defaulter (24% vs. 18%) and failure (10% vs. 1%) rates among the former group. Table [6](#T6){ref-type="table"} presents treatment outcomes for new cases of TB across the years. A steady rise in treatment success and decline in defaulter rate was noticed particularly among patients on SCC. Treatment failure showed a remarkable decline in both SCC and LCC groups. Death rate did not show much variation across the years. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Treatment outcomes across the years for new cases of tuberculosis ::: **1994 n (%)** **1995 n (%)** **1996 n (%)** **1997 n (%)** **1998 n (%)** **1999 n (%)** **2000 n (%)** -------------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- ---------------- PTB+ on SCC Treatment success 5 (42) 40 (34) 45 (38) 122 (47) 338 (60) 811 (69) 805 (74) Default\* 5 (42) 61 (51) 50 (42) 77 (30) 114 (20) 225 (19) 194 (17) Failure \- 3 (3) 2 (2) 2 (1) \- 8 (1) 10 (1) Death \- 4 (3) 2 (2) 10 (4) 24 (4) 44 (4) 33 (3) Transferred out 2 (17) 11 (9) 20 (17) 46 (18) 89 (16) 88 (7) 50 (5) PTB+ on LCC Treatment success 188 (38) 384 (48) 564 (56) 271 (45) 308 (52) 231 (72) 40 (68) Default\* 190 (38) 298 (38) 304 (30) 246 (40) 240 (40) 72 (22) 15 (25) Failure 23 (5) 18 (2) 8 (1) 8 (1) 6 (1) 3 (1) \- Death 17 (3) 21 (3) 18 (2) 20 (3) 19 (3) 9 (3) 3 (5) Transferred out 78 (16) 71 (9) 107 (11) 63 (10) 24 (4) 6 (2) 1 (2) PTB- on SCC Treatment success 1 (8) 15 (23) 18 (20) 32 (29) 67 (30) 153 (39) 189 (47) Default\* 8 (67) 40 (60) 67 (74) 55 (50) 85 (38) 116 (29) 112 (28) Death 1 (8) 6 (9) 2 (2) 10 (9) 8 (4) 25 (6) 20 (5) Transferred out 2 (17) 5 (8) 3 (3) 12 (11) 65 (29) 101 (26) 81 (20) PTB- on LCC Treatment success 107 (44) 88 (31) 84 (34) 78 (31) 199 (37) 182 (50) 34 (47) Default\* 101 (42) 163 (57) 127 (51) 141 (57) 280 (51) 135 (37) 25 (34) Death 4 (2) 5 (2) 6 (2) 5 (2) 16 (3) 20 (6) 1 (1) Transferred out 30 (12) 30 (11) 31 (13) 27 (11) 48 (9) 25 (7) 13 (18) EPTB on SCC Treatment success 4 (21) 19 (26) 25 (44) 29 (32) 43 (44) 101 (51) 215 (56) Default\* 11 (58) 43 (59) 29 (51) 30 (33) 29 (30) 68 (35) 110 (28) Death 2 (11) 6 (8) 1 (2) 3 (4) 1 (1) 4 (2) 7 (2) Transferred out 2 (11) 5 (7) 2 (4) 28(31) 24 (25) 24 (12) 53 (14) EPTB on LCC Treatment success 80 (35) 274 (49) 559 (52) 387 (36) 341 (34) 222 (43) 69 (50) Default\* 122 (52) 203 (36) 321 (30) 412 (39) 553 (56) 253 (49) 58 (42) Death 2 (1) 3 (1) 3 (0.3) 7 (1) 8 (1) 14 (3) \- Transferred out 28 (12) 85 (15) 198 (18) 260 (24) 91 (9) 24 (5) 11 (8) \* Patients with missing information on treatment outcome were assumed to have defaulted and analysed as defaulters. PTB+ = smear positive pulmonary tuberculosis; PTB- = smear negative pulmonary tuberculosis; EPTB = extra-pulmonary tuberculosis SCC = short course chemotherapy (2 months on RHZ+/- E or S, followed by 6 months on EH or RH) LCC = long course chemotherapy (2 months on EH +/-S, followed by 10 months on EH; TH had also been used as an alternative to EH until its recent withdrawal). R = Rifampicin,; H = Isoniazid; Z = Pyrazinamide; E = Ethambutol; S = Streptomycin; T = Thioacetazone. ::: When we adjusted the outcome measures for treatment results by various potentially confounding variables, significantly higher treatment success was exhibited among female patients, those 15--24 years, patients treated with SCC, those on re-treatment for relapse, PTB+ cases, those treated during 2000 and those treated in the Lemmo district health facilities (Table [7](#T7){ref-type="table"}). Meanwhile, male patients, patients aged 45--54, PTB- cases, those on LCC, return after default cases and those treated in Duna district health facilities exhibited significantly higher default rate. ::: {#T7 .table-wrap} Table 7 ::: {.caption} ###### Adjusted odds ratios for various factors that might affect treatment outcomes among registered tuberculosis patients ::: **Treatment success (all TB)** **Default (all TB)** --------------------------- -------------------------------- ---------------------- --------- ------ ------------------- --------- **Sex** Male 45.5 1.00 37.5 1.00 Female 49.4 1.15(1.08--1.23) \<0.001 34.0 0.88(0.82--0.94) \<0.001 **Age group (years)** 0--14 42.6 0.83 (0.75--0.92) \<0.001 41.2 1.17 (1.06--1.30) 0.002 15--24 (reference group) 51.1 1.00 33.4 1.00 25--34 47.5 0.84(0.77--0.91) \<0.001 34.4 1.07 (0.98--1.17) 0.14 35--44 45.7 0.76 (0.68--0.84) \<0.001 36.1 1.14 (1.03--1.28) 0.02 45--54 44.9 0.69 (0.60--0.79) \<0.001 37.7 1.26 (1.09--1.46) 0.002 55--64 51.4 1.04 (0.84--1.29) 0.71 35.4 1.04 (0.81--1.27) 0.90 ≥ 65 37.4 0.67 (0.48--0.93) 0.02 43.7 1.27 (0.92--1.726 0.15 **Patient category** New 47.0 1.00 35.9 1.00 Transferred-in 59.5 1.09 (0.89--1.36) 0.40 35.7 1.19 (0.95--1.48) 0.13 Return after default 33.9 0.68 (0.51--0.92) 0.01 52.9 1.57 (1.19--2.07) 0.002 Treatment failure 54.3 1.30 (0.91--1.86) 0.15 23.6 0.72 (0.48--1.09) 0.12 Relapse 62.9 1.51(1.01--2.26) 0.05 16.9 0.50 (0.30--0.82) 0.006 **TB type** Pulmonary positive 55.3 1.00 28.7 1.00 Pulmonary negative 37.5 0.55 (0.50--0.60) \<0.001 43.7 2.06 (1.87--2.26) \<0.001 Extra-pulmonary 42.4 0.51 (0.47--0.55) \<0.001 41.1 1.94 (1.78--2.11) \<0.001 **Treatment regimen** Short course chemotherapy 55.5 1.00 27.7 1.00 Long course chemotherapy 45.0 0.68 (0.62--0.75) \<0.001 40.9 1.46 (1.33--1.61) \<0.001 **Treatment Centre\*\*** Hossana Hospital 27.0 1.00 40.0 1.00 Lemmo district HF 60.4 5.14 (4.62--5.72) \<0.001 20.6 0.28 (0.25--0.31) \<0.001 Shashogo district HF 50.8 2.81 (2.41--3.28) \<0.001 44.1 1.18 (1.01--1.37) 0.04 Misha district HF 60.3 4.78 (4.23--5.39) \<0.001 33.0 0.58 (0.51--0.65) \<0.001 Gibe district HF 60.7 4.78 (4.20--5.45) \<0.001 26.2 0.51 (0.45--0.59) \<0.001 Soro district HF 50.5 2.70 (2.39--3.06) \<0.001 42.5 1.11 (0.98--1.26) 0.09 Duna district HF 30.2 0.77 (0.55--1.08) 0.12 67.3 5.18 (3.71--7.24) \<0.001 Badewacho district HF 41.1 1.64 (1.44--1.86) \<0.001 54.4 2.11 (1.86--2.40) \<0.001 **Year of treatment** 1994 38.4 1.00 42.7 1.00 1995 40.6 1.18 (0.99--1.39) 0.06 39.9 1.17 (1.00--1.37) 0.05 1996 48.7 1.29 (1.10--1.51) 0.002 33.3 0.83 (0.71--0.97) 0.02 1997 38.9 0.72 (0.61--0.85) \<0.001 39.5 1.08 (0.93--1.27) 0.32 1998 41.9 0.74 (0.63--0.87) \<0.001 43.8 1.20 (1.03--1.40) 0.02 1999 56.7 1.37 (1.16--1.62) \<0.001 30.0 0.59 (0.5--0.70) \<0.001 2000 59.8 1.41 (1.18--1.69) \<0.001 25.8 0.54 (0.45--0.65) \<0.001 OR = odds ratio; CI = confidence intervals; HF = Health facilities \* All the variables in the table are included in the model. \*\* All treatment centres in a district grouped together as a unit. ^1^Percentage out of total registered for treatment. Patients with missing outcome record were assumed to have defaulted and analysed as defaulters. ::: Trend over time --------------- DOTS was initiated in 1996 in a hospital and one health centre, with potential population coverage of 25% (defined as population living within 2 hours walking distance from a health facility; estimated at 250,000 for the hospital and 25000 for the health centre). The number of health facilities providing DOTS increased to 10 in 1997, 30 in 1999 and 41 in 2001, making the population coverage by DOTS 31%, 58% and 75% respectively. The proportion of patients treated with SCC increased from 7% in 1994 to 27% in 1998, 58% in 1999, and 97% in 2001. 95% (39/41) of the treatment centres had at least isoniazid, rifampicin, pyrazinamide and ethambutol at the time of visit for data collection. Reagents for Acid-fast stain were available in nine of the 13 diagnostic centres functioning at the time. Simultaneously, treatment success for new PTB+ patients (on SCC and LCC together) increased from 38% in 1994 to 56% in 1998, 70% in 1999 and 73% in 2000 (χ^2^~trend~, p \< 0.001). Defaulting among new PTB+ patients declined from 38% in 1994 to 30% in 1998, 20% in 1999 and 18% in 2000 (χ^2^~trend~, p \< 0.001). Treatment failure decreased from 5% in 1994 to 1% in 2000. The proportion of reported deaths remained unchanged over years with some variations in the range of 2--5%. Treatment at small and large centres ------------------------------------ Thirty-one out of 40 (77.5%) peripheral treatment centres had significantly higher treatment success, and 28 (70%) of them had significantly lower defaulter rate compared to the zonal hospital (details not shown). Further comparison by districts revealed that six of the seven rural districts had significantly better treatment success while three out of seven had lower defaulter rate than the zonal hospital (Table [7](#T7){ref-type="table"}). Discussion ========== The results of our study show that there has been a continuous increase in treatment success and eventual decline in defaulter and failure rates in parallel to the expansion and decentralisation of DOTS to the lower treatment units. Treatment with SCC increased from a very low coverage in 1994 to almost full coverage in 2001 partly due to the expansion of DOTS (with SCC as one of its elements), and partly due to the regional TLCP commitment to SCC both in DOTS and non-DOTS areas. However, trends in case detection and notification showed inconsistencies over years. In the hierarchy of study designs, observational studies come after the randomised controlled trials \[[@B17]\]. Nevertheless, in the real field condition, where randomised trials are not logistically feasible, such observational study designs may provide important and valid information. Our study design does not allow comparison of treatment outcomes between the DOTS and non-DOTS areas. However, this study gives us a valuable information on the program performance before and along the course of expanding DOTS in this resource-constrained rural setting. Four years after the introduction of DOTS in the zone, it has been possible to achieve treatment success of 73% in 2000 for new PTB+ patients. Faring well towards the WHO/IUATLD recommended 85%, this confirms the finding of other studies \[[@B7],[@B18]\] that the DOTS strategy works well in resource-constrained settings with low overall health coverage. Increased coverage by SCC, improved access to care through decentralisation of the service and improved patient follow-up with the introduction of DOTS have most likely played a significant role in improving the treatment outcomes. Higher treatment success and lower default among female patients confirms previous study findings \[[@B11]\]. However, the proportion of females among patients registered for TB treatment was found to be consistently low, and it was exceptionally lower among patients older than 45 years. This may reflect a genuine gender difference in the TB epidemiology \[[@B19]-[@B21]\]. However, the possibility of this being a reflection of gender differentials in access to health care within the society may need to be ruled out by further studies. Patients treated at the peripheral treatment centres exhibited better treatment outcomes compared to those treated in the zonal hospital, which implies better follow-up of cases or better access to the TB care services. Despite significant decline in defaulter rate over time, patients that received re-treatment as \"return after default\", were much more likely to default again compared to new patients. This group of defaulters seems refractory to the conventional approach of treatment supervision; social and cultural factors that might play a role need to be explored. More death among patients on SCC was an indication of the policy that severely sick patients and those co-infected with HIV be a priority for treatment by SCC and these groups were more likely than others to die. Though lower among patients on SCC, the high proportion of failure among patients retreated for previous failure might signal the emergence of multi-drug resistant TB (MDR-TB). Earlier studies have shown that MDR has so far been below 1% \[[@B22],[@B23]\]. The proportion of PTB+ was fairly stable in the range of 40--50% across the years, which indicates that the diagnosticians uniformly observed the diagnostic algorithms recommended by the NTLCP. The case detection rate (CDR), as estimated by the proportion of expected incident cases notified, showed an encouraging increase from 45% in 1994 to 115% in 1999, well above the CDR recommended by the WHO. The most plausible explanation for this might be that the vigorous implementation of the program during the earlier years of introduction of DOTS might have enabled detection of a big pool of prevalent cases of TB. But the pace in case detection could not be maintained during 2000 and 2001 when the CDR fell to 94% and 67% respectively. Reduction in the incidence of TB could be a possible explanation, but one cannot justify such a fast decline in a very short time. This calls for some case detection improvement initiatives to be in place as the proportion of TB cases detected and cured under DOTS is one of the health-related indicators of the millennium development goals \[[@B24]\]. Quality of the laboratory services and the possibility of under-reporting are among factors to be explored further. Two major areas of program weakness need to be addressed. First, 94% of the patients were registered as new cases, which suggests that some re-treatment cases had been wrongly classified as new. Further, treatment outcomes have not been recorded for about one-fifth of the patients. Improving the record keeping system seems to be an immediate priority. Second, the proportion of PTB+ patients who had follow-up sputa examined for AFB was low. The threat of anti-TB drug resistance is imminent, and it is important to strengthen patient follow-up with sputum examinations. About one-third of the peripheral diagnostic facilities lacked reagents for Ziehl-Neelsen stain during the time of visit, and this might partly explain the low performance. Conclusion ========== The introduction and expansion of DOTS in Hadiya has led to a significant increase in treatment success and decrease in default and failure rates. The smaller institutions exhibited better treatment outcomes compared to the larger ones including the zonal hospital. The high number of patients with missing information in the unit registers is perhaps an issue that needs to be addressed as urgently as possible. Further studies are recommended to see the impact of the programme on the prevalence and incidence of tuberculosis. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= EBS was the principal investigator and participated in the design of the study, conducted the study, performed data entry and analysis, and wrote the manuscript. BL was the project co-ordinator and participated in the design, data analysis and write-up of the manuscript. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Map of Ethiopia with the study area highlighted**. Ethiopia is administratively divided into nine regional states and two city administrations, and the Southern Nations, Nationalities and Peoples\' Regional State (SNNPRS) accounts for one-fifth (13 million) of the total population of the country. The study area (Hadiya zone) has got a population of 1.2 million. ::: ![](1471-2458-5-62-1) ::: Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/5/62/prepub> Acknowledgements ================ We would like to thank the Southern Nations, Nationalities and Peoples\' Regional Health Bureau, the regional and zonal TLCP and the staff of respective health facilities involved in the study for their genuine support during the field data collection. The Centre for International Health, University of Bergen funded this study.

PubMed Central

2024-06-05T03:55:58.424237

2005-6-6

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173119/", "journal": "BMC Public Health. 2005 Jun 6; 5:62", "authors": [ { "first": "Estifanos B", "last": "Shargie" }, { "first": "Bernt", "last": "Lindtjørn" } ] }

PMC1173120

Background ========== Almost 21% of the female population experience major depression in their life \[[@B1]\]. Twice as many females experience some form of depression when compared to male \[[@B2]\]. Female depression has higher risk on first onset, can last longer, and often recur \[[@B3],[@B4]\]. There is established relationship between life stress and depression \[[@B5]\]. Infertility is a stressful event in life of human being. In a comparison to patients with other medical conditions, psychological symptoms associated with infertility are similar to those related to cancer, hypertension, and cardiac rehabilitation \[[@B6]\]. Infertile women, in comparison with control group, showed higher scores on the depression and anxiety scales \[[@B7],[@B8]\]. Evidentially, depression has some neurobiological and hormonal mechanisms. Changes in ovarian hormones are likely necessary, but not sufficient for inducing depression \[[@B9]\]. The risk of relapse after treatment is 2.5 times higher in women; on the other hand, non-remission rate is similar in both sexes \[[@B10]\]. It appears that the sex difference cannot explain the sex-ratio relapse rate. As a result, there should be multiple factors including sex for inducing depression overall and in infertile subjects \[[@B3]\]. Many studies have been considered to determine risk factors of depression in infertility. In some studies, risk factors include female gender, repeated treatment cycles, unsuccessful treatments, low socioeconomic state, foreign nationality, lack of partner\'s support for women, life events for female, previous depression, and 2 to 3 years history of infertility \[[@B8],[@B11],[@B12]\]. However, some of these factors such as short period of infertility and husband cooperation have not been confirmed in other studies \[[@B12],[@B13]\]. The above-mentioned studies were designed as cross-sectional studies. So far, there is no prospective cohort or before-after study to find predictors of depression in infertile patients after treatment except a single one \[[@B14]\]. Lok et al showed that severity of depression following a failed treatment was positively associated with the duration of infertility. However, the post-treatment BDI scores were not associated with type of infertility treatment received, age, education, cause of infertility and number of previous treatments received \[[@B14]\]. Screening depression is a difficult task. The reason being is that there are numerous questionnaires for interpretation of psychological symptoms. Beck Depression Inventory (BDI) was developed and revised by Beck et al \[[@B15],[@B16]\]. This 21-item self-report questionnaire was intended to assess the severity of current depressive symptomatology in the psychiatric population. It is written on a fifth or sixth-grade reading level. It requires minimal time and no special training to administer. The BDI has been used, extensively, in clinical diagnosis and research \[[@B15],[@B16]\]. Versions in other languages are confirmed and used, as well \[[@B17]-[@B19]\]. In our study, Persian version of BDI has been employed. We analyzed Beck Depression Inventory score at the beginning and at the end of treatment cycle to determine which factors may influence the score of BDI, after treatment of infertility. Methods ======= Between April 2003 and 2004, 350 infertile women, whom had been visited in the Infertility Ward of Shariati Hospital in Tehran, participated in a prospective before-after study. Exclusion criteria were out of Tehran residency, definite diagnosis of mood disorders (with or without current treatment), medications reportedly associated with depression (e.g. steroidal contraceptives), and medical conditions associated with depression such as thyroid dysfunction and diabetes mellitus \[[@B20]\]. All subjects signed informed consent. Demographic data collected included age, education, type of infertility, duration of infertility, cause of infertility, and history of previous treatment (expressed as number). Candidates were asked to complete BDI (BDI1) in their first visit, during waiting periods, at the outpatient clinic. The questionnaire was completed by an assistant for uneducated subjects. Accordingly, BDI score was calculated. Index score of ≤9 is considered to be within normal range, a score of 10 to 15 shows minimal depressive symptomatology, a score of 16--31 points toward mild depression, a score of 32--47 is in favor of moderate depression and a score of \>47 indicates severe depression \[[@B16]\]. Due to rules of ethics, 10 patients who had the score of 47 or above were informed about their condition. Eight of these 10 patients were excluded from study since it was essential for them to seek psychiatric consultation as soon as possible. We continued the study with 342 patients. The treatment consisted of either intrauterine insemination (IUI) or assisted reproductive technology (ART). IUI candidates were excluded from the study. Therefore, the sample size was 257. At day 16 after embryo transfer, a positive result of treatment was defined as beta hCG (β-hCG) ≥200 IU/ml. One to Three weeks after receiving result of β-hCG, patients were asked to complete the second BDI (BDI2). This study was approved by ethics committee of Tehran University of Medical Sciences. The statistical analysis was performed using the Statistical Package for Social Sciences, version 10.0 (SPSS Inc., Chicago, IL). The primary outcome measure was the score of BDI after treatment. Data was expressed as means +/-SD and percentile of total. The tests being used were Wilcoxon Signed Ranks test, Mann-Whitney test, one way ANOVA, multiple regression analysis, and chi-square test. Results ======= A total number of 257 women were entered into the study after completing BDI1. Six (2.3%) patients refused answering BDI2. As a result, there were 251 paired complete questionnaires. Demographic characteristics of patients are given in Table [1](#T1){ref-type="table"}. We divided the patients into four groups based on cause of infertility. The prevalence was lowest in unexplained infertility (10%) and highest in female infertility (43%). The education level was divided into six groups. A large number of participants were at intermediate education level (35.1%) and a few of them were uneducated (2.4%). The number of previous attempts for treatment of infertility (IUI and/or ART) defers from 0 (24.7%) to 11 times (0.4%). Positive result of β-hCG was achieved in 61 patients (24.3%) (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Demographic characteristics of subjects ::: Characteristic No. Mean(± SD) or % ----------------------------- ------------------------ ----- ----------------- Age (y) 251 28.9 ± 5.5 Duration of infertility (y) 251 6.9 ± 4.5 No. of previous treatment 251 1.3 ± 1.4 Education No education 6 2.4 Primary school 36 14.3 Junior school 60 23.9 High school 12 4.8 Intermediate education 88 35.1 University 49 19.5 Type of infertility Primary 230 91.6 Secondary 21 8.4 Cause of infertility Male 86 34.3 Female 108 43 Combined 32 12.7 Unexplained 25 10 Result of β-hCG Positive 61 24.3 Negative 190 75.7 ::: The differences of somatic variables were analyzed between pregnant and non-pregnant subgroups. Pregnant subjects were younger (*P*\< 0.000), and had shorter duration of infertility (*P*= 0.005). There were no statistically significant difference in terms of type of infertility, education, cause of infertility, mean number of previous treatment, and mean score of BDI1 (Table [2](#T2){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of demographic characteristics between pregnant and non-pregnant subjects ::: Pregnant (n = 61) Non-pregnant (n = 190) *P* ----------------------------- ----------------------- ------------------- ------------------------ -------- Mean age (y) 26.6 ± 4.2 29.6 ± 5.7 \<.000 Duration of infertility (y) 5.5 ± 3.7 7.4 ± 4.6 .003 Primary infertility (%) 58 (95.1) 172 (90.5) NS Cause of infertility (%) Male 22 (36.1) 64 (33.7) NS Female 27 (44.3) 81 (42.6) Both 7 (11.5) 25 (13.2) Unknown 5 (8.2) 20 (10.5) Education (%) No education 0 6 (3.2) NS Primary school 6 (9.8) 30 (15.8) Guiding school 11 (18) 49 (25.8) High school 3 (4.9) 9 (4.7) Completed high school 29 (47.5) 59 (31.1) University 12 (19.7) 37 (19.5) Previous treatment 1.2 ± 1.4 1.3 ± 1.4 NS BDI1 score 14.6 ± 11.3 14.5 ± 9.8 NS ::: The BDI score at the beginning of the study (BDI1) was within normal range (0--15) in 61% of patients. In 31% the score was 16--31, suggestive of mild depression, and only 8% of the women had BDI score higher than cutoff score (≥32), suggestive of moderate and severe depressive symptoms. After knowing the results of β-hCG, BDI score changed. After treatment, 47% of the whole sample was within normal range. In pregnant and non-pregnant subjects, normal score of BDI2 was found 95.1% and 31.6%, respectively. The difference in percentile between pregnant and non-pregnant patients were statistically significant (*P \<*.000). Our main outcome was evaluation of factors that can influence BDI2 score. When BDI2 score was considered as dependent variable, in a multiple regression analysis, it was positively correlated with BDI1 score (r = 0.61, *P*\< .000) and duration of infertility (r = 0.15, P \< .000). Negative correlation was found between BDI2 score and result of therapy (r = - .51, *P*= .000). Eighty six percent of variation of BDI2 score could be predicted by the above three variables. There was no correlation between BDI2 score and age, education, number of previous treatment, and type of infertility. The mean score of BDI2 was evaluated in relation to cause of infertility. The mean score in male factor was less than other groups. The difference between groups was statistically significant (Table [3](#T3){ref-type="table"}). ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Mean score of BDI1 and BDI2 in relation to cause of infertility ::: Cause of infertility Mean of BDI1 Mean of BDI2 ---------------------- --------------- --------------- -------------- ------------ Male 11.2 ± 8.9 14.6 ± 11.3 *P*\< .000\* Z = -4.168 Female 15.8 ± 11 21.1 ± 13.2 *P*\< .000\* Z = -4.745 Combined 16.7 ± 9.2 21.9 ± 14.3 *P*= .006\* Z = -2.735 Unknown 15.2 ± 9.5 19.9 ± 10.5 *P*= .02\* Z = -2.303 *P*= .006\*\* *P*= .002\*\* Total 14.3 ± 10.2 18.9 ± 12.8 *P*\< .000 Z = -7.196 \*: difference between variables in row \*\*: difference between variables in columns ::: Mean of BDI2 score rose after unsuccessful treatment and dropped after a successful treatment (Table [4](#T4){ref-type="table"}). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Comparison of mean BDI1 and BDI2 in pregnant and non-pregnant subjects ::: pregnant Non-pregnant *P* ------------------ --------------------- ---------------------- --------------------- BDI1 (mean ± SD) 13.6 ± 11.3 14.5 ± 9.8 .28 (Z = -1.079) BDI2 (mean ± SD) 6.2 ± 5.4 22.9 ± 11.8 \<.000 (Z = -9.180) *P* \<.000 (Z = -5.845) \<.000 (Z = -11.481) ::: We observed the factors that affect BDI2 score in pregnant and non-pregnant groups, separately. In pregnant subjects, BDI2 score was positively correlated with BDI1 score (r = 0.82, *P*\< .000). In non-pregnant patients, BDI2 score was positively correlated with BDI1 score (r = 0.8, *P*\< .000), and duration of infertility (r = 0.17,*P*= .001). We studied BDI1 score, as well. In a multiple regression analysis, BDI1 score was negatively correlated with education (r = -0.26, *P*\< 0.000). There was no association between BDI1 score and age, duration of infertility, previous fertility, and number of previous cycles of treatment. Only 26% of variation of BDI1 could be predicted by the factor of education. In subgroups of infertility cause, the mean score of BDI1 in male factor was less than other groups. The difference between the groups was statistically significant (Table [3](#T3){ref-type="table"}). Discussion ========== The purpose of this study was to describe the subgroups of infertile women at risk of depression after treatment of infertility. This study can be valuable because it is using a self-reported inventory that differs in both cost and time from psychiatric structured interview. Nevertheless the shortcomings of this inventory, it has been widely used in research. Sensitivity and specificity of Beck Depression Inventory is not high but is reasonable. Beck et al suggested that a score greater than 9 points to depression symptomatology \[[@B16]\]. BDI score were also categorized in subgroups. The score of 9 and less shows normal range, a score of 10--15 indicates at least mild to moderate depressive symptoms, and a score of 16 and above indicates clinical depression \[[@B21]\]. We chose the second version since it screens subgroups of depression more precisely. In this study, the prevalence of depressive symptomatology as assessed by BDI1 score ≥ 16 was 39%, while the prevalence of moderate to severe depression(BDI ≥ 32) In an interesting study, the questionnaires were sent through internet to estimate degree of some psychological characteristics such as depression. Based on this study, authors found out that more than one quarter of patients could be considered moderately or severely depressed \[[@B22]\]. In screening depression in pregnancy, researchers chose a score greater than 9 and 43% of their population scored above this cutoff. As they designated BDI score greater than 15 as abnormal, nevertheless, the score exceeded the cutoff value in 19% of patients. In that study, a cutoff value greater than 15 yields a sensitivity of 0.83, a specificity of 0.89, a positive predictive value of 0.50, and a negative predictive value of 0.98 \[[@B21]\]. One of the best ways to determine the prevalence of a disease or symptom in relation to a risk factor is to compare a case group with a well-chosen control group. Domar et al showed that the prevalence of depression was 25.8% in infertile women compared with 13.2% in women who were waiting for a routine gynecologic examination \[[@B8]\]. These percentiles were reported to be 36.7% compared to 18.4% in another study \[[@B23]\]. Therefore, it is important that self-report questionnaires used for screening a disease should have a cutoff point reaffirmed either by a gold standard test or control group. The main outcome of this study was determining factors that may influence BDI2. The score after treatment (BDI2) was correlated with the score of BDI before treatment and duration of infertility and with the result of therapy. Hence, Table [4](#T4){ref-type="table"} shows no statistically significant difference between mean score of BDI1 in the two groups, we assume that the factors affecting BDI1 in the pregnant group and those in the non-pregnant group are the same. As BDI1 had a strongest affect on BDI2, a regression analysis considering BDI1 as a dependent variable was performed. It was found out that BDI1 score decreases as the level of education increases. In a stepwise regression, we failed to show any other factors to have relation with BDI1 score. There have been studies in which correlation between BDI score and demographic characteristics were evaluated. Demyttenaere et al showed that age, duration of infertility, and numbers of previous IVF attempts are factors affecting BDI score, but there was no data about level of education \[[@B24]\]. On the other hand, Beutel et al reported the correlation between education and depression (Spearsman correlation = -0.15 *P*\< .05) \[[@B11]\]. This inverse relationship between education and BDI1 score was considerable in our study (r = -0.26). This relationship can be explained by cultural views. It seems that high-educated people have objectives other than fertility to focus on. Data about age and risk of depression is not conclusive. Some studies have reported associations been them \[[@B11],[@B24],[@B25]\], while other studies could not show any relationships between these two \[[@B14],[@B26]\]. We suggested that other variables could act as intermediate variables between age and BDI1. Because of the strength of theses variables, the effect of age was hidden in multivariate analysis. The second important factor was the result of therapy. Apparently, subjects who are at risk of depression are more prone to develop depression at the time of exposure to life events. The third factor that had influence on BDI2 was duration of infertility. The association of BDI score after treatment with duration of infertility in subgroup with failed treatment was showed in another study \[[@B14]\]. In our study, BDI2 is associated positively with duration of infertility. However, the strength of this association is low (r = 0.15). It seems that such a low influence cannot play a significant role in practice. The assumption on whether the association between duration of infertility and post-treatment BDI score is a straight association or depends on intermediate variables such as number of treatment failure experiences needs more studies \[[@B27]\]. The perception of woman about the cause of infertility may change her BDI score. Pressure on infertile women with female factor infertility is high. The risk of depressive symptomatology is lower when a woman thinks that the problem is male factor. Combined factor indicates a severe problem, so it is reasonable that BDI score rises by this diagnosis (Table [3](#T3){ref-type="table"}). This type of cultural view has been showed in countries with family-based societies \[[@B28]\]. We emphasize that a number of meaningful relationships were introduced in other studies which were not mentioned here. Thus, in prospective studies other variables should be included and other more sensitive tools should be used. The other limitation of this study was using BDI for uneducated subjects that might induce a bias. It is clear that designing studies using interviews instead of questionnaires will be more valuable. Conclusion ========== Our findings suggest a plan for screening women who are prone to depression after infertility treatment. It is obvious that there is no significant difference among them before treatment (Table [2](#T2){ref-type="table"}). Answering the following two simple questions can conduct the physician to screen these women. 1\. Is my patient at risk of depression (due to BDI1 score) at the beginning of the treatment? 2\. What will be the result of therapy? Therefore, the physician can predict subjects who may need psychological supportive programs at the beginning of treatment cycles. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= AK, AA and MA drafted the manuscript. AK performed the statistical analysis. AK, AAA and FR participated in the study design and coordination. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-244X/5/25/prepub> Acknowledgements ================ We gratefully acknowledge the participation of Dr. Ehsan Akbari Hamed and the infertility ward crew of Shariati Hospital: Nahid Abbassi, Roghaeh Amrollahi, and Mehrnoosh Amini. This study was supported by Vali-e-Asr Reproductive Health Research Center, Tehran University of Medical Sciences.

PubMed Central

2024-06-05T03:55:58.430849

2005-5-24

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173120/", "journal": "BMC Psychiatry. 2005 May 24; 5:25", "authors": [ { "first": "Afsaneh", "last": "Khademi" }, { "first": "Ashraf", "last": "Alleyassin" }, { "first": "Marzieh", "last": "Aghahosseini" }, { "first": "Fatemeh", "last": "Ramezanzadeh" }, { "first": "Ali Ahmadi", "last": "Abhari" } ] }

PMC1173121

Background ========== Duodenal ulcer perforations are a common cause of peritonitis. The classic, pedicled omental patch that is performed for the \'plugging\' of these perforations was first described by Cellan-Jones in 1929 \[[@B1]\], although it is commonly, and wrongly attributed to Graham, who described the use of a free graft of the omentum to repair the perforation in 1937 \[[@B2]\]. In this, a strand of omentum is drawn over the perforation and held in place by full thickness sutures placed on either side of the perforation, and this procedure has become the \"gold standard\" for the treatment of such perforations. However, occasionally, large perforations of the duodenum may be encountered in which there exists the threat of post-operative leakage following closure by this simple method \[[@B3],[@B4]\]. Here, other surgical options such as partial gastrectomy, jejunal serosal patch, jejunal pedicled graft, free omental plug, suturing of the omentum to the nasogastric tube, proximal gastrojejunostomy, or, even, gastric disconnection may be deemed necessary for adequate closure \[[@B3]-[@B8]\]. Very little data is available in literature regarding the definition, incidence, and the management of large perforations of duodenal ulcers. This paper represents our experience with the management of this subset of duodenal ulcer perforations over a period of three years from January 2001 to December 2003. Methods ======= A total of 162 patients underwent emergency surgery for duodenal ulcer perforations at our hospital over a period of three years (January 2001 to December 2003). The case files of all these patients were analyzed, and the patients were sorted into four groups according to the size of the perforation noted intra-operatively -- Group 1 (less than1 cm perforation); Group 2 (1 cm to 2 cm); Group 3 (2 cms to 3 cms); and, Group 4 (more than 3 cms perforation). No cases of anterior and posterior ulcers, or multiple perforations were encountered while reviewing the operative notes. The technique of omentopexy was essentially the same in all the cases -- a total of three sutures were placed onto the normal, healthy duodenum on either side of the perforation, a strand of omentum was placed directly onto the perforation, and the sutures were knotted above this. No attempt was made to close the perforation prior to placing the omentum as a graft. The case files of all the patients were retrospectively analyzed for patient particulars, intra-operative findings, surgery performed, post-operative stay, morbidity and mortality. The groups were then compared with each other in terms of age, leak rates, hospital stay, morbidity, mortality and the surgery performed. Statistical analysis was done using the *chi-square*and the *t- test*by an independent comparison of each group singly against another by a statistician who was blinded to the study. A *p*value of \< 0.05 was taken as significant. It was found that the perforations between 1 cm and 3 cm in size (Groups 2 and 3 as above) behaved in a similar manner statistically, and therefore, the patients these two groups were combined to give a single group. We finally ended up with 3 groups of perforations -- Group A (less than1 cm perforations), Group B (perforations between 1 cm and 3 cms in size), and Group C (more than 3 cms perforation). Results ======= Of the total of 162 patients that underwent emergency surgery for duodenal ulcer perforations at our hospital over three years, there were 148 males (91.36 %) and 14 female (8.64 %) patients, giving a male to female ratio of 10.57 : 1. The average age of the patients was 40.63 years (range 15 -- 82 years), with an almost equal age of occurrence for males (40.52 years) and females (41.78 years). All the patients were divided into three groups as explained above. Group A was deemed to be the small perforation group, Group B was called \'large\' perforations, and Group C, \'giant\' perforations. The majority of patients came under the \'small\' perforation group, but there were 38 patients (23.46 %) with large perforations as per our definition. These patients had a higher age of presentation (47.18 years) than the patients with smaller perforations (39.46 years). Giant perforations, or perforations greater than 3 cms in size were seen only 2 cases, accounting for a small percentage (1.23 %) of all cases seen. When the small perforation group was compared with the larger perforations, it was found that the large perforations had a higher morbidity (*x*^*2*^= 37.4503, *p*\< 0.05), leak rate (*x*^*2*^= 4.9117, *p*\< 0.05), and hospital stay (*t value*5.117, *p*\< 0.001) and that this difference was statistically significant. This therefore, lends support to the popular opinion that large perforations have a worse outcome. Overall, the commonest surgery performed was the Cellan-Jones omental patching -- in 119 of the 122 cases in Group A; 30 of the 38 patients in Group B. When the results of omental patch were compared between the two groups, no significant difference was found in the leak rates (*x*^*2*^= 2.8698; *p*\> 0.5) and mortality (*x*^*2*^= 1.4732; *p*\> 0.1), thereby implying that this was an equally effective method for the closure of larger perforations also. Jejunal serosal patch using a loop of the jejunum, and antrectomy (4 cases each) were the other surgeries performed in Group B, when closure with the omentum was thought to be unsafe by the operating surgeon. Five (12.5 %) patients of this group had a leak following closure of the perforation; 3 following omental patch and 1 each after performance of jejunal serosal patch and antrectomy, whereas only 3 cases developed leak in Group A (2 after omental patch and 1 after truncal vagotomy and pyloroplasty). Two cases had \'giant\' perforations extending onto the pylorus -- in one, resection and Billroth II reconstruction was performed, and in the other, jejunal serosal patch. The patient who underwent resection had presented late, and succumbed to septicaemia on the very first post-operative day. The other remained well and was discharged on the 11^th^post-operative day. Overall, the patients with large perforations (Group B) had significantly increased hospital stay, leak rates, and morbidity (Table [1](#T1){ref-type="table"}). The hospital stay was almost double for these patients (13.65 days versus 6.93 days). Although the overall morbidity was 48.76 %, it was much higher in the larger perforations (groups B and C). The common morbidity encountered was chest infections (39 cases), but wound infection (12 cases), biliary leak (08 cases), intra-abdominal abscesses (06 cases), burst abdomen (06 cases), renal failure (02 cases), DIC (04 cases), jaundice and upper gastrointestinal bleeding (01 case each) were also recorded. The mortality in this series was 8.64 % (14 cases), and again, it was significantly higher in perforations more than 1 cm in size (*x*^*2*^= 3.8940; *p*\< 0.05). Table [1](#T1){ref-type="table"} gives the details of all the three groups. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Patient data ::: **Group A -- \'Small\' (Less than 1 cm)** **Group B -- \'Large\' (1 cm -- 3 cm)** **Group C -- \'Giant\' (More than 3 cm)** ---------------------------------- ------------------------------------------- ---------------------------------------------------------------------- -------------------------------------------------------- **Number of cases** 122 (75.31 %) 38 (23.46 %) 02 (1.23 %) **Average age** 39.46 years 47.18 years 37.50 years **Male/Female** 109 : 13 37 : 1 2 : 0 **Average Duration of Symptoms** 2.5 days 3.18 days 3.50 days **Surgery Performed** Omental Patch 119 \*\* Pyloroplasty 03 \* Omental Patch 30 \*\*\* Jejunal Serosal Patch 04 \* Antrectomy 04 \* Antrectomy and Billroth II 01 Jejunal Serosal Patch 01 **Post-operative Leak** 03 (2.46 %) 05 (13.16 %) \- **Morbidity** 41 37 01 **Post-operative Hospital Stay** 6.93 days 13.65 days 6.00 days **Mortality** 07 (5.74 %) 06 (15.79 %) 01 (50 %) Each \* indicates one post-operative leak ::: Discussion ========== Duodenal ulcer perforation is a common surgical emergency in our part of the world. The overall reported mortality rate varies between 1.3 to nearly 20 % \[[@B9]-[@B11]\] in different series, and recent studies have shown it to be around 10 % \[[@B11]\]. Factors such as advancing age, concomitant disease, preoperative shock, size of the perforation, delay in presentation and operation, have all been defined by various authors to be risk factors for mortality in such a situation \[[@B9]-[@B11]\]. Although the size of a perforation is an important measure in determining the outcome, a review of literature failed to reveal, any accepted definition of either small or giant perforations of duodenal ulcers. Neither could we come across any specific recommendations regarding the management of giant / large perforations, which are said to be \"difficult\" to manage and have anecdotally been associated with high leak rates and mortality. This is in contrast to the well accepted and documented definition of giant duodenal ulcers (more than 2 cms in size), which may or may not perforate, but are usually considered to be an indication for definitive, elective ulcer surgery \[[@B8],[@B12]\]. Commonly, duodenal ulcer perforations are less than 1 cm in greatest diameter, and as such, are amenable to closure by omentopexy \[[@B3]\]. Our experience does seem to validate this, and this subset of \'small\' perforations does seem to have the best outcome. It is the perforations that are larger that have been the cause of much confusion in their definition and management. The size of such \'giant\' sized perforations has arbitrarily been defined by various authors as being greater than 0.5 cms \[[@B7]\], 1 cm \[[@B3],[@B4]\], or 2.5 cms \[[@B6]\] in greatest diameter, but we failed to uncover any specific size in available English language literature beyond which to label these perforations as \"giant\". These perforations are considered particularly hazardous because of the extensive duodenal tissue loss and surrounding tissue inflammation, which are said to preclude simple closure using omental patch, often resulting into post-operative leak or gastric outlet obstruction \[[@B3],[@B4]\]. The tendency to leak may further be aggravated by the high intraluminal pressures, extrusion of the duodenal mucosa through the closure, and, autodigestion by the pancreatic enzymes and bile, thereby further compromising an already sick patient \[[@B13]\]. Our data seems to suggest that based on the size, duodenal perforations can be classified into three main groups (1) *small*perforations that are less than 1 cm in size, and have the best outcome; (2) *large*perforations, that have a size between 1 cm and 3 cms; and, (3) *giant*perforations that exceed 3 cm size. The usage of the word \'giant\' for a duodenal perforation should be restricted to such large defects, where omentopexy may be deemed unsafe, and other options may be thought to be necessary. In the absence of any specific definition and guidelines regarding the management of such large / giant perforations in literature, different authors have recommended varied surgical options from time to time, based on their experience and research. These have included resection of the perforation bearing duodenum and the gastric antrum in the form of a partial gastrectomy, with reconstruction as either a Billroth I or II anastomosis, or the more morbid procedure of gastric disconnection in which vagectomy, antrectomy, gastrostomy, lateral duodenostomy and feeding jejunostomy are performed, with restoration of intestinal continuity electively after 4 weeks of discharge \[[@B8]\]. Others have recommended conversion of the perforation into a pyloroplasty, or, closure of the perforation using a serosal patch or a pedicled graft of the jejunum, or, the use of a free omental plug to patch the defect, and even, suturing of the omentum to the nasogastric tube \[[@B3]-[@B8]\]. Proximal gastrojejunostomy and / or vagotomy may be added to these procedures to provide diversion and a definitive acid reducing procedure respectively \[[@B8]\]. However, as can be appreciated, each of these procedures not only prolongs the operating time, but also requires a level of surgical expertise that may not be available in the emergency \[[@B6]\]. In addition, each of these procedures has it own morbidity that may add up significantly to alter the final outcome of the patient, and more importantly, none of them is immune to the risk of leak in the post-operative period, which has been the main concern against performing the omental patch in larger perforations \[[@B3],[@B4]\]. The results of omentopexy in small and large sized perforations in the present series give statistically similar results. The leak rates and mortality of the two groups after omentopexy remain comparable, thereby suggesting that this may be considered as the procedure of choice in all perforations upto a size of 3 cms. The procedure is simple and easy to master, and, avoids the performance of a major resection in a patient who is already compromised. In fact, Sharma et al also reported the success of the omental plug in perforations of duodenal ulcers more than 2.5 cms in size; only, they preferred using a free graft of the omentum rather than a pedicled one \[[@B6]\]. We feel that mobilization of the omentum on its pedicle from the colon, and placement of sutures into the normal duodenum away from the perforation makes the performance of omental patch safe even in the presence of large sized perforations. In the present series, only 2 cases were defined to be \'giant\' according to the size (more than 3 cm) that we have defined -- one underwent antrectomy and Billroth II reconstruction, the other, a jejunal serosal patch. The first patient (antrectomy) succumbed to the ongoing septicaemia on the very first post-operative day, but the other patient survived. This is the group of patients with truly giant perforations who need to be analyzed further to determine the best course of action i.e. resectional versus non-resectional surgery. However, the less number of patients in this group did not allow us to reach any definite conclusion regarding their ideal management. Further study is needed to optimize our efforts to this target group. Conclusion ========== Duodenal perforations should be classified as small, large or giant according to their size encountered at laparotomy. In the emergency setting, such patients are often seriously ill and it is not advisable to perform major surgical procedures on them. The Cellan-Jones omental patch is simple, can be performed in a relatively short time, and remains dependable even for the closure of large sized perforations (i.e. perforations upto 3 cms in size). The addition of a feeding jejunostomy and placement of a tube drain in the Morrison\'s space may offer a further sense of \'\'security\'\' to the operating surgeon, keeping by open the option of maintaining the nutrition of the patient as well as creating a controlled duodenal fistula in case of a post-operative leak. The word \'\'giant\'\' should be reserved only for perforations that exceed 3 cm in diameter. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= **SG**carried out acquisition, statistical analysis and interpretation of the data and drafting of the initial manuscript **RK**conceptualized the paper, and was involved in the interpretation of data, drafting of the manuscript, and revising it critically for the intellectual content **RS**carried out drafting of the manuscript, and revised it critically for the intellectual content till the final version was reached **AKA**helped in the revisions of the intellectual content and gave final approval of the version to be published. All authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2482/5/15/prepub>

PubMed Central

2024-06-05T03:55:58.434016

2005-6-25

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173121/", "journal": "BMC Surg. 2005 Jun 25; 5:15", "authors": [ { "first": "Sanjay", "last": "Gupta" }, { "first": "Robin", "last": "Kaushik" }, { "first": "Rajeev", "last": "Sharma" }, { "first": "Ashok", "last": "Attri" } ] }

PMC1173122

Background ========== The most important phosphate-containing calculi involved in urinary stone disease are carbonate apatite, brushite, and struvite. Overall, phosphate stones account for 12--20% of all urinary stones and rank first in the list of recurrent calculi \[[@B1]\]. The most definitive therapy involving any type of phosphate stone is calculus removal by shock-wave lithotripsy, percutaneous stone removal, or open surgery (especially in children). However, while technologies have successfully treated patients with phosphate calculi, they are associated with injury to the urinary tissue and retained stone fragments have resulted in recurrences. Stone dissolution through chemolysis remains an option for urologists even in the advent of more sophisticated modalities in the treatment of urolithiasis. Chemolysis via acidification of the urine and antibiotic therapy (especially for infection stones) are important adjuvant modalities for phosphate calculi therapy. Hitherto, a many solutions have been tested in the quest for more effective dissolution agents. The most effective solution among them is Renacidin (R), which is a buffer consisting mainly of citrate and gluconate \[[@B2]-[@B4]\]. These solutions are delivered directly into the kidney by ureteric catheter or percutaneous nephrostomy, and are successful in the dissolution of struvite. But the injury to urothelium restrains their wide use. So, solutions with more dissolution efficiency and less side effects are needed urgently and may be used widely. Our in vitro study was designed to evaluate the effectiveness of two new complex solutions (using D-gluconic acid-lactone, D-gluconic-acid and other ingredients which makes them different from R) in the chemolysis of calcium phosphate stones. Material and methods ==================== Stones specimen preparation --------------------------- Eighteen stones composed mainly of calcium phosphate were taken from 18 patients who underwent kidney stone surgery. These stones were large enough (weight range from 0.514--0.928 g) to be fragmented and matched equally into six groups. Each stone was analyzed using the Merckognost at Urinary Stone analysis Kit. The chemical composition of these stones was varied but there was a predominance of calcium phosphate in almost all the stones and calcium oxalate was less than 10% (Table [1](#T1){ref-type="table"}, Table [2](#T2){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Chemical analysis of 18 stones for stone dissolution. There was a predominance of calcium phosphate in almost all the stones and calcium oxalate was less than 10%. ::: Stone% Ca Oxalate NH~4~ PO~4~ Mg Uric Acid Tri-calcium phosphate Calcium oxalate Struvite -------- ---- --------- ------- ------- ---- ----------- ----------------------- ----------------- ---------- 1. 45 2 2 57 7 2. 60 45 10 9 3. 60 10 4 55 10 4. 45 15 45 5 4 5. 35 5 35 6. 80 11 3 65 6 7. 50 10 60 6 8. 40 2 7 25 5 7 9. 5 5 45 15 20 8 10. 15 20 15 40 6 11. 45 5 3 45 10 12. 35 20 2 35 13. 50 20 5 25 14. 40 10 35 15 15. 45 15 35 9 16. 40 15 0.5 35 5 17. 75 5 15 18. 15 5 45 10 15 ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Mean chemical composition of 18 stones, showing that there was a predominance of calcium phosphate in almost all the stones and calcium oxalate was less than 10%. ::: Composition ± *s*(%) --------------- ------------- Ca^++^ 42.5 ± 20.7 Oxalate 9.6 ± 6.2 NH~4~ 1.5 ± 2.1 PO~4~ 40.1 ± 13.9 Mg 1.7 ± 4.9 Uric acid 3.3 ± 4.9 Cystine 0.0 ± 0.0 Ca^++^oxalate 3.4 ± 3.7 ::: Preparation of solvent ---------------------- Six different solvents were used for chemolysis of phosphate stones. The molecular formula and the molecular weights of some drugs used in these solvents are provided in Table [3](#T3){ref-type="table"}. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Molecular formulae and molecular weights of chemical drugs used in our experiment ::: Experimental drug Molecular formula Molecular weight -------------------------------- -------------------------------- ------------------ Citric acid C~6~H~6~O~7~·H~2~O 210.14 Sodium citrate Na~3~C~6~H~6~O~7~·2H~2~O 294.11 D-Gluconic Acid Lactone C~6~H~10~O~6~ 178.1 Magnesium citrate Mg~3~(C~6~H~5~O~7~)~2~·14H~2~O 703.4 Magnesium Carbonate (MgCO~3~)~4~·Mg(OH)~2~·5H~2~O 485.8 Disodium EDTA C~10~H~14~O~8~N~2~Na~2~·2H~2~O 372.24 Potassium citrate, tribasic K~3~C~6~H~5~O~7~·H~2~O 324.34 Potassium phosphate, monobasic KH~2~PO~4~ 136.09 ::: 1\. Physiologic sodium chloride solution (Phys). 2\. Artificial urine (Art, pH5.7, according to Griffith et al \[[@B5]\]) consisting of urea (25 g/L), sodium chloride (4.6 g/L), potassium-dihydrogen-phosphate (2.8 g/L), sodium sulfate (2.3 g/L), potassium chloride (1.6 g/L), ammonium chloride (1.0 g/L), calcium chloride dehydrate (0.1 g/L), sodium oxalate (0.05 g/L), and sodium citrate (0.01 g/L), dissolved in 1000 ml. distilled water. 3\. 0.03 M disodium ethylenediaminetetraacetic acid buffered to pH 8.5 with triethanolamine (EDTA, pH 8.5) 4\. 10% Renacidin (R, pH 3.9), consisting of citric acid (28.2 g), gluconic acid (5.0 g), calcium carbonate (1.0 g), magnesium bicarbonate (14.5 g), citrate magnesium (2.5 g), dissolved in 1000 ml. distilled water. 5\. Test solution 2 (S2, pH 4.0), prepared using citric acid (18.0 g), citrate magnesium (1.0 g), calcium carbonate (0.5 g), magnesium carbonate (7.5 g), D-gluconic acid-lactone (3.0 g), dissolved in 100 ml. distilled water. The solution was kept at 37? for 3 days, then diluted with 140 ml. distilled water. The concentrations of these ingredients are shown in Table [4](#T4){ref-type="table"}. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Concentrations of chemical drugs contained in S1 and S2 ::: Chemical drug Concentration (mmol/L) ----------------------------- ------------------------ Citric acid 357.1 Magnesium citrate 5.9 Magnesium Carbonate 64.3 D-Gluconic Acid Lactone 70.2 Calcium carbonate 20.8 D-Gluconic Acid 63.8 Sodium citrate 302.7 Potassium citrate, tribasic 302.7 Calcium chloride, anhydrous 454.5 Calcium carbonate is difficult to be dissolved in water, but in acidic solutions of S1 or S2, it can be dissolved thoroughly. So we give the concentration of Calcium as above. In diluted solutions these cations were diluted correspondingly, such that their pH did not change. ::: 6\. Test solution 1 (S1, pH 3.9), prepared using citric acid (18.0 g), citrate magnesium (1.0 g), calcium carbonate (0.5 g), magnesium carbonate (7.5 g), D-gluconic-acid (3.0 g), dissolved in 100 ml. distilled water, and processed as described for S2. The concentrations of these ingredients are shown in Table [4](#T4){ref-type="table"}). We also altered the pH of these solutions to 4, 5, 7, and 8.5 to test their effectiveness and added different cations (Na^+^, by adding 302.7 mmol/L citrate sodium 10 ml. into 100 ml. of the initial solution; K^+^, by adding 302.7 mmol/L citrate potassium 10 ml. into 100 ml. of the initial solution; Ca^2+^, by adding 454.5 mmol/L calcium chloride anhydrous 10 ml. into 100 ml. of the initial solution). S1 and S2 were formulated according to different dilutions (1:1, 1:2, 1:3, 1:4). The pH of these changed solutions had no detectable variation. After stones were immersed in these solvents for 24, 48 and 72 h, reduction in weight, percentage weight loss, and dissolution rate were determined. Results were analyzed using the Multiple Comparisons Test with the alpha set at 0.05. Stones remaining after the dissolution period were weighed (gm), and final weights were applied to the following formula: ![](1471-2490-5-9-i2.gif) The formula used to determine the dissolution rate was: ![](1471-2490-5-9-i3.gif) ### Chemolysis Every stone was placed in a vessel containing solvents, which were held at a constant temperature of 37? using a thermostat. Solvents were exchanged 2 times per day at 6 ml/100 mg per stone each time. The pH values were constantly controlled by a pH meter (Mettler Toledo 320-S, sensitivity 0.01). The weight of the calculus was measured continuously by Electronic Semi-micro-, Analytical and Precision Balances (Sartorius BP211D-OCE, sensitivity 0.01 mg) and documented online with a computer system (PC windows 486, Software Wedge for window, version 1.1). The analysis represents six stones for each experiment, repeated three independent times. Results ======= Urological calcium phosphate calculi were poorly dissolved by Phys and Art, and they had a low dissolution rate in EDTA at pH 8.5. The most effective solutions were R, S1 and S2 with 24 h mean dissolution rates: 5.05 ± 0.15 mg/hr (S1), 4.52 ± 0.64 mg/hr (S2), 4.53 ± 0.46 mg/hr (R); 72 h mean dissolution rates: 5.75 ± 0.44 mg/hr (S1), 5.2 ± 0.63 mg/hr (S2) and 4.55 ± 0.46 mg/hr (R) (![](1471-2490-5-9-i1.gif) ± *s*, *P*\< 0.01, R, S1 and S2 vs Phys, Art and EDTA; *P*\< 0.05, S1 vs R, LSD-test). The mean percentage weight loss at 72 h was: 40.5 ± 3.67 % (R), 52.1 ± 15.75 % (S1) and 44.4 ± 7.37 % (S2) (![](1471-2490-5-9-i1.gif) ± *s*, *p*\< 0.01 R, S1 and S2 vs Phys, Art and EDTA, LSD-test). Figure [1](#F1){ref-type="fig"} shows the average dissolution rates of the 6 different solutions at the end of 72 h. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Dissolution rates of phosphate calculi in vitro at 24 and 72 h using different solutions.**Phys = physiologic sodium chloride solution, pH 7.0. Art = artificial urine, pH 5.7. EDTA = 0.03 M disodium EDTA+TEA, pH 8.5. R = renacidin, pH 4.0. S1 = test solution 1, pH 4.0. S2 = test solution 2, pH 3.9. ^a^*P*\< 0.01 vs Phys and Art. ^b^*P*\> 0.05 vs EDTA, Phys and Art. ^c^*P*\< 0.05 vs R·![](1471-2490-5-9-i1.gif) ± *s*·n = 18 for each group. For statistics, see results section. ::: ![](1471-2490-5-9-1) ::: Table [5](#T5){ref-type="table"} Shows mean weight loss and mean percentage weight loss of stones dissolved by the 6 solutions at the end of 24 and 72 h. (![](1471-2490-5-9-i1.gif) ± *s*, ^a^*P*\< 0.01 Phys, Art and EDTA vs R, S1 and S2; ^b^*P*\< 0.05 EDTA vs S1, S2 and R.^c^*P*\< 0.05 R vs S1). ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Mean weight and percentage weight decrease after 24 and 72 h between groups dissolved with 6 different solutions. Phys = physiologic sodium chloride solution, pH 7.0. Art = artificial urine, pH 5.7. EDTA = 0.03 M disodium EDTA+TEA, pH 8.5. R = renacidin, pH 4.0. S1 = citrate complex 1, pH 4.0. S2 = citrate complex 2, pH 3.9. ± *s*, ^a^*P*\< 0.01 vs R, S1 and S2.^b^*P*\< 0.05 vs S1, S2.^c^*P*\< 0.05 vs S1. n = 18 for each group. For statistics, see results section. ::: Groups Weight loss (mg) Percentage weight loss (%) -------- ------------------ ---------------------------- ----------------- ------------------ Phys 3.52 ± 0.73^a^ 18.15 ± 13.15^a^ 0.00 ± 0.00^a^ 1.33 ± 1.53^a^ Art 2.48 ± 0.37^a^ 7.44 ± 1.9^a^ 0.00 ± 0.00^a^ 0.00 ± 0.00^a^ EDTA 43.78 ± 6.23^ab^ 153.6 ± 20.84^ab^ 4.67 ± 2.08^ab^ 19.00 ± 5.19^ab^ R 114.4 ± 8.92^c^ 346.14 ± 26.56^c^ 13 ± 1.35^c^ 40.5 ± 3.67^c^ S2 127 ± 23.09 408.11 ± 62.94 13.5 ± 2.76 44.4 ± 7.37 S1 138.5 ± 31.09 439.5 ± 102.00 5.77 ± 1.29 52.1 ± 15.75 ::: The dissolution rates were pH dependent. Even Phys and Art were effective to some extent at pH 4. The addition of 1 mol/L sodium hydroxide 10 ml into S1 or S2 100 ml, though not enough to lead to a detectable pH rise, would make the solutions cloudy and greatly reduce their effectiveness. At pH 5.0 they were nearly ineffective, as also observed at pH 7.0, 8.5. However, the dissolution rate of EDTA increased at an elevated pH value. At pH 8.5, EDTA approached a maximum dissolution rate with 72 h mean 1.56 ± 0.05 mg/hr (Figure [2](#F2){ref-type="fig"}). ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Effectiveness of six solvents at different pH (mg/hr).**Phys = physiologic sodium chloride solution, Art = artificial urine, EDTA = 0.03 M disodium EDTA+TEA, R = renacidin, S1 = test solution 1, S2 = test solution 2. ::: ![](1471-2490-5-9-2) ::: The addition of Na^+^, K^+^or Ca^2+^(302.7 mmol/L Sodium citrate 10 ml, 302.7 mmol /L Potassium citrate 10 ml or 454.5 mmol/L Calcium chloride dihydrate 10 ml into initial solution 100 ml, respectively) caused no significant change in dissolution rate. Diluted solutions demonstrated an interesting result. Diluted twice, S1 and S2 were more effective than their initial solutions with 72 h mean dissolution rates: 6.04 ± 1.36 mg/hr (S1), 5.60 ± 1.23 mg/hr (S2) (![](1471-2490-5-9-i1.gif) ± *s*, *P*\< 0.05, S1 and S2 vs R). The effectiveness of three times diluted S1 or S2 was the same as the initial solution. When diluted further, the effectiveness was reduced gradually until nearly 0 after solutions were diluted five times (Figure [3](#F3){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Concentration and effective of S1 and S2 ::: ![](1471-2490-5-9-3) ::: Discussion ========== Chemolysis is useful for eliminating cystine stones as well as for cases in which lithotripsy or endourology is considered to be difficult or risky. It proved to be a useful method for reducing staghorn stones before performing lithotripsy \[[@B6]\]. The type of chemolytic solution is dependent upon the composition of the stone and must be regarded as an effective adjuvant treatment \[[@B7]\]. Calcium phosphate can be dissolved with Suby or R, but the treatment is often tedious and time consuming. Calcium oxalate, the major urinary stone component, cannot be dissolved by these solutions. EDTA and other strong calcium chelators cannot be used because of their local toxicity. Certain enzymes can digest the organic matrix of the stone \[[@B8]\]. The first attempts to dissolve calcium stones were done by Hellstrom and Albright in 1930. They used citric acid for phosphate calculi, but this proved irritating to tissues. Suby and Albright modified the solution by adding magnesium oxide and sodium carbonate (Suby\'s G solution) \[[@B9],[@B10]\] to reduce injury to the rabbit bladder mucosa even though adding this cation reduced the speed of dissolution of struvite Mg (NH~4~) PO~4~. A contradictory conclusion was been made by a Dutch study \[[@B11]\] that magnesium in R promotes stone dissolution by cation exchange with calcium in apatite \[[@B12]\]. Our S1 and S2 solutions were similar to R although actual formulas were different. First, we added D-gluconic acid-lactone or D-gluconic-acid as new chelators which bind with calcium well by their special trait. Second, we used new ingredients and ingredients at different weights. The higher density of our solutions than R may be one of the factors which enhanced stone dissolution rates, but it may prevented further dissolution. When diluted twice, the solutions may have gained more space to accommodate cations than the initial solutions and achieve better effectiveness (*P*\< 0.05, S1 and S2 vs R). The further diluted solutions had a lower concentration suggesting that less ingredient took part in the reaction thus accounting for a lower dissolution rate. So, we think twice diluted S1 and S2 may be more useful than their initial solutions. The dissolution of the majority of the stones in this study can be attributed on the basis of their reaction equation: 3Ca^+2^+ 2 Citrate→CaCitrate→Ca^2+^+ CaCitrate The citrate in the solution binds with the calcium component of the stone producing calcium citrate, thus preventing crystallization. The dissolution process may be brought about by the combination of the above reaction. As the solubility of calcium phosphate is very pH dependent, the acidification of the urine by the incorporation of citric acid produce a pH between 3.0--4.0, the ideal pH for dissolution as described by Albright et al \[[@B9],[@B10]\]. When the pH is elevated, the citrate will react with cations thus losing its effectiveness. On the other hand, excessive alkalization may lead to the formation of calcium phosphate calculi. Added minor quantities of cations (Na^+^, K^+^or Ca^2+^) in S1 and S2 apparently causes no significant change to their dissolution rate. Conclusion ========== The study has shown that solutions S1 and S2 can dissolve calcium phosphate stones effectively in vitro at a precise dilution of their chemical components. Based on these findings, it is suggested that S1 and S2 may become useful complements to modern techniques of stone fragmentation such as extracorporeal shock wave lithotripsy and percutaneous surgery. Their roles may be suited for the treatment of infectious stones with a CaOx content of less than 10%. However, their safety profiles should be further investigated in order to support their use in subsequent human trials. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= Xiang-bo Zhang participated in the design of the study, in the sequence alignment, carried out the experiment, performed the statistical analysis and drafted the manuscript. Zhi-ping Wang conceived of the study, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Jian-min Duan, Jian-zhong Lu, Bao-liang Ma, participated in the sequence alignment. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2490/5/9/prepub> Acknowledgements ================ We are indebted to Zhi-ping Wang, Professor and director of the institute of urinary, the 2nd Hospital of LanZhou University, for his help in conducting the study and for reviewing the manuscript. We thank Jian-min Duan, Jian-zhong Lu, Bao-liang Ma, staff at the institute of urinary, the 2nd Hospital of Lanzhou University, for generously supplying the room for our experiment.

PubMed Central

2024-06-05T03:55:58.435674

2005-5-22

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173122/", "journal": "BMC Urol. 2005 May 22; 5:9", "authors": [ { "first": "Zhang", "last": "Xiang-bo" }, { "first": "Wang", "last": "Zhi-ping" }, { "first": "Duan", "last": "Jian-min" }, { "first": "Lu", "last": "Jian-zhong" }, { "first": "Ma", "last": "Bao-liang" } ] }

PMC1173123

Background ========== Aseptic loosening is the most common cause for long-term failure of total hip replacement (THR). The success rate after revision surgery is much lower than that after the primary operation. Wear, which contributes to aseptic loosening \[[@B1]-[@B4]\], has been shown to correlate with the contact pressure of the UHMWPE acetabular cup \[[@B5],[@B6]\]. An increase of the abduction angle of the acetabular cup has been shown to increase contact pressure and wear \[[@B7],[@B8]\]. Different geometrical cup designs, polyethylene thicknesses and clearances between the cup and head have been demonstrated to modulate implant survival \[[@B2],[@B7],[@B9]-[@B15]\]. However, the combined effect of these parameters on polyethylene stress is still unclear. For instance, the different geometries of the acetabular cups may modify the contact pressure in different ways depending on the abduction angle. Parameters reducing contact pressure reduce plastic deformation and wear, and thus may diminish the risk for aseptic loosening. Pre-clinical testing of hip implants prior to marketing is known to be important \[[@B16]\], since failure to perform proper testing may lead to unsatisfactory clinical results. Therefore, the geometries and orientations of acetabular cups should be optimized. Finite element (FE) modeling enables the evaluation of stress distribution throughout the THR prostheses \[[@B8],[@B17]\] and an assessment of geometries and material properties which would be difficult or time-consuming to test experimentally. In this study, two UHMWPE acetabular cups with different geometries and good long term clinical records, according to Scandinavian hip registers, were studied \[[@B18],[@B19]\]. However, the cups belong to the same hip prosthesis system and their individual survivorships are not described separately in the registers. FE modeling and cyclic testing of hip implants in a simulator were used to test the hypotheses that the geometry, thickness and abduction angle of the cup, as well as the clearance between the femoral head and the cup could affect acetabular cup stress upon loading. Information from this study can be used for the future development of THR implants prior to their surgical installation. Methods ======= Laboratory tests ---------------- Two UHMWPE acetabular cups\* with different designs (Waldemar LINK GmbH & Co., Hamburg, Germany: IP, n = 4 and Lubinus eccentric (with a snap-fit primary locking mechanism), n = 8) were tested experimentally under a cyclic load \[[@B20]\]. Axial loading was used to investigate the plastic deformations occurring in the cups and to minimize wear. Before the tests, the cups were fixed in stainless steel holders with bone cement (Palacos^®^R-40 cum Gentamicin, Schering-Plough Europe, Brussels, Belgium). To speed up the experimental tests, two combinations of head, cup and cement were tested at the same time which led overall to six tests (Table [1](#T1){ref-type="table"}, Fig. [1](#F1){ref-type="fig"}). Testings were carried out using a dynamic tester (Instron 8874, Instron Corporation, Canton, MA) at a 5 Hz frequency for 5 million cycles. The load profile was a Paul gait curve with a peak load of 3 kN, and abduction angles of 45° and 60° were used. Diluted bovine serum supplemented with EDTA and antibacterial agents was used as lubricant \[[@B21]\]. The serum was filtered through a 0.2 μm filter and had a total protein content of 25 mg/ml. Temperature (37°C) and pH (7) were regularly recorded. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Summary of the parameters obtained from experimental tests and numerical simulations. ::: Experimental tests Finite element analyses -------------------------------------- ----- -------------------- ------------------------- ----------- ----------- ----------- ----------- Compression (μm) 45° Mean: 115 (n = 1) Mean: 129 (n = 2) 98 132 \- \- Range: 125--132 60° Mean: 123 (n = 1) Mean: 140 (n = 2) 107 133 \- \- Range: 129--150 Penetration rate (μm/million cycles) 45° Mean: 11.8 (n = 2) Mean: 9.5 (n = 4) \- \- \- \- Range: 11.6--11.9 Range: 8.9--10.2 60° Mean: 14.7 (n = 2) Mean: 11.2 (n = 4) \- \- \- \- Range: 14.2--15.2 Range: 10.0--11.9 Von Mises stress 45° \- \- Fig. 3 Fig. 3 Fig. 3 Fig. 3 60° \- \- \" \" \" \" Contact pressure and/or area 45° Fig. 6 Fig. 6 Figs. 4-8 Figs. 4-8 Figs. 4-8 Figs. 4-8 60° \- \- \" \" \" \" 45° and 60° refer to the abduction angle. Experimental maximum compressions were analyzed from the initial loading cycles for two cup-cement combinations (12 acetabular cups, 6 experimental tests) at the same time (Fig. 1). Penetration rates of the cups, *i.e*. plastic deformations, were calculated after 5 million loading cycles. ::: ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Hip implant simulator and tested acetabular cups**. Two types of acetabular cups (up right, IP and Lubinus eccentric) were tested in a simulator (left) under cyclic axial loading. Two cups, fixed in a metal backing with bone cement, were tested at the same time. ::: ![](1475-925X-4-32-1) ::: During the tests, compressions of whole measurement systems, *i.e*. compressions of femoral heads into the cup-cement combinations during each cycle, were recorded. Maximum compression values during the initial cycle were used for the validation of the finite element models. As the compression was applied according to Fig. [1](#F1){ref-type="fig"}, the final value for one cup-cement combination was approximated to be half of the recorded value, and consequently six compression values were obtained (Table [1](#T1){ref-type="table"}). After 5 million cycles, the penetration rates (μm / million cycles) of the acetabular cups, *i.e*. plastic deformations, were analyzed using a coordinate tester Dea Global C 091508 (Dea, Tourin, Italy). The penetration rates were evaluated as the displacement of the center of the spherical surface of the cups, which were mathematically estimated after the cyclic tests. Finite element modeling ----------------------- The femoral head-acetabular cup-cement complexes, used in experimental tests, were processed for finite element analyses. In addition, two new designs were reconstructed to investigate the effect of clearance and polyethylene thickness on the contact pressure. Thus, altogether four 3-D meshes consisting of 15834 -- 16072 hexahedral elements (Fig. [2](#F2){ref-type="fig"}) were created: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Finite element meshes of the present study**. Four acetabular cups, cement mantle and femoral head were reconstructed by using hexahedral elastic elements. ::: ![](1475-925X-4-32-2) ::: 1\) Lubinus eccentric (inner diameter of acetabular cup = 28.5 mm, experimentally tested), 2\) Lubinus concentric, (inner diameter of acetabular cup = 28.5 mm, virtual design), 3\) IP1 (inner diameter of acetabular cup = 28 mm, experimentally tested), 4\) IP2 (inner diameter of acetabular cup = 28.5 mm, virtual design). The outer diameter of all acetabular cups was 52 mm and the diameter of the femoral head was 28 mm. Lubinus concentric was reconstructed to analyse differences between eccentric and concentric cups and, consequently, the effect of polyethylene thickness. The effect of clearance was studied by increasing the inner diameter of the IP cup (IP1 → IP2) to correspond to the diameter of Lubinus eccentric/concentric. Young\'s moduli (*E*) for femoral head, acetabular cups and bone cement, as determined experimentally, were 193 GPa, 0.69 GPa and 2.74 GPa, respectively. A friction coefficient of 0.05 was assumed for the contact between acetabular cup and femoral head \[[@B22]\]. The cup-cement interface was fixed and the boundary constaint was applied for the outer surface of the cement layer to simulate the metal backing used in the experimental tests. First, the compressions of the femoral heads into the acetabular cup -- cement combinations were analyzed using 3 kN loading force and compared with corresponding experimental findings. Second, the effect of abduction angle on the stresses of acetabular cups was investigated with different loading forces in order to simulate different body weights performing regular daily activities and exceptionally high impact loads. In addition to the experimentally used abduction angles of 45° and 60°, simulations were also conducted using an angle of 80°. Von Mises stress distributions in the cups and contact pressures both at cup-head and cup-cement interfaces were studied. Abaqus code (v6.3, Hibbitt, Karlsson & Sorenssen, Inc., Pawtucket, RI, USA) was used for the FE simulations. Results ======= Experimentally, the maximum compressions of the femoral heads during the initial loading cycle using the IP and Lubinus eccentric acetabular cups were 115 μm and 129 μm, respectively, at the 45° abduction angle (Table [1](#T1){ref-type="table"}). The corresponding values obtained in the FE simulations were 98 μm and 132 μm. At the abduction angle of 60°, experimental maximum compression values during the initial cycle using the IP and Lubinus eccentric cups were 123 μm and 140 μm, respectively. The corresponding values obtained in the FE simulations were 107 μm and 133 μm. The measured penetration rates of the cups were 11.8 μm (11.6 -- 11.9) and 9.5 μm (8.9 -- 10.2) per million cycles for IP and Lubinus eccentric, respectively, at the abduction angle of 45° (Table [1](#T1){ref-type="table"}). The corresponding values at the 60° angle, 14.7 μm (14.2 -- 15.2) and 11.2 μm (10.0 -- 11.9), were significantly higher (p \< 0.05, Wilcoxon signed ranks test, n = 6 + 6). The penetration rates were significantly (p \< 0.05, Mann-Whitney U-test, n = 4 + 8) higher for IP than for Lubinus eccentric. FE simulations disclosed that there were different stress distributions (von Mises stress) in Lubinus eccentric and IP cups (Fig. [3](#F3){ref-type="fig"}). Due to the varying geometries, the direction of the stress was different. The decrease of the thickness of the Lubinus acetabular cup (eccentric *versus*concentric) did not affect the stress distribution, but the increase in the gap between the head and the IP cup had a major influence on the von Mises stresses. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Stress distribution in the acetabular cups**. Von Mises stresses of a) Lubinus eccentric, b) IP1, c) Lubinus concentric and d) IP2 models at 45°, 60° abduction angles. ::: ![](1475-925X-4-32-3) ::: FE simulations showed that the abduction angle (45° -- 80°) had only a minor influence on the contact pressure between the head and the Lubinus eccentric cup (Figs. [4a](#F4){ref-type="fig"} and [5a](#F5){ref-type="fig"}). In contrast, the contact pressure on the surface of IP cup increased notably as a function of the abduction angle (Figs. [4b](#F4){ref-type="fig"} and [5a](#F5){ref-type="fig"}). However, the larger contact area between the IP cup and head induced lower contact pressure at low loading forces on the surface of IP compared to the corresponding situation with Lubinus eccentric (Fig. [6](#F6){ref-type="fig"}). At high loads, the edge of the IP acetabular cup experienced very high contact pressures, especially at high abduction angles (60° and 80°). The difference in the contact pressure between the eccentric and concentric Lubinus models at the inner surface of the acetabular cup was not significant (Figs. [4a, c](#F4){ref-type="fig"} and [5](#F5){ref-type="fig"}). As the increase of the radius of the IP acetabular cup from 28 mm to 28.5 mm reduced the contact area, the peak contact pressure increased (Figs. [4b, d](#F4){ref-type="fig"} and [5](#F5){ref-type="fig"}). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Contact pressure at the inner surface of acetabular cups at 45° 60° and 80° abduction angles.**a) Lubinus eccentric, b) IP1, c) Lubinus concentric and d) IP2. ::: ![](1475-925X-4-32-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Peak contact pressure at the cup-head interface as a function of loading force and abduction angle**. a) Lubinus eccentric and IP1, b) Lubinus concentric and IP2. ::: ![](1475-925X-4-32-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Contact area between humeral head and acetabular cup**. Experimentally analyzed (a) and numerically simulated (b) contact area between Lubinus eccentric and IP acetabular cup and femoral head. After cyclic tests of hip implants in a simulator, the surfaces of UHMWPE cups were covered with a thin layer of diluted colourant (Tuschierpaste blau, Emil Alberts GmbH, Gevelsberg, Germany), and femoral heads were positioned in contact with the acetabular cups in the loading direction. After removing the femoral heads, the presence of blue color indicated the contact area. The contact area in the FE-simulations was analyzed under 3 kN axial load which was the maximum load in the experimental cyclic tests. ::: ![](1475-925X-4-32-6) ::: At the interface between the acetabular cup and the cement, the contact pressure increased as a function of abduction angle in all models (Figs. [7](#F7){ref-type="fig"} and [8](#F8){ref-type="fig"}). Due to the larger contact area between cup and head and the smaller deformation of the cup, the contact pressure remained lower on the outer surface of the IP cup at all abduction angles (Figs. [7a, b](#F7){ref-type="fig"} and [8a](#F8){ref-type="fig"}). A decreased thickness of the acetabular cup (Lubinus eccentric → concentric) (Figs. [7a,c](#F7){ref-type="fig"} and [8](#F8){ref-type="fig"}) and the increased clearance between IP cup and femoral head (Figs. [7b,d](#F7){ref-type="fig"} and [8](#F8){ref-type="fig"}) elevated contact pressures on the outer surfaces of the cups. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **Contact pressure at the outer surface of acetabular cups at 45°, 60° and 80° abduction angles**. a) Lubinus eccentric, b) IP1, c) Lubinus concentric and d) IP2 ::: ![](1475-925X-4-32-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### **Peak contact pressure at cup-cement interface as a function of loading force and abduction angle**. a) Lubinus eccentric and IP1, b) Lubinus concentric and IP2. ::: ![](1475-925X-4-32-8) ::: Discussion ========== In the present study, 3-D finite element analyses and experimental cyclic tests were used for the estimation of the combined effects of acetabular cup geometry and orientation on the stresses in cemented UHMWPE. It was found that IP and Lubinus eccentric acetabular cups, manufactured by Link^®^, were subjected to considerably different stress distributions and contact pressures both on the inner and outer surfaces of the cups. The abduction angle and thickness of the cup, as well as the clearance between the femoral head and the cup, contributed significantly to stress experienced by the acetabular cup. Experimental and theoretical compressions during one loading cycle were consistent with each other. The slight discrepancies (2 -- 15%) were probably due to the fact that the processing of UHMWPE induces uncertainties in the diameters of the cups, whereas in the finite element simulations, optimal diameters, supplied by the manufacturer, were used. On the other hand, the initial fit of the head-cup-cement complex may not have been perfect in all of the experimental configurations, and this may have induced higher compression values. The experimentally determined penetration rates of the present study supported the conclusions based on the FE analyses. The experimental tests showed that the penetration rate increased significantly as a function of the abduction angle. This was consistent with the FE analyses which suggested that the contact pressure increased as a function of the abduction angle, even though this increase with the Lubinus model was relatively low. Even though the peak contact pressure remained lower on the inner surface of the IP acetabular cup compared to that on the Lubinus eccentric when simulating loading forces of less than 3 kN, the experimentally determined penetration rate was faster for the IP cup. The contact area between the IP cup and the femoral head was larger than that between Lubinus eccentric and the head (Fig. [6](#F6){ref-type="fig"}), which probably induced this discrepancy. The minor clearance and consequently high contact area have been shown to increase also volumetric wear \[[@B10],[@B15]\]. Previously, in simulator tests, wear rates of Lubinus eccentric cups were lower than those of IP cups and some other models \[[@B23]\]. This advantageous effect might be due to the formation of a protective protein layer on the cup surface as a result of the more closed lubrication environment in the Lubinus eccentric design. The behavior of contact pressure and penetration rate of the present study were consistent with earlier studies \[[@B8],[@B9],[@B12]\]. Jin *et al*. (1999) investigated contact pressure and area in eight combinations of femoral head and UHMWPE cup, and found similar behavior of contact pressure as a function of load as observed in the present study \[[@B9]\], although their model was 2-D axisymmetric and was not affected by the edge of the cup. Patil *et al*. (2003) studied the effect of abduction and anteversion angle of acetabular cup on the contact stress and wear of UHMWPE \[[@B8]\]. FE simulations of their study showed that peak contact stresses increased as a function of abduction angle, as was also found in the present study. Wear rates after testing in the hip simulator and linear head penetrations of the clinical study \[[@B8]\] as well as plastic deformations of the cups in the present study supported those findings. Some studies have suggested that the cup orientation plays minor role on the polyethylene wear, but more important is to determine the shape of the wear distribution \[[@B10],[@B24]\]. In the present study, the distribution of von Mises stress and contact pressure were highly influenced by the cup geometry. In general, slight differences between the previous studies and the present study are probably due to differences in the acetabular cup geometries, material properties and the loading protocols used in the experimental tests and analytical or numerical analyses \[[@B8],[@B25],[@B26]\]. Abduction angle clearly differentiated the acetabular cups. Lubinus eccentric seemed to be relatively forgiving in dealing with the variations in the abduction angle, whereas at 60° and 80° abduction angles, the IP cup experienced high peak contact pressures, especially at high loading forces. Evidently, geometrical factors were behind this kind of behaviour. The snap-fit primary locking mechanism in Lubinus cup eliminated the effect of the abduction angle. In this model the contact area between the ball and cup did not change as a function of the abduction angle. This minimizes the variations in the contact pressure and indicates that the survival of Lubinus acetabular cup may not be so operation sensitive. On the other hand, even though the geometry of IP cup permits a large range of motion, the contact area between IP cup and head changed as a function of the abduction angle, modifying the contact pressure. Furthermore, the edge of the IP cup was the most critical area in elevating the contact pressure, pointing to an increased risk for the cup failure \[[@B27]\]. The thickness of the acetabular cup and the clearance between cup and femoral head may have significant clinical relevance \[[@B9]-[@B12],[@B14],[@B15]\]. For instance, in the present study, a 13% decrease of the thickness at the upper part of the acetabular cup (Lubinus eccentric *versus*concentric) increased the peak contact pressure at the cup-cement interface at a 45° abduction angle by 20% (3 kN load). Consequently, eccentricity may decrease the risk for cup rotation. The increase of the clearance between IP acetabular cup and femoral head from 0 to 0.25 mm elevated peak contact pressure at cup-head interface by 27% and at the cup-cement interface by 65% (3 kN load). Also earlier studies have shown that the increased clearance and decreased thickness raise peak contact pressure \[[@B9],[@B12]\]. These percentual values suggest that rather minor changes in the geometrical parameters of UHMWPE acetabular cups may have significant consequences (e.g. rotation, wear, loosening) in long-term use. In the present study, all of the materials and boundaries were kept similar to permit a comparative evaluation of acetabular cups with different geometrical designs. In the clinical setting, however, the bone support and joint capsule will affect the stresses to which the acetabular cups are subjected \[[@B28]\]. Also, the contact area between the ball and the cup increases during the so-called running-in period, which typically lasts less than a year \[[@B29]\]. The bearing surface of the polyethylene accommodates to the head radius through deformation, wear and creep, so that the effective clearance approaches zero. Therefore, the contact pressure and fracture risk decrease substantially. After the first year, the penetration rate of the head into the cup is fairly constant \[[@B29],[@B30]\]. In long-term clinical use, this phenomenon may modify the behavior of the acetabular cups tested in the present study. Finally, cement fixation was assumed to extend up to the edge of the acetabular cups. However, in the clinical situation, the upper edge of the cup is not always fully supported by the cement mantle, especially at low abduction angles (\<45°), inducing variations in the stress and wear concentrations of the cups. A previous finite element model with elasto-plastic polyethylene cup proposed a perfect plastic yield after 8 MPa von Mises stress \[[@B15]\], which is lower than the maximum von Mises stress observed in the present study. However, the radiation-crosslinked UHMWPE tested in this study has a yield strength of 20 MPa or more (ASTM F648) \[[@B31]\] and the von Mises stress of IP cup at the abduction angles of 45° and 60° did not exceed 20 MPa under any load. In this study the cup was assumed to be elastic to enable analysis of potential peak stresses also at very high impact loads, not only during normal walking. Under low loads (\< 3 MPa), the von Mises stress did not exceed 20 MPa in any cup design. In addition, if von Mises stresses are calculated for the IP acetabular cup at the 3 kN load, which was the maximum value used in the experimental tests of the present study, only approximately 8 MPa peak von Mises stress is reached at an abduction angle of 45°. Plasticity, nonlinear elasticity or viscoelasticity \[[@B8]\] of UHMWPE would have changed absolute pressure values, but not the conclusions of the present study. FE modeling provides a rapid and inexpensive estimate of implant-related factors (e.g. geometry of the acetabular cup) and surgery-related factors (e.g. abduction angle of the implant), but the final judgement is based on laboratory tests and clinical data. The hip arthroplasty registers \[[@B18],[@B19]\], used to obtain clinical reference data, do not differentiate between Lubinus eccentric and IP acetabular cups, but the present study points to differences in the performances of these cups. The results of this study indicate that it would be beneficial if cups of different designs were recorded separately in the registers, rather than combining cups from a single hip system. All experimental and numerical data presented in this study address the importance of optimizing the geometry and orientation of the acetabular cup before and during the operation, respectively. These are factors which may affect significantly failure of THR. Conclusion ========== FE analyses of the present study showed that the geometrical design, thickness and abduction angle of acetabular cup, as well as the clearance between cup and head modify the mechanical stresses experienced by UHMWPE. These factors should be taken into account in the future development of THR and the FE technique is a powerful tool for this purpose. Authors\' contributions ======================= RKK was involved in the design of the study, developed the FE models and calculated numerical predictions, as well as drafted the manuscript. AK participated in the design of the study, experimental tests and data analyses, as well as in the statistical analyses. YTK participated in the design of the study and coordination as well as helped in the manuscript preparation. SSS was involved in conceiving the study, coordination and manuscript preparation. RL participated in conceiving the study, coordination, experimental tests, FE model development and manuscript preparation. All authors have read and approved the final manuscript. Note ==== \*Partially cross-linked polyethylene cups are processed according to ISO 5834-II and ASTM F648 standards. Cups are compression molded and gamma irradiated in a vacuum to a level of 27 kGy. Acknowledgements ================ Financial support from the Academy of Finland, Sigrid Jusélius Foundation, Helsinki, Finland, and the Invalid Foundation, Helsinki, Finland, is acknowledged. The authors thank MSc Petteri Väänänen for conducting the experimental measurements and Esa Miettinen for technical support.

PubMed Central

2024-06-05T03:55:58.438381

2005-5-17

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173123/", "journal": "Biomed Eng Online. 2005 May 17; 4:32", "authors": [ { "first": "Rami K", "last": "Korhonen" }, { "first": "Arto", "last": "Koistinen" }, { "first": "Yrjö T", "last": "Konttinen" }, { "first": "Seppo S", "last": "Santavirta" }, { "first": "Reijo", "last": "Lappalainen" } ] }

PMC1173124

Background ========== High Frequency Chest Compression (HFCC) \[[@B6],[@B9],[@B19],[@B29]\] is a useful and popular therapy for clearing bronchial airways of excessive or thick mucus since it does not require patients to do any directed efforts for respiration while therapy is given, unlike other airway clearance techniques such as active cycle of breathing or autogenic drainage \[[@B17]\]. A HFCC machine pumps air into an inflatable jacket worn by patient. By means of a surrogate piston, sine waveform compression pulses with frequencies ranging from \~5 Hz to \~21 Hz are supplied to the thorax of a patient through the jacket. These pulses squeeze and vibrate the patient\'s thorax at prescribed frequencies. These actions help in the evacuation of mucus through changing the rheological property of the mucus and airflow oscillation. King et al. \[[@B13]\], Krumpe et al. \[[@B15]\], and Tomkiewicz at al. \[[@B28]\], showed that HFCC pulses decrease the viscosity of mucus and helps evacuation. A more important consequence is the respiratory airflow oscillated during use of HFCC, which results from variation of intrapleural pressure. Lapin \[[@B17]\] and Warwick \[[@B30]\] pointed out that airflow is the most important factor for mucus transport since airflow produces the shear stresses for evacuation of mucus. Although several models have been developed to simulate respiration by other researchers and their models successfully worked for their own purposes, these models are not appropriate for HFCC simulation because these models were either linearized \[[@B7],[@B8]\] or because they assumed that respiration was driven by a mechanical ventilator at the mouth \[[@B5],[@B20]\]. The model presented in this paper provides reliable simulation results on fast change of intrapleural pressure (*P*~*pl*~) altered by HFCC because this model is described with nonlinear equations and *P*~*pl*~is selected as the driving force of respiration. The nonlinear characteristics of the respiratory system are not easily noticed during quiet tidal breathing, but the airflow signal oscillation measured at the mouth during use of HFCC is a strong indicator of the nonlinear characteristics of the respiratory system. In this study, we simulated the respiratory system with a computational model that carefully reflected its nonlinear characteristics. The model is an electrical circuit analogue, in which nonlinear resistors (R\'s), capacitors (C\'s), and inductors (L\'s) represent airway resistance, lung capacitance, and inertance of air, respectively. Just like in a living organ, the driving force of this model is *P*~*pl*~which is a superposition of HFCC pulses on the spontaneous breathing effort. We compared the simulation data to the *in vivo*data to demonstrate and understand the characteristics of airflow signal oscillation. Methods ======= Conceptual model of the respiratory system ------------------------------------------ Our modelling and simulation for reproducing airflow signals required simplification of airway structure in the lung. The geometrical and dimensional structures of the airways were proposed by several researchers, among which Horsefield et al. \[[@B10]\] and Weibel \[[@B31]\] dissected and measured the human lungs and airways, and more recently, and Tawhai et al. \[[@B11]\] and Kitaoka et al. \[[@B14]\] proposed algorithmic approaches for reconstructing the branching structure. We employed Weibel\'s morphometry of the lung, which provides the geometries and the dimensions based on the symmetric dichotomous structure of airway branching when the lung volume is assumed to be 75% of the total lung capacity (*TLC*). According to his morphometry, the trachea is defined as airway generation 0 and it is separated into two geometrically identical daughter branches. Each daughter is repeatedly branched up to 22 times, thus the lung is considered to have 24 (0--23) airway generations and the number of airway branches total \~1.7 × 10^7^. In each airway generation, there are 2^*z*^(z is a generation number) identical branches whose lengths and radii are provided and the dimensions of airway branches in each generation differ from generation to generation. He also suggested that airway generations 0--16 comprise the conducting zone, whereas airway generations 17 -- 23 make up the respiratory zone in which gases are exchanged. Based on Weibel\'s morphometry of the lung, we simplified the geometry of the airways. For airway generation *Z*in conducting zone, a bundle of 2^*z*^identical airway branches are considered as the big tube whose cross sectional area equals to 2^*z*^times the cross sectional area of a single airway branch. The tubes for each airway generation are represented by RCL T-networks shown in fig. [1 (b)](#F1){ref-type="fig"}, and evaluation of the R\'s, C\'s, and L\'s in the RCL T-network are explained in equations (4) -- (6). Meanwhile, the respiratory zone is considered as a big lump, the alveolar space, since alveolar ducts and sacs are scattered throughout the respiratory zone \[[@B8]\]. Although the upper airway is not presented in Weibel\'s morphometry, it is one of the chief sites for airway resistance. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### \(a) The conceptual model based on Weibel\'s morphometry of the lung \[31\]. The lung and airways are comprised of five regions, namely, the alveolar space, the conducting airway zone, the pleural cavity, the thorax outside the lung, and outside the thorax. (b) RCL T-network that represents a region (airway generation) in the conducting airway zone. (c) Electrical circuit analogue converted from (a). For acronyms in the figure, see list of abbreviations. ::: ![](1475-925X-4-34-1) ::: Our conceptual model as a summary of the simplification, \'the upper airway + 17 conducting airway generations + the alveolar space\', is demonstrated in fig. [1 (a)](#F1){ref-type="fig"}, and circuit analogue converted from fig. [1 (a)](#F1){ref-type="fig"} is shown in fig. [1 (c)](#F1){ref-type="fig"}. Dimensions of airway branches ----------------------------- Since the dimensions of every single airway branch in the lung vary during respiration, it is crucial to track them to extract proper values of R\'s, C\'s, and L\'s. The radii of the airway branches in airway generation *Z*are determined by transmural pressure (*P*~*tmZ*~), pressure difference between the pleural cavity and inside airway generation *Z*. In order to determine *P*~*tmZ*~, we need to rely on the study about transpulmonary pressure (*P*~*tp*~), pressure difference between the pleural cavity and alveolar space. According to Salizar et al. \[[@B26]\], *P*~*tp*~is a function of lung volume (*LV*) and *TLC*. With given *LV*, *P*~*tp*~is obtained by, *P*~*tp*~= -*log*(1 - *LV*/ *TLC*) × 7.22. (1) If *LV*is functional residual capacity (*FRC*) and no airflow exists in the airways, the lung is at rest. In this situation, air pressure at any site is the same and atmospheric (zero), therefore *P*~*tmZ*~and *P*~*tp*~are the same. Our simulation begins with assuming that the lung was at rest. The equation of Lambert et al., or Lambert\'s tube law \[[@B16]\], is a function of *P*~*tmZ*~, and gives the ratio of the cross sectional area (*A*~*Z*~) to the maximum cross sectional area of an airway branch (*A*~*maxZ*~) in airway generation *Z*. That is, *A*~*Z*~/ *A*~max\ *Z*~= 1.0 - (1.0 - *α*~0~)(1.0 - *P*~*tmZ*~/ *P*~0~)^-*N*^ (2) where *α*~0~, *α*~0~\' and *N*are constants for each airway generation given from Lambert\'s tube law, and *P*~0~= (*α*~0~-1)*N*/*α*~0~\'. As mentioned earlier, the dimensions of the airway branches provided by Weibel\'s morphometry are based on 75 % of *TLC*. Let *A*~*Z75*~be Weibel\'s cross sectional areas in airway generation *Z*. By using ![](1475-925X-4-34-i1.gif) *A*~*maxZ*~for each airway generation can be found. Since *A*~*maxZ*~does not change whether or not the lung is at rest, the equation is valid to find out *A*~*Z*~during respiration. The length of an airway branch is assumed to vary in away that; *l*~*z*~/ *l*~*FRC*~= *r*~*z*~/ *r*~*FRC*~ (3) where *l*~*z*~and *r*~*z*~are the length and the radius of an airway branch in airway generation *Z*, respectively. *l*~*FRC*~and *r*~*FRC*~are the length and the radius on *FRC*, respectively. Equivalent circuit elements modelling ------------------------------------- The RCL T-network consists of two resistors (*R*~*gZ*~), two inductors (*L*~*gZ*~), a capacitor (*C*~*gZ*~), and a DC voltage source (*P*~*tm\_ini*~). *R*~*gZ*~and *L*~*gZ*~are the airway resistance and the inertance of air in the entire airway generation *Z*, respectively. *C*~*gZ*~, the airway capacitance of the entire airway generation *Z*, represents inflation and deflation of the non-rigid airway wall during respiration. *P*~*tm\_ini*~is the initial value of *P*~*tmZ*~. This initial pressure counterbalances *P*~*pl*~, which is negative when the lung is at rest. The values of the circuit elements in the RCL T-network of airway generation *Z*can be obtained by totalling the 2^*z*^airway branches. Therefore, *R*~*gZ*~is given by ![](1475-925X-4-34-i2.gif) Similarly, *L*~*g*~*z*is given by ![](1475-925X-4-34-i3.gif) and *C*~*gZ*~is *C*~*gZ*~= *C*~*sZ*~× *N*~*z*~\[ml/cmH~2~O\], (6) where *R*~*sZ*~, *L*~*sZ*~, and *C*~*sZ*~indicate the values of a single airway branch in airway generation *Z*. *N*~*z*~is 2^*z*^, the number of airway branches in generation *Z*. *R*~*sZ*~depends on the classic Poiseuille equation and the *Zeta*correction factor proposed by Pedley et al. \[[@B21]\]. The *Zeta*correction factor is given by ![](1475-925X-4-34-i4.gif) and ![](1475-925X-4-34-i5.gif) *R*~*e*~is Reynolds number ![](1475-925X-4-34-i6.gif), *V*is air velocity, *ρ*is the density of air, and *η*is the viscosity of air. Because electric current flows at the speed of light in an electrical circuit, it is necessary to compensate for the incomparably slower behaviour of airflow in the airways by placing inductors. Inertance of air in a single airway branch (*L*~*sZ*~) depends on the dimension of the airway branch \[[@B5]\] and it is given by ![](1475-925X-4-34-i7.gif) To compute the values for *C*~*sZ*~in equation (2), relation between *A*~*Z*~and *P*~*tmZ*~in airway generation *Z*, is used again. That is, ![](1475-925X-4-34-i8.gif) ![](1475-925X-4-34-i9.gif) The alveolar space is represented by a capacitor in the circuit analogue. Alveolar capacitance (*C*~*as*~) is the ratio of the alveolar volume (Δ*AV*) to Δ*P*~*tp*~. Lung volume (*LV*) consists of airway volume and alveolar volume, and airway volume is negligible compared to alveolar volume. Therefore, ![](1475-925X-4-34-i10.gif) which implies that ![](1475-925X-4-34-i11.gif) represents *C*~*as*~. It can be obtained by equation (1); however, the equation did not consider the hysteresis of *P*~*tp*~-volume curves that is caused by several proposed reasons \[[@B1]\]. To overcome this, we defined *C*~*as*~as ![](1475-925X-4-34-i11.gif) \**k*(*k*\<0), and *k*was continuously changed over the time course of the simulation. The values of *k*were empirically determined to achieve the acceptable shape of the hysteresis, which is shown in fig. [2](#F2){ref-type="fig"}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### *P*~*tp*~-volume curves of the lung model during use of HFCC. Although the initial *LV*(*FRC*) is 2700 ml, *LV*during use of HFCC is smaller. The curves became jagged due to HFCC pulses. ::: ![](1475-925X-4-34-2) ::: A resistor and an inductor characterize the upper airway, from the nasal/oral cavity to larynx. The equations for the resistance of the upper airway (*R*~*ua*~) that Jackson *et al.*\[[@B12]\] validated are described below: ![](1475-925X-4-34-i12.gif) where *F*~*ua*~is the airflow rate in the upper airway. As Marchal et al. \[[@B18]\] estimated, the value of the inductor that represents inertance of air is 0.00003 \[cmH~2~O·s^2^/ml\]. Numerical methods for nonlinear circuit analogue ------------------------------------------------ Although the number of elements is manageable and the structure of circuit is fairly simple, analysis of the circuit analogue in fig. [1 (c)](#F1){ref-type="fig"} is not trivial since the values of all the energy-storing (C\'s and L\'s) elements as well as resistive elements (R\'s) change at every sequence of the simulation time-step. To exemplify a general idea of numerical methods for the nonlinear circuit analogue, analysis of a simple nonlinear second-order system in fig. [3](#F3){ref-type="fig"} was demonstrated. The system of fig. [3](#F3){ref-type="fig"} is described by an equation: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### A nonlinear second-order system in the form of an electrical circuit. *i*is electrical current, and *v*~*C*~is voltage difference between each end of *C*. The value of the resistor (R) in this figure is dependant on *i*as well as *v*~*C*~whereas the capacitor (C) and the inductor (L) are dependent only on *v*~*C*~. V~in~(*t*) is a voltage source for this simple circuit system. Nonlinear circuit elements in this circuit imply the nonlinearity of fig. 1 (c). ::: ![](1475-925X-4-34-3) ::: ![](1475-925X-4-34-i13.gif) where *v*~*C*~is the voltage difference between each end of the capacitor. Note that *R*(•) is a function of *i*and *v*~*C*~, and *L*(•) and C(•) are functions of *v*~*C*~just like in fig. [1 (c)](#F1){ref-type="fig"}. Using backward Euler approximation \[[@B22]\], equation (14) is converted to a difference equation: ![](1475-925X-4-34-i14.gif) where Δt is time difference between sequence \[n\] and \[n-1\]. Equation (15) is the same as ![](1475-925X-4-34-i15.gif) By the definition of backward Euler approximation, the current of the circuit ![](1475-925X-4-34-i16.gif). Then equation (16) can be restated as an equation; ![](1475-925X-4-34-i17.gif) In equation (17), the right hand side consists of all known values, and the left hand side is a function of *v*~*C*~\[n\]. Suppose that *v*~*C*~\[n\] is *x*, equation (17) can be expressed as *f*(*x*) = c, (18) where c is a constant. Equation (18) can be easily solved using an iteration method \[[@B24]\]. To do this computation, MATLAB (Mathworks, Natick, MA) codes were written. Protocols for measurement airflow signals ----------------------------------------- In this study, *The Vest*™ (Advanced Respiratory, St.Paul, MN, USA; now named Hill-Rom Co.,Inc.), which delivers sine waveform compression pulses, was used for application of HFCC to a subject for measuring and recording the airflow signals at the mouth. The general usage of HFCC device was early described in the review of Hansen et al. \[[@B9]\] and the typical respiratory airflow during use of HFCC is shown in fig. [4](#F4){ref-type="fig"}. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### \(a) The airflow and (b) the oesophageal pressure measured by Fink et al. \[6\] (modified with the permission), which shows typical respiratory airflow during use of HFCC. The positive numbers of airflow rate represent inspiration and the negative numbers represent expiratory airflow. Oesophageal pressure reflects the intrapleural pressure (*P*~*pl*~) that is altered by HFCC pulses (tiny peaks). ::: ![](1475-925X-4-34-4) ::: The subject sat upright on a chair for measuring and recording the airflow signals at the mouth with an in-house built electronic spirometer. The subject worn a nose clip and breathed through a mouthpiece. After the HFCC device is properly set up, the compression pulses were applied and then the subject made several slow and large, but not to *TLC*, breaths. During the breaths the subject hold his glottis open until data collection was completed. This protocol was followed for the low (5 Hz), high (21 Hz), and medium frequencies (15 Hz) of HFCC pulses. Before each frequency recording of airflow signals, the subject rested for one minute. To ensure that the subject had adapted to HFCC pulses and had reached a steady state, only the last ten seconds of the one- minute breathing were recorded and analyzed for our study. Results ======= For the simulation, the parameters of the lung were determined to represent a normal healthy lung. The *TLC*, *FRC*, and *RV*of the model lung are 6000 ml, 2700 ml, and 1000 ml, respectively. The ambient atmospheric pressure is assumed to be zero, and the air density and viscosity are 0.00113 g/cm^3^and 0.00019 g/cm·s, respectively. It was assumed that there would be no airflow in the airways and that initially the *LV*would be the same as the *FRC*. Change of *P*~*pl*~is the primary driving force of respiration and is initially -4.32 cmH~2~O, which is the counterbalance to initial *P*~*tp*~given by equation (1). Fig. [4](#F4){ref-type="fig"} is the airflow the oesophageal pressure measured by Fink et al. \[[@B6]\], which shows typical respiratory airflow during use of HFCC. Fig. [5](#F5){ref-type="fig"} shows the subject\'s airflow signals during ten seconds at the three different frequencies. In this figure, each airflow signal is also viewed as the low-pass filtered and the high-pass filtered curves. The high-pass filtered curve indicates the fast airflow signal oscillation generated by HFCC pulses (HFCC airflow signal) whereas the low-pass filtered curve is the airflow during the spontaneous breathing effort (spontaneous airflow signal) of the subject. We divided one cycle of the respiration into four phases. Phase I is the portion of the inspiration phase when the spontaneous airflow signal is greater than the amplitudes of HFCC airflow signal. In this phase *LV*increases. During phase II, the spontaneous airflow signal stays within the amplitudes of the HFCC airflow signal. Phase II is the pause before expiration begins. The amplitudes of HFCC airflow signals change considerably as the phase moves from I to II. To emphasize the difference of HFCC airflow signals, phase II were consciously prolonged and they are longer than phase II in fig. [6](#F6){ref-type="fig"}. In phase III, passive or active expiration begins and the amplitudes of HFCC airflow signals decrease to about phase I amplitudes. The low-pass curve gets greater to the negative direction curve than the high-pass curve. The next and last phase is phase IV. Phase IV is the resting period before inspiration, begins. During this phase the amplitudes of HFCC airflow signals again become greater than the spontaneous airflow signal. To reproduce similar airflow signals, a cycle of respiration from phase IV to phase IV was simulated using our computational model. The simulation results with the three frequencies are shown in fig. [6](#F6){ref-type="fig"}. ::: {#F5 .fig} Figure 5 ::: {.caption} ###### The airflow signals measured at the mouth of the subject, the high-pass filtered (HFCC airflow signal) and the low-pass filtered (spontaneous airflow signal) curves. The subject was using HFCC with (a) 5 Hz, (b) 15 Hz, and (c) 21 Hz. Regardless of the frequencies, larger spontaneous airflow signals result in smaller HFCC airflow signals. Since it is difficult to breathe hard during 21 Hz, spontaneous airflow signals in (c) are smaller than in (a) and (b). Phase I is the portion of the inspiration phase when spontaneous airflow signals are greater than the amplitudes of HFCC airflow signals. During phase II, spontaneous airflow signals stay within the amplitudes of HFCC airflow signals. In phase III, expiration begins, and the amplitudes of airflow oscillation decrease to about the amplitudes during phase I. Spontaneous airflow signals get greater to the negative direction than HFCC airflow signals. Finally phase IV, the resting period before inspiration, begins, and the amplitudes of HFCC airflow signals again become greater than spontaneous airflow signals. To emphasize the difference of HFCC airflow signals, phase II were consciously prolonged and they are longer than phase II in fig. 6. ::: ![](1475-925X-4-34-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### Simulated respiration, spontaneous airflow signals and HFCC airflow signals. The frequencies of HFCC were set (a) 5 Hz, (b) 15 Hz, and (c) 21 Hz. Definitions for Phases I, II, III, and IV in this figure are the same as those in the text and fig. 5. ::: ![](1475-925X-4-34-6) ::: From the measured values and the simulation of airflow signals at the mouth, we could observe two important phenomena during respiration pertaining to the airflow signal oscillation generated by HFCC. First, the amplitudes of HFCC airflow signals in phases I and III were smaller than those in phases II and IV. Second, the amplitudes in phases I and III became even smaller as the spontaneous airflow signal became greater. We used the simulation results to investigate how the nonlinearity of airway resistance, lung capacitance, and inertance of air characterized the respiratory airflow. The simulation was repeated after setting one of the three properties as a linear constant value. Fig. [7](#F7){ref-type="fig"} compares the linear values with the nonlinear values of the three properties. The linear values are means of the nonlinear values during the simulation. Fig. [8](#F8){ref-type="fig"} presents the simulation results of the lung model under the imaginary assumptions. Fig. [8 (a)](#F8){ref-type="fig"} is normal respiratory airflow at 15 Hz, and fig. [6 (b)](#F6){ref-type="fig"} and fig. [8 (a)](#F8){ref-type="fig"} are from the same simulation data. Fig. [8 (b)](#F8){ref-type="fig"} demonstrates the predicted airflow signals at the mouth with linear airway resistance, which indicates the amplitudes of HFCC airflow signals do not vary significantly. Fig. [8 (c)](#F8){ref-type="fig"} is the airflow when lung capacitance is set to a linear value. Just like fig. [8 (a)](#F8){ref-type="fig"}, HFCC airflow signals are the largest when spontaneous airflow signals are close to zero. Therefore, it can be presumed that the nonliearity of lung capacitance does not play a role in the non-uniformity of HFCC airflow signals. And neither is inertance of air. Fig. [8 (d)](#F8){ref-type="fig"}, which shows the simulation data with linear inertance of air, is almost identical to fig. [8 (a)](#F8){ref-type="fig"}. This indicates that the inertance of air is also not a factor in the non-uniformity of HFCC airflow signals. Fig. [9](#F9){ref-type="fig"} shows the HFCC airflow signals as a function of spontaneous airflow signals based on the same simulation data shown in fig. [8](#F8){ref-type="fig"}. Fig. [8 (a), (c)](#F8){ref-type="fig"}, and [8 (d)](#F8){ref-type="fig"} indicate that larger spontaneous airflow signals result in smaller oscillations of HFCC airflow signals whereas HFCC airflow signals do not seem to be related to spontaneous airflow signals in fig. [8 (b)](#F8){ref-type="fig"}. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### \(a) Airway resistance, (b) lung capacitance, and (c) inertance of air during the simulation used for fig. 8. The linear (fixed) values are the mean values of each nonlinear (varying) value. ::: ![](1475-925X-4-34-7) ::: ::: {#F8 .fig} Figure 8 ::: {.caption} ###### Simulated spontaneous airflow signals and HFCC airflow signals, (a) when the values of airway resistance, lung capacitance, and inertance of air are nonlinear (varying), (b) when the value of airway resistance is linear (fixed), (c) when the value of lung capacitance is linear (fixed), and (d) when the value of inertance is linear (fixed). The frequencies of each plot are 15 Hz, so (a) is the same as fig. 6 (b). In (b), the amplitudes of HFCC airflow signals almost do not vary. ::: ![](1475-925X-4-34-8) ::: Discussion ========== The number of the airway branches is estimated to be about 17 million \[[@B31]\], and the dimensions of each airway branch vary while the lung is being inflated and deflated during respiration. To reduce the computational burden, models of the respiratory system are often simplified by ignoring the nonlinear natures of the respiratory system \[[@B7],[@B8]\]. Linearization entails considerable flexibility in numerical analysis since linear circuits are computationally much cheaper than nonlinear circuits, either on a time-domain or a frequency-domain basis. However, even during the slowest breathing, airway resistance, lung capacitance, and inertance of air change due to the variation of airway dimensions as well as turbulence. In particular, increase of the airway resistance due to turbulence is so drastic that inaccurate evaluation of airway resistance may result in a misleading simulation result. Our simulation results demonstrated that such errors are more likely when HFCC intervention is applied. Evidence for this was that the result from a linear model was significantly different from that from a nonlinear model (fig. [8](#F8){ref-type="fig"} and [9](#F9){ref-type="fig"}). As parts of the respiratory system, lung capacitance and inertance of air are also nonlinear although it was observed that their nonlinearity did not cause the non-uniform amplitudes of HFCC airflow signals. However, this does not imply that the nonlinear characteristics of lung capacitance and inertance of air have no role in the simulation of the respiratory system. Dimensional changes of the airway branches in the lung are responsible for the nonlinearity of lung capacitance and inertance of air. When dimensions of the airways are involved for simulation and prediction, the nonlinearity is very important for accurate results. For example, in the study of Sohn et al. \[[@B27]\], our model is used for estimation of air velocity, which is airflow rate ÷ cross sectional area. Since the cross sectional areas of airway branches do not vary linearly, lung capacitance and inertance of air also should not be linear in order to avoid discrepancy. Ideally, studies about airflow in the airways would be best resolved by CFD (Computational Fluid Dynamics), however, using a CFD approach in this study presents several problems that cannot be overcome by modern technology. First, the turbulence mechanism is not completely known \[[@B23]\]. Even during the slowest breathing manoeuvre, turbulence exists in the proximal airways, and ignoring turbulence would not give true simulation results. Another obstacle is that the whole lung cannot be taken into consideration even with the latest supercomputing power. A CFD approach to simulate dynamics of airflow interacting in a huge number of airway branches in the lung requires extremely massive computation. There is no way to deal with it, if any, its computation time would be incredibly long. It should be also pointed out that CFD techniques for airflow in non-rigid wall tubes are not yet mature. Although not perfect, our circuit analogue model allows us to effectively simulate the nonlinear characteristics of the respiratory system. It is well known that respiratory system impedance consists of airway impedance and chest wall (tissue) impedance \[[@B2]-[@B4]\]. Since the driving force of airflow in our model is the intrapleural pressure altered by HFCC pulses transferred from the body surface, our model does not necessarily incorporate chest wall (tissue) impedance. However, it is not clear how effectively the chest wall transfers HFCC pulses to the pleural cavity. Other factors should also be considered for impedance between the jacket of HFCC and the pleural cavity -- such as clothes, posture, and tightness of the jacket. Currently, based on observations of Milla et al. \[[@B19]\], we assume that HFCC pulses on the body surface are transferred to the pleural cavity without any distortion of pulsation waveforms although some attenuation may exist. The HFCC device was initially developed only for cystic fibrosis patients who normally have healthy lungs \[[@B29]\]; consequently we modelled a healthy lung as the first step of the research. As this medical treatment becomes widely applicable to other lung diseases, we are planning to develop models that can be used to simulate HFCC on various lung diseases in the future. ::: {#F9 .fig} Figure 9 ::: {.caption} ###### HFCC airflow signals are demonstrated as a function of spontaneous airflow signals based on the same simulation data shown in fig. 8. Nonlinear airway resistance (a, c, and d) produces the non-uniform amplitudes of HFCC airflow signals. ::: ![](1475-925X-4-34-9) ::: Conclusion ========== In this study, the airflow signals measured at the mouth during use of HFCC are viewed as a composite of two causes: the spontaneous breathing effort and HFCC pulses. However, since the respiratory system is nonlinear, airflow signals at the mouth are not a mere superposition of the effects from these two causes. In laboratory measurements, the amplitudes of the airflow signal oscillation varied considerably despite the uniformity of the HFCC pulses. After confirming that the simulation results matched up with the observations, we analyzed the simulation data to explain the observed inconsistency in the HFCC airflow signal amplitudes. We found that the amplitudes of HFCC airflow signals behave as a function of spontaneous airflow signals. This is due to the nonlinearity of the respiratory system, particularly variations in airway resistance. The findings in this paper may not be immediately applicable for HFCC therapy, but they do lead to ways to better prescribe HFCC therapy. Most importantly, the usefulness of our computational simulation and the model-based approach as a tool to understand clinical observations of HFCC was well demonstrated in this paper. List of abbreviations ===================== HFCC: High Frequency Chest Compression *FRC*: Functional Residual Capacity *TLC*: Total Lung Capacity *RV*: Residual Volume *LV*: Lung volume *AV*: Alveolar volume R: resistor C: capacitor L: inductor *A*~*Z*~: Airway cross sectional area in airway generation *Z* *A*~*maxZ*~: Maximum Airway cross sectional area in airway generation *Z* *A*~*Z75*~: Airway cross sectional area in airway generation *Z*when *LV*is 75% of *TLC* *C*~*as*~: Alveolar capacitance *C*~*gz*~: Total airway capacitance in airway generation *Z* *C*~*sZ*~: Single airway capacitance in airway generation *Z* *F*~*inZ*~: Incoming airflow in airway generation *Z* *F*~*outZ*~: Outgoing airflow in airway generation *Z* *F*~*ua*~: Airflow in the upper airway *L*~*gz*~: Total inductance (inertance of air) in airway generation *Z* *L*~*sZ*~: Single inductance (inertance of air) in airway generation *Z* *l*~*Z*~: The length of an airway branch in airway generation *Z* *N*~*Z*~: Number of airway branches in airway generation *Z* *P*~*alv*~: Alveolar pressure *P*~*awZ*~: Airway pressure in airway generation *Z* *P*~*pl*~: Pleural pressure or intrapleural pressure *P*~*tm\_ini*~: The initial value of transmural pressure *P*~*tmZ*~: Transmural pressure in airway generation *Z* *P*~*tp*~: Transpulmonary pressure *P*~*tp\_ini*~: The initial value of transmural pressure *R*~*gz*~: Total airway resistance in airway generation *Z* *R*~*sZ*~: Single airway resistance in airway generation *Z* *r*~*z*~: The radius of an airway branch in airway generation *Z* Authors\' contributions ======================= KS designed all the necessary computational and experimental procedures, analysed data to reach the conclusion, and also prepared the texts and figures in the manuscript. WJW conceived this study, helped organizing the manuscript, and advised all aspects of physiology and medicine for this study. YWL and JL helped measuring the respiratory airflow. JEH advised all aspects of technology and engineering for this study.

PubMed Central

2024-06-05T03:55:58.440825

2005-5-19

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173124/", "journal": "Biomed Eng Online. 2005 May 19; 4:34", "authors": [ { "first": "Kiwon", "last": "Sohn" }, { "first": "Warren J", "last": "Warwick" }, { "first": "Yong W", "last": "Lee" }, { "first": "Jongwon", "last": "Lee" }, { "first": "James E", "last": "Holte" } ] }

PMC1173125

This book is intended for biologists, medical physicists, biomechanical engineers and clinicians who are involved in bone studies. For these practitioners, the book is a comprehensive review of physical measurement techniques applied to bone, with useful emphasis on practical aspects such as protocols of measurement, sample preparation and sources of errors. Chapters were written by leading scientists in the relevant fields, and overall, the outcome is a well-written, comprehensive and useful textbook that will definitely serve the biomedical community. The book contains 18 chapters classified into four major sections: Introduction, Invasive Techniques, Ionizing Radiation Techniques, and Non-Ionizing Techniques. As typically occurs with books that are collections of chapters from several authors, the quality is not uniform throughout. For example, while some figures are clear and illustrative, other figures suffer poor quality or were reproduced at a too low resolution. The important points of strength and weakness are summarized below. The introduction includes a comprehensive review of bone anatomy and physiology, in which each skeletal component -- bone cells, matrix and mineral -- is separately addressed. It would have been useful here to also refer to tissues that connect to bone (tendons, ligaments, cartilage) and produce mechanical loads which act on bone (muscles) and particularly, to the interfaces between bone and these soft tissues. A review of the pathophysiology, etiology, epidemiology, and economic impact of osteoporosis follows, but other, less common bone diseases are only briefly mentioned. A useful chapter on biology safety considerations, with focus on bone studies, is also included in the introduction. I found this chapter particularly helpful as a guide for students or laboratory technicians who become involved in preparing bone sections for histopathology or mechanical testing. Typically, the information on how to avoid accidents such as when dealing with bone saws, chemicals for demineralizing, fixing and staining bone samples, radiation, and biological waste, is posted in internal laboratory or institutional regulations. This is the first textbook chapter that integrates this information specifically for experiments with bone, and I will certainly make it an obligatory reading material for those involved in bone studies in my laboratory. The last chapter in the introduction concerns instrument evaluation. While appreciating the important information provided on measurement errors, precision, and accuracy, I found this chapter to be too general in nature, and was not convinced that it should have been included in a textbook that focuses on bone measurements. For example, the classical Bland and Altman paper in Lancet (1986) \[[@B1]\] which is discussed in detail as related to bone mineral density measurements, originally referred to flow and oxygen saturation measurements, and can be generalized for bone as for any other application. Accordingly, this chapter could have fit a textbook on biomedical instrumentation, but it does not contain contents that are specific enough for bone studies. The second section, on invasive techniques, puts heavy emphasis on the chapter which concerns mechanical testing of bone. Unfortunately, the contents of this chapter substantially overlap information already provided in the Bone Mechanics Handbook \[[@B2]\] which is a classic in this field. I was disappointed to find out that the tabulated data of mechanical properties for cancellous bone, cortical bone and individual trabeculae mostly summarize publications from the 1970\'s and 1980\'s (with few references to publications from the 1990\'s). Publications from 2000\'s (e.g. \[[@B3],[@B4]\]) are not covered. I would have expected this new book to cover a more recent literature than that covered in the Bone Mechanics Handbook published on 2001 \[[@B2]\] or in Prof. J.D. Currey\'s book \"Bones: structure and mechanics\" published on 2002 \[[@B5]\], particularly because of the growth in attention to bone mechanical properties as osteoporosis becomes a major epidemic. The issue of bone viscoelasticity, which is discussed in detail in \[[@B2]\], is only briefly mentioned here in terms of the dependence of the elastic modulus of cortical bone on the strain rate. On the contrary, I found the subsequent chapters on bone histomorphometry, microscopy and related techniques very useful, specifically because these chapters include detailed procedures for bone histological staining as well as for immunohistochemistry, immunofluorescence, confocal microscopy and scanning electron microscopy analyses. The third section concerns ionizing radiation techniques and specifically, absorptiometric methods to asses bone mineral density such as DXA, quantitative CT, peripheral QCT and micro-CT. Again, the parts which describe the fundamentals of radiation physics, interactions of x-rays with matter and radiological instrumentation may be too general for this type of book, but the applications to assessment of bone quality in humans and animal models are useful. A remarkable drawback in this section is that nearly all attention is given to bone mineral density, and very little information is provided on morphological parameters of bone that are demonstrated by the CT-based imaging modalities. Specifically, measurements of trabecular separation, trabecular thickness, bone volume fraction, trabecular number, trabecular connectivity etc., and accuracy of these measurements using clinical QCT and micro-CT are not addressed. Relations between these morphologic parameters, which are important characteristics of cancellous bone, are also not mentioned. Although the chapter on MRI that is included in the fourth section (non-ionizing techniques) partially concerns these issues, the reader is lacking information on morphological parameters that can be measured specifically by CT-based methods, and on the relation of accuracy to CT resolution for each morphological parameter. The last section is dedicated to non-ionizing techniques for bone studies. These include imaging methods such as MRI and ultrasound, as well as finite element and animal models. The chapter on ultrasound measurements of bone is particularly worth mentioning as a well-written review on the technological aspects, accuracy and precision and clinical applications. The chapter on finite element analysis covers up-to-date modeling approaches based on CT cross-sectional images at the whole bone level, or based on micro-CT or micro-MR images at the trabecular level. Geometrical model reconstructions from cross-sectional images and bone mechanical properties relevant to structural modeling are described in adequate detail, but meshing techniques and the effect of mesh density and element selection on the predicted stresses and deformations in bone were not addressed (qualitatively or quantitatively). This may disappoint a reader who was hoping to use this chapter as a practical, rather than a theoretical guide for model development. A second drawback relates to the discussion on models at the trabecular levels, where generic models of the trabecular architecture are ignored. Generic modeling of trabecular bone, which utilizes the theory of cellular solid mechanics, is a complementary modeling approach to specimen-specific (micro-CT or micro-MR image based) modeling. Generic models allow to study the mechanics of \'typical\' trabecular bone samples \[[@B6],[@B7]\], as opposed to specimen-specific models which may include abnormalities or anatomical variations that are specific to the studied specimens \[[@B7]\]. The concluding chapter in the last section concerns animal models to study bone, and particularly, osteoporosis. I found this chapter to be extremely useful for experimentalists who weigh alternative models for an experimental design of an animal study of osteoporosis. Having considered the points of strength and weakness in the different sections of this book as discussed above, I certainly recommend individual researchers and laboratories who deal with bone studies to hold a copy of this book. It should be noted that for those investigators focused on biomechanics of bone, it is not a substitute for the classics \"Bone Mechanics Handbook\" \[[@B2]\] and \"Bones\" \[[@B5]\], but rather, a complimentary book with particular added value in biology-oriented topics of safety considerations, microscopical techniques and animal model specifications. Bioengineering students, medical physics students and perhaps biology students at the graduate levels will also find this book useful for obtaining insight into the broad field of bone analysis. Abbreviations ============= DXA = dual x-ray absorptiometry CT = computed tomography QCT = quantitative computed tomography MRI = magnetic resonance imaging MR = magnetic resonance

PubMed Central

2024-06-05T03:55:58.444216

2005-6-14

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173125/", "journal": "Biomed Eng Online. 2005 Jun 14; 4:37", "authors": [ { "first": "Amit", "last": "Gefen" } ] }

PMC1173126

Background ========== An elevated immunoglobulin G (IgG) index and the presence of oligoclonal bands (OCB) in the cerebrospinal fluid (CSF) are a hallmark of multiple sclerosis (MS) \[[@B1],[@B2]\]. Although this finding is not specific for MS, 72 % percent of patients present with an elevated IgG index and even 98 % show an oligoclonal distribution of IgG bands exclusively in the CSF \[[@B3],[@B4]\]. Intrathecal IgG is thought to be the product of B lymphocytes residing in the brain of MS patients after they have crossed the blood brain barrier in an activated state with the help of various co-stimulatory signals \[[@B5]\]. Instead of undergoing apoptosis, the B cells expand clonally within the central nervous system (CNS) giving rise to a persistent antibody production \[[@B6]\]. Despite intense investigations, no single antigen against which the antibodies might be directed has been isolated so far. In contrast, the intrathecal antibody response covers a large number of CNS and non-CNS antigens as well as various pathogens \[[@B7]-[@B14]\], including the viral antigens such as measles, rubella and varicella zoster \[[@B15]\]. In up to 96 % of MS patients an intrathecal antibody production against at least one of the three antigens has been found \[[@B3],[@B16]\]. Little is known, however, about the clinical significance of these findings. Previous studies attempting to evaluate the long-term evolution of intrathecal viral antibodies were hampered by technical shortcomings such as few sensitive detection methods and an absence of correction for blood-CSF-barrier disturbance \[[@B17],[@B18]\]. Hence, the findings of these preliminary studies were partially contradictory with regard to the stability of the CSF antibody production \[[@B17],[@B18]\]. Furthermore, no correlation to the clinical course could be demonstrated \[[@B17],[@B18]\]. Here we report the results of a follow-up study on 70 MS patients from which at least two CSF analyses including cell count, IgG index, local IgG synthesis, antibody specific index and oligoclonal bands were available. The clinical implications of the immunological findings are discussed. Methods ======= Patients -------- The study used 70 consecutive patients with definite MS according to the criteria of Poser \[[@B19]\] and a primary relapsing course. All patients had at least two spinal taps, mostly during disease exacerbations. Three lumbar punctures (LP) were performed in 26 patients, four in 12 patients and five in 7 patients. Patients were characterized clinically by age, sex, disease duration, course of the disease and the expanded disability status scale (EDSS), documented at the time of the first LP. Furthermore, the progression index was calculated as the ratio of the EDSS and the disease duration for each patient. Patients with corticosteroid treatment in the last four weeks or with immunomodulatory or immunosuppressive therapy in the last 3 months prior to the first LP were excluded. Cerebrospinal fluid and serum sample pairs were analyzed for cell count in the CSF, oligoclonal bands in serum and CSF, local IgG synthesis, IgG index and antibody index for the following antigens: measles, rubella and varicella zoster virus. We focused only on these specific antigens because of the frequent detection of these antigens within CSF in the case of MS described in prior studies \[[@B3]\]. Cell count ---------- The CSF cell count was determined immediately after LP. For this purpose, 90 μl of CSF were stained with 10 μl dye containing 20% crystal violet solution, 20 % glacial acetic acid and 60% H~2~O. Cells were enumerated in a Fuchs-Rosenthal counting chamber. IgG index --------- The intrathecal IgG production was quantitated by the IgG index. For this purpose, albumin and IgG were measured in matched serum and CSF pairs by nephelometry according to the manufacturer\'s instructions using commercially available kits (antiserum against human albumin or IgG, respectively) and the BN 100 nephelometer (both Dade Behring GmbH, Marburg, Germany). The IgG index was calculated according to the following formula: (CSF IgG × serum albumin) / (serum IgG × CSF albumin). An IgG index above 0.7 was indicative of an intrathecal IgG synthesis \[[@B20]\]. Local IgG synthesis ------------------- We quantitated the local IgG synthesis within the CSF with the help of a method described by Reiber and colleagues \[[@B3]\]. Therefore, we used the following formula: Ig~loc~= \[Q~Ig~- Q~lim~(Ig)\] × Ig(serum) (Q = quotient CSF: serum). Details of the calculation have been described previously \[[@B21]\]. Oligoclonal bands (OCB) ----------------------- All samples were analyzed directly after performing the LP without freezing, to minimize post-sampling changes. The presence of oligoclonal bands in the CSF indicated an intrathecal IgG production. For detection of OCB, isoelectric focusing was performed on matched serum and CSF sample pairs. The serum and CSF samples were diluted to the same IgG concentration and run on polyacrylamide gel precoated with ampholytes (pH between 4.5 and 10.0) at increasing voltage according to the manufacturer\'s instructions (Servalyt Precotes, Serva Electrophoresis GmbH, Heidelberg, Germany). Subsequently, a silver staining with commercially available reagents was done as indicated by the manufacturer (Serva Electrophoresis GmbH). The patterns were interpreted qualitatively by comparing the presence or absence of OCB in the serum and CSF. Antibody specific index (AI) ---------------------------- During the whole study period, antibody production against measles, rubella and varicella zoster virus was analysed in matched serum and CSF pairs by commercially available, specific enzyme immune assays (Enzygnost by Dade Behring) on an ELISA II processor using the alpha-method (Dade Behring). The AI was calculated according to the following formula: AI = Q spec. / Q lim. with Q spec. = CSF antibodies / serum antibodies and Q lim. being calculated from the CSF / serum albumin ratio according to the Reiber formula \[[@B21]\]. AI larger than 1.4 indicate an intrathecal antibody synthesis against the given antigen. Statistical analyses -------------------- This study was carried out as retrospective case-control study. The mean values of cell count, IgG index, local IgG synthesis and AI were compared for the whole group at the different time points (LP 1 to 5) by non-parametric tests for dependent samples (Friedman test). The percentage of patients positive for OCB or with a positive AI against measles at the 1^st^and 2^nd^LP, respectively, was compared with the Fisher\'s Exact test. Different clinical subgroups, i.e. those with a high or low EDSS, were analyzed for their mean measles AI with the help of a median split. The mean measles AI in the subgroups was compared with a non-parametric test for independent samples (Mann Whitney U test). Results ======= Patient characteristics ----------------------- A total of 70 patients, 45 women and 25 men, were included in this study. The mean age of the patients was 38 years (range 21 to 67 years). All patients had a primary relapsing course of the disease. The mean disease duration was 4.5 years with a range of one to 37 years. The mean EDSS at study entry was 2.9 (standard deviation SD 1.33). The mean time from the first LP to the second was 15 months with a range from 1 to 81 months. The mean time from the first to the last LP was 22 months (range: 1 to 106 months) in all patients. In 30 patients the range between the first and second LP was at least 12 months. In 24 patients the time between these LPs was less than 6 months. At the occasion of the second LP, 22 patients received an immunomodulatory or immunosuppressive treatment (interferon beta, glatiramer acetate, intravenous immunoglobulin, azathioprine). It has to be stressed that the study population is inhomogeneous. If the LPs were sorted by the follow-up years in which they had been arranged, no new information could be observed. This might be due to the wide range of time between the different LPs in this patient cohort (1 to 106 months). The data are not shown. Mean IgG index, local IgG synthesis and AI ------------------------------------------ There was no significant change of the mean IgG index, local IgG synthesis or AI for any of the viral antigens investigated over time (p = 0.414 for IgG index, p = 0.578 for local IgG synthesis, p = 0.673 for measles AI, p = 0.984 for rubella AI, p = 0.941 for varicella zoster AI, Friedman test). The data are summarized in Table [1](#T1){ref-type="table"}. In addition, there was no significant change in the mean cell count (p = 0.291, Friedman test, data not shown). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Mean and SD of IgG index, local IgG synthesis (Ig~loc~) and antibody index (AI) for measles, rubella and varicella zoster virus at each successive lumbar puncture (LP). The number of patients in each category of serial LP is given. All included patients received at least two serial LPs (first and second). Additional LPs were only obtained in the mentioned number of patients. ::: LP no. No. of patients with the respective LP IgG index Ig~loc~(mg/L) AI measles AI rubella AI varicella zoster -------- ---------------------------------------- ------------ --------------- ------------ ------------- --------------------- 1. 70 1.3 ± 0.75 33 ± 5,0 3.4 ± 3.29 4.6 ± 7.11 3.6 ± 5.86 2. 70 1.3 ± 1.58 25 ± 4,1 3.5 ± 3.67 4.8 ± 7.67 3.6 ± 5.85 3. 26 1.4 ± 0.99 46 ± 1,1 3.9 ± 5.59 5.4 ± 9.28 4.7 ± 7.28 4. 12 1.2 ± 0.73 28 ± 0,9 2.8 ± 2.09 6.7 ± 14.78 4.4 ± 8.74 5. 7 1.0 ± 0.39 19 ± 4,8 2.1 ± 0.89 1.9 ± 1.28 1.9 ± 1.46 ::: OCB and AI positive patients ---------------------------- As expected, the presence of OCB in the CSF was more sensitive in the detection of an intrathecal IgG production than a positive IgG index \[[@B22],[@B23]\]. For instance, at the 1^st^LP 94% of patients had positive oligoclonal CSF bands whereas 88% of patients had a positive IgG index. Once positive the OCB bands remained positive during the follow-up. The percentage of patients who were positive for oligoclonal bands in the CSF increased after the 1^st^LP from 94 to 100 %. Likewise, the relative number of patients with an intrathecal antibody synthesis against measles virus increased significantly over time (Figure [1](#F1){ref-type="fig"}). No obvious change in rubella and varicella zoster antibody positive patients over time was observed. This significant increase in patients with an intrathecal antibody production against measles was seen in both patient subgroups irrespective of whether or not they were treated with any immunomodulatory treatment mentioned above at the 2^nd^LP or not (Table [2](#T2){ref-type="table"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Percentage of patients with an intrathecal antibody synthesis against measles, rubella, and varicella zoster virus at five successive spinal taps (LP, n = 70 for LP 1 and 2, n = 26 for LP 3, n = 12 for LP4 and n = 7 for LP5). There was a significant increase (*p*\<0.05) in patients positive for antibody synthesis against measles virus over time. There were no significant changes in rubella or varicella zoster antibody synthesis. ::: ![](1743-8454-2-3-1) ::: ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Percentage of patients who had oligoclonal bands in the CSF or a positive AI for measles at the respective lumbar puncture (LP). ::: 1^st^LP 2^nd^LP p value -------------------- ------------ --------- --------- ---------- Untreated patients OCB 96 100 not done measles AI 68 73 0.013 Treated patients OCB 91 100 not done Measles AI 55 68 0.001 All patients OCB 94 100 not done Measles AI 64 71 \<0.001 ::: Individual long-term follow-up studies -------------------------------------- In 7 patients, data from 5 consecutive spinal taps were available. Except for patient no. 57, all data were derived from patients with a secondary progressive course of the disease and collected over a period of 1 to 6 years. Although the AI for the various antigens was not identical, 4 of the patients were stable over time with regard to their status as antibody producers (patients no. 16, 21, 41, 52). Of these patients, one was positive for measles alone, one for measles and rubella, one for all three viral antigens and one patient was antibody negative. The introduction of an immunomodulatory or immunosuppressive treatment at the 2^nd^LP (interferon beta patients 16, 21 and 41, azathioprine patient no. 52) had no impact on the antibody production. Two additional patients experienced an expansion of their antibody response from one (rubella virus in both cases) to three antigens, despite treatment initiated at the 2^nd^LP (interferon beta in patient no. 5 and azathioprin in patient no. 34). Correlation with clinical features ---------------------------------- There was no significant difference in the mean AI against measles in patients with high or low disability (p = 0.995, Mann Whitney U test) at the 1^st^LP. High disability was defined as an EDSS score above the median of 3.0. Furthermore, there was no difference in the mean measles AI in patients with long and short disease duration (median 2 years, p = 0.304). The EDSS scale and disease duration was used to calculate the progression index for each patient. A progression index of more than 1 (median), indicating an increase of at least 1 point on the EDSS scale per year was regarded as high. The mean AI against measles was significantly higher in patients with a high progression index than in those with a low progression index (p = 0.038). In the subgroup with a high progression index the mean measles AI was 4.2, whereas in the low progression index group a mean measles AI of 2.6 was measured. In the low progression index subgroup 53% of patients had an intrathecal antibody production against measles, but almost 74% were positive for measles in the high progression index subgroup. There were no significant differences in the local IgG synthesis when comparing the groups with high or low disability at the 1^st^LP or the groups with a high or low progression index (p = 0.71, p = 0.962, respectively). In these respective groups, the IgG index was not changed (p = 0.165 high disability group; p = 0.828 high progression index). Discussion ========== There are three main findings of our study: i) Once an intrathecal antibody production is established it is maintained over time; ii) For the first time we provide evidence for an expansion of antibody reactivity in the CSF of MS patients; and iii) disease progression and B cell activation may be linked. Persistent IgG production ------------------------- In our study patients positive for oligoclonal IgG bands in the CSF remained positive at the follow-up. These findings are in accordance with previous data indicating a clonally stable IgG production in the CSF over long periods \[[@B24]\]. This was true despite the introduction of immunomodulatory or immunosuppressive treatment. In 132 MS patients serial CSF analysis before and after 2 years of interferon beta treatment or placebo failed to reveal differences in the IgG index or OCB in either treatment group \[[@B25]\]. Even after autologous hematopoietic stem cell transplantation identical OCB compared to pre-treatment CSF analysis were found \[[@B26],[@B27]\]. Similarly in our study, there was no effect of treatment on IgG production as documented by the IgG index or on the local IgG synthesis in the CSF. In addition to the well-known finding of a stable OCB production, we demonstrated for the first time that the intrathecal antibody production is stable over time as well. This means, once an intrathecal antibody response against measles, rubella or varizella zoster virus was established it persisted at the follow-up analyses for at least 4 years. Although persistent IgG and antibody production is a well-known phenomenon in multiple sclerosis, the mechanisms remain unknown. It has been demonstrated recently in an experimental model of B cell migration over the blood brain barrier that activated human B cells readily cross the blood brain barrier with the help of adhesion molecules and chemokines \[[@B5]\]. Longevity of B cells and plasma cells is a common feature \[[@B28]\], and the microenvironment of MS lesions promotes the persistence and activation of B cells \[[@B29]\]. Although repeated antigenic stimulation favors longevity of B cells, persistance of antigen is not required for the survival of B cells \[[@B30]\]. Instead, high expression of anti-apoptotic proteins may be involved in the survival of B cells. Interestingly, increased expression of the apoptosis-inhibitory proteins Bcl-2 and FLIP were detected in B cells of MS patients \[[@B6]\]. Alltogether, microenvironment and antiapoptotic signaling might explain in part the persisting IgG and antibody production in the CSF of MS patients even in the absence of persistent antigen exposure. Expansion of antigen response ----------------------------- More intriguing was the finding that the number of patients positive for intrathecal OCB increased over time from 94 to 100%. It has been assumed that in early disease activated B cells might not yet have become resident in the CNS. Progression from a monoclonal to an oligoclonal CSF pattern has been reported in multiple sclerosis patients \[[@B31]\]. An increase in CNS clones after the initial stage of the disease has been suggested as a possible explanation. Further evidence for an expanding B cell activity during the course of the disease comes from a CSF analysis in which the prevalence of mature plasma cells was found to be higher in patients with longer disease duration \[[@B32]\]. In our study, an increasing number of patients showed a newly established intrathecal antibody response against measles virus: whereas 64% of patients were positive for measles at the 1^st^LP, 71% became positive at the 2^nd^and 86% at the 5^th^LP. Furthermore, in two out of seven patients with repeated CSF analysis, there was an expansion of the B cell response from one to three viral antigens by the 5^th^LP. To our knowledge, this is the first follow-up report on intrathecal polyspecific antibody production in multiple sclerosis patients. This follow-up for the first time provides evidence for an expansion of antibody reactivity over time in multiple sclerosis patients. However, this expansion of antibody is in contrast to the stable local IgG synthesis over time. One possible explanation for a polyspecific intrathecal antibody production might be a bystander activation of B cells with a given antibody reactivity. This is possible in the presence or absence of T cells. If certain cytokines -namely tumor necrosis factor alpha- is present, IgG secretion by activated B cells may occur even in the absence of T cells \[[@B33]\]. Why preferentially measles, rubella, varicella zoster and to a lesser extent other virus reactive B cells become activated during the disease process remains unclear. Association between B cell activity and clinical aspects of MS -------------------------------------------------------------- Although the evidence of B cell activity in the CNS is a common feature in MS, its clinical significance remains to be elucidated. Early studies on fluctuations in the CSF antibody production failed to show an association between antibody titers and the clinical course, possibly due to methodological shortcomings \[[@B17],[@B18]\]. In contrast, a recent publication revealed an association between a very high IgG index and a rapidly progressing course of the disease \[[@B34]\]. Further evidence for an impact of B cell products on disease severity comes from a publication by Villar and colleague. They found an intrathecal IgM synthesis to be associated with a worse clinical outcome \[[@B35]\]. Accordingly, a predominance of B cells over T cells and macrophages in the CSF has been shown to be associated with a more rapid disease progression, but not with a longer disease duration or a higher disability score \[[@B36]\]. Interestingly, when our patients were divided into two subgroups according to the rate of disease progression, there were more measles positive patients in the rapidly progressive subgroup. Furthermore, the AI for measles was significantly higher in those with a rapid progression of disability. Conclusion ========== In summary, our data indicate that once an intrathecal antibody production has been established it is stable over time. Furthermore, an expansion in antibody specifity occurred in a proportion of patients. The presented data support the hypothesis that a strong B cell activation is associated with a worse clinical outcome in MS. List of abbreviations ===================== AI -- antibody specific index CNS -- central nervous system CSF -- cerebrospinal fluid EDSS -- expanded disability status scale Ig -- immunoglobulin LP -- lumar pucture MS -- multiple sclerosis OCB -- oligoclonal bands Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= H-FP gave the idea of this study and participated in the design of the study. The statistical analysis and the design of the study was performed by DR. Both authors read and approved the final manuscript.

PubMed Central

2024-06-05T03:55:58.445531

2005-6-27

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173126/", "journal": "Cerebrospinal Fluid Res. 2005 Jun 27; 2:3", "authors": [ { "first": "Hela-Felicitas", "last": "Petereit" }, { "first": "Dirk", "last": "Reske" } ] }

PMC1173127

Background ========== Chronic inflammation and airway smooth muscle dysfunction are consistent features of asthma responsible for disease progression and airway remodeling \[[@B1]\]. The increase in bronchial smooth muscle, both hypertrophy \[[@B2]\] and hyperplasia \[[@B3]\], plays a critical role in the development of airway hyperreactivity (AHR), the hallmark of asthma. Airway smooth muscle cells (ASMC) may also play a secretory or immunomodulatory role by producing pro-inflammatory cytokines, chemokines, polypeptide growth factors, extracellular matrix proteins, cell adhesion receptors, and co-stimulatory molecules, which perpetuate submucosal inflammation \[[@B4],[@B5]\]. These mediators may act on the ASM itself in an autocrine manner as well to further contribute to the asthma phenotype \[[@B6]\]. Therefore, smooth muscle itself may be capable of initiating and maintaining airway inflammation. Also, ASMC have been shown to undergo cell migration, which could contribute to airway remodeling \[[@B7]\]. Thus, regulation of airway smooth muscle hypertrophy and migration may be a new target for treatment of asthma \[[@B7],[@B8]\]. It is well known that IgE plays a critical role in the pathogenesis of asthma in the early and late phases by interacting with its two receptors, the high affinity receptor (FcεRI) and the low affinity receptor (FcεRII) \[[@B9]\]. IgE plays a key role in bronchial hyperresponsiveness and smooth muscle hyperreactivity \[[@B8]\]. Crosslinking of the high affinity IgE receptor (FcεRI) on mast cells leads to cellular degranulation and the release of various proinflammatory mediators and cytokines contributing to bronchoconstriction. The low affinity IgE receptor (CD23) (FcεRII) has been identified on B cells, monocytes, follicular dendritic cells, Langerhan\'s cells, eosinophils, and platelets \[[@B10]\]. Upregulation of the CD23 receptor is thought to increase allergic responses in the bronchial mucosa through the enhancement of antigen uptake and presentation \[[@B8]\]. The receptor has two isoforms that differ only in their cytoplasmic domains \[[@B11]\]. CD23a is constitutively expressed on B cells and is associated with endocytosis of IgE coated particles, and CD23b is induced by IL-4 and is also found on non- B cells such as T cells, Langerhan\'s cells, monocytes, macrophages, platelets, and eosinophils \[[@B12],[@B13]\]. IL-4 causes CD23 induction on B cells through CD40 \[[@B12]\]. CD23b mediates phagocytosis of soluble IgE complexes. An autocatalytic process involving cleavage of membrane bound CD23 by a matrix metalloprotease yields a series of soluble elements (sCD23) which increase IgE production via the CD21 receptor on B cells \[[@B13],[@B14]\]. The CD23 receptor has been shown to be upregulated on monocytes and alveolar macrophages in a T helper cell type 2 (TH2) environment and may contribute to chronic inflammation in asthma through this mechanism \[[@B15]\]. It has been shown that IL-4 and GM-CSF induce CD23 expression on monocytes, and GM-CSF primes monocytes for cellular activation and secretion of IL-1 upon subsequent exposure to IgE-containing immune complexes \[[@B8]\]. CD23 is also involved in antigen presentation to B cells as well as cellular interactions between B and T cells \[[@B12]\]. In previous studies, Hakonarson et al. \[[@B16]\] demonstrated the expression of CD23 (FcεRII) messages and a low level of the protein in airway smooth muscle cells. In two patients who died of status asthmaticus, CD23 expression was also markedly upregulated on ASMC. The CD23 expression was inducible with human atopic sera or IgE immune complexes in naïve (control) ASMC, and this upregulation was blocked when pretreated with anti-CD23 blocking antibody. The authors concluded that IgE coupled activation of CD23 contributes largely to its upregulation \[[@B5],[@B14]\]. In a corresponding experiment with rabbit ASMC subjected to control and atopic serum, they were able to demonstrate, through Western blot analysis, a markedly enhanced expression of CD23 in the ASMC sensitized with atopic serum or IgE immune complexes. They were able to achieve significant inhibition of upregulation by pretreatment with anti-CD23 mAb. They hypothesized that IgE was responsible for the upregulation of the low affinity IgE receptor \[[@B17]\]. Hakonarson, et al. have also demonstrated that ASMC in vitro exposed to human atopic sera results in an initial increase in TH2 cytokines including IL-5 and GM-CSF followed hours later by production of IL-1 and TH1 cytokines \[[@B18],[@B19]\]. Recently, phase I trials have been completed on IDEC-152, an IgG1 anti-CD23 antibody, for patients with mild to moderate persistent asthma. The drug was well tolerated by participants, and a dose-dependent decrease in mean IgE values was reported \[[@B20]\]. T helper cell type 2 (TH2) mediated inflammatory cytokines, such as IL-4, IL-13, and IL-5, as well as other enzymes and chemokines are active in the asthmatic patient. GM-CSF has been shown to be involved in asthma pathogenesis and in vivo can induce TH2 differentiation independent of IL-4 \[[@B21]\]. Tryptase and prostagladin D2 (PGD2) were chosen as major mast cell mediators, and more recently the PGD2 receptor gene (PTGDR) has been shown to be an asthma susceptibility gene \[[@B22]-[@B24]\]. It is possible that many factors are responsible for the upregulation of the low affinity IgE receptor in addition to and independent of IgE. The purpose of this study was to identify specific mediators released in the asthmatic patient that are responsible for the upregulation of CD23 on human airway smooth muscle cells independent of IgE. Methods ======= Cell culture and flow cytometry ------------------------------- Alpha-smooth muscle isoactin positive Human ASMC (Cambrex, Walkersville, MD) in T-75 flasks were starved for 24 hours in 0.1% (vol/vol) fetal bovine serum (FBS) containing medium M199 (Cellgro, Herndon, VA) supplemented with 1% (vol/vol) antibiotic/antimycotic solution (Sigma Chemical Co., St Louis, MO). The cells were then stimulated with either vehicle (bovine serum albumin, BSA, vehicle for cytokines (1 mg/ml), in M199; ethanol (EtOH), vehicle for LTD4 (6% final concentration) and PGD2 (0.001--0.01% final concentration); M199, vehicle for tryptase), an individual mediator, or a mediator in combination with GM-CSF at their optimum concentrations for 24 hours. The doses of cytokines used were up to four time ED50 including: IL-4 (0.04--1 nM), GM-CSF (0.07--0.8 nM), IL-13 (0.4 nM), IL-5 (0.01--0.07 nM), IL-13 (0.3--2.2 nM), PGD2 (1--10 μM), LTD4 (1--10 μM), tryptase (30 nM, a concentration sufficient to induce ASMC proliferation) (Sigma). Dose ranging studies were performed to determine the optimum concentration of IL-4 and GM-CSF on the expression of CD23, and the doses chosen were IL-4 (0.5 nM) and GM-CSF (0.4 nM) (Figure [1](#F1){ref-type="fig"}). All cytokines were obtained from R & D Systems Inc. Minneapolis, MN except GM-CSF which was obtained from Sigma. The cells were then harvested with a soft rubber edged scraper, centrifuged for 5 minutes at 1000 rpm (200 g), washed and resuspended in 1% BSA in phosphate buffered saline (PBS) and fixed with 70% ETOH. After washing twice more, the cells were resuspended in 1% BSA in PBS. Finally, they were filtered through a 40 μm nylon mesh to obtain single cell suspension and stained with (20 μl) of PE (phycoerythrin)-CD23 (EBVCS-5, BD Biosciences, San Jose, CA) or PE-mouse IgG~1~for 15 minutes in the dark to facilitate staining for flow cytometry. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **Upregulation of CD23 by IL-4 and GM-CSF**. Dose-ranging studies were performed to determine the optimum concentrations of IL-4 and GM-CSF. Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml), IL-4 (0.125. 0.25, 0.5, or 1.0 nM) or GM-CSF (0.1, 0.2, 0.4, or 0.8 nM) for 24 h. The cell lysates in RIPA buffer were subjected to western blot analysis for CD23. Mouse anti-human CD23 monoclonal antibody (clone M-L233, BD Biosciences, 1 μg/5 ml) was used as the primary antibody and anti-mouse horseradish peroxidase linked antibody as the secondary antibody (Amersham). The immunoreactive protein bands were detected by enhanced chemiluminescence light (ECL) (Amersham). ::: ![](1476-7961-3-6-1) ::: Protein analysis ---------------- A commercially available bicinchoninic acid (BCA) kit (Pierce, Rockford, IL) was used for protein analysis according to the manufacturer\'s instructions. The optical densities were read using a Bio-Kinetics EL-312 Microplate reader. Indirect immunofluorescence --------------------------- Indirect immunofluorescence stainings were performed with anti-smooth muscle-α isoactin antibody (Sigma) and anti-human CD23 antibody (M-L233, 1 μg/ml, BD Biosciences), which are specific monoclonal antibodies and either a FITC or TRITC fluorochrome, conjugated second antibody. Fixed huASMC were incubated with the above antibodies diluted in PBS with 3% BSA for 60 min at room temperature. The cells were then washed three times with PBS for 10 minutes for each wash. Non-specific binding was blocked by incubating cells with 3% BSA in PBS for 60 minutes. The blocking solution was then removed and cells were incubated with FITC- or TRITC- fluorochrome conjugated antibody for 45 minutes in the dark to facilitate staining. Cells were then washed with PBS three times. Finally, one drop of Fluoromount-G (Southern Biotechnology Inc., Birmingham, AL) was added. Western blot ------------ Standard Western blot analyses were performed to detect anti-STAT6 (1:500, Calbiochem, San Diego, CA) polyclonal rabbit, anti-p-STAT-6 (1:500, Calbiochem) polyclonal rabbit. Human ASMC lysates in radio-immunoprecipitation assay (RIPA) buffer were transferred onto Hybond-ECL nitrocellulose membranes and were immunoblotted with monoclonal anti-human CD23 (1:500 dilution, clone M-L233, 1 μg/5 ml, BD Biosciences), polyclonal anti-IL-4Rα (1:500 dilution, Santa Cruz), monoclonal anti-IL-2Rγc (1:250 dilution, R&D Systems, Inc.). The nitrocellulose membranes were incubated with a 1:1,000 dilution of anti-rabbit or anti-mouse horseradish peroxidase linked whole antibody (Amersham, Piscataway, NJ) in PBS-T for 1 hour at room temperature. Paxillin monoclonal antibody (1:500 dilution, Transduction Laboratories) was used as a positive isotype control for CD23, and fibronectin polyclonal antibody (1:250, Sigma) was used as a positive control for the remaining antibodies. The immunoreactive protein bands were detected by enhanced chemiluminescence light (ECL) (Amersham). Statistical analysis -------------------- Data were analyzed with Prism 4 software (GraphPad, San Diego, CA). One-way analysis of variance (ANOVA) was used. Results are expressed as mean ± SEM. A P value less than 0.05 was considered statistically significant. Results ======= CD23 protein expression is upregulated in huASMC by IL-4, GM-CSF, or IL-4/GM-CSF -------------------------------------------------------------------------------- Previous studies have shown that IgE immune complexes in atopic serum caused an increase in CD23 expression in ASMC \[[@B16]\]. To determine if other humoral factors in atopic serum effect CD23 expression in human ASMC, we have tested the effect of the relevant cytokines, arachidonic acid metabolites, and the mast cell enzyme tryptase. Flow cytometry was performed to evaluate differences in cell populations after stimulation of the huASMC for 24 hours with either individual mediators IL-4 (0.5 nM), GM-CSF (0.4 nM), IL-13 (0.4 nM), IL-5 (0.4 nM), PGD2 (10 μM), LTD4 (10 μM), tryptase (30 nM) or a combination of IL-4, IL-5, and IL-13 each with GM-CSF. Within the huASMC stimulated by IL-4, GM-CSF or the combination of IL-4/GM-CSF, two populations of cells were detected distinguishable by cell size. While the smaller cells did not show a significant expression of CD23, many of the larger cells showed increased expression of CD23. In the example in Figure [2](#F2){ref-type="fig"}, 66% of the larger cells (gate D) showed an increase in cell expression of CD23 when compared to the controls. As stated previously, the functions of ASMC are heterogeneous including proliferation and synthesis. Previous studies have shown, on flow cytometry of ASMC stimulated in vitro with IL-1β and TNF-α, only 20--60% of ASMC produce GM-CSF. The ASMC producing GM-CSF include some which also have increased proliferative properties. This suggests that considerable heterogeneity exists in the phenotypic expression of the ASMC in culture \[[@B25]\]. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Upregulation of CD23 by IL-4, GM-CSF and IL-4/GM-CSF**. Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with IL-4 (0.5 nM)/GM-CSF (0.4 nM) for 24 h. The FACS analysis showed that the smaller cells passed through Gate A and larger cells passed through Gate D. Background noise was eliminated using the BSA-stimulated control cells that were labeled with PE-anti-CD23 (EBVCS), represented by C in Gate A, and F in Gate D. The FACS results of a representative experiment showed 66% of the larger cells (Gate D) had an increase in cell expression of CD23 when compared to the controls. ::: ![](1476-7961-3-6-2) ::: In addition to the combination of IL-4/GM-CSF inducing increased expression of CD23, both IL-4 and GM-CSF alone independently increased the expression of CD23 in huASMC. The percentage of cells with increased fluorescence intensity above the control was 25.1 ± 4.2% (IL-4), 15.6 ± 2.7% (GM-CSF) and 32.9 ± 13.9% (IL-4/GM-CSF combination). On the other hand, IL-5, IL-13, cysteinyl leukotrienes, and tryptase did not induce CD23 expression (Table [1](#T1){ref-type="table"}). ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Il-4, GM-CSF, and IL-4/GM-CSF Increase CD23 Expression on ASMC (n = 3) ::: ------------------------------------------------------------------------------ Cytokine (24 hours) \#Cells in Gate D\ \%Cells with Increased CD23\ (n = 3) Expression Above the Control -------------------------- -------------------- ------------------------------ BSA (1 μg/ml) 1,343 ± 122 0 IL-4 (0.5 nM) 1,413 ± 197 25.1 ± 4.2\* GM-CSF (0.4 nM) 1,346 ± 243 15.6 ± 2.7\* IL-4/GM-CSF (0.4/0.5 nM) 1,324 203 32.9 ± 13.9\* IL-5 (0.4 nM) 1,130 ± 251 0 IL-13 (0.4 nM) 1,316 ± 269 0 PGD2 (1 μM) 1,521 ± 123 0 PGD2 (10 μM) 1,159 ± 204 0 LTD4 (10 μM) 2,037 ± 375 0 Ethanol (6% vol/vol) 2,507 ± 200 0 Tryptase (10 μM) 2,385 ne ± 405 0 ------------------------------------------------------------------------------ Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml), IL-4 (0.5 nM), GM-CSF (0.4 nM), IL-4 (0.5 nM)/GM-CSF (0.4 nM), IL-13 (0.4 nM), IL-5 (0.4 nM), PGD2 (10 μM), LTD4 (10 μM), tryptase (30 nM) for 24 h. Results are mean ± SEM of the percentage of cells with an increased fluorescence intensity above the control (n = 3). Control values: BSA-stimulated, PE-anti-CD23 labeled, 10.9 ± 1.4 %; BSA-stimulated, PE-mouse IgG~1~, non-immune, 0.5 ± 0.1 % (n = 3). \*denotes significant increase in CD23 expression above the BSA control value. ::: Expression of CD23 in response to IL-4, GM-CSF, IL-4/GM-CSF is accompanied by changes in huASMC morphology ---------------------------------------------------------------------------------------------------------- Western blot analysis of huASMC stimulated with IL-4, GM-CSF, or Il-4/GM-CSF for 24 h showed an increase in CD23 expression compared to BSA vehicle control (Figure [3](#F3){ref-type="fig"}). Indirect immunofluorescence was used also to identify any morphological changes associated with the cytokine stimulation and upregulation of CD23 (Figure [4A--D](#F4){ref-type="fig"}). Those cells stimulated with the combination of IL-4/GM-CSF demonstrated CD23 expression along with changes in cell morphology including depolymerization of isoactin fibers, cell spreading, and membrane ruffling (Figure [4B](#F4){ref-type="fig"}). These changes in phenotype are consistent with flow cytometry results in that the larger cells expressed CD23 (Figure [4D](#F4){ref-type="fig"}). In contrast, the control BSA stimulated population showed no changes in cell cytoskeletal structure and morphology (Figure [4A](#F4){ref-type="fig"}) or specific staining for CD23 (Figure [4C](#F4){ref-type="fig"}). ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **Western blot analysis of CD23 after stimulation of IL-4, GM-CSF, IL-4/GM-CSF**. Alpha-smooth muscle isoactin positive huASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml) (vehicle control), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.5 nM/0.4 nM) for 24 h. The cell lysates in RIPA buffer were subjected to western blot analysis for CD23. Mouse anti-human CD23 monoclonal antibody (clone M-L233, BD Biosciences, 1 μg/5 ml) was used as the primary antibody and anti-mouse horseradish peroxidase linked antibody as the secondary antibody (Amersham). The immunoreactive protein bands were detected by enhanced chemiluminescence light (ECL) (Amersham). Paxillin mouse monoclonal IgG~1~(Transduction Laboratories) was used as an irrelevant isotype control. ::: ![](1476-7961-3-6-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Expression of CD23 in response to IL-4/GM-CSF is accompanied by changes in huASMC morphology**. Alpha-smooth muscle isoactin positive huASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with either BSA or IL-4 (0.5 nM)/GM-CSF (0.4 nM) for 24 h, and stained with either anti-smooth muscle-isoactin (A & B) or anti-CD23 antibody (C & D). Those cells stimulated with the combination of IL-4/GM-CSF demonstrated CD23 expression (D) and changes in cell morphology including depolymerization of isoactin fibers, cell spreading, and membrane ruffling (B). Cells stimulated with BSA (vehicle for IL-4/GM-CSF) alone did not increase the expression of CD23 (C) nor changes in phenotype (A). These findings were confirmed by three independent observers. ::: ![](1476-7961-3-6-4) ::: To confirm activity of protein synthesis, the protein content of the control and the experimental groups of cells were compared using a BCA protein analysis kit. Human ASMC were starved for 24 hours in 0.1% FBS containing medium M199 and then stimulated with BSA (1 μg/mL), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4 (0.5 nM)/GM-CSF (0.4 nM) for 24 hours. The protein content was increased by 19% in the IL-4/GM-CSF treated cells above that of the control (Table [2](#T2){ref-type="table"}). The increase in protein concentration with IL-4 alone was not statistically significant. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### IL-4/GM-CSF Combination Increases Protein Content in huASMC. ::: Cytokine (nM) mg/10^6^cells (n = 3) ------------------------- ----------------------- BSA (vehicle) 1.17 ± 0.08 IL-4 (0.5) 1.18 ± 0.01 GM-CSF 1.12 ± 0.03 IL-4 (0.5)/GM-CSF (0.4) 1.39 ± 0.02 \* Alpha-smooth muscle isoactin positive huASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then stimulated with BSA (1 μg/ml)), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.4/0.5 nM) for 24 h. The cell lysates in RIPA buffer were analyzed for protein content using a commercially available BCA kit (Pierce). The optical density was read using a Bio-Kinetics EL-312 Microplate reader. Results are mean ± SEM (n = 3). \*denotes value significantly different from the BSA vehicle treated control. ::: Stimulation of huASMC with IL-4 induces phosphorylation of STAT-6 and expression of IL-2Rγc ------------------------------------------------------------------------------------------- IL-4 binds the IL-4R with high affinity, and signaling through IL-4 causes enhanced expression of IL-4R \[[@B21]\]. The induction of these genes is mediated through signal transduction molecules including signal transducer activator of transcription (STAT-6). The binding of IL-4 to its receptor complex induces the formation of an IL-4 receptor complex which consists of IL-4Rα and the common gamma chain (γc) of the receptors for IL-2, IL-4, IL-7, IL-9, IL-15, and IL-21 \[[@B21]\]. It has not been previously reported that airway smooth muscle cells express the IL-2Rγc, the signaling unit of the IL-4 receptors. Western blot analysis of IL-4Rα and IL-2Rγc in huASMC lysates showed the presence of these receptor components on huASMC. Figure [5](#F5){ref-type="fig"} shows abundant expression of IL-4Rα and a low level expression of IL-2Rγc protein on huASMC. After stimulation of huASMC with IL-4 (0.4 nM) for 24 h, a two fold increase in γc expression was observed compared to the BSA vehicle control (Figure [6](#F6){ref-type="fig"}). ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **IL-4Rα and IL-2Rγc Expression in huASMC**. A: IL-4Rα and IL-2Rγc expression in huASMC at baseline. Unstarved huASMC lysates were subjected to western blot analysis for IL-4Rα and IL-2Rγc using polyclonal anti-IL-4Rα (1:500 dilution, Santa Cruz), or monoclonal anti-IL-2Rγc (1:125 dilution, R&D Systems, Inc.). The nitrocellulose membranes were incubated with a 1:1,000 dilution of anti-rabbit or anti-mouse horseradish peroxidase linked whole antibody (Amersham). The immunoreactive protein bands were detected by ECL (Amersham). IL-2Rγc is minimally expressed in huASMC while IL-4Rα is expressed abundantly in huASMC. ::: ![](1476-7961-3-6-5) ::: ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Upregulation of IL-2Rγc Expression in huASMC by IL-4 and IL-4/GM-CSF**. Alpha-smooth muscle isoactin positive Human ASMC (Clonetics) in T-75 flasks were starved for 24 h in 0.1% FBS containing medium M199. The cells were then either stimulated with BSA (vehicle) (1 μg/ml), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.5 nM/0.4 nM) for 24 hours. The IL-4 and IL-4/GM-CSF stimulated cells had increased IL-2Rγc expression compared to the BSA (vehicle) group. Fibronectin polyclonal rabbit antibody (Sigma) (1:250) was used as an irrelevant isotype control. ::: ![](1476-7961-3-6-6) ::: To confirm that IL-4 was activating the IL-4Rα during the stimulation of huASMC, we examined the phosphorylation of downstream STAT-6 by western blot. Human ASMC were starved for 24 h and then stimulated with IL-4 (0.4 nM) for fifteen minutes. Results of Western blot revealed an approximately a four fold increase in intensity of the band for phosphorylated-STAT-6 in IL-4 stimulated cells when compared to BSA control (Figure [7](#F7){ref-type="fig"}). This supports the role of an IL-4 mediated signal transduction pathway involvement in CD23 upregulation in huASMC. ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **Phosphorylation of STAT-6 by IL-4 and IL-4/GM-CSF in huASMC**. Starved huASMC were incubated with either BSA vehicle control (1 μg/ml), IL-4 (0.5 nM), GM-CSF (0.4 nM), or IL-4/GM-CSF (0.5 nM/0.4 nM) for 15 minutes. Standard Western blot analyses were performed to detect STAT-6 and phosphorylated-STAT-6 (p-STAT-6) using a anti-STAT-6 polyclonal rabbit antibody (Calbiochem) and anti-p-STAT-6 polyclonal rabbit antibody (Calbiochem). Anti-rabbit horseradish peroxidase linked antibody was used as the secondary antibody. Protein bands were detected by ECL. STAT-6 was abundantly expressed by all four groups, while p-STAT-6 was only expressed in the IL-4 and IL-4/GM-CSF groups. Fibronectin polyclonal rabbit antibody (Sigma) was used as an irrelevant isotype control and was abundantly expressed in all four groups. ::: ![](1476-7961-3-6-7) ::: Discussion ========== In our study, we have demonstrated that CD23, the low affinity IgE receptor, is upregulated on human airway smooth muscle cells by the cytokines IL-4, GM-CSF, and the combination of IL-4/GM-CSF. This upregulation of CD23 by the combination of IL-4 and GM-CSF was accompanied by an increase in cell volume and protein content, cytoskeletal depolymerization, cell spreading and membrane ruffling. Because ASMC require a doubling time of 48 hours, the increase in protein content could not be attributed to an increase in cell number. Also, the increase in cell number in gate D seen with LTD4 was likely secondary to the effects of ethanol (Table [1](#T1){ref-type="table"}). Stimulation of huASMC by IL-4 caused an activation of STAT-6 and an increase in γc expression. Collectively, our findings suggest that CD23 expression can be stimulated by IL-4 and GM-CSF cytokines independent of IgE in huASMC and the upregulation of CD23 may play a role in cell migration and hypertrophy. Previous studies have demonstrated an increase in CD23 expression in alveolar monocytes after stimulation with IL-4 and GM-CSF \[[@B10]\]. In that study, the use of the individual mediators alone did not increase the CD23 levels to that of asthmatic patients suggesting a possible synergistic role between IL-4 and GM-CSF. Our findings are consistent with these in that the combination of IL-4 and GM-CSF was most effective in upregulating CD23 in huASMC. Not all TH2 cytokines are involved in this process; IL-5 (0.4 nM) and IL-13 (0.4 nM) had no effect on CD23 expression. Cysteinyl leukotriene LTD4 (10 μM), prostaglandin PGD2 (10 μM) and tryptase (30 nM) did not induce CD23 expression on huASMC \[[@B26]\]. We evaluated the effect of stimulation of huASMC with IL-4 on phosphorylation of STAT-6 via the IL-4R which would confirm the presence of the receptor in huASMC. STAT-6 is a critical mediator of IL-4 stimulated gene activation, and it is regulated by both tyrosine and serine kinases \[[@B27]\]. It has been shown in a mouse model that STAT-6 binds the CD23a murine promoter, and STAT -/- mice stimulated with IL-4 are unable to upregulate CD23. This suggests STAT-6 is a critical mediator for IL-4 induced upregulation of CD23 \[[@B28]\]. IL-4 along with CD40 mediated signals are responsible for upregulation of CD23 on B cells \[[@B14]\]. In this study, we have confirmed the expression of the IL-4Rα and a low level of common gamma chain in huASMC and that after stimulation of huASMC with IL-4, there was a two fold increase in γc chain expression (Figure [6](#F6){ref-type="fig"}). Phosphorylation of STAT-6 after stimulation with IL-4 for 15 min confirms that IL-4 has bound and activated the IL-4 receptor complex (Figure [7](#F7){ref-type="fig"}). It has been shown that the common gamma chain is a functional β chain of the IL-4 receptor complex in certain cells \[[@B27]\], and our data suggest that this is the case in huASMC. Interestingly, IL-13 (4 nM, a concentration sufficient to simulate huASMC proliferation, unpublished observation) did not upregulate CD23. For proliferation of ASMC by IL-13, IL-4Rα and IL-13Rα1 are required for signal transduction and downstream activation of p44/42 extracellular regulated kinases (ERK, unpublished data). Apparently, IL-4Rα and IL-13Rα1 engagement is not sufficient for CD23 expression, further supporting the role of γc chain in CD23 expression by IL-4. The signs of signal transduction in response to IL-4, and the increase in protein content of the cell in response to IL-4 and GM-CSF combination (Table [2](#T2){ref-type="table"}) represent activation of transcription and translation of CD23 in this case. Coupling of GM-CSF and its receptor complex is known to activate ERK that may have contributed to the synergistic effect of GM-CSF on CD23 expression. CD23 expression was associated with changes in cell morphology including depolymerization of isoactin fibers, cell spreading, and membrane ruffling (Figure [4B](#F4){ref-type="fig"} &[4D](#F4){ref-type="fig"}). Actin in ASMC is in a dynamic state and undergoes polymerization-depolymerization during the contraction-relaxation cycle \[[@B29],[@B30]\]. Membrane ruffling and cell migration involve signaling pathways including PI3-kinase, Rac and other Rho family G protein members in a variety of cell types, including vascular smooth muscle cells. Rac has an essential role in cell migration and regulation of the actin cytoskeleton \[[@B31],[@B32]\]. Moreover, ASMC are capable of switching their phenotypes from contractile to synthetic phenotype that is mediated by Rho kinases \[[@B32],[@B33]\]. In summary, we have demonstrated that the low affinity IgE receptor can be induced on huASMC by specific cytokines including IL-4, GM-CSF, and the combination of IL-4/GM-CSF. The combination of IL-4/GM-CSF also induced morphologic changes in the ASMC that may contribute to the synthetic function or migration. In addition, IL-4 and IL-4/GM-CSF stimulation of huASMC increased the protein content of the cell, suggesting hypertrophy. Conclusion ========== T helper type 2 cytokines including IL-4 have major role in asthma pathogenesis. GM-CSF is a hemopoetic growth factor, mostly released by activated monocytes and T cells. Additional sources of GM-CSF include epithelium of asthmatic airways \[[@B34]\] and human airway smooth muscle cells \[[@B6],[@B35]\]. Therefore, the effect of GM-CSF on CD23 expression can be both via paracrine and autocrine mechanisms. Previous studies by Hakonarson et al. \[[@B16],[@B17]\] have shown that upregulation of the CD23 receptor has been associated with proasthmatic changes in agonist-mediated ASM constrictor and relaxant responsiveness. Our study suggests that CD23 expression is associated with elements of hypertrophy (i.e. an increase in cell volume and protein content), thus consistent with their findings. Inhaled corticosteroids, the mainstay in treatment of asthma, effectively reduce inflammation and remodeling of the epithelium and basement membrane. However, no agents have been proven effective in reducing smooth muscle mass in asthmatic patients. Recent study results on anti-CD23 therapy showed decrease in serum IgE. Further studies to intervene the upregulation of CD23 expression by cytokines IL-4 and GM-CSF may open a new avenue to target smooth muscle hypertrophy, an important element of severe asthma \[[@B2]\]. Competing interests =================== The author(s) declare that they have no competative intrests. List of abbreviations ===================== huASMC: Human airway smooth muscle cells FceRI: High-affinity receptor for IgE FceRII (CD23): Low-affinity receptor for IgE AHR: Airway hyperreactivity PGD2: Prostaglandin D2 LTD4: Leukotriene D4 PE: Phycoerythrin FBS: Fetal bovine serum BSA: Bovine serum albumin PBS: Phosphate bufferred saline BCA: Bicinchoninic acid FITC: Fluorescein Isothiocyanate TRITC: Tetramethyl Rhodamine Iso-Thiocyanate RIPA: Radio-immunoprecipitation assay ECL: Enhanced chemiluminescence light Rα: Receptor alpha γc: Common gamma chain FACS: Fluorescent Activated Cell Sorter STAT: Signal Transducer Activator of Transcription TH2: T helper cell type 2 ERK: Extracellular regulated kinases Authors\' contributions ======================= JTB participated in designing experiments and performing CD23 analysis by flow cytometry. RKG carried out flow cytometry, immunofluorescence studies, and drafted the manuscript. HM carried out western blot analyses. DBL supervised all aspects of the project. All authors have read and approved the final manuscript. Acknowledgements ================ This work was supported in part by the grant from the Le Bonheur Children\'s Medical Center, NIH-HL56812, Children\'s Foundation Research Center, Molecular Resource Center, and American Academy of Allergy, Asthma and Immunology. The authors would like thank Jan Aldrich for her technical assistance.

PubMed Central

2024-06-05T03:55:58.447562

2005-5-20

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173127/", "journal": "Clin Mol Allergy. 2005 May 20; 3:6", "authors": [ { "first": "Joseph T", "last": "Belleau" }, { "first": "Radha K", "last": "Gandhi" }, { "first": "Holly M", "last": "McPherson" }, { "first": "D Betty", "last": "Lew" } ] }

PMC1173128

Background ========== Allergic asthma ranks in Germany among frequent occurred occupational diseases. Cow hair allergy in Finland is one of the most frequent reasons for the development of occupational asthma \[[@B1]\], but epidemiologically in Germany still insufficiently explored. The prevalence of respiratory complaints in Finnish farmers with cow husbandry is 27% for rhinitis respectively 12% for asthma in consequence of occupational exposure \[[@B2],[@B3]\]. Until now there is only limited insight into the risk factors with regard to this disease. The prevention of sensitization in stock farming to cow hair and epithelial allergens including lipocalins is often difficult due to intense exposure, with traditional milking still requiring close animal contact. The German Professionals insurance laws require a complete elimination of allergens for the recognition of an allergic airway disease as a compensable occupational disease. Complete avoidance of occupational exposure of cow hair allergens is only possible if stock farming is abandoned. This is, however, often not feasible in small dairy plants because of the resulting loss of income. In the present case report we discuss the efficiency of the strategy of exposure minimization to allergens by the use of a milking robot, i.e., an automatic milking system (AMS) as an alternative measure to total elimination of allergens for the prevention of occupational asthma caused by cow hair allergy. Case presentation and discussion ================================ A 37-year-old female farmer reported to suffer from daily asthmatic complaints by milking of cows with a conventional device. Exposure associated respiratory symptoms increased continuously during a period of 12 years. Allergic bronchial asthma, caused by sensitization against cow dust-derived allergens, was diagnosed by a positive skin prick test (Bencard^®^) and by the detection of IgE antibodies against *Bos d 2*at 74.9 kIU/l (Pharmacia, UniCAP^®^). Moreover, nonspecific bronchial hyperresponsiveness was verified by a positive histamine inhalation challenge test. The diagnosis of occupational asthma was confirmed by a separate specific inhalation challenge test using a 10% extract of cow hair allergens which led to a 330% increase of airway resistance (Raw). The German Professionals insurance laws require complete elimination of allergen exposures for the compensation of occupational allergic asthma as occupational disease \[[@B4]\]. Legal consequence would usually be to abandon dairy farming. However, in our case presented, dairy farming substantially contributed to family-income and the compensation offered would be insufficient for subsistence. Hyposensitization is the most effective measure in the prevention of allergic symptoms in patients with asthma. In that farmer the hyposensitization was not feasible due to permanent exposure to allergens. This farmer did not wear a helmet during the milking process because of claustrophobic panic attacks (fear of suffocation). Moreover an immunotherapy with recombinant lipocalins is currently not available. Respirator helmets have been predominantly used as effective measures for prevention of the inhalation of fumes and dusts \[[@B5]\]. Investigations about the effectiveness of respirator helmets against high-molecular allergens to prevent allergic asthma are rare. Taivainen et al. (1998) demonstrated that due to the use of a dust respirator helmet with a P2-class filter, dairy farmers with occupational asthma induced by cow dander showed less rhinitic symptoms and an improved peak flow rate \[[@B6]\]. However, workers using respirator helmets were not prevented against a progressive behavior of occupational asthma \[[@B7],[@B8]\]. Thus it is to doubt whether respirator helmets are efficient in farmers with occupational asthma to provide a complete protection against allergens \[[@B9]\]. To enable further dairy farming, the farmer installed a milking robot in 1999, i.e., an automatic milking system (AMS). At present circa 1300 AMS are installed, predominantly in the Netherlands, followed by Germany. AMS were developed in the course of increasing automation of the agricultural production. AMS consist of a milking box with the milking equipment and an electronic controlling system, an automatically operating dosing unit of a specific food-concentrate as well as a software controlled management system. The cows are decoyed from a specific food-concentrate to enter the milking box. A transponder, fixed on the cow neck, initiates milking operations automatically controlled by software and laser. The milking robot operates in the cowshed, which is air-shielded from the control-centre. After the installation of AMS the farmer was involved only in terms of monitoring the milking process. Figure [1](#F1){ref-type="fig"} shows the job and the current exposure situation of the farmer. It is apparent that the novel milking technique reduced exposure significantly by some 90% (from over 2 hours to approx. 10 min.) without a decline of milk production. Beyond allergen exposure reduction, the physical workload was lessened and the time available for activities to improve coping was increased -- these factors are increasingly recognized as relevant for the prognosis of asthma \[[@B10]\]. Table [1](#T1){ref-type="table"} shows the influence of the AMS technique on the current occupational exposure of the farmer. According to German Professionals insurance laws all conceivable measures must be taken into account even AMS is an expensive technical device to prevent occupational asthma. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Milking process monitoring by an air-shielded control-centre. ::: ![](1476-7961-3-8-1) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Parameters influenced by automatic milking system (AMS) ::: -------------------------------------------------------------------------------------------------------- **Parameter** **Power of parameter** **Influence of AMS** ----------------------- ------------------------ ------------------------------------------------------- Daily exposure period ↓↓↓ Drop of more than 2 hours to approx. 10 min. (\> 90%) Exposure intensity ↓↓↓ Increasing space to cow Physical workload ↓↓ Monitoring tasks Timing flexibility ↑↑ • No commitment to fixed milking times\ • Free working plan Quality of life ↑ More time available for other activities -------------------------------------------------------------------------------------------------------- Explanation: ↓ = reduction of power, ↑ = improvement of power ::: After installation of the milking robot the farmer reported significantly less frequent allergic and asthmatic symptoms. Asthma medication was reduced and required only once daily in the evening. Improvement was noted also in terms of lung-function and decreased total serum IgE: serum IgE decreased from 1332 IU/ml before milking robot installation to 572 and 474 IU/ml two and three years after installation, respectively. FEV~1~improved by 20% and vital capacity by 11%, comparing values one year before with two years after installation. A CAP-RAST score \"5\" versus *Bos d 2*was determined after the intervention. Preceding EAST-Test results vs. *Bos d 2*were not suitable for comparison. Current discussions address threshold limit values for allergens that may prevent allergic symptoms \[[@B11]\]. These have been proposed also for the cow dust-derived allergen *Bos d 2*\[[@B12]\]. Table [2](#T2){ref-type="table"} shows examples for successful measures of exposure minimization at workplaces to reduce sensitizations. Recent studies have already demonstrated that allergen exposure reduction at workplaces is successful in reducing sensitizations, e.g., to latex proteins \[[@B13],[@B14]\], laboratory animals \[[@B15]\] or acid anhydrides \[[@B16]\]. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Examples for successful measures of exposure minimization at workplaces in reducing of sensitizations ::: **Exposure areas** **Allergen** **Measure of exposure minimization** **Study** ------------------------------ --------------------------------- ------------------------------------------------------------------ ----------------------------------------- Hospitals Latex proteins Replacement of powdered gloves through powder-free gloves Edelstam et al. 2002, Tarlo et al. 1994 Animal research laboratories Allergens to laboratory animals Technical improvement of housing cabinets for animals Thulin et al. 2002 Electrical industry Acid anhydrides Installation of a closed system in the production of epoxy resin Drexler et al. 1999 ::: Conclusion ========== The case presented and the evidence obtained from the literature indicates that the strategy of exposure minimization to allergens at workplaces can be an effective alternative to total elimination, which may not be feasible for economical reasons. In farmers with cow dust allergy a milking robot is an appropriate technical measure to minimize allergen-exposure particularly with regard to the reduced animal contact. Independently from this, such milking robots also increasingly used to enhance productivity (usually in herds comprising more than 50 animals). The efficacy of milking robots in terms of primary and secondary prevention should be evaluated also epidemiologically in a suitable cohort study, preferably in countries with intensive dairy farming. List of abbreviation ==================== AMS; Automatic milking system Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= GK was the principal investigator. GK examined the patient, performed the investigations and drafted the manuscript. HCB revised the manuscript. WU revised the manuscript and evaluated the epidemiological aspects. HD conceived the case report. All authors read and approved the final manuscript.

PubMed Central

2024-06-05T03:55:58.449923

2005-6-22

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173128/", "journal": "Clin Mol Allergy. 2005 Jun 22; 3:8", "authors": [ { "first": "Gintautas", "last": "Korinth" }, { "first": "Horst Christoph", "last": "Broding" }, { "first": "Wolfgang", "last": "Uter" }, { "first": "Hans", "last": "Drexler" } ] }

PMC1173129

Background ========== The average annual incidence rate of hepatitis A virus (HAV) infection varies significantly worldwide ranging from 9.1 per 100,000 in the United States and Europe to 500 per 100,000 in the Central Asian Republics \[[@B1],[@B2]\]. According to the latest available data, the incidence of HAV infection remains as high as 150 cases per 100,000 population in the Central Asian Republics and Kazakhstan in 2001 \[[@B3],[@B4]\]. Although HAV infection incidence has decreased slightly compared to the figures for the last decade, the persistence of high incidence in Kazakhstan is mainly due to problems related to poor supplies of good-quality drinking water, sanitation and hygiene practices \[[@B1],[@B5],[@B6]\]. It has been documented that the incidence of HAV infection is closely linked to poor socio-economic living conditions \[[@B3]\]. HAV infection is usually a mild illness, but can be severe if it occurs concurrently with other diseases or comorbid conditions \[[@B7]-[@B10]\]. In Britain, and North America, a patient with HAV infection is hospitalized only if (a) (s)he is too sick with diarrhea, nausea, and vomiting to be managed appropriately with antiemetics and oral rehydration at home, (b) (s)he has hepatic and/or renal failure (sometimes seen with HAV-infection in persons with underlying alcoholic liver disease or with chronic hepatitis caused by other viral agents) \[[@B11]\], or (c) (s)he has a clinical presentation that is truly anomalous (*e.g*., meningoencephalitis has been reported in association with HAV infection but is exceedingly rare \[[@B12]\]). However, in many republics of the former Soviet Union, HAV infection has been a cause for routine hospitalization, as mandated in administrative directives. The premise on which many such patients are hospitalized should be reconsidered because relatively few patients with HAV infection become seriously ill, no treatment to influence the underlying course of the disease exists, and neither admission to hospital nor isolation from others in the home lessens the spread of the infection \[[@B13]\]. Because HAV infection is neither a severe illness nor one with important sequelae, it has been observed that the rates of infection of family members of patients with mild HAV managed at home or in hospital were similar \[[@B13]-[@B15]\]. Previous studies using data from the region reported that the hospitalization of patients after the appearance of jaundice was unjustified, and that the quality of treatment between inpatient and outpatient settings would be similar \[[@B13],[@B14]\]. However, in severe cases (\<10% of HAV infections), patients are more likely to be managed in a hospital and to receive more intensive treatment, including ultrasound and computerized tomography exams, as well as liver biopsies \[[@B7],[@B8],[@B10]\]. In such cases, the assumption of similar quality of inpatient and outpatient treatment does not hold. Using data gathered from patients with mild HAV infection in Almaty, Kazakhstan, we conducted a cost-comparison analysis of the policy of mandatory hospitalization and an alternative policy of outpatient management for patients with mild HAV infection. The purposes of this study were (a) to determine the mean treatment costs associated with managing cases of mild HAV infection in an inpatient setting under a policy of mandatory hospitalization, (b) to compare those costs with costs of treating patients with mild HAV infection using outpatient management, assuming a similar quality of treatment under inpatient and outpatient conditions, and (c) to determine total direct medical and nonmedical costs as well as productivity losses associated with mild HAV infection and then compare these costs both from the perspective of the resource-allocating authority (*i.e*., the Ministry of Health) and from a broader societal perspective. Materials and methods ===================== Data sources ------------ Data were collected using survey questionnaires that involved interviews with patients and medical practitioners between November 2000 and February 2001. Information was collected on patient\'s age, sex, principal diagnosis, duration of illness, use of services, and cost data. Patients were interviewed to collect information regarding out-of-pocket and direct nonmedical expenses, such as expenses paid for transportation to outpatient clinics and indirect costs associated with the number of days absent from work. Interviews with medical personnel and hospital administrators were conducted to determine the costs and quantify the amount of health and nonhealth staff time spent per day in caring for patients with mild HAV infection. A sample size of 451 subjects (200 inpatients and 251 outpatients) was used to conduct a cost-comparison analysis. We enrolled patients presenting to either of two hospitals in Almaty, Kazakhstan, in a geographical area where mandatory hospitalization for mild HAV infection was still practiced. We also enrolled patients in 19 outpatient clinics in geographical areas where mandatory hospitalization was no longer practiced. Patients selected for the study included members of both sexes, all age groups, and varied in ethnicity. Case definition --------------- A case of mild HAV infection was defined as a clinical presentation with manifestations of acute hepatitis and serologic confirmation of HAV infection in a person who reported to one of the selected hospitals or outpatient clinic sites. The clinical manifestations of HAV infection included sudden onset of fever, anorexia, fatigue, and/or abdominal discomfort followed within a few days by jaundice and hepatomegaly \[[@B16]-[@B18]\]. Serologic confirmation of mild HAV infection was determined by elevated alanine serum transferase (AST) not in excess of 5 times the upper limit of normal of the reference value, and by presence of detectable IgM anti-HAV Criteria for inclusion in this analysis were (a) having mild HAV infection as a primary or sole diagnosis confirmed by laboratory; (b) being between 1 and 65 years of age and being either hospitalized or managed at one of the study sites by a health care worker during the study period; and (c) providing voluntarily signed consent that included permission to review medical records. Persons who had multiple concurrent diseases or comorbid conditions and those who had not been diagnosed and tested positive for mild HAV infection more than 30 days before the study period were excluded from the study. Cost description ---------------- Costs for direct medical and non-medical services provided for patients with mild HAV infection were calculated based on interviews with medical providers, hospital administrators, health officials and patients. Units of direct medical services recorded included the number of days spent in hospital, health staff time allocated to treatment, and costs for those services specifically identified as part of the management of patients with mild HAV infection in hospital and outpatient clinics. Time spent by health care personnel was obtained based on reviews of medical records. Costs were based on estimates of costs for time spent by hospital- or clinic-personnel per day in caring for patients with mild HAV infection: in hospital, physicians spent approximately 54 minutes (min.), nurses approximately 89 min., and nonhealth staff about 40 min.; in the outpatient clinics, physicians spent approximately 26.4 min., nurses approximately 18 min., and nonhealth staff a negligible amount of time. In this analysis, we are considering only the base salary for the health care personnel; physicians are paid US\$15.84 per 8-hour workday, US\$4.8 per day for nurses, and US\$2.4 per day for nonhealth staff obtained from the budget of each health facility and based on the national contract for health personnel. Although physicians earn more than their base salary through various bonus payments and through informal payments from patients, consistent estimates of these payments could not be captured and were not included in this analysis \[[@B6],[@B19],[@B20]\]. Since an estimated 80% to 90% of physician income is obtained informally \[[@B20]\], it is difficult to track the flow of incomes for doctors, and we were able to use only the base salary for our estimates. The cost per patient-hospital day included the costs of purchase and repair of medical equipment and buildings, administrative and operational costs in the hospitals allocated to the treatment of mild HAV patients. To cover the full capital and recurrent cost, cost per patient-hospital day was determined based on average national estimates provided by hospital administrators \[[@B21]-[@B23]\]. Although hospital costs cover capital and recurrent costs such as the cost of building and medical equipment, laboratory services, health and nonhealth personnel, and hospital administrative costs, we estimated separately the costs for laboratory and personnel services for managing patients in hospital based on medical records and interviews with medical practitioners. Costs for hospital meals and prescription drugs were obtained based on patient self-reported out-of-pocket expenses. The cost of hospital meals was estimated for the respective number of inpatient hospital days. The number of hospital days was determined based on patient medical records. Because all patients paid for hospital meals and prescription drug charges where incurred, we calculated the cost of hospital meals and prescription drugs separately. Total treatment costs were calculated as the sum of direct medical and non-medical costs which included hospital costs, costs for time spent with patients by physicians, nurses, and other staff, laboratory and prescription drug costs, as well as non-medical costs such as hospital meals and transport. Because there are no long-term sequelae of mild HAV infection \[[@B7],[@B10]\], no home care was needed for such patients; therefore, the cost of home care was not calculated. Productivity losses were calculated as indirect costs associated with lost workdays due to illnesses of HAV infection. The number of lost workdays were determined based on self-reports of missed work. The exchange rate used was US\$1 = 148 Kazakhstan Tenge (KZT), a rate that has not varied more than 15% since January 2001. As of May 5, 2005, the exchange rate was US\$1 = 131 KZT. Treatment outcome ----------------- Having measured the direct medical and nonmedical costs, as well as productivity losses of managing mild HAV infection, we defined effective treatment or management for a mild HAV infection as an episode of care (which included all services provided) that led to a complete recovery. As has been observed epidemiologically elsewhere, we assumed that the rates of infection of family members of patients with mild HAV infection managed at home or in hospital were similar \[[@B13],[@B15]\]. Statistical analyses -------------------- We completed our analyses in five stages: First, we estimated the mean annual treatment cost per case of mild HAV infection when managed in inpatient and outpatient settings. Using SAS statistical programs, comparison between inpatient and outpatient groups were made using PROC TTEST of the null hypothesis of equal group variances, and we determined the difference in mean treatment cost per case using t tests of differences in means. The confidence interval (CI) for the difference between two sample means was obtained using a \"pooled\" estimate of the standard deviation \[[@B24]\]. Second, we stratified the samples by four different age groups, given significant differences in mean age between the two groups. We categorized patients into four age groups: ≤ 18, 19 -- 26, 27 -- 34, and ≥ 35 years to estimate the mean annual treatment cost per case of mild HAV infection by age group. We used PROC GLM to conduct one-way analysis of variance (ANOVA) using Bonferroni t tests of pair of means within age groups to test the null hypothesis of equality of the group means among different age groups. Third, we estimated the number of lost workdays for both inpatient and outpatient group. We calculated the productivity losses by multiplying the number of lost workdays by employed patients aged 18 -- 64 years or by adult parents or guardians of children with mild HAV infection, and by daily urban wage rates. In this analysis, we did not consider replacement costs for unemployed patients. Fourth, we estimated the total annual cost of mild HAV infection as the sum of the annual treatment costs consisting of direct medical and nonmedical costs as well as productivity losses resulting from lost workdays. Finally, we conducted sensitivity analyses to determine which parameters exert the greatest influence on the results of this analysis and establish the range of total annual cost of mild HAV infection. A *p*-value ≤ 0.05 was considered to be statistically significant. Epi Info software \[[@B25]\] was used for data entry. Data from questionnaire forms were entered twice into a computer database to reduce the chances of data-entry errors and were analyzed using SAS statistical programs \[[@B26]\]. Results ======= Table [1](#T1){ref-type="table"} presents sociodemographic characteristics of the inpatient and outpatient groups. The mean age of the inpatient group was 22 years; the mean age of the outpatient group was 12 years. The inpatient group reported a higher annual household income of US\$898.2 (2001 US\$) compared with US\$502.3 for the outpatient group. Several other comparisons reflected disparities in age of the two sample groups: 26.5% of the inpatient group was employed, compared with 6.7% of the outpatient group; 21% of the inpatient group were married, compared with 8.9% of the outpatient group; 32% of the inpatient group smoked cigarettes, compared with 9.9% of the outpatient group. The average wage rate per patient in urban areas was 14,617 KZT (US\$98.76) per month compared to 5000 KZT (US\$33.78) per patient in rural areas. There were 27.21 lost workdays for employed inpatients and 6.19 lost workdays for family members caring for minor patients in hospital; employed outpatients had an average of 15.73 lost workdays (lost workdays: 33.40 vs. 15.73). The average length of hospital stay for a patient with mild HAV infection was 9.7 days, compared with 3.4 visits per patient in the outpatient group. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Sociodemographic characteristics of patients with mild HAV infection, Almaty, Kazakhstan, 2001 ::: Variables All patients (n = 451) Inpatients (n = 200) Outpatients (n = 251) -------------------------------------- ------------------------ ---------------------- ----------------------- Gender (%) Male 50.9 50.0 51.4 Female 49.1 50.0 48.6 Marital status (%) Married 17.8 21.0 8.9 Unmarried 85.2 79.0 91.1 Education level (%) Never attended or kindergarten 13.2 0.5 23.4 School 1--8 years 55.8 72.0 42.7 Grades 9--11 years 16.3 27.5 17.3 High school graduate 1.1 \- 2.0 College 1--3 years or graduate 13.6 \- 14.5 Ethnicity (%) Kazakhs 40.6 36.0 44.2 Russians 49.2 54.5 45.0 Uzbek 2.4 2.0 2.8 Others 7.8 7.5 8.0 Employment status (%) Employed 16.4 26.5 6.7 Unemployed 83.6 73.5 93.3 Tobacco use (%) Smoker 20.3 32.0 9.9 Nonsmoker 79.7 68.0 90.1 Urban status (%) Urban 99.8 100.0 99.6 Rural 0.2 \- 0.4 Home ownership (%) Owned 98.2 99.5 97.2 Rented 1.8 0.5 2.8 Age group (%) ≤ 18 66.3 37.5 89.2 19 -- 26 22.6 42.5 6.8 27 -- 34 6.7 12.5 2.0 ≥ 35 4.4 7.5 2.0 Mean Age (years) (SE) ^¶^ 17 (0.41) 22 (0.49) 12 (0.43) Mean household income (US\$) (SE)^¶^ 886.6 (60.03) 898.2 (61.65) 502.3 (28.38) ^¶^Standard error (SE) are values in parenthesis. ::: Table [2](#T2){ref-type="table"} presents the mean treatment costs per patient with mild HAV infection in inpatient and outpatient settings. The mean total direct medical and nonmedical cost of managing mild HAV infection was estimated at US\$157.4 (2001 US\$) per inpatient (95% CI = US\$151 -- 163), compared with US\$22.5 per outpatient (95% CI = US\$21 -- 23). The hospital costs (US\$81) were calculated by taking the product of the average number of hospital days (9.7 days) and the cost per HAV patient-hospital day (US\$8.4) obtained from the survey. Of the US\$81 of hospital costs, as much as 40% (US\$32) of the hospital costs was allocated to the costs for the purchase and repair of medical equipment and buildings and 60% (US\$49) was spent on recurrent hospital administrative and operational costs including overhead and utilities. Laboratory costs accounted for 6.9% (US\$11.0) of the mean treatment costs for inpatients compared with 35.6% (US\$8.0) of those costs for outpatients. We estimated that about 19% (US\$30) of the mean treatment costs were allocated to hospital meals. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Mean treatment costs of managing cases of mild HAV infection for inpatient and outpatient settings, Almaty, Kazakhstan (2001 US\$) ::: Inpatient Outpatient ---------------------------- ----------- ------------ ---------------- ------ ------ --------------- Hospital cost 81.0 1.78 (89.0, 96.0) \- \- \- Physicians\' time 18.0 0.49 (17.0, 19.0) 3.0 0.08 (2.80, 3.20) Nurses\' time 9.0 0.17 (8.7, 9.30) 1.6 0.07 (1.46, 1.70) Non-health staff 2.0 0.04 (1.92, 2.10) \- \- \- Laboratory 11.0 0.35 (10.0, 12.0) 8.0 0.11 (7.8, 8.2) Prescription drugs 6.4 0.21 (6.19, 7.01) 5.9 0.33 (5.25, 6.55) Hospital meals 30.0 0.57 (29.0, 31.0) \- \- \- Transport \- \- \- 4.0 0.20 (3.60, 4.40) Total inpatient/outpatient 157.4 3.04 (151.0, 163.0) 22.5 0.42 (21.0, 23.0) Productivity losses^¶^ 150.9 2.12 (146.7,155.1) 71.5 4.75 (66.8, 76.3) Total societal cost^§^ 308.3 4.90 (298.7, 317.9) 94.0 4.28 (85.6, 102.4) Note: Data may add exactly due to rounding ^†^Standard error (SE) ^‡^95% Confidence interval (95% CI) ^¶^Productivity losses calculated based on the number of lost workdays reported and average urban wage rate. ^§^Total societal costs include treatment costs and productivity losses ::: From the perspective of the Ministry of Health, the mean direct medical costs of managing mild HAV infection was estimated at US\$127.4 per patient in hospital, compared with US\$18.5 in an outpatient setting. From the societal perspective, the mean total cost of direct medical and nonmedical services as well as of productivity losses was estimated at US\$308.3 per patient in hospital compared with US\$94.0 per patient in an outpatient setting. Of the US\$157.4 total for direct medical and nonmedical services, a patient in hospital paid approximately US\$36.4 out of pocket costs for drugs and hospital meals, compared with US\$9.9 for such costs paid by a patient in an outpatient setting. Results from the PROC TTEST indicated that the difference between the mean treatment costs for management of mild HAV infection among the inpatient and outpatient groups was US\$134.6 (2001 US\$), 95% CI = (US\$ 128.0 -- 140.0) and statistically significant at (*p*≤ 0.0001). When we controlled for the effect of age on medical cost by stratifying the samples by age group, we found that the mean treatment costs in the outpatient setting was still considerably lower than those in the inpatient setting. For example, the mean treatment costs of managing cases of mild HAV infection for outpatients aged 19 to 26 years was estimated at 13.9% of that for the inpatients of the same age. The one-way ANOVA analysis indicated that the mean treatment costs per inpatient decreased as age increased (Figure [1](#F1){ref-type="fig"}). The mean treatment cost per inpatient for the age group ≥ 35 years was estimated at US\$144 compared with US\$163 for the age group ≤ 18 years (*F*-value = 1.40; *p*= 0.25). Results from this analysis indicated that although there were differences in population means among different age groups, these differences were not statistically significant. We also noted that the mean treatment cost per outpatient in the age group ≥ 35 years was estimated at US\$28 versus US\$22 for the age group ≤ 18 years (*F*-value = 2.05; *p*= 0.10). Again, results from this analysis indicated that although there were differences in population means, these differences were not statistically significant. Mean treatment costs as well as productivity losses per outpatient increase slightly as age increases (Table [3](#T3){ref-type="table"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Mean treatment costs per mild HAV inpatient by age group ::: ![](1478-7547-3-4-1) ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Mean, median total cost and standard error of managing mild HAV infection, by age group and treatment setting (inpatient vs. outpatient), Almaty, Kazakhstan (2001 US\$) ::: Total cost\* per inpatient Total cost\* per outpatient ---------- ---------------------------- ----------------------------- ----- ----- ------------- ----- ≤ 18 75 320 (32.42) 319 224 92 (8.64) 97 19 -- 26 85 303 (6.72) 304 17 96 (4.75) 97 27 -- 34 25 302 (11.25) 304 5 111 (11.60) 111 ≥ 35 15 290 (12.88) 292 5 93 (12.95) 101 All 200 308.3 (4.90) 304 251 94 (4.28) 98 \* Note the total cost includes the treatment costs as well as productivity losses ^§^Sample size ^†^Standard error (SE) ::: Nationally, we estimated about 21,600 cases of mild HAV infection occurred during 2001. Of these cases, an estimated 19,008 employed patients with HAV infection would report lost workdays due to illness considering the 12% unemployment rate in Kazakhstan \[[@B6]\]. Assuming no replacement costs of unemployed sick workers, we calculated the productivity losses by multiplying the average lost wages of US\$150.9 per inpatient by the cases of employed inpatients. We estimated the productivity losses at US\$2.87 million (95% CI = US\$2.79 million -- US\$2.95 million). Using 21,600 cases of mild HAV infection, an average of 9.7 hospital days and 33.4 lost workdays per inpatient, we calculated a total of 209,520 hospital days and 721,440 lost workdays. Annual mild HAV infection treatment costs for inpatients were estimated at US\$3.39 million (95% CI = US\$3.26 million -- US\$3.52 million). These costs could be lowered by managing patients with mild HAV infection in an outpatient setting for as low as US\$486,000 (95% CI = US\$453,600 -- US\$496,800). From the societal perspective, the annual total cost of mild HAV infection for patients treated in hospital was estimated at US\$6.26 million, 95% CI = (US\$6.05 -- US\$6.47). The treatment costs represent 51% of the total costs whereas productivity losses account for 49% of the total costs. The total cost of managing mild HAV infection in an outpatient was estimated at US\$1.85 million, 95% CI= (US\$1.72 -- US\$1.95). The annual total cost of mild HAV infection could be reduced by almost 70% from US\$6.26 million to US\$1.85 million by encouraging patients with mild HAV infection to be treated in outpatient rather than in inpatient settings. Using an average saving (US\$214) per inpatient and 21,600 cases of mild HAV infection, the potential savings at the national level from changing the treatment setting from inpatient to outpatient would be about US\$4.62 million (2001 US\$) per year. Discussion ========== Our findings suggest that the mean treatment costs per patient managed in hospital were almost seven times the costs per patient in an outpatient setting. Our results were consistent with previous reports that inpatient care is more expensive than outpatient care settings \[[@B6]\]. Even though no other study has directly estimated the costs of mild HAV infection in Kazakhstan, we observed how expensive the inpatient setting in the region was, relative to the outpatient setting. Health systems in the Central Asian Republics, including Kazakhstan, have not been designed and managed with attention to the criteria of *technical efficiency*(the production of a defined set of services at the lowest cost within the health system; also known as operational efficiency) or *allocative efficiency*(the distribution of resources across varied range of services in an optimum manner). These health systems have tended to overinvest services provided in inpatient settings; hospitals utilize approximately 75% of the Kazakhstan health budget compared to 10% of the budget for outpatient care \[[@B6]\]. Many disease entities managed with outpatient treatments in European Union countries (including viral hepatitis and tuberculosis) are managed on an inpatient basis in Kazakhstan \[[@B6]\]. In order to provide optimal care within the budget limits imposed by the allocation of hospital resources to various emerging infectious diseases like human immunodeficiency virus (HIV) infection, health authorities should consider choosing health care practices known to be beneficial on the basis of evidence of cost-effectiveness \[[@B27]\]. Inefficiency in the use of public funds has resulted in the increasing practice of informal or \'under-the-table\' payments in the health system \[[@B6]\]. Informal out-of-pocket expenses are increasingly required due to declining funds for the health sector. In the late 1990s, a 65% reduction in government revenues resulted in insufficient health sector funding that led to delays in wage payments and shortage of medical supplies \[[@B20]\]. Although the benefit package in Kazakhstan has covered inpatient drug costs, many hospitals in practice cannot afford to supply them. Therefore, patients often pay for food and drugs while they are in hospital \[[@B6]\]. These expenditures are informal payments that patients are asked to pay out-of-pocket. Our results were consistent with the findings of Ensor and Savelyeva, that estimates of patient per capita payment for prescription drugs in hospital range between US\$ 8.5 and US\$10.4 \[[@B20],[@B28],[@B29]\]. In Kazakhstan, private practice has been permitted since 1991, and more physicians are becoming semi-autonomous practitioners in group practices funded through a patient capitation fund. Many local health authorities permit health facilities to establish charge departments in order to obtain additional revenue \[[@B20]\]. User charges for medical services by public sector health organizations were legalized in 1995. Now, health care facilities charge for services, and as a result, outpatients make official co-payments for the treatment \[[@B6]\]. In addition to these official charges, recent surveys indicated that informal payments to health care practitioners continue to exist, ranging from US\$5 for consultations to thousands for an operation \[[@B6],[@B20],[@B30],[@B31]\]. In this paper, estimates of out-of-pocket expenses were obtained by interviewing patients. Higher costs in inpatient settings may reflect unnecessary use of hospital resources. Other factors that contributed to higher costs in inpatient services may be due to the decrease in government revenues that used to subsidize the hospital services, low official salaries of health care staff, and growth in private health care services \[[@B6]\]. Because of these latter factors, medical practitioners seek augmentation of income via the informal payments \[[@B20]\]. User fees for those who can afford them or cost-sharing strategies could address the problems of under-funding in health systems, stabilize these informal payments, and free up public funds. By balancing the mix of resources and increasing the proportion of spending on services managed in outpatient settings, Ministries of Health in Central Asia might be able to achieve optimal allocation of resources in the health care system. Our findings may be conservative for three reasons. First, in estimating mean treatment costs, we did not consider informal payments to health care staff. Second, our estimates of productivity losses were based on 264 working days per year (22 working days per month) and did not include the value of housekeeping productivity losses. Third, we used lower estimates of the number of patients with mild HAV as 150 per 100,000; the upper estimates of this figure could be as high as 500 per 100,000. This study has some limitations. First, we did not evaluate post-treatment health outcomes after hospital discharge, including completion of outpatient visits. However, there is a wealth of literature that has demonstrated a lack of long-term health outcome differences between persons restricted to bedrest or institutionalized and persons forced to exercise or allowed to engage in activities *ad libitum*when symptomatic with acute HAV infection \[[@B32],[@B33]\]. Second, because of lack of reliable data, we were unable to obtain the capital costs of outpatient facilities including the cost of space, maintenance for facility and medical equipment, overhead, and utilities associated with the treatment of mild HAV infection; therefore, our cost estimates for the outpatient might be biased toward the lower end. Despite these potential concerns, these limitations would not invalidate the findings of this study. In sensitivity analyses, we varied some parameters to establish the range of annual cost estimates of mild HAV infection. We observed that annual cost estimates are sensitive to the following four variables: value of medical services, number of lost workdays, average wage rates, and incidence of HAV infection cases. In these analyses, the annual cost estimates of mild HAV infection varied from US\$4.37 to US\$24.66 million (Table [4](#T4){ref-type="table"}). Considering patients with multiple concurrent diseases or patients diagnosed with severe HAV, cost of medical services can vary significantly. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Total annual cost estimates of mild HAV infection for inpatients in sensitivity analysis, Kazakhstan, (2001 US\$) in millions ::: Analysis Treatment cost Productivity losses Total cost ------------------------------------------------------ ---------------- --------------------- ------------ Base case assumption 3.39 2.87 6.26 Value of medical services ^†^ 4.89 2.87 7.76 Rural wage rate (KZT 5000 per month) ^§^ 3.39 0.98 4.37 Greater adjustment of hepatitis A under-reporting^¶^ 12.59 12.07 24.66 Note the total cost includes the treatment costs as well as productivity losses ^†^an estimated informal payment is included. When 80--90% informal payment is taken into account, the treatment costs were estimated at US\$226.2 per inpatient. ^§^reported wage rate based on the survey results ^¶^annual incidence of hepatitis A as high as 500 per 100,000 (reported in the Central Asia Republics in 1998) ::: Future research should consider conducting cost-effectiveness analyses (CEA) taking into account the quality of post-treatment outcomes rather than assuming similar treatment outcomes among inpatients and outpatients. In conducting these analyses, researchers should consider two possible scenarios: (a) the treatment outcome for the inpatient setting is better than that for the outpatient setting (in this case, the inpatient setting is more costly and more effective than the outpatient setting, so an incremental cost-effectiveness analysis should be conducted); and (b) the treatment outcome for the outpatient setting is as good as or better than that for the inpatient setting (thus, because the cost of managing mild HAV infection for outpatients is lower than for inpatients, cost savings should be calculated). Conclusion ========== From a societal perspective, the annual total treatment costs and productivity losses associated with mild HAV infection ranged from US\$6.05 million to US\$6.47 million for inpatients compared to a range of US\$1.72 million to US\$1.95 million for patients in the outpatient setting. The results of this study show that compliance with a change in the policy of hospitalization for patients with mild HAV infection could produce savings of US\$4.62 million at the national level in Kazakhstan. Appendix ======== Before beginning this study, we obtained institutional review board (IRB) approval (Protocol No. 2708) from the Centers for Disease Control and Prevention (CDC) Assurances-Human Subjects Office. In addition to CDC IRB approval, we also obtained an approval letter from the Ethics Committee Review from the Ministry of Health in Kazakhstan. Prior to administering questionnaires and interviews with patients, we explained the purposes of the study and obtained informed consent forms signed by all adult patients with mild HAV infection. Parents or guardians of all child participants with mild HAV infection under 18 years of age signed the informed consent forms for children. This study was conducted in collaboration with the Data for Decision Making and Policy Branch of the Division of International Health (DIH), Epidemiology Program Office (EPO) of CDC, the CDC Central Asian Regional Office (CDC/CAR), and the Almaty City Sanitary and Epidemiological Stations of the Ministry of Health in Almaty, Kazakhstan. Acknowledgements ================ The authors thank colleagues at CDC, the CDC Central Asian Regional Office, and the Ministry of Health and health care practitioners in the Republic of Kazakhstan for their support in collecting the data. The study was conducted entirely at the Division of International Health, Epidemiology Program Office, Centers for Disease Control and Prevention. The opinions expressed in this article do not necessarily represent those of the agencies of the United States Government

PubMed Central

2024-06-05T03:55:58.451351

2005-5-12

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173129/", "journal": "Cost Eff Resour Alloc. 2005 May 12; 3:4", "authors": [ { "first": "Abdiaziz S", "last": "Yassin" }, { "first": "Michael", "last": "Favorov" }, { "first": "Edmond", "last": "Maes" }, { "first": "Ramses", "last": "Sadek" }, { "first": "Aliya", "last": "Jumagulova" }, { "first": "Victor", "last": "Merker" }, { "first": "Tatiana", "last": "Surdina" }, { "first": "Terence", "last": "Chorba" } ] }

PMC1173130

Introduction ============ In spite of continuous improvements in revascularization techniques, there remain patients ineligible for coronary artery bypass graft surgery (CABG) and coronary angioplasty. Those patients frequently suffer from severe angina, CCS (Canadian Cardiovascular Society) class III to IV; their complaints may not be managed even with optimized drug treatment. An estimated 0.5--1% of catheterized patients belongs to this category. Taking into account the above numbers of catheterizations, in Germany alone there are up to 6000 patients per year suffering from angina-related pain without an option for revascularization. Those patients master their daily lives only by an excessive use of short-acting nitrates and avoiding physical activity. Regular use of morphine is not uncommon. A vicious circle is established: The pain limits physical activity thus inducing further decrease of physical fitness. Frequent hospitalizations, visits to the attending physician and the need for intensive care (when a myocardial infarction is suspected) cause immense costs for the health care system. Based on the \"gate control theory\" \[[@B1]\] Wall and Sweet \[[@B2]\] were the first to use electric stimulation of spinal nerves or dorsal roots in treating chronic pain. This new technique called spinal cord stimulation (SCS) caused a significant decrease in ischemia-related pain and an increased perfusion of capillaries in skeletal muscle and skin through stimulating different types of neurons and fibers in the spinal cord. In recent years the number of studies showing decreasing symptoms, improved quality of life and less signs of myocardial ischemia (mostly measured by ECG) in SCS-treated patients rose continuously \[[@B3]-[@B13]\]. However, the exact mode of action and the influence on myocardial perfusion are yet unknown. Very recently Jessurun and co-workers could demonstrate an improved myocardial perfusion in TENS (transcutaneous electric neurostimulation) treated patients with syndrome X after 4 weeks \[[@B14]\]. Besides a study with 9 patients and a 6 week follow up \[[@B15]\], there exist no long-term data about myocardial ischemia under SCS in patients with coronary artery disease. In our trial, we therefore investigated not only the development of symptoms, quality of life and exercise capacity but also the myocardial perfusion measured by MIBI-SPECT (single photon emission computer tomography) in a one year follow up of 31 patients with CAD and intractable angina. Materials and methods ===================== Patients -------- Between March 1999 and May 2003 31 patients were included in the study. Criteria for accepting a patient was a refractory angina CCS III-IV in spite of optimized medical therapy (ACE inhibitors, aspirin, β-blockers, statins, calcium channel blockers, nitrates). All patients had a coronary angiogram in the last 3 months before screening and were discussed in detail for possibility of CABG and/or PCI (percutaneus coronary intervention) by a specialized team of cardiologists and heart surgeons. Only if none of these revascularization techniques was considered as suitable was the patient included in the trial. Since relieving of pain is the main aim of SCS treatment we only included patients with angina CCS III-IV without relevant symptoms of heart failure. Hence, patients with a left ventricular ejection fraction less than 40% or a NYHA class greater than II were excluded from the study. All inclusion and exclusion criteria are listed in table [1](#T1){ref-type="table"}. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Inclusion and exclusion criteria ::: ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Inclusion criteria** Chronic coronary artery disease (CAD) Evidence of ischemia in MIBI-SPECT examination Coronary angiogram in the last 3 months Angina pectoris, CCS III-IV Optimal medical treatment (Beta-blocker, ACE-inhibitor or AT1-antagonist, Ca^2+^-antagonist, ASS, CSE-inhibitor, long-acting nitrate \[LAN\]) Not suitable for percutaneous coronary intervention (PCI) or aortocoronary bypass surgery (CABG) after discussion with a team of specialists (cardiologists and heart surgeons) **Exclusion criteria** Myocardial infarction or unstable in the last 3 months Pacemaker with unipolar electrode Relevant valvular heart disease Symptomatic heart failure \> NYHA II Left ventricular ejection fraction (LVEF) \< 40% ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ::: Implantation of the electrode ----------------------------- The electrode was placed in the epidural space at the level of Th4-5 through a small skin incision under local anesthesia, and then it was moved up to C7-Th1 subcutaneously. Correct placement was defined by test-stimulation and interviewing the patient. Patients should experience paraesthesia in the same dermatome normally affected by angina. After final placement, the electrode was connected to an external stimulator, which allowed the patient to test the effects of SCS for 3--4 days. Afterwards the final stimulator (Medtronic Itrel III) was implanted subcutaneous in the lateral abdomen and connected to the electrode. The stimulator was controlled through an external device by the patient allowing to change the amplitude of the stimulus or to switch it on/off. During the course of the study, examinations were scheduled after four weeks, three months (with MIBI-SPECT and SAQ), twelve month (with MIBI-SPECT and SAQ) and whenever wished by the patients. Exercise capacity ----------------- The bicycle ergometry protocol began at a workload of 25 Watt and increases in 25 Watt increments every 2 minutes. Exercise was terminated by the patient, when angina pectoris occurred or he was exhausted. Criteria for early termination of exercise were ischemic ST-segment depression \> 2 mm; ventricular or supraventricular tachycardia; and a decrease or abnormal elevation in systolic blood pressure. In our study in none of the patients termination was necessary because of tachycardia or abnormal blood pressure. The six-minute walk test was performed using an internal hallway with a marked distance of 50 meters. Patients were instructed to walk the distance at their own pace but to cover as much ground as possible. They were allowed to stop and rest during the test if needed. Nitrate intake and Seattle Angina Questionnaire (SAQ) ----------------------------------------------------- The SAQ is a self-administered, disease-specific measure for patients with CAD that has previously been demonstrated to be valid, reproducible, and sensitive to clinical change \[[@B16]-[@B18]\]. The SAQ measures five dimensions of coronary artery disease: patients\' physical limitations caused by angina, the frequency of and recent changes in their symptoms (angina frequency and angina stability), their satisfaction with treatment and the degree to which they perceive their disease to affect their quality of life. Each scale is transformed to a score of 0 to 100, where higher scores indicate better function (e.g. less physical limitation, less angina, and better quality of life). The amount of short acting nitrates (SAN) during one week was reported by the patients at every visit in the clinic. For our analysis we used average SAN consumption after 3 and 12 months. Myocardial scintigraphy (MIBI-SPECT) ------------------------------------ SPECT- examinations were run on two days, one day under resting conditions, the second day after stress, using 740 MBq 99-Technetium-MIBI. Stress was induced either by physical exercise (bicycle ergometry) or by dobutamine injection (10, 20, 30, 40 μg per /kg/ min). Scanning was run one hour after the injection in the stress examinations and two hours after resting-conditions. To compare conditions before SCS implantation and during follow up stress scans were made at a comparable rate pressure product. The camera was a three-head gamma camera with a LEHR- (low-energy-high-resolution) -collimator (prism 3000, Picker/Philips; matrix 128 × 128, 40 steps with 3° each, 20 seconds per step). For generating tomograms in short and long axis views, a filtered backprojection technique with additional semiquantitative polarmap analysis was used. SPECT examinations took place prior to the implantation of the SCS system as well as three and twelve months after. Myocardial scintigraphy is usually used to access regional perfusion, e.g. after a heart attack or PCI. However, if the total myocardial perfusion in one patient is to be examined over a course of time, one frequently encounters the reduction of ischemia in one area combined with increased ischemia in another, e.g. improved perfusion in the anterior myocardium with reduced perfusion in the posterior myocardium. This makes it difficult to judge whether conditions have improved or actually worsened. To meet this problem, results were interpreted with the „Emory Cardiac Toolbox" (Emory Medical Device Group). This software compares the perfusion found in the patient to a virtual normal perfusion, setting the normal nuclide distribution- that is the normal perfusion- as 1. If a patient has a score of 0.5 (50% of normal radionuclide distribution and myocardial perfusion) in the initial scan and reaches a score of 0.8 (80% of normal radionuclide distribution and myocardial perfusion) in follow-up examinations this means total perfusion has improved, even if perfusion of some areas has decreased. The score is not determined in direct comparison of two SPECT-scans, but by comparing each scan to virtual normal results. This software makes it possible to monitor the development of myocardial perfusion over a course of time. In spite of this method, accessing the total perfusion remains more complex than judging regional perfusion (e.g. after PCI or heart attack). We therefore decided to limit our interpretation of data on the development of myocardial perfusion in the course of the study as follows: A score improved at least 0.2 was judged an improvement; a score lowered by 0.2 was regarded as a worsened state. A score of ± = 0.1 was defined as unchanged. We left out any semi-quantitative scores with standard deviation as we felt they would over-interpret the data. Statistical analysis -------------------- Data are presented as mean ± 95 % confidence interval. Differences among groups were compared by an unpaired t-test. Significance was assigned to a value of p \< 0.05. Results ======= Patients -------- The mean age was 65 years (± 11). As angina is more common in men, there were more men (n = 26) than women (n = 5) enrolled. 26 Patients suffered from a 3-vessel coronary artery disease, 5 from a 2-vessel disease. All patients had undergone bypass surgery (n = 24) and/or PCI (stent) (n = 28); five patients already underwent transmyocardial laser revascularization. Detailed patient characteristics are presented in table [2](#T2){ref-type="table"}. Drug therapy had been optimized; apart from morphine intake it was not changed significantly during the trial\'s course (see table [3](#T3){ref-type="table"}). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Patient characteristics at baseline ::: ---------------------------------------------- -------------------- Gender: male/female 26 / 5 Mean Age (range), years 65 (35--79) Diagnosis 3-vessel coronary artery disease 26 (84%) 2-vessel coronary artery disease 5 (16%) History, number of patients and percentage Myocardial infarction 23 (74%) Heart Failure \>NYHA II 0 (0%) Previous coronary intervention (PTCA/Stent) 28 (90%) Previous CABG 24 (77%) TMLR 5 (16%) Cerebrovascular Disease (stroke and TIA) 5 (16%) Symptomatic peripheral vascular disease 4 13%) Hypertension 16 (52%) Diabetes mellitus 7 (23%) Hyperlipoproteinemia 23 (74%) Adipositas (BMI \>30) 8 (26%) Nicotine abuse past/current 24 / 6 (77% / 19%) ---------------------------------------------- -------------------- NYHA = New York Heart Association, PTCA = percutaneous transluminal coronary angioplasty, CABG = coronary artery bypass graft, TMLR = transmyocardial laser revascularization, TIA = transitoric ischemic attack, BMI = body mass index ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Medication of the patients at baseline and one year follow up ::: **Baseline (n = 31)** **1 Year (n = 27)** ----------------------------------- ----------------------- --------------------- **Beta-blocker** 31 (100%) 27 (100%) **ACE-Inhibior/AT~1~-Antagonist** 29 (94%) 26 (96%) **Ca^++^-Antagonist** 26 (84%) 24 (89%) **CSE-inhibitor** 31 (100%) 27 (100%) **Long-acting nitrate** 31 (100%) 25 (93%) **Morphine derivate** 12 (39%) 1 (4%) Medication of the patients at baseline (n = 31) and one year follow up (n = 27). Before one year follow up SCS device was explanted in 3 patients (2 during test phase, one after 8 months). One patient died before last visit. ::: 2 patients did not feel a positive effect of SCS on their symptoms in the test period (after implantation of the electrodes and using the external stimulator) and therefore dropped out during the initial phase of the study. SCS device was explanted in another patient after 8 months because of a lack of anti-anginal effect, too. Unfortunately, after explantation of the SCS device none of the 3 patients joined further follow up visits or MIBI-SPECT examinations. One patient died because of a liver cell cancer one month before one year visit. Cancer was not known at time of inclusion in the study. Exercise capacity ----------------- The physical fitness of patients improved as measured by six-minute walk test and bicycle ergometry already after 3 months (142.74 ± 22.13 vs. 224.83 (± 23.96) meters and 67.50 ± 7.11 vs. 96.43 ± 11.74 watt) and maintained at this level after one year (250.56 ± 25.24 meters; 98.08 ± 8.12 watt, see table [4](#T4){ref-type="table"}). The 3 months follow up was finished by 29, the one year follow up by 27 patients (see above). ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### SAN consumption per week and results of bicycle ergometry and 6-minute walk test at baseline and follow up ::: **Baseline** **3 Months** **1 Year** -------------------------------- ------------------ -------------------- -------------------- **SAN Consumption**amount/week 12.35 (± 1.6) 3.38 (± 0.96)\* 2.78 (± 0.90)\* **Bicycle Ergometry**watt 67.50 (± 7.11) 96.43 (± 11.74)\* 98.08 (± 8.12)\* **6-Minute Walk**meters 142.74 (± 22.13) 224.83 (± 23.96)\* 250.56 (± 25.24)\* \* = p \< 0.05 compared to baseline. Results are listed in mean ± 95% confidence interval. ::: Symptoms, nitrate intake and Seattle Angina Questionnaire (SAQ) --------------------------------------------------------------- Apart from the three patients who decided to have the electrodes explanted as they felt no relieve in symptoms, all other patients felt a significant decrease of symptoms due to treatment with SCS (28 out of 31). Eleven out of twelve patients treated with morphine at baseline no longer needed morphine at all; one patient was at least able to reduce the dose from 60 mg morphine sulphate (MST^®^) to 20 mg per day. Two patients no longer needed long-acting nitrates (see table [3](#T3){ref-type="table"}). The use of short-acting nitrates (SAN) was lowered significantly, as well (12.35 ± 1.6 vs. 3.38 ± 0.96 and 2.78 ± 0.90 times per week, see table [4](#T4){ref-type="table"}). The Seattle Angina Questionnaire showed a significant increase in quality of Life after 3 months under SCS treatment (SAQ score 39 at baseline vs. 73). After one year the improved quality of life maintained (score 75). Scores for physical limitation 33 vs. 47 \[3 months\] and 48 \[1 year\]), angina stability (30 vs. 51 and 53) and frequency (26 vs. 53 and 54) improved significantly, as well (see figure [1](#F1){ref-type="fig"}). ::: {#F1 .fig} Figure 1 ::: {.caption} ###### Results from the Seattle Angina Questionnaire (SAQ). \* = p \< 0.05 compared to baseline. Results are in mean ± 95% confidence interval (error indicator). ::: ![](1468-6708-6-7-1) ::: Myocardial perfusion (MIBI-SPECT) --------------------------------- As described above, data were interpreted with the „Emory Cardiac Toolbox" software. The examinations 3 months after SCS implantation (n = 29) showed unchanged SPECT results in most patients (n = 16). Only 6 patients showed a better myocardial perfusion while in 7 ischemia worsened (figure [2](#F2){ref-type="fig"}). Interestingly, a shift towards an improved myocardial perfusion could be found after one year. Of 27 examined patients a worsened situation compared to baseline was found only in one patient. 16 patients showed clearly less myocardial ischemia. Results of the MIBI-SPECT analysis with the \"Emory Cardiac Toolbox\" software are shown in figure [2](#F2){ref-type="fig"}. Black arrows indicate the shift of the patients between the three groups (worse, equal or improved myocardial perfusion) during 3 and 12 months follow up. An exemplary scintigram prior to and twelve months after implantation is shown in figure [3](#F3){ref-type="fig"}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### Results of the MIBI-SPECT analysis with the \"Emory Cardiac Toolbox\" software. Black arrows indicate the shift of the patients between the three groups (worse, equal or improved myocardial perfusion) during 3 and 12 months follow up. ::: ![](1468-6708-6-7-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### Scintigram of one patient prior to (left) and twelve months after (right) implantation. Each first line shows a stress scan, the second line the appendant rest scan. The third line shows again a stress scan etc. After one year clearly reduced hibernation resp. prolonged ischemia of the anterior LV wall (arrows vertical long axis views). Less pronounced reduction of the ischemia of the lateral LV wall (arrow heads short axis views). ::: ![](1468-6708-6-7-3) ::: Complications ------------- No serious complications were caused by the implantation of the neurostimulators. Neither injuries of the spinal cord nor infections occurred. The major problem in the beginning were a dislocation of the electrodes (n = 4). This could be remedied without any difficulties by surgical re-intervention. As the operating surgeons became more experienced, electrode dislocations requiring surgical intervention did not happen any longer. Small dislocations could be handled easily by re-programming the stimulation parameters. Discussion ========== This is the first clinical report with a long term follow up of myocardial perfusion measured by MIBI-SPECT in patients with multivessel coronary artery disease under spinal cord stimulation (SCS). Like in prior investigations we additionally measured symptoms, quality of life and exercise capacity of the patients, as well. There was an early reduction of angina related symptoms under SCS treatment. Patients showed not only a significant reduction of SAN consumption (table [4](#T4){ref-type="table"}) but also less use of morphine derivates already in the first weeks under neurostimulation therapy (table [3](#T3){ref-type="table"}). The analysis of the self-administered Seattle Angina questionnaire (SAQ) resulted in improved scores for physical limitation, angina frequency, angina stability and quality of life after 3 months with unchanged results after one year (figure [1](#F1){ref-type="fig"}). In opposite to the early functional improvement of the patients (symptoms, physical activity, quality of life), myocardial ischemia measured by MIBI-SPECT was not decreased after 3 months. 22 patients showed unchanged or even more ischemia compared to baseline before SCS implantation (figure [2](#F2){ref-type="fig"}). After one year -- while functional parameters (exercise capacity, quality of life etc.) did not show a further change -- myocardial perfusion was remarkable improved in many patients (16 of 27). Only one patient still had more cardiac ischemia than in baseline examination (figure [2](#F2){ref-type="fig"}). Interestingly, the score for physical limitation (SAQ) of this patient did not improve very much, as well (55 \[baseline), 62 \[3 months\], 58 \[1 year\]). Since a direct effect of SCS on myocardial perfusion should be established early after beginning of the neurostimulation, one would have expected to see alterations in the MIBI-SPECT after 3 months. The improved myocardial perfusion found after one year might therefore be due to the development of collateral vessels associated with increased physical fitness. Two recent studies reported alterations in myocardial blood flow already after 4 or 6 weeks \[[@B22],[@B23]\]. Both found a homogenization of myocardial perfusion measured by positron emission tomography (PET) in patients with multivessel disease or syndrome X. They suggested that the change in myocardial blood flow can be denoted as a steal phenomenon at the microcirculatory level eventually mediated by a direct effect of SCS on coronary vasomotion. However, they could not find an improved resting myocardial blood flow in their studies, as well. Hautvast and co-workers even found a decreased flow reserve under dipyridamole stress \[[@B22]\]. Thus, our study is the first to show that myocardial perfusion can improve under neurostimulation in patients with coronary multivessel disease. Since we applied a different method to measure myocardial ischemia, we cannot provide data regarding a possible homogenization of cardiac blood flow in the early phase of spinal cord stimulation. But our data suggest that the main effect of SCS treatment is anti-symptomatic and changes in myocardial ischemia are rather secondary and might be due to increased physical activity. This hypothesis is supported by data of Belardinelli and co-workers, who demonstrated that moderate exercise training improved myocardial perfusion measured by thallium scintigraphy in patients with chronic coronary artery disease \[[@B19]\]. If this decrease of myocardial ischemia does also improve the prognosis of the patients cannot be answered yet. Nevertheless, the impressive improvement of angina-related symptoms, the exercise ability and last but not least the increase in quality of life of 90% (28/31) of our patients in a one year follow up supports the data of other studies which treated patients with CAD and refractory angina with spinal cord stimulation. In a recent study, which included 32 patients with a follow up of 65 months, SCS relieved angina effectively also after long-term treatment without development of tolerance. SAQ scores still were improved after years \[[@B20]\]. Maintenance of increased exercise ability for years might be an ongoing stimulus for coronary collateralization. In conclusion electrical spinal cord stimulation is a valuable therapeutic option for the treatment of refractory angina pectoris in patients not suitable for coronary revascularization procedures. Additionally, the symptomtic relief may allow more physical activity which might induce coronary collateralization resulting in an improved myocardial perfusion. Study limitations ----------------- Since it is difficult to conduct placebo-controlled trials with electrical neuromodulation, because there is no alternative for the paraesthesia induced by the neurostimulators, our study has -like others- no placebo group. However, measurement of myocardial perfusion with MIBI-SPECT is an objective parameter not impressionable by the patient \[[@B21],[@B22]\]. As placebo effect is generally believed to decrease with time \[[@B23]-[@B26]\] the sustained relief of symptoms over one year in this study seem to contradict the estimation that placebo effects are the major mode of action of SCS treatment. To finally prove whether the altered myocardial perfusion is due to direct SCS effects or increased exercise ability with improved collateralization a control group without SCS but with exercise training would be necessary.

PubMed Central

2024-06-05T03:55:58.455574

2005-5-19

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173130/", "journal": "Curr Control Trials Cardiovasc Med. 2005 May 19; 6(1):7", "authors": [ { "first": "Holger", "last": "Diedrichs" }, { "first": "Carsten", "last": "Zobel" }, { "first": "Peter", "last": "Theissen" }, { "first": "Michael", "last": "Weber" }, { "first": "Athanassios", "last": "Koulousakis" }, { "first": "Harald", "last": "Schicha" }, { "first": "Robert HG", "last": "Schwinger" } ] }

PMC1173131

I see no important disagreement between me and the commentators, but disagreement does exist on these topics. I would classify opposing views into two major categories. On one side, some writers assert that causal inference can and should be done without counterfactuals. In my view, Dawid \[[@B1],[@B2]\] is a moderate on that side, as he deploys devices that are isomorphic to counterfactuals, so the difference seems mostly one of labeling and emphasis (which is not always unimportant). More radical counterfactual deniers include Shafer \[[@B3]\], who appears mostly upset because counterfactual models continue to weave into the foundation of statistics, econometrics, sociology, and health sciences, while his approach \[[@B4]\] appears destined for the dustbin, along with other truly noncounterfactual theories of causation. Pearl \[[@B5]\] gives a succinct account of the failings of these theories (reference 5, section 7.5), noting (as do Greenland and Brumback \[[@B6]\]) that the causal models entering into scientific teaching and application (including causal graphs, causal \"pies,\" and structural equations) have mappings into counterfactual formalisms. The target of my article was the other side, those who take counterfactuals uncritically or superficially, without paying enough attention to what these hypothetical quantities are supposed to mean. My article originated as a chapter in a WHO volume \[[@B7]\]. This volume arose from a conference which seemed a festival of counterfactual abuse, rife with talk of cause-of-death removal as if it were an intervention. It is disheartening if not frightening to witness discussion of global health policy framed in such terms. In this context, the concerns expressed by Dawid \[[@B1],[@B2]\] about counterfactuals seem reserved. Counterfactual abuse can be diminished by connecting potential outcomes to interventions. Susser and Schwartz \[[@B8]\] point out that this connection is needed for lifestyle risk factors (smoking, physical inactivity, etc.) just as for social factors. I agree; risk-factor epidemiology could better serve public health if it addressed what could be done, rather than estimating effects of the unattainable (like removal of all tobacco exposure), without regard to how change is brought about. Becoming more realistic involves more than just operationalizing the exposure (right-hand) side of the structural equation; one also needs to expand the left-hand side to consider the full spectrum of intervention effects, such as all effects of smoking cessation (e.g., weight gain, depression). The traditional narrow focus on a few prominent endpoints (like cancer and cardiovascular disease), encouraged by the case-control viewpoint, has discouraged grappling with the multivariate complexity of outcomes as well as exposures. Worse, in the smoking context, there may be a bias against acknowledging that one of our most damaging population exposures (tobacco) may bring worthwhile benefits to a non-negligible portion of the population -- not just medical benefits like Parkinsonism prevention, but also psychologic benefits like enhanced sense of well-being, which are hard to measure and weigh against costs. If we really believe in informed consent, then we must inform the public about how lifestyle choices are not just about lifespan maximization, but are also choices of how to live and die. This view will not sit well with those for whom good sensations are evil if the sensations do not come from sanctified sources like religious faith, licensed entertainment, or prescription drugs.

PubMed Central

2024-06-05T03:55:58.457904

2005-5-19

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173131/", "journal": "Emerg Themes Epidemiol. 2005 May 19; 2:2", "authors": [ { "first": "Sander", "last": "Greenland" } ] }

PMC1173132

Background ========== Diethylcarbamazine citrate (DEC) has been used in the treatment and control of lymphatic filariasis (caused by the nematodes *Wuchereria bancrofti*, *Brugia malayi*and *B. timori*) since 1947 and it continues to play an important role, being one of the drugs used in the Global Programme for the Elimination of Lymphatic Filariasis \[[@B1]\]. However, despite this long period of use, DEC\'s mode of action is still poorly understood. Particularly intriguing is the marked contrast between its rapid action *in vivo*and the lack of significant activity *in vitro*. *In vivo*, the response is rapid: within a few minutes of treatment, peripheral blood microfilariae counts drop dramatically \[[@B2]\]. The poor *in vitro*activity indicates that DEC probably requires some host factor for its activity, and previous work has highlighted the role of the innate immune system and leucocytes independent of T cells and complement in the activity of DEC \[[@B3],[@B4]\]. DEC also has anti-inflammatory properties, as a result of its interference with arachidonic acid metabolism \[[@B4]\]. The products of the arachidonic acid metabolic pathway, eicosanoids, have a number of biological effects, including inhibition of platelet aggregation; regulation of leucocyte activation and adherence; mediation of granulocyte chemotaxis and degranulation; and promotion of vasodilatation \[[@B5]\]. It is well known that DEC inhibits enzymes of the 5-lipoxygenase pathway, leukotriene synthases \[[@B6],[@B7]\]. Additionally, *in vitro*, DEC blocks endothelial cell production of the cyclooxygenase (COX) pathway products prostaglandin (PG) E~2~, prostacyclin (PGI~2~) and thromboxane A~2~but has no effect on platelet prostanoid production \[[@B8]\]. In addition, the drug increases the rate and degree of microfilariae adherence to granulocytes, with eosinophil adhesion in particular being augmented \[[@B9]-[@B11]\]. Nevertheless, a role for some of these activities has yet to be demonstrated *in vivo*and so we have used a mouse model to identify the host factors responsible for the rapid efficacy of DEC. The arachidonic acid pathway includes lipoxygenase and cyclooxygenase enzymes. The COX pathway has similarities with the nitric oxide (NO) pathway, since both have constitutive and inducible isoforms of their enzymes and are key regulators of inflammatory responses \[[@B12],[@B13]\]. The COX and NO pathways are known to interact with each other, with there being \'cross-talk\' between NO/PGE~2~and iNOS/COX which is generally stimulatory but may also be inhibitory \[[@B14],[@B15]\]. Therefore, we have used a combination of pharmacological inhibitors and gene-knockout technology to elucidate the role of these two pathways in DEC\'s activity *in vivo*. Materials and methods ===================== Parasites and mice ------------------ Microfilariae of *Brugia malayi*were obtained from TRS Laboratories (Georgia, USA), suspended in RPMI 1640 with 5% FCS, and 300000 parasites in a volume of 200 μl were injected intravenously into mice. Systemic parasitaemia was allowed to equilibrate for 24 hours, then heparinised blood samples were taken by tail bleeding and parasitaemia was measured. Mice were allocated into age- and size-matched groups and treated as described below. All animals were kept in the Biological Services Unit of the University of Liverpool in accordance with Home Office regulations and were fed and watered *ad libitum*. BALB/c mice were kept under standard conditions, and the 129/SV and targeted knockout of the iNOS gene (iNOS^-/-^, kindly provided by Prof. F.Y. Liew, University of Glasgow) strains in filter-top cages. Action of DEC against microfilariae *in vivo*in mice ---------------------------------------------------- Three BALB/c mice infected with *B. malayi*microfilariae were treated with a single, oral dose of DEC 100 mg/kg \[[@B3]\] (Sigma, U.K) in distilled water and the parasitaemia monitored from five minutes to two weeks post treatment. To investigate the role of the arachidonic acid metabolic pathway in the mode of action of DEC, indomethacin (10 mg/kg in 1% ethanol), water-soluble dexamethasone (3 mg/kg in water, both obtained from Sigma, U.K.), or vehicle was given by intra-peritoneal (i.p.) injection to microfilaraemic male BALB/c mice 30 minutes before oral DEC administration (100 mg/kg, three mice per treatment group). One animal was kept as an untreated control. Heparinised blood samples were taken at intervals post treatment for measurement of parasitaemia. Experiments were repeated three times. The requirement for inducible NO in DEC\'s efficacy was determined in iNOS^-/-^mice. DEC (100 mg/kg) or vehicle were administered orally to three female iNOS^-/-^mice or their background strain, 129/SV. Mice were tail-bled at regular intervals post-treatment for evaluation of parasitaemia. To test the efficacy of another anti-filarial drug, ivermectin, in these mice, ivermectin phosphate (1 mg/kg in 1% DMSO) was administered by i.p. injection. This experiment was repeated three times. Expression of COX-1, COX-2 and iNOS in DEC-exposed peritoneal exudate cells --------------------------------------------------------------------------- Male 129/SV and iNOS^-/-^mice were injected i.p. with 10 mg/kg DEC in endotoxin-free water or 100 μl of endotoxin-free water (three mice in each group). After 30 minutes, peritoneal exudate cells were collected in sterile PBS with 1 g/L glucose, 1% bovine serum albumin and 1 U/ml heparin. The cells were pelleted and lysed in 1 ml TRI reagent (Sigma, U.K.) then protein extracted according to the supplied protocol. For Western blot analysis, 10 μg of each protein were separated on a 7.5 % denaturing SDS polyacrylamide gel and blotted on to 0.45 μM pore size PVDF membrane (Immobilon P, Micropore, U.K.). After blocking overnight at 4°C in block buffer (1% casein in PBS/0.1% Tween) and washing in PBS/0.1% Tween, membranes were incubated for 1 hour in rabbit anti-mouse COX-1, COX-2 or iNOS polyclonal IgG (Cayman Chemical Co., Alexis Corporation, U.K.) diluted to 1 in 5000 in block buffer. The anti-COX antibodies showed no cross-reactivity with the opposite isoform, whilst the anti-iNOS antibody showed only 5% cross-reactivity against nNOS and none against eNOS. Membranes were then washed and incubated for 1 hour in goat anti-rabbit IgG conjugated to horse radish peroxidase (Nordic, The Netherlands) diluted to between 1 in 20000 and 1 in 100000, depending on the primary antibody, followed by further washing. The electrochemiluminescent reagent SuperSignal West (Pierce Perbio, U.K.) was used to visualise the bands on X-ray films. Statistical analysis -------------------- Parasitaemia data were expressed as mean percentage of pretreatment microfilariae or as a percentage of untreated control microfilaraemias per 100 μl of blood and were analysed by the two-tailed Student\'s *t*-test. P values of \< 0.05 were considered to be significant. Results ======= Action of DEC against microfilariae *in vivo*in mice ---------------------------------------------------- In BALB/c mice treated with DEC alone, microfilaraemia levels were reduced by five minutes with a sustained reduction for at least 60 minutes post-treatment (Fig. [1](#F1){ref-type="fig"}). However, by 24 hours after treatment, microfilarial levels had partially recovered and two weeks later they had returned to levels approaching those pre-treatment (Fig. [1](#F1){ref-type="fig"}). Subsequent experiments focused on the rapid activity of DEC over the first one to two hours. Neither vehicle, indomethacin nor dexamethasone by itself had any effect on microfilaraemia in BALB/c mice (data not shown). However, in mice pre-treated with indomethacin or dexamethasone, microfilaraemias were reduced by only 11% (dexamethasone) or 44% (indomethacin) of untreated controls at 60 minutes post DEC administration (Fig. [2](#F2){ref-type="fig"}). The differences from the DEC-only group were statistically significant for all time points for indomethacin (P \< 0.004) and for 15 and 30 minutes post-treatment for dexamethasone (P \< 0.017) pre-treatments. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **DEC causes rapid sequestration of *B. malayi*microfilariae in BALB/c mice.**BALB/c mice intravenously injected with *B. malayi*microfilariae were dosed orally with 100 mg/kg DEC and microfilaraemia monitored from 5 to 60 minutes post treatment, then at 24 hours and two weeks. ::: ![](1475-2883-4-4-1) ::: ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Indomethacin or dexamethasone pre-treatment reduces efficacy of DEC in BALB/c mice infected with *B. malayi*microfilariae.**Indomethacin (10 mg/kg), dexamethasone (3 mg/kg) or vehicle was administered 30 minutes before oral dosing with DEC (100 mg/kg). Symbols are means of three mice for the DEC plus dexamethasone group (triangles), seven mice for the DEC plus indomethacin group (white circles) and four mice for the DEC-only group (black circles). Significantly different results from the DEC-only group are denoted by \* (P \< 0.017), \*\* (P = 0.001) or \*\*\* (P = 0.000). ::: ![](1475-2883-4-4-2) ::: DEC administration also rapidly reduced microfilaraemias in 129/SV mice but, in contrast, had no effect on microfilariae levels in iNOS^-/-^mice, in which microfilaraemia was maintained at pre-treatment levels for at least 2 hours (Fig. [3a](#F3){ref-type="fig"}), with no significant differences from untreated iNOS^-/-^controls (P \> 0.887 for all time points). In contrast, ivermectin was effective in both 129/SV and iNOS^-/-^mice (Fig. [3b](#F3){ref-type="fig"}), although it had a slower onset of action than DEC. However, by 24 hours no microfilariae were detected in either strain of mouse given ivermectin. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **DEC is ineffective against *B. malayi*microfilariae in the absence of iNOS.**Efficacy of (a) a single, oral dose of DEC (100 mg/kg) or (b) a single, i.p. dose of ivermectin phosphate (1 mg/kg) in 129/SV and iNOS^-/-^mice infected with *B. malayi*microfilariae. Black symbols represent 129/SV mice, white symbols iNOS^-/-^. Squares indicate DEC administration, triangles ivermectin administration and circles untreated controls. Symbols represent mean results from at least three or four mice, except in the case of those treated with ivermectin (two animals) from two combined experiments which were representative of a further repeat. Significantly different microfilaraemias between 129/SV and iNOS^-/-^mice after DEC administration are denoted by \* (P = 0.001) or \*\* (P = 0.000). ::: ![](1475-2883-4-4-3) ::: Expression of COX-1, COX-2 and iNOS in DEC-exposed peritoneal exudate cells --------------------------------------------------------------------------- Thirty minutes after administration of endotoxin-free water to 129/SV and iNOS^-/-^mice, peritoneal exudate cells were expressing COX-1 protein, whereas those from DEC-exposed animals contained markedly less COX-1 (Fig. [4](#F4){ref-type="fig"}). Interestingly, there seemed to be a higher level of COX-1 remaining in the iNOS^-/-^than the 129/SV macrophages after DEC treatment. Neither COX-2 nor iNOS protein was detected in any of the 129/SV or iNOS^-/-^groups (not shown). ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Western blot detection of COX-1 protein from peritoneal exudate cells.**COX-1 protein was detected in 129/SV and iNOS^-/-^peritoneal exudate cells thirty minutes after i.p. injection of endotoxin-free water (control) or DEC (10 mg/kg). Proteins (10 μg) were separated on a 7.5% denaturing SDS polyacrylamide gel, transferred to PVDF membrane, incubated with rabbit anti-mouse COX-1, then goat anti-rabbit IgG-horse radish peroxidase conjugate and detected by chemiluminescence. ::: ![](1475-2883-4-4-4) ::: Discussion ========== Here we have used a murine model to elucidate the processes within the mammalian host that contribute to DEC\'s rapid *in vivo*action. The involvement of two interacting pathways, the cyclooxygenase and inducible nitric oxide pathways, were shown to mediate the activity of DEC *in vivo*. Treatment of mice with DEC resulted in a rapid reduction in microfilaraemia. This reduction, however, was transient and microfilaraemia began to recover 24 hours after treatment, with almost full restoration to pre-treatment levels two weeks after treatment. This has been previously observed in other models \[[@B16],[@B17]\] and suggests that the disappearance of the microfilariae from the peripheral circulation and their sequestration in the central vascular system occur independently of parasite killing. A prolonged course of DEC treatment of *B. malayi-*infected mice led to sustained reductions in circulating microfilariae for at least 30 days \[[@B18]\]. Our results confirm previous findings showing that an important target for DEC is the arachidonic acid metabolic pathway. Inhibition at the first stage in the pathway by dexamethasone, which inhibits phospholipase A2, almost completely abolished the activity of DEC, whereas inhibition of the cyclooxygenase enzymes COX-1 and COX-2 by indomethacin reduced its efficacy by 56%, indicating that in addition to its well documented inhibition of the 5-lipoxygenase pathway \[[@B6],[@B7]\], DEC acts on the cyclooxygenase pathway. We have shown that at least one way it does this *in vivo*is by the loss of COX-1 protein within 30 minutes of administration. The lack of activity of DEC in mice deficient in iNOS identifies a novel enzyme system involved in the *in vivo*activity of DEC. Previously we have shown that *B. malayi*microfilariae are susceptible to nitric oxide *in vitro*\[[@B19]\]. However, we found no evidence that DEC itself up-regulated iNOS activity either *in vitro*(not shown) or *in vivo*, in agreement with Rajan et al. \[[@B20]\], who did not find any induction of NO release from murine macrophages or rat endothelial cells treated with DEC. It therefore seems probable that iNOS exerts an effect on DEC activity via its interaction with cyclooxygenases, an idea supported by the reduced loss of COX-1 protein in peritoneal exudate cells derived from iNOS^-/-^mice. Several studies have shown that NO and iNOS interact with COX enzymes to cause an increase in enzymatic activity \[[@B21]\] and consequently increased prostaglandin synthesis \[[@B22]-[@B25]\], although large amounts of endogenous NO inhibited COX expression and activity in murine macrophages \[[@B26]\]. One explanation of the differential effects of NO on COX activity may relate to effects on different COX isoforms. For example, NO can activate COX-1 in fibroblasts but inhibit COX-2 in the same cell \[[@B15]\]. Although our studies do not distinguish between the role of COX-1 and COX-2 in DEC\'s activity, the rapid activity of DEC sequestration and the depletion of COX-1 protein suggest a role for COX-1. COX-1 but not COX-2 is essential for the early production of prostaglandins from macrophages and mast cells \[[@B27],[@B28]\]. Further studies on mice deficient in COX isoforms or the use of isoform-specific pharmacological inhibitors could address this question. Several polymorphisms in the human iNOS gene have been described that are associated with a variety of diseases, including malaria \[[@B29]-[@B31]\] and hypertension \[[@B32]\]; it would be interesting to know if these or other polymorphisms affected responsiveness to DEC therapy. Our findings could help expand our understanding of the mechanisms involved in the cellular processes leading to sequestration and the subsequent killing of parasites. In addition to the elevation of granulocyte adherence, platelets have also been shown to bind to and kill microfilariae \[[@B33]\]. In view of the well know effects of NO and prostaglandins on platelet function and evidence to suggest the presence of inducible NO in human platelets \[[@B34],[@B35]\], the role of platelets in parasite sequestration and killing should be re-evaluated *in vivo*. Filarial parasites also produce and release prostanoids, including PGE~2~, PGI~2~and PGD~2~\[[@B36]-[@B41]\], which result in inhibition of platelet aggregation \[[@B40]\], vasodilatation of the blood vessels and immune suppression, and may contribute to the long persistence of these parasites in their natural hosts \[[@B41]\]. This prostanoid production is also inhibited by DEC \[[@B8]\]. Significantly, they do not produce thromboxane A~2~\[[@B36]\]. In contrast to mammalian systems, in which eicosanoid formation is often in response to agonist-induced stimulation, microfilariae produce prostanoids constitutively \[[@B36]\], but the mechanisms by which they do so have not yet been described in detail, although a glutathione *S*-transferase of *O. volvulus*synthesizes PGD~2~from PGH~2~\[[@B39]\]. It is not clear if DEC acts predominantly against the prostanoids of the worm or of the host. The lack of any direct effect of dexamethasone and indomethacin on microfilaraemia suggests that these drugs either do not influence parasite prostaglandins *in vivo*or that if they do, they are not involved in DEC-mediated sequestration. Further studies that involve inhibition of the key parasite enzymes would be required to determine the role of parasite-derived prostanoids in DEC activity. Recent studies have reported a direct activity of DEC against *Wuchereria bancrofti*microfilariae that results in exsheathment, organelle damage and cytolysis \[[@B42]\], which occur both *in vitro*and *in vivo*and suggest that DEC may have a direct effect on worms in addition to its interaction with host-derived pathways as reported here. Much remains to be discovered of the mode of action of DEC. What mechanisms lead to parasite killing following sequestration in the central vasculature; and how does this relate to the paradoxical appearance of microfilariae in the peripheral circulation following the \'DEC provocative test\'? What is the role of host immunity and effects on adult worms in the long-term efficacy of DEC? This model should be a powerful tool to address these questions and others to further unravel the mysteries of this elusive drug. Conclusion ========== Inducible nitric oxide synthase and the cyclooxygenase pathway were found to be essential for DEC\'s activity *in vivo*. Along with its well-documented activity on the lipoxygenase pathway, DEC administered *in vivo*reduced the amount of the host\'s COX-1. Further elucidation of DEC\'s mechanism of action with this murine model could provide a clearer understanding of the interaction of the nitric oxide and cyclooxygenase pathways and the cellular and molecular events at the site of sequestration. List of abbreviations ===================== DEC, diethylcarbamazine citrate; COX, cyclooxygenase; i.p., intra-peritoneal; PG, prostaglandin; PGI~2~, prostacyclin; NO, nitric oxide; iNOS, inducible nitric-oxide synthase. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= HFM assisted with the *in vivo*experiments, performed the Western blot detection, analysed and interpreted the results, conducted statistical analysis and wrote the manuscript. LDP collected the parasitaemia data and assisted with the *in vivo*experiments. MJT conceived the study, performed the *in vivo*experiments, interpreted the results and advised on the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ MJT is supported by a Senior Wellcome Trust Fellowship in Basic Biomedical Science. LDP was funded by the University of Liverpool Research Development Fund. We thank Prof. F.Y. Liew (University of Glasgow) for providing the iNOS^-/-^mice.

PubMed Central

2024-06-05T03:55:58.458351

2005-6-2

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173132/", "journal": "Filaria J. 2005 Jun 2; 4:4", "authors": [ { "first": "Helen F", "last": "McGarry" }, { "first": "Leigh D", "last": "Plant" }, { "first": "Mark J", "last": "Taylor" } ] }

PMC1173133

Background ========== Protease inhibitor based antiretroviral therapy (PI-ART), defined as the combination of at least two nucleoside analogues with at least one protease inhibitor (PI) \[[@B1]\], was introduced in 1996 and has greatly reduced the incidence of HIV-related morbidity and mortality in the industrialised world \[[@B2],[@B3]\]. PI-ART would thus be expected to have a positive effect on health-related quality of life (HRQL). On the other hand, HRQL might be negatively affected by strict adherence requirements as well as by short and long-term adverse effects \[[@B4],[@B5]\]. Several studies (cross-sectional or longitudinal) have focused on the HRQL of HIV-positive individuals in different stages of the HIV infection and under different treatment regimes. The results have varied but in general HIV infection affects several physical, psychological and social dimensions of HRQL and patients with symptomatic disease and/or an AIDS-defining complication are more severely affected than those with other comparable chronic diseases \[[@B6]-[@B8]\]. HRQL has been shown to be related to the CD4 value, viral load and symptoms, so that patients with a more advanced state of HIV infection reported poorer HRQL \[[@B9]-[@B15]\]. Moreover, symptoms, physical function, role function and sexual function deteriorated over time, while emotional domains were unchanged or improved \[[@B9],[@B10],[@B16]\]. Only a few studies on the influence of long-term (\>1 year) PI-ART on the quality of life have yet been published. Nieuwkerk *et al*. \[[@B17]\] compared the development of HRQL during three different PI based regimes and concluded that in terms of HRQL, patients with higher CD4 values at start experienced less benefit from the treatment. Burgoyne *et al.*\[[@B18],[@B19]\] followed HRQL in 41 patients with different treatment status over a period of four years and found no overall change of HRQL and that HRQL was less sensitive to CD4 changes than to symptom changes as well as that change in HRQL was somewhat related to change in social support. The main aim of the present study was to investigate HRQL before and after two years of first generation PI-ART. The following research questions were addressed: (a) Does HRQL change during two years of PI-ART? (b) Do viral response, adherence, subjective experience of adverse effects and initial CD4 count predict changes in HRQL during two years of PI-ART? Methods ======= Data collection --------------- ### Subjects The study was performed at the Department of Venhälsan at South Stockholm General Hospital, Sweden. A convenient sample of 72 subjects (70 men and 2 women) in an advanced state of HIV infection and who were among the first patients to receive PI-ART in Sweden responded to the HRQL instrument described below before the start of PI-ART (*pre PI-ART*). HIV infection was documented by at least two laboratory tests (two repeated ELISA tests or one ELISA test and one Western Blot). The patients were treated according to best clinical practice and the participant\'s physician chose the drug combination of the PI-ART (at least 2 nucleoside analogue reverse transcriptase inhibitors and either indinavir (n = 55) or ritonavir (n = 17) at start). Approximately two years after the introduction of PI-ART (*follow-up*; mean 25.1 months, standard deviation (SD) 2.8 months, post initiation), 54 of the 72 subjects (75 %) completed the *follow-up*measure. Thirteen patients had died, four had changed clinic and one did not receive the *follow-up*questionnaire. ### The Swedish Health-Related Quality of Life Questionnaire (SWED-QUAL) The patients completed the SWED-QUAL at the *pre PI-ART*and two-year *follow-up*visits. SWED-QUAL was developed by Brorsson *et al.*\[[@B20]\] from the measures used in the US Medical Outcomes Study (MOS) \[[@B21]-[@B24]\]. The questionnaire, which is designed to measure HRQL, consists of 70 items, of which 63 forms two single-item and 11 multi-item dimension scales of Likert type: physical functioning (7 items), mobility (1 item), satisfaction with physical ability (1 item), role limitations due to physical health (3 items), pain (6 items), emotional well-being: positive affect (i.e. positive feelings; 6 items), emotional well-being: negative affect (i.e. negative feelings; 6 items), role limitations due to emotional health (3 items), sleep problems (7 items), satisfaction with family life (relations with parents, siblings, children etc.; 4 items), relation to partner (6 items), sexual functioning (4 items) and general health perception (9 items). In the present study, the relation to partner section was slightly modified to make it suitable for the investigated group (i.e. the word \"spouse\" was replaced by \"partner\"). Each scale is transformed into a 0--100 index; the higher the score, the better the perceived HRQL. In a general population sample, the internal-consistency reliability coefficients (Cronbach\'s α) ranged from 0.79 to 0.89 \[[@B20]\]. Brorsson *et al.*have also reportedpreliminary support for the construct validity \[[@B20]\]. The instrument has been used in our previous study of HIV-positive subjects in Sweden \[[@B25]\]. A factor analysis (see below) was performed on a merged material from three groups of patients: (a) the group described above at *pre PI-ART*(n = 72), (b) a previously described group of protease inhibitor naïve HIV patients (n = 73) \[[@B25]\] and (c) an unpublished material of 164 HIV-negative men who have sex with men visiting our information and screening clinic for sexually transmitted diseases (own unpublished data). ### Adherence/adverse effects At the *follow-up*, the patients\' adherence and drug-related adverse effects were examined. Adherence was assessed by asking the following question, modified from Morisky *et al.*\[[@B26]\]: \"How many times during the last month did you skip doses of your HIV medication for the following reasons: because you felt so good that it did not matter, because you felt worse due to the medication or because you forgot?\" For each of the three given reasons the patients were asked to rate the number of missed doses on a scale with five alternatives (1 = 0 doses, 2 = 1--2 doses, 3 = 3--5 doses, 4 = 6--9 doses or 5 = \>9 doses). Patients were classified as non-adherent if they indicated \>9 missed doses in at least one of the three mentioned categories (i.e. indicating 90% adherence or less) or if they indicated missed doses in all three categories. The subjective experience of medication-related adverse effects was addressed in one question, which asked the patients to rate their global experience of adverse effects during the past month on a 10 cm visual analogue scale (VAS) ranging from \"none at all\" to \"severe\". ### Laboratory measures Lymphocyte subsets were determined, using routine flow cytometry \[[@B27]\], at least every 3--4 months. The level of HIV-1 RNA in plasma was quantified at the Swedish Institute for Infectious Disease Control by a commercially available reverse transcriptase polymerase chain reaction (PCR)-based assay (HIV Monitor; Roche Diagnostic Systems, Branchburg New Jersey, USA). The level of quantification at the time of the study was 500 viral copies ml^-1^. ### Viral outcome On the basis of the participants\' long-term virologic outcome to PI-ART, they were subdivided into viral responders and viral non-responders. Viral responders were defined as participants who, after the first 3 months of PI-ART, had either HIV RNA less than 500 copies ml^-1^in more than 75% of the analysed samples or a continuous decrease in HIV RNA to below 500 copies ml^-1^before 18 months. All other participants were regarded as viral non-responders \[[@B28]\]. ### Demographic and medical data Demographic and medical data were obtained from specific items in the questionnaires and from the medical files. Statistical methods ------------------- Statistical calculations were performed with the assistance of the personal-computer program SPSS for Windows, version 11.0.0. The criterion for statistical significance was *p*\< 0.05. Since some parameters did not fulfil the assumptions of a normal distribution, the Wilcoxon signed ranks test \[[@B29]\] was used to compare differences between two related groups (i.e. *pre PI-ART*versus *follow-up*). In order to achieve data reduction to create composite scores for use in multiple regression analyses of the material, a rotated component matrix analysis \[[@B30]\] was performed on the scores from ten of the 13 SWED-QUAL scales (i.e. those measuring physical and emotional health). The following scales were used in the factor analysis: physical functioning, mobility, satisfaction with physical ability, role limitations due to physical health, pain, emotional well-being: positive affect, emotional well-being: negative affect, role limitations due to emotional health, sleep problems and sexual functioning. The analysis resulted in two factors: (a) physical HRQL composite score, i.e. PCS, comprising the averaged scores of the SWED-QUAL scales physical functioning, mobility, satisfaction with physical ability, pain, role limitations due to physical health and sexual functioning, and (b) emotional HRQL composite score, i.e. ECS, comprising the averaged scores of the SWED-QUAL scales role limitations due to emotional health, positive affect, negative affect and sleep problems (Table [1](#T1){ref-type="table"}). These two factors explained 65 % of the total variance. ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Rotated factor loading (varimax) of the principal component analysis (n = 309) ::: **Rotated factor loadings** ------------------------------------------ ----------------------------- ------ **SWED-QUAL Scale** Physical functioning 0.92 0.07 Mobility 0.83 0.04 Satisfaction with physical ability 0.79 0.24 Pain 0.56 0.45 Role limitations due to physical health 0.66 0.46 Sexual functioning 0.63 0.32 Role limitations due to emotional health 0.33 0.75 Emotional well-being, positive affect 0.08 0.81 Emotional well-being, negative affect 0.08 0.85 Sleep problems 0.32 0.70 ^PCS^Physical HRQL composite score; ^ECS^Emotional HRQL composite score ::: A set of hierarchical multiple linear regression analyses \[[@B31]\] were performed to investigate whether the initial CD4 value, viral response, adherence and subjective experience of adverse effects predicted the change in HRQL. The two analyses (one for PCS and one for ECS) were conducted in two forced steps with the *follow-up*PCS and ECS as dependent variables. The *pre PI-ART*PCS and ECS, respectively, were entered in the first step, while the three variables viral response, adherence and subjective experience of adverse effects were entered in the second step together with the initial CD4 value. Ethical approval ---------------- The study was approved by the Local Ethical Committee at Huddinge University Hospital. Information about the study was given to the subjects in connection with an ordinary, scheduled visit to the clinic. Results ======= Demographic and medical data ---------------------------- At *pre PI-ART*, the mean age of the 72 patients was 41 (SD 9, range 23 -- 65) years, 59 were of Swedish origin, 61 had prior antiretroviral treatment with only nucleoside analogue reverse transcriptase inhibitors, 32 had a CD4 count of \<200 × 10^6^cells l^-1^and 28 had an AIDS diagnosis. Further demographic and medical data are shown in Table [2](#T2){ref-type="table"}. ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Demographic and medical data regarding a Swedish sample of HIV-infected persons before the start of PI-ART (*n*= 72) ::: **No. of subjects** **Percentage** ----------------------------------------------------------------- -------- --------------------- ---------------- ------------ **Mode of transmission** Male to male 69 96 Female to male/male to female 3 4 **Working status** Full or part time work/studies 36 50 Sick leave or disability pension 31 43 Not stated 5 7 **Education** Compulsory school 13 18 Upper secondary school/high school 27 38 University studies 27 38 Not stated 5 7 **Having a partner**: Yes/no 30/41 42/57 not stated 1 1 **CD4 count**, cells × 10^6^l^-1^: median 127 range 1--660 **HIV RNA**, log~10~copies ml^-1^: median 4.87 range 2.70--6.18 **Time since first start of antiretroviral treatment**, months: mean 33 SD 31.6 ^PI-ART^Protease inhibitor based antiretroviral therapy; ^SD^Standard deviation ::: Changes from *pre PI-ART*to *follow-up* --------------------------------------- From the *pre PI-ART*to the *follow-up*, the CD4 count of the surviving patients increased from a median (Md) of 150 (range 1 -- 660) × 10^6^cells l^-1^to 325 (range 5 -- 840) × 10^6^cells l^-1^(p \< 0.001) and the HIV-1 RNA level decreased from 4.90 (range 2.70 -- 6.00) log~10~copies ml^-1^to 2.70 (range 2.70 -- 6.31) log~10~copies ml^-1^(p \< 0.001). Thirty patients had switched PI during the first two years of PI-ART. A total of 36 subjects were classified as viral responders and 18 as viral non-responders. ### The patients\' emotional well-being deteriorated during the first two years of PI-ART The *pre PI-ART*and *follow-up*results of the composite scores and the single SWED-QUAL scales are shown in Table [3](#T3){ref-type="table"}. The ECS decreased from *pre PI-ART*to *follow-up*(Md 71.2, interquartile range (IQR) 47.1 -- 86.9 versus Md 63.6, IQR 41.0 -- 77.3; p \< 0.01;). A comparison of the single SWED-QUAL scales at *pre PI-ART*with the same scales at *follow-up*revealed statistically significant decreases in the SWED-QUAL scales role limitations due to emotional health (*pre PI-ART*Md 88.9, IQR 66.7 -- 100 versus *follow-up*Md 77.8, IQR 50.0 -- 100; p \< 0.05) and emotional well-being: negative affect (*pre PI-ART*Md 62.5, IQR 37.5 -- 94.8 versus *follow-up*Md 50.0, IQR 25.0 -- 75.0; p \< 0.01; Table [3](#T3){ref-type="table"}). The effect sizes of the change were 0.34 for the PCS, 0.36 for role limitations due to emotional health and 0.37 for emotional well-being: negative affect. ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### The Swedish Health-Related Quality of Life Questionnaire (SWED-QUAL) results before and after 2 years of protease inhibitor based antiretroviral therapy (*pre PI-ART*and *follow-up*respectively). The higher the score the better the health-related quality of life (n = 54) ::: **Score** ------------------------------------------ ------------------- ------------------- -------- Physical HRQL composite score (PCS) 79.3 (64.5--92.5) 77.2 (58.4--92.9) NS Emotional HRQL composite score (ECS) 71.2 (47.1--86.9) 63.6 (41.0--77.3) \<0.01 Physical functioning 95.2 (81.0--95.2) 95.2 (84.5--100) NS Mobility 100 (66.7--100) 100 (66.7--100) NS Satisfaction with physical ability 66.7 (33.3--83.3) 66.7 (33.3--100) NS Pain 100 (53.8--100) 100 (84.7--100) NS Role limitations due to physical health 66.7 (55.6--100) 88.9 (50.0--100) NS Sexual functioning 75.0 (41.7--91.7) 58.3 (41.7--91.7) NS Role limitations due to emotional health 88.9 (66.7--100) 77.8 (50.0--100) \<0.05 Emotional well-being, positive affect 72.9 (47.9--83.3) 62.5 (37.5--83.3) NS Emotional well-being, negative affect 62.5 (37.5--94.8) 50.0 (25.0--75.0) \<0.01 Sleep problems 67.9 (39.3--87.5) 57.1 (36.6--75.0) NS Satisfaction with family life 72.9 (49.5--90.6) 72.9 (55.7--85.4) NS Relation to partner 87.5 (75.0--100) 79.2 (52.1--95.8) NS General health perception 55.6 (36.1--75.0) 63.9 (38.9--80.6) NS ^a^Wilcoxon signed ranks test, change from *pre PI-ART*to *follow-up*; ^NS^Non significant; ^IQR^Interquartile range; ^HRQL^Health-related quality of life ::: ### Subjective experience of adverse effects contributed to deteriorated emotional HRQL at follow-up Forty-seven (87 %) of the 54 patients were classified as adherent and seven (13 %) as non-adherent. The mean VAS score for subjective experience of adverse effects at *follow-up*was 3.39 (SD 2.95, Md 2.57, IQR 0.79 -- 5.84) cm on a 10 cm scale. Two hierarchical multiple regression analyses were performed to investigate whether the change in physical and emotional HRQL composite scores could be predicted by initial CD4 count, viral response, adherence and subjective experience of adverse effects (Table [4](#T4){ref-type="table"}). Only the ECS model showed a statistically significant R^2^change; that is, the subjective experience of adverse effects predicted a decrease in the emotional HRQL. ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Hierarchical multiple linear analyses of the influence of initial CD4 value and the variables viral response, subjective experience of adverse effects and adherence on the change in PCS and ECS from before start (*pre PI-ART*) to after two year of treatment (*follow-up*) ::: ***Follow-up*PCS** ***Follow-up*ECS** ---------------------------- ------------------------- -------------------- ------------------------- -- *Pre PI-ART*PCS 0.722\*\*\* \- *Pre PI-ART*ECS \- 0.681\*\*\* R^2^ 0.605\*\*\* 0.456\*\*\* F 78.26 42.72 d.f. (regression;residual) 1;51 1;51 **Step 2** **Partial correlation** **Partial correlation** Viral response 0.202 0.076 Adherence 0.167 0.235 Adverse effects 0.091 -0.366\* *Pre PI-ART*CD4 value -0.043 -0.024 R^2^-change 0.028 0.128\* F-change 0.91 3.63 d.f. (regression;residual) 4;47 4;47 ^PCS^Physical HRQL composite score; ^ECS^Emotional HRQL composite score; ^PI-ART^Protease inhibitor based antiretroviral therapy; ^*a*^partial correlation of the final model (step 2); \*p \< 0.05; \*\*\*p \< 0.001 ::: Internal consistency of the SWED-QUAL scales -------------------------------------------- Cronbach\'s α reliability estimates for the 11 SWED-QUAL multi-item scales for the *pre PI-ART*and *follow-up*ranged between 0.74 and 0.92 indicating good internal consistency. The internal consistency for the two HRQL composite scores was for the PCS α 0.88 and 0.85 (*pre PI-ART*and *follow-up*, respectively) and for the ECS α 0.91 and 0.89 (*pre PI-ART*and *follow-up*, respectively). Discussion ========== In this study we used the SWED-QUAL instrument to investigate the HRQL of 54 relatively advanced HIV patients before and after two years of first generation PI-ART in a setting where protease inhibitors were introduced. We also studied the patients\' viral outcome, self reported adherence and subjective experience of adverse effects and the relationship between these variables and the development in HRQL. To minimise multiple comparisons, a set of linear regression models were used. To increase the power to detect relations as a result of these models, we created two SWED-QUAL composite scores (PCS and ECS) through a data reduction procedure. The main findings from the present study were that the physical HRQL remained stable while the emotional HRQL deteriorated for two years of first generation PI-ART and that subjective experience of adverse effects was the strongest predictor of the deterioration in emotional HRQL ratings. The present study thus suggests that first generation PI-ART interrupted the progressive negative development of the physical domain of HRQL that has been reported in studies performed before the introduction of PI-ART \[[@B9],[@B10],[@B16],[@B32],[@B33]\]. Our findings are consistent with those of Goujard *et al.*\[[@B34]\] and Burgoyne *et al.*\[[@B18]\] who also failed to detect any changes in HRQL after one and a half and four years, respectively, when monitoring patients in a period after PI-ART had become available. The somewhat surprising decrease in emotional health found in the present study has, to our knowledge, not been reported in other longitudinal studies of PI-ART. Longitudinal studies performed before the introduction of PI-ART have, in general, showed stable or improved mental/emotional health over time \[[@B9],[@B10],[@B16],[@B35]\]. We also found stable scores in the emotional domain of HRQL in a previous pre PI-ART era study, where patients with no or only single drug antiretroviral therapy were followed for two years (own unpublished data). However, a study investigating the HRQL of patients receiving didanosine monotherapy or in combination with delavirdine (not approved in Europe due to side effects) showed slightly declining mental health scores for up to two years after the start of the trial \[[@B33]\]. After the introduction of PI-ART, improvements in depressive symptoms and mental health was reported in a shorter time period, i.e. after up to one year following the addition of a PI to existing antiretroviral therapy \[[@B36],[@B37]\]. Similarly, Rabkin *et al.*\[[@B38]\] found a reduction in psychological distress and clinical depression during a two-year period when PI-ART became widely available. In the latter study, however, PI-ART was introduced to the cohort continuously during the follow-up, resulting in a mean time with PI-ART shorter than two years (i.e. the analysis did not evaluate patients before and after start of PI-ART in a consistent manner). From a longer term perspective, however, emotional HRQL was found to be stable over a four-year period \[[@B18]\]. When we further investigated the relation between the change in HRQL and the variables viral response, adherence, subjective experience of adverse effects and baseline CD4 value, our study showed that it was the subjective experience of adverse effects that contributed most to the deterioration in the emotional HRQL. It should be stressed that our measure of adverse effects was a global single item, where the patients were asked to rate their total perception of adverse effects during the previous month and that we did not sub-analyse this experience further into different symptoms. However, the negative impact of perceived side effects/symptoms on HRQL, is confirmed in a great number of quantitative investigations \[[@B12],[@B18],[@B37]-[@B42]\]. Also, the results from our quantitative study agreed with those of a qualitative study that aimed to elicit the patients\' real-life descriptions of their experience of combination therapy. Erlen & Mellors \[[@B43]\] found that side effects were one of the major problems associated with the therapy. Furthermore, symptoms or adverse events have been shown to be related to medication adherence \[[@B44],[@B45]\]. In order to improve HRQL and adherence, it is therefore crucial to find treatment combinations and strategies that minimise these negative effects and to individualize treatment. The simplified and less toxic treatment regimes available today, with antiretroviral therapy based on three nucleoside analogue reverse transcriptase inhibitors or nucleoside analogues combined with non-nucleoside analogue reverse transcriptase inhibitors or ritonavirboosted PIs once or twice daily \[[@B1]\], may be less liable to have a negative impact on HRQL. This was indicated in a randomised study comparing two nucleoside analogue reverse transcriptase inhibitors combined with either a PI or efavirenz, over a one year period and where the combination with efavirenz had a better influence on HRQL than the combination with the PI \[[@B46]\]. Similarly, Carrieri *et al.*\[[@B37]\] found a positive development of HRQL in patients switching to a non-PI-containing therapy, as compared to patients continuing stable PI-ART. However, these aspects need to be further investigated and there is a need for further study of the effect of different treatment regimes on HRQL, in experienced as well as naïve patients. Certain study limitations should be emphasized when interpreting the present results. Firstly, the small sample may mean that the study failed to detect further relations between the investigated variables. Secondly, a large proportion of the investigated patients were in an advanced state of their HIV infection and many had previous experience of antiretroviral monotherapy. The inferences drawn from this study may not be applicable to ART naïve patients starting antiretroviral therapy in a less advanced state of the disease. Today treatment is started when certain laboratory criteria are fulfilled and before symptoms, including AIDS-defining disease, are at a high risk of appearing. Thirdly, our study population may not be totally representative of all patients receiving PI-ART. The majority of our patients were well-educated males with a homosexual route of transmission. The influence on HRQL may be different for female patients or patients with other educational status or routes of transmission. It should be noted that we chose to consider viral outcome of the treatment in a longitudinal context, i.e. viral response defined by repeated HIV RNA measures over up to 18 months. The impact on the development of HRQL might have been different if only one single measure of viral load at *follow-up*had been taken into account. Conclusion ========== In this sample of mainly advanced patients, the emotional dimension of HRQL had deteriorated for two years after the start of first generation PI-ART, and the subjective perception of adverse effects made a major contribution to this decrease. The results of the present study show the importance of studying HRQL in a situation where there is a desperate need for life-saving new therapies. Therefore, consistently taking HRQL into account when treating HIV patients is of the utmost importance. Finding treatment combinations and strategies with the least negative long-term influence on HRQL is essential at a time when those having access to antiretroviral combination therapy have a dramatically increased life span. Furthermore, considering the current numerous treatment options together with the fluctuations in HRQL over time, further short and long-term investigations of HRQL in patients receiving antiretroviral therapy are of the utmost importance. Authors\' contributions ======================= LEE, ES and GN conceived of the study. All authors made substantial contributions to conception, planning and design. LEE participated in the coordination, carried out the statistical analysis and interpretation of data and drafted the manuscript. GAB participated the coordination and acquisition of the study and helped draft the manuscript. ES participated in the acquisition of the study and helped draft the manuscript. GN participated in the interpretation of data and helped draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We would like to thank the nurses and physicians at the Department of Venhälsan, South Stockholm General Hospital for their help in administering the instruments to the participants, Eva-Lena Fredriksson, Department of Venhälsan, South Stockholm General Hospital for her great assistance in the co-ordination of the study and J. Petter Gustavsson, Department of Nursing, Karolinska Institutet for helpful discussions. We would also like to acknowledge the assistance of the participants, who gave time and effort to respond to the questionnaires. This study was supported by grants from the Board of Research for Health and Caring Sciences at Karolinska Institutet.

PubMed Central

2024-06-05T03:55:58.460198

2005-5-4

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173133/", "journal": "Health Qual Life Outcomes. 2005 May 4; 3:32", "authors": [ { "first": "Lars E", "last": "Eriksson" }, { "first": "Göran A", "last": "Bratt" }, { "first": "Eric", "last": "Sandström" }, { "first": "Gun", "last": "Nordström" } ] }

PMC1173134

Background ========== The question of whether disease cases are clustered in space has received considerable attention in the literature \[[@B1]-[@B4]\]. Although many statistical tests for disease clusters have been proposed, most tests suffer from multiple testing problems due to one or two unknown parameters that must be set prior to their applications. For example, Cuzick and Edwards\'s procedure \[[@B5]\] has an unknown number *k*of nearest-neighbours and Besag and Newell\'s method \[[@B6]\] has an unknown number of cases *k*for the size of the cluster. As far as we know, the spatial scan statistic proposed by Kulldorff \[[@B7],[@B8]\] and Tango\'s maximized excess events test \[[@B9],[@B10]\] are exceptions and take multiple testing into account in the sense that we have only to specify the maximum possible cluster size. Especially, Kulldorff\'s circular spatial scan statistic has been applied to a wide variety of epidemiological studies for cluster detection (for example, see \[[@B11]-[@B13]\]). In recent power comparisons of disease clustering tests, his scan statistic has been shown to be the most powerful for detecting localized clusters \[[@B14],[@B15]\]. It should be noted, however, that the power estimates provided reflect the \"power to reject the null hypothesis for whatever reason\" and that the probability of both rejecting the null hypothesis and detecting the true cluster correctly is a different matter. As the circular spatial scan statistic uses a \"circular window\" with variable size to define the potential cluster area, it is difficult to correctly detect noncircular clusters such as those along a river. Most geographical areas are noncircular. Furthermore, in our experience in applying SaTScan program \[[@B16]\] to various data, even if the null hypothesis is rejected, the circular spatial scan statistic tends to detect a larger cluster than the true cluster by absorbing surrounding regions where there is no elevated risk. It should be noted that although Kulldorff originally made no assumptions about the shape of the scanning window in his paper \[[@B8]\], a circular scanning window has been used in almost all purely spatial applications especially for the availability of software and computational speed. Recently, Patil and Taillie \[[@B17]\] and Duczmal and Assunção \[[@B18]\] proposed non-circular spatial scan statistics based on the likelihood ratio test formulated in the same way as in the circular spatial scan statistic. To avoid undertaking computationally infeasible searches, they considered different approaches. Patil and Taillie \[[@B17]\] used the notion of \"upper level set\" to reduce the size of windows to be scanned and proposed \"upper level set scan statistic\". However, they do not discuss how to select the level *g*which defines the upper level set and do not provide any illustrations of their method nor any results of comparison with the circular scan statistic. Duczmal and Assunção \[[@B18]\], on the other hand, have applied a simulated annealing method in which they try to examine only the most promising windows using a graph-based algorithm to obtain the local maxima of a certain likelihood function over a subset of the collection of all the connected regions. Their method seems to be very complicated but they do not show any programmable procedure of their method. In our experience using their program (personal communication to Professor Duczmal via email) which is executable with the Borland C++ Builder 6, their scan statistic, in most cases, detected a cluster of peculiar shape that was much larger than the true cluster by absorbing not only surrounding regions with non-elevated risk but also faraway regions with non-elevated risk. An example of such properties of Duczmal and Assunção\'s procedure is shown later in comparison with the circular spatial scan statistic and the proposed flexible spatial scan statistic. That is why we did not include both the Patil and Taille method and Duczmal and Assunção\'s procedure in our simulation for comparison. In this paper, we propose an alternative *flexibly shaped spatial scan statistic*(\'flexible spatial scan statistic\' hereafter) in which the detected cluster is allowed to be flexible in shape while at the same time the cluster is confined within relatively small neighborhoods of each region. The performance of the flexible spatial scan statistic is compared with that of the circular spatial scan statistic using Monte Carlo simulation. In comparing performance we examined not only the usual power but also the newly introduced bivariate power distribution classified by the number of regions detected as the most likely cluster and the number of hot-spot regions included in the most likely cluster. The proposed flexible spatial scan statistic is illustrated with some simulated disease maps for the Tokyo Metropolitan area. Methods ======= Consider the situation where an entire study area is divided into *m*regions (for example, county, enumeration districts, etcetera). The number of cases in the region *i*is denoted by the random variable *N*~*i*~with observed value *n*~*i*~, *i*= 1, \..., *m*. Under the null hypothesis *H*~0~of no clustering, the *N*~*i*~are independent Poisson variables such that *H*~0~: *E*(*N*~*i*~) = *ξ*~*i*~, *N*~*i*~\~ Pois(*ξ*~*i*~), *i*= 1, \..., *m* (1) where Pois(*e*) denotes Poisson distribution with mean *e*and the *ξ*~*i*~are the null expected number of cases in the region *i*. To specify the geographical position of each region, we will use the coordinates of the administrative population centroid. Under this situation, the circular spatial scan statistic imposes a circular window **Z**on each centroid. For any of those centroids, the radius of the circle varies from zero to a pre-set maximum distance *d*or a pre-set maximum number of regions *K*to be included in the cluster. If the window contains the centroid of a region, then that whole region is included in the window. In total, a very large number of different but overlapping circular windows are created, each with a different location and size, and each being a potential cluster. Let **Z**~*ik*~, *k*= 1,\..., *K*, denote the window composed by the (*k*- 1)-nearest neighbours to region *i*. Then, all the windows to be scanned by the circular spatial scan statistic are included in the set *Z*~1~= {**Z**~*ik*~\| 1 ≤ *i*≤ *m*, 1 ≤ *k*≤ *K*} (2) A flexible scan statistic we propose, on the other hand, imposes an *irregularly shaped*window **Z**on each region by connecting its adjacent regions. For any given region *i*, we create the set of irregularly shaped windows with *length k*consisting of *k*connected regions including *i*and let *k*moves from 1 to the pre-set maximum *K*. To avoid detecting a cluster of *unlikely peculiar shape*, the connected regions are restricted as the subsets of the set of regions *i*and (*K*- 1)-nearest neighbours to the region *i*where *K*is a pre-specified maximum length of cluster. In total, as in the circular spatial scan statistic, a very large number of different but overlapping arbitrarily shaped windows are created. Let **Z**~*ik*(*j*)~, *j*= 1,\..., *j*~*ik*~denote the *j*-th window which is a set of *k*regions connected starting from the region *i*, where *j*~*ik*~is the number of *j*satisfying **Z**~*ik*(*j*)~⊆ **Z**~*ik*~for *k*= 1,\..., *K*. Then, all the windows to be scanned are included in the set *Z*~2~= {**Z**~*ik*(*j*)~\| 1 ≤ *i*≤ *m*, 1 ≤ *k*≤ *K*, 1 ≤ *j*≤ *j*~*ik*~} (3) In other words, for any given region *i*, the circular spatial scan statistic consider *K*concentric circles, whereas the flexible scan statistic consider *K*concentric circles plus all the sets of connected regions (including the single region *i*) whose centroids are located within the *K*-th largest concentric circle. So, the size of *Z*~2~is far larger than that of *Z*~1~which is at most *mK*. Details of the algorithm that we adopted to find all these arbitrarily shaped windows within a pre-specified maximum length *K*are given in the Appendix. Under the alternative hypothesis, there is at least one window **Z**for which the underlying risk is higher inside the window when compared with outside. In other words, we are considering the following hypothesis: *H*~0~: *E*(*N*(**Z**)) = *ξ*(**Z**), for all **Z**, *H*~1~: *E*(*N*(**Z**)) \> *ξ*(**Z**), for some **Z** (4) where *N*() and *ξ*() denote the random number of cases and the null expected number of cases within the specified window, respectively. For each window, it is possible to compute the likelihood to observe the observed number of cases within and outside the window, respectively. Under the Poisson assumption, the test statistic, which was constructed with the likelihood ratio test \[[@B8]\], is given by ![](1476-072X-4-11-i1.gif) where **Z**^*c*^indicates all the regions outside the window **Z**, and *n*() denotes the observed number of cases within the specified window and *I*() is the indicator function. The window **Z**\* that attains the maximum likelihood is defined as the *most likely cluster*(MLC). To find the distribution of the test statistic under the null hypothesis, Monte Carlo hypothesis testing \[[@B19]\] is required. In this paper, *p*-value of the test is based upon the null distribution of likelihood ratio test statistic with a large number (we used 999) of Monte Carlo replications of the data set generated under the null hypothesis. It should be noted that, in the same manner as the circular spatial scan statistic, the flexible spatial scan statistic is also able to locate secondary clusters that do not overlap the most likely cluster but are still statistically significant. Results ======= Illustrations and powers ------------------------ In this section, we will compare the flexible spatial scan statistic with the circular spatial scan statistic. As an entire study population, we will use *m*= 113 regions comprising the wards, cities and villages in the area of Tokyo Metropolis and Kanagawa prefecture in Japan (Figure [1](#F1){ref-type="fig"}). The variability of regional populations for *m*= 113 regions is: 25 percentile = 56, 704, median = 142, 320 and 75 percentile = 200, 936. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### **An entire study population for simulation studies**. The 113 regions comprising wards, cities and villages in the area of Tokyo Metropolis and Kanagawa prefecture in Japan. The region number used in the text is shown. Especially, The region numbers of four hot-spot clusters **A-D**are **A**= {14, 15, 20}, **B**= {14, 15, 20, 26}, **C**= {14, 15, 26, 27}, and **D**= {73, 74, 75, 76, 78}, respectively. ::: ![](1476-072X-4-11-1) ::: ### Hot-spot clusters We will consider the following four hot-spot clusters where the expected total number of cases ![](1476-072X-4-11-i2.gif) is set to be 200 under the null hypothesis. 1\. Cluster **A**= {14, 15, 20} 2\. Cluster **B**= {14, 15, 20, 26} 3\. Cluster **C**= {14, 15, 26, 27} 4\. Cluster **D**= {73, 74, 75, 76, 78} where the region included in a hot-spot cluster is called a \"hot-spot region\" (hot-spot region numbers are shown in Figure [1](#F1){ref-type="fig"}). The relative risk within any cluster **R**is set to three, i.e., *H*~1~: *N*(**R**) \~ Pois(*θξ*(**R**)), *θ*= 3.0 (6) The cluster **A**is considered here as an example of a circular cluster that can be in the set of the circular windows and is expected to be identified by the circular spatial scan statistic more often than by the flexible spatial scan statistic. The other clusters are examples of noncircular clusters that are not in the set of the circular windows and thus cannot be identified correctly by the circular spatial scan statistics. For example, consider the region *i*~0~= 15 as the starting region and the set of (*K*- 1)-nearest neighbours to the region 15, which is listed as follows in the ascending order of distance from the region 15: 15, 14, 20, 12, 4, 26, 13, 27, 16, 40, 19, 42, 10,\..., In this case, circular windows are {15}, {15, 14}, {15, 14, 20}, {15, 14, 20, 12}, \... When the starting region is 14 or 20, the corresponding set of (*K*- 1)-nearest neighbours is 14, 15, 20, 4, 16, 13, 19, 12, 5, 1, 17, 10, 26, 3, 27,\..., and 20, 14, 15, 19, 16, 4, 17, 26, 40, 13, 5, 12, 1, 27,\..., respectively. In both cases, cluster **B**and **C**are easily found to be not in the set of circular windows. The cluster **D**is considered as an example of a long and narrow cluster as is shown in Figure [1](#F1){ref-type="fig"}. ### Illustrative example As an illustration, we will apply the circular spatial scan statistic, the flexible spatial scan statistic and Duczmal and Assunção\'s spatial scan statistic to the disease map shown in Figure [2](#F2){ref-type="fig"} which is a random sample of *n*= 235 cases assuming the cluster model **C**. Circles are drawn only for the regions whose observed-expected ratio (standardized risk ratio) is statistically significantly larger than 1 at *α*= 0.05. The radius of the circles is set inversely proportional to the upper tail *p*-value. The number shown in Figure [2](#F2){ref-type="fig"} indicates the region number. Figure [2](#F2){ref-type="fig"} obviously suggests the clusters occurring in the area including regions {14, 15, 26, 27, 33}. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **A random sample from cluster model C**. Dots describe the centroids of regions with some cases. Circles are drawn only for the regions whose standardized risk ratios are statistically significantly larger than 1 at *α*= 0.05 and the region number is placed in stead of dot. The radius is set inversely proportional to the tail probability. ::: ![](1476-072X-4-11-2) ::: Before applying the three spatial scan statistics, we have to specify a common maximum length *K*for the most likely cluster. This makes comparisons to a certain extent fair. In this example, we chose two kinds of maximum length *K*= 15 and *K*= 20 since it is not unreasonable to assume that an actual cluster size will be less than one third or one fourth of the size of the whole study area. Irrespective of the value of *K*, the circular spatial scan statistic detected the regions {14, 15} as MLC with log likelihood ratio = 20.1, *p*= 1/(999 + 1) = 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 3.47. This is shown in Figure [3(a)](#F3){ref-type="fig"}. The flexible spatial scan statistic, regardless of the value *K*, detected the regions {14, 15, 26, 27, 33} as MLC with log likelihood ratio = 29.7, *p*= 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 3.41. This is shown in Figure [3(b)](#F3){ref-type="fig"}. Duczmal and Assunção\'s method, on the other hand, detected a cluster of peculiar shape that is much larger than the true cluster. In the case of *K*= 15, their scan statistic detected an area consisting of *K*= 15 connected regions {14, 15, 24, 26, 27, 31, 32, 33, 48, 54, 69, 77, 78, 90, 110 } as MLC with log likelihood ratio = 31.8, *p*= 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 2.40. This is shown in Figure [4(a)](#F4){ref-type="fig"}. Figure [4(b)](#F4){ref-type="fig"} shows the most likely cluster {14, 15, 26, 27, 31, 32, 33, 48, 60, 61, 62, 67, 69, 77, 78, 80, 89, 90, 108, 110 } detected by Duczmal and Assunção\'s scan statistic for *K*= 20 where the length of MLC is also the same as *K*= 20 and log likelihood ratio = 36.0, *p*= 0.001 and the estimated relative risk is ![](1476-072X-4-11-i3.gif) = 2.26. In the case of *K*= 15, the results of the three scan statistics are summarized in Table [1](#T1){ref-type="table"}. Although the most likely cluster detected by Duczmal and Assunção\'s scan statistic has the largest log likelihood ratio among three scan statistics, it has detected MLC surprisingly larger than the true cluster. ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **The most likely cluster detected by the circular and the flexible spatial scan statistic**. (a) Detected by the circular spatial scan statistic for both *K*= 15 and *K*= 20 and (b) by the flexible spatial scan statistic for both *K*= 15 and *K*= 20, when applied to a random sample from the cluster model **C**= {14, 15, 26, 27}. ::: ![](1476-072X-4-11-3) ::: ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **The most likely cluster detected by the Duczmal and Assunção\'s scan statistic**. (a) Detected for *K*= 15 and (b) for *K*= 20, when applied to a random sample from the cluster model **C**= {14, 15, 26, 27}. ::: ![](1476-072X-4-11-4) ::: ::: {#T1 .table-wrap} Table 1 ::: {.caption} ###### Regions detected as the most likely cluster by three procedures. Regions detected as the most likely cluster by the circular scan, the flexible scan and Duczmal and Assunção\'s scan, with the maximum length of cluster set to be *K*= 15 for the simulated random sample from the cluster model C where the hot spot cluster is assumed to be the set of connected four regions {14, 15, 26, 27} with the assumed relative risk *θ*= 3.0. For details, see text. ::: region no. population observed no. cases expected no. cases relative risk estimated (true) Log likelihood ratio (LLR) and estimated relative risk for the most likely cluster ------------ ------------ -------------------- -------------------- -------------------------------- ------------------------------------------------------------------------------------ ------------ ------------ 14 319,687 14 3.794 3.69 (3.0) \* \* \* 15 529,485 21 6.283 3.34 (3.0) \* \* \* LLR = 20.1 = 3.47 26 139,077 6 1.650 3.64 (3.0) \* \* 27 165,564 6 1.964 3.05 (3.0) \* \* 33 105,899 4 1.257 3.18 (1.0) \* \* LLR = 29.7 = 3.41 24 466,347 8 5.534 1.44 (1.0) \* 31 197,677 3 2.346 1.27 (1.0) \* 32 349,050 5 4.142 1.20 (1.0) \* 48 58,635 1 0.696 1.43 (1.0) \* 54 3,808 1 0.045 22.12(1.0) \* 69 119,575 3 1.419 2.11 (1.0) \* 77 177,742 5 2.109 2.37 (1.0) \* 78 125,127 2 1.485 1.34 (1.0) \* 90 194,866 5 2.312 2.16 (1.0) \* 110 21,535 1 0.256 3.91 (1.0) \* LLR = 31.8 = 2.41 ::: Using a PC(Windows XP, CPU pentium 4, 3.2 GHz), the execution time of the flexible spatial scan statistic in this example is 14 seconds for *K*= 15 and 379 seconds for *K*= 20 which is certainly greater than that for the circular spatial scan statistic (less than 1 second for both *K*= 15 and *K*= 20). ### Power comparison In the power comparison, we chose *K*= 15. To compare the power of the flexible spatial scan statistic with that of the circular spatial scan statistic based upon Monte Carlo simulation, we will introduce a new bivariate power distribution *P*(*l*, s) classified by the *length l*of the significant MLC and the number *s*of hot-spot regions included in the most likely cluster: ![](1476-072X-4-11-i4.gif) where *l*≥ 1 and *s*≥ 0. Based on *P*(*l*, *s*), we examined the following powers, 1\. the usual power, i.e., *P*(*+,+*) = ∑~*l*≥1~∑~*s*≥0~*P*(*l*, *s*), 2\. the joint power *P*(*l*, *s*), especially *P*(*s*\*, *s*\*) where *s*\* is the length of the hot-spot cluster assumed in the simulation. 3\. the marginal power distribution of *s*(≥ 0), *P*(+, *s*) = ∑~*l*≥1~*P*(*l*, *s*) and its conditional power *P*(+, *s*)/*P*(+,+), 4\. the marginal power distribution of *l*(≥ 1), *P*(*l*, +) = ∑~*s*≥0~*P*(*l*, *s*). The powers are calculated for tests of nominal *α*levels of 0.05 and for the expected total number of cases 200 under the null hypothesis, which are based on Monte Carlo simulation using Poisson random numbers. For each simulation, 1,000 trials were carried out. The resultant power distribution *P*(*l*, *s*) × 1000 is shown in Tables [2](#T2){ref-type="table"}, [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"} for each of the four cluster models, respectively, in the form of cross table classified by *l*(*\"length\"*in tables) and *s*(*\"include\"*in tables). ::: {#T2 .table-wrap} Table 2 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model A. Comparison of bivariate power distribution *P*(*l*, *s*) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster **A**= {14, 15, 20}. Nominal *α*-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ------- --- --- --- ----- ----- Flexible (*K*= 15) Circular (*K*= 15) Length *l* Include *s*\ Total Length *l* Include *s*\ Total hot-spot regions hot-spot regions 0 1 2 3 0 1 2 3 1 0 0 0 1 0 0 0 2 0 0 0 0 2 1 0 0 1 3 0 0 0 142 142 3 0 0 0 738 738 4 0 0 0 116 116 4 0 0 0 134 134 5 0 0 0 137 137 5 0 0 0 39 39 6 0 0 0 149 149 6 0 0 0 12 12 7 0 0 0 165 165 7 0 0 0 9 9 8 0 0 0 131 131 8 0 0 0 1 1 9 0 0 0 84 84 9 0 0 2 3 5 10 0 0 0 27 27 10 0 0 0 2 2 11 0 0 0 11 11 11 0 0 0 4 4 12 0 0 0 2 2 12 0 0 0 12 12 13 0 0 0 0 0 13 0 0 0 14 14 14 0 0 0 0 0 14 0 0 0 3 3 15 0 0 0 0 0 15 0 0 0 6 6 Total 0 0 0 964 964 Total 1 0 2 977 980 usual power = 0.964 usual power = 0.980 -------------------- --------------------- ------- ------------ --------------------- ------- ------- --- --- --- ----- ----- ::: ::: {#T3 .table-wrap} Table 3 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model B. Comparison of bivariate power distribution *P*(*l*, *s*) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster **B**= {14, 15, 20, 26}. Nominal *α*-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- --- ----- ----- ----- Flexible (*K*= 15) Circular (*K*= 15) Length *l* Include *s*\ Total Length *l* Include *s*\ Total hot-spot regions hot-spot regions 0 1 2 3 4 0 1 2 3 4 1 0 0 0 1 0 0 0 2 0 0 0 0 2 0 0 0 0 3 0 0 0 0 0 3 0 0 0 523 523 4 0 0 0 0 127 127 4 0 0 0 65 0 65 5 1 0 0 0 157 158 5 0 0 0 23 0 23 6 0 0 0 0 205 205 6 0 0 0 7 66 73 7 0 0 0 2 198 200 7 0 0 0 0 15 15 8 0 0 0 1 151 152 8 0 0 0 0 32 32 9 0 0 0 5 85 90 9 0 0 0 1 15 16 10 0 0 0 1 24 25 10 0 0 0 0 7 7 11 0 0 0 0 17 17 11 0 0 0 2 3 5 12 0 0 0 0 5 5 12 0 0 0 2 63 65 13 0 0 0 0 0 0 13 0 0 0 0 96 96 14 0 0 0 0 0 0 14 0 0 0 0 30 30 15 0 0 0 0 0 0 15 0 0 0 0 22 22 Total 1 0 0 9 969 979 Total 0 0 0 623 349 972 usual power = 0.979 usual power = 0.972 -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- --- ----- ----- ----- ::: ::: {#T4 .table-wrap} Table 4 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model C. Comparison of bivariate power distribution *P*(*l*, *s*) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster **C**= {14, 15, 26, 27}. Nominal *α*-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- ----- ----- ----- ----- Flexible (*K*= 15) Circular (*K*= 15) Length *l* Include *s*\ Total Length *l* Include *s*\ Total hot-spot regions hot-spot regions 0 1 2 3 4 0 1 2 3 4 1 0 0 0 1 1 0 1 2 0 0 0 0 2 0 0 351 351 3 0 0 0 0 0 3 2 0 4 0 6 4 0 0 0 0 138 138 4 0 0 3 0 0 3 5 0 0 0 3 147 150 5 2 0 2 0 0 4 6 1 0 0 2 200 203 6 1 0 0 0 0 1 7 0 1 0 4 147 152 7 0 0 0 81 0 81 8 0 0 2 9 107 118 8 0 0 10 18 38 66 9 0 0 0 10 71 81 9 0 0 2 0 26 28 10 1 0 2 5 28 36 10 0 0 0 29 3 32 11 0 0 0 0 10 10 11 0 0 1 13 1 15 12 0 0 0 0 2 2 12 0 0 2 4 60 66 13 0 0 0 0 0 0 13 0 0 0 5 62 67 14 0 0 0 0 0 0 14 0 0 0 10 27 37 15 0 0 0 0 0 0 15 0 0 0 6 37 43 Total 2 1 4 33 850 890 Total 6 0 375 166 254 801 usual power = 0.890 usual power = 0.801 -------------------- --------------------- ------- ------------ --------------------- ------- ----- ------- --- --- ----- ----- ----- ----- ::: ::: {#T5 .table-wrap} Table 5 ::: {.caption} ###### Comparison of the circular and the flexible spatial scan statistic for the cluster model D. Comparison of bivariate power distribution *P*(*l*, *s*) × 1000 between the circular spatial scan statistic and the flexible spatial scan statistic for the hot-spot cluster **D**= {73, 74, 75, 76, 78}. Nominal *α*-level is set as 0.05 and 1000 trials are carried out. For more details, see text. ::: -------------------- --------------------- ------- ------------ --------------------- ------- ----- ----- ------- ---- --- ---- ---- ----- ---- ----- Flexible (*K*= 15) Circular (*K*= 15) Length *l* Include *s*\ Total Length *l* Include *s*\ Total hot-spot regions hot-spot regions 0 1 2 3 4 5 0 1 2 3 4 5 1 0 0 0 1 6 0 6 2 1 0 0 1 2 3 5 0 8 3 0 0 0 0 0 3 0 0 0 14 14 4 1 0 0 1 0 2 4 1 0 4 5 0 10 5 0 1 0 3 1 242 247 5 0 0 2 1 0 0 3 6 1 0 0 1 2 162 166 6 1 0 0 1 363 0 365 7 2 3 0 5 5 93 108 7 0 0 1 0 56 0 57 8 1 2 1 6 7 53 70 8 0 0 2 2 28 0 32 9 0 2 0 1 5 38 46 9 0 0 2 2 10 0 14 10 0 2 0 1 1 18 22 10 1 0 0 3 3 0 7 11 0 0 0 2 2 5 9 11 0 0 0 0 3 11 14 12 0 0 1 0 0 1 2 12 0 0 0 2 3 8 13 13 0 0 0 0 0 0 0 13 0 0 0 1 1 16 18 14 0 0 0 0 0 0 0 14 0 0 1 0 0 5 6 15 0 0 0 0 0 0 0 15 0 1 0 0 1 7 9 Total 6 10 2 20 23 612 673 Total 12 6 12 31 468 47 576 usual power = 0.673 usual power = 0.576 -------------------- --------------------- ------- ------------ --------------------- ------- ----- ----- ------- ---- --- ---- ---- ----- ---- ----- ::: #### 1) Usual power Both tests have the same size 0.043 (distribution of length of significant MLC is omitted) and are shown to have high powers for the hot-spot clusters considered here. The flexible spatial scan statistic generally has higher power except for the model **A**(circular cluster) where, however, the difference is small. #### 2) Joint powers at (*s*\*, *s*\*) and at its neighbours Table [2](#T2){ref-type="table"} shows the good characteristics of the circular spatial scan statistic. Namely, the circle-based scan statistic could detect circular hot-spot cluster **A**with length *s*\* = 3 considerably more accurately with power 738/1000 compared to 142/1000 of the flexible spatial scan statistic. Tables [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}, on the other hand, show that the power of the circular spatial scan statistic in detecting exactly noncircular hot-spot clusters is 0/1000 due to the circular window. However, the circular spatial scan statistic is seen to be able to include some of the hot-spot regions into MLC reasonably well. For example, when applied to the noncircular cluster **B**with length *s*\* = 4, three or four regions including three hot-spot regions can be detected as the most likely cluster with relatively high power (523 + 65)/1000 = 0.588 (Table [3](#T3){ref-type="table"}). When applied to the model **D**with length *s*\* = 5, the similar high power 363/1000 can be observed at (*l*, *s*) = (6, 4) (Table [5](#T5){ref-type="table"}). The flexible spatial scan statistic, on the other hand, has no such high power at a single point (*l*, *s*) near (*s*\*, *s*\*). However, the characteristic of the flexible spatial scan statistic is that the support of the power distribution is distributed in a relatively narrow range of / on the line *s*= *s*\*,i.e, we have *s*\* ≤ *l*≤ 12 in the four cluster models considered here. #### 3) Marginal power *P*(+, *s*) and its conditional marginal power *P*(+, *s*)/*P*(+, +) Regarding the marginal power *P*(+, *s*\*) at *s*= *s*\*, the flexible spatial scan statistic is shown to have much higher power than the circular spatial scan statistic for the case of noncircular clusters (Tables [3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}, [5](#T5){ref-type="table"}). Furthermore, the conditional marginal power *P*(+, *s*)/*P*(+, +) of the flexible spatial scan statistic is 964/964 = 1.000, 969/979 = 0.990, 850/890 = 0.955 and 612/673 = 0.909 for the cluster **A-D**, respectively. These results indicate that the identified MLC by the flexible spatial scan statistic includes the hot-spot cluster with quite high probability. For the noncircular clusters, the mode of *P*(+, *s*) of the circular spatial scan statistic is around *s*= *s*\* - 1 or *s*= *s*\* - 2. #### 4) Marginal power distribution *P*(*l*, +) For the flexible spatial scan statistic, the probability that the length of significant MLC is less than *s*= *s*\* is shown to be zero or quite small and the maximum length is around 10 to 12. the circular spatial scan statistic, on the other hand, tends to detect a much longer cluster than expected from the hot-spot cluster assumed in the simulation. For example, the probability that the length of MLC for the cluster B with length *s*\* = 4 is greater than or equal to 12 is 213/1000 = 0.213 compared with 5/1000 = 0.005 for the flexible spatial scan statistic. The probability that the length of MLC for the cluster **C**with length *s*\* = 4 is greater than or equal to 12 is 213/1000 = 0.213 compared with 2/1000 = 0.002 for the flexible spatial scan statistic. This tendency is shown even in the circular cluster **A**where the same probabilities are 0.035 vs. 0.002. ### Cost comparison Based upon the bivariate power function *P*(*l*, *s*), we can compute the following expected total cost incurred by incomplete identification of the true cluster: *C*= *C*~2~{*rE*(*s*\* - *S*) + *E*(*L*- *S*)}, *r*= *C*~1~/*C*~2~ (8) where *C*~1~and *C*~2~denote the average cost of missing one region in the true cluster and that of incorrectly detecting one region not in the true cluster, respectively. *L*and *S*denote the random variable of *l*and *s*, respectively. Two expected numbers *E*(*s*\* - *S*) and *E*(*L*- *S*) for four kinds of clusters **A-D**are shown in Table [6](#T6){ref-type="table"}. In general, we can assume *r*\> 1. For example, the ratio *C*/*C*~2~is shown for the case of *r*= 1 and *r*= 2, respectively, in Table [6](#T6){ref-type="table"}. However, in this example, irrespective of the value of *r*(\> 1), the circular spatial scan statistic is shown to have lower cost for detecting circular cluster **A**but to have higher cost for detecting non-circular clusters **B-D**. ::: {#T6 .table-wrap} Table 6 ::: {.caption} ###### Cost comparison Expected number of undetected regions included in the true cluster *E*(*s*\* - *S*), expected number of detected regions not in the true cluster *E*(*L*- *S*) and the ratio of costs *C*/*C*~2~(*r*= 1, 2) incurred by incomplete identification of the true cluster. The spatial scan statistic with low values is better. ::: Hot-spot Cluster Scan statistic *E*(*s*\* - *S*) *E*(*L*- *S*) the ratio *C*/*C*~2~ ----------------------------- ------------------- ------------------ --------------- ---------------------- ------- **A**= {14, 15, 20} Flexible (K = 15) 0.108 2.951 3.059 3.167 Circular (K = 15) 0.065 0.722 0.787 0.852 **B**= {14, 15, 20, 26} Flexible (K = 15) 0.097 2.548 2.645 2.742 Circular (K = 15) 0.735 2.525 3.260 3.995 **C**= {14, 15, 26, 27} Flexible (K = 15) 0.492 2.243 2.735 3.227 Circular (K = 15) 1.736 3.153 4.889 6.625 **D**= {73, 74, 75, 76, 78} Flexible (K = 15) 1.774 1.088 2.862 4.636 Circular (K = 15) 2.770 1.709 4.479 7.249 ::: Limitations of current work --------------------------- Needless to say, the results derived here are based upon a small Monte Carlo simulations study and thus the characteristic observed in the current work could change a little bit depending on the cluster model adopted. We assumed here only one hot spot cluster and did not consider the case of several hot spot clusters. Therefore, we need a further simulation study to compare the performance of the two spatial scan statistics under several different clusters. Regarding the algorithm adopted for the flexible spatial scan statistic, we set the restriction that irregularly shaped windows **Z**with length *k*(≤ *K*) are constructed from members of the (*K*- 1)-nearest neighbours to the starting region. It seems that this restriction plays an important role in preventing the flexible spatial scan statistic from reaching out for and absorbing faraway regions with non-elevated risk. However, to avoid undertaking computationally infeasible searches, the flexible spatial scan statistic has to be set with an upperbound for *K*. This depends on the disease map under study and the capability of the computer. The current practical upperbound is around *K*= 30 for the reason that the execution time of our current algorithm will take more than a week if *K \>*30 for the number of regions *m*= 200 \~ 300. However, it seems to be unlikely that the length of the true cluster would be larger than 10 \~ 15 percent of the total number of regions. So, we think that our current algorithm can be applied to many epidemiological studies with small to moderate cluster sizes. However, for larger cluster sizes, a more sophisticated algorithm to increase the upperbound for *K*is needed. Regarding data type, the proposed spatial scan statistic can only be applied to regional count data whereas the circular spatial scan statistic can be applied to not only count data but also individual point data. However, at least in disease surveillance, most of the data that people analyze is aggregated, so the method covers most real-world situations. Finally, one of the reviewers commented that using small areas as basis for clustering without any attempt to incorporate heterogeneity in background rates is a fundamental flaw of all existing scanning methods. In general, we know that disease risks over study regions are heterogeneous to a certain extent and the null hypothesis of complete spatial randomness is not true. However, statistical hypothesis testing is based upon the null hypothesis which is not true. Likewise, we will use complete spatial randomness as the null hypothesis as indicated in equation (1) since we are interested in rejecting the null hypothesis and detecting the local clusters with excess risk. If we are interested in estimating a clustering mechanism, we should use some modeling approach rather than spatial scan statistics. Discussion ========== In this paper, we proposed a flexibly shaped spatial scan statistic to detect arbitrarily shaped clusters by amalgamating administrative units. The flexible spatial scan statistic is, via Monte Carlo simulation, shown to have reasonably high powers compared with the circular spatial scan statistic when examined by a newly introduced bivariate power distribution *P*(*l*, *s*). The simulation reveals that the circular spatial scan statistics shows a high level of accuracy in detecting circular clusters exactly and reasonably good power for including some hot-spot regions into the most likely cluster. The flexible spatial scan statistic exhibits no such high power regarding exact identification of clusters but the support of the power distribution is shown to be concentrated in a relatively narrow range of length *l*on the line *s*= *s*\*, indicating that an observed significant most likely cluster contains the true cluster with quite high probability. The circular spatial scan statistic, on the other hand, is shown to have zero powers for detecting exactly noncircular clusters that cannot be captured by any circular window. The circular spatial scan statistic is also shown to have a tendency to detect a larger cluster than the true cluster assumed in the simulation even for the case when the true cluster is circular. Furthermore, by introducing the two kinds of cost due to incomplete detection of the true cluster, we could summarize these characteristics in terms of minimizing expected total cost. One of the reviewers suggested a similar cost comparison using the number of people that are incorrectly classified rather than the number of regions since the cost of misclassifying a large region is at least for disease surveillance purposes higher than that of misclassifying a region with smaller population. We think that would be an interesting additional simulation study worth conducting. However, since it can be expected that the result of such a cost comparison strongly depends on the spatial configuration of regions with different population size in the neighborhood of and within the true cluster and thus it requires careful design for creating suitable cluster models from which we can intuitively infer the result to a certain extent, we would like to leave such a simulation study in our future work. The surprising result that Duczmal and Assunção\'s scan statistic detected quite large and unlikely peculiar shaped clusters that had the largest likelihood ratio among the three scan statistics might cast a doubt on the validity of the model selection based upon maximizing the likelihood ratio (5). Such a doubt can also be seen in some simulation results of the circular spatial scan statistic that had non-negligible probabilities of detecting much longer clusters than the true cluster. The flexible spatial scan statistic, on the other hand, is shown not to detect such an unexpected long cluster probably because it has the restriction that our windows are constructed only from members of the (*K*- 1)-nearest neighbours to the starting region. Nevertheless, these undesirable properties produced by maximum likelihood ratio might suggest the use of a different criterion for model selection. For example, we might consider a penalized likelihood where we consider a penalty for the *complexity of the cluster shape*, which is also worth future research. All the computations and simulations have been conducted on a PC with Windows XP. For users who are interested in applying the flexible spatial scan statistic, we can provide the software FleXScan \[[@B20]\]. Conclusion ========== The circular spatial scan statistics shows a high level of accuracy in detecting circular clusters exactly and reasonably good power for including some hot-spot regions into the most likely cluster. The proposed flexible spatial scan statistic is shown to have good usual powers plus the ability to detect the noncircular hot-spot clusters more accurately than the circular spatial scan statistic. However, the proposed spatial scan statistic work well for small to moderate cluster size, say up to 30. For larger cluster sizes, the method is not practically feasible and a more efficient algorithm is needed. Appendix: algorithm to find the set *Z*~2~defined in equation (3) ================================================================= There are probably several procedures to find the set *Z*~2~that is defined as the set of arbitrarily shaped windows **Z**within a pre-specified maximum length *K*. The algorithm that we used is described as follows: Step 1. First we set an *m*× *m*matrix *A*= (*a*~*ij*~) such as ![](1476-072X-4-11-i5.gif) and set *Z*~2~= *φ*and *i*~0~= 0 Step 2. Let *i*~0~← *i*~0~+ 1 and *i*~0~(= 1, 2,\..., *m*) be the starting region. Then we create the set ![](1476-072X-4-11-i6.gif) consisting of (*K*- 1)-nearest neighbours to the starting region *i*~0~and *i*~0~itself, i.e., ![](1476-072X-4-11-i7.gif) = {*i*~0~, *i*~1~, *i*~2~,\..., *i*~*K*-\ 1~}, where *i*~*k*~is the *k*-th nearest to *i*~0~. Step 3. We consider all the set **Z**⊂ ![](1476-072X-4-11-i7.gif), which includes the starting region *i*~0~. For any given such set **Z**, repeat the following steps 4--7. Step 4. We divide the set **Z**into two disjoint sets: **Z**~0~= {*i*~0~} and **Z**~1~which contains the other regions of **Z**. Step 5. We make two new sets ![](1476-072X-4-11-i8.gif) and ![](1476-072X-4-11-i9.gif). ![](1476-072X-4-11-i8.gif) consists of the regions of **Z**~1~that are connected to some regions of **Z**~0~. On the other hand, ![](1476-072X-4-11-i9.gif) consists of the regions of **Z**~1~that are not connected to any regions of **Z**~0~. Then we replace **Z**~0~and **Z**~1~by ![](1476-072X-4-11-i8.gif) and ![](1476-072X-4-11-i9.gif), respectively. Step 6. We repeat the step 5 recursively until either **Z**~0~or **Z**~1~becomes null first. Step 7. We make a decision as follows. **Z**is said to be \"connected\" when **Z**~1~becomes null first and \"disconnected\" when **Z**~0~becomes null first. If we can find **Z**\"connected\", **Z**is added to the set *Z*~2~. If we find **Z**\"disconnected\", **Z**is discarded. Step 8. Repeat the steps 2--7 until we finally get the set *Z*~2~consisting of arbitrarily shaped windows **Z**whose maximum length is *K*. Now we shall give an example using regions in the Tokyo Metropolitan area shown in Figure [1](#F1){ref-type="fig"}. Let the starting region *i*~0~= 14. Then, the regions in the set of (*K*- 1)-nearest neighbours to the region 14 are listed as follows in the ascending order of distance to the region 14, i.e., **W**~14~= {14, 15, 20, 4, 16, 13, 19, 12, 5,\...}. Suppose that we take a subset **Z**= {14, 15, 20, 26}. In the first step, we have **Z**~0~= {14}, **Z**~1~= {15, 20, 26}. Since *a*~14,15~= *a*~14,20~= 1 and *a*~14,26~= 0, we then have **Z**~0~= {15, 20}, **Z**~1~= {26}. Further, because *a*~15,26~= *a*~20,26~= 1, these sets are replaced by **Z**~0~= {26}, **Z**~1~= *φ*. So, we can find that the set **Z**= {14, 15, 20, 26} is \"connected\" and can be a member of *Z*~2~. If we take a subset **Z**= {14, 15, 20, 5}, we can find **Z**is \"disconnected\" because *a*~14,5~= *a*~15,5~= *a*~20,5~= 0, **Z**~0~= *φ*and **Z**~1~= {5} at the final stage. Authors\' contributions ======================= TT proposed the flexibly shaped spatial scan statistic and the bivariate power distribution. KT considered the algorithm given in the appendix, programmed the C++ code and carried out the power simulations. TT wrote the first draft of the manuscript. Both authors interpreted the results and wrote the final version of the paper. Acknowledgements ================ This research was supported in part by Grant-in-Aid for Scientific Research (Grant No. 16300091) from the Ministry of Education, Culture, Sports, Science and Technology of Japan.

PubMed Central

2024-06-05T03:55:58.463681

2005-5-18

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173134/", "journal": "Int J Health Geogr. 2005 May 18; 4:11", "authors": [ { "first": "Toshiro", "last": "Tango" }, { "first": "Kunihiko", "last": "Takahashi" } ] }

PMC1173135

Introduction ============ One of the earliest observations in the study of circadian rhythms was that continuous light (LL) lengthens circadian period in most nocturnal animal species \[[@B1]\]. \"Aschoff\'s Rule\" posits that there is a positive log-linear relationship between the LL intensity and period \[[@B2]-[@B5]\]. In all these studies LL was white light, in one study full-spectrum light \[[@B4]\]. However, we found that mice had slightly longer circadian periods when the monitoring device was a passive infrared (ir) proximity sensor compared to a system using ir beams that crossed the cage. The only obvious difference between the systems was that the proximity sensor had a small, red light-emitting diode (LED) that came on immediately after discerned motion and stayed on for several seconds after motion was not discernable. The first question to be raised is whether a dim red LED can affect the circadian system of mice. The circadian rhythm of locomotor activity in rats is entrained by red light \[[@B6]\]. However, several studies which examined the spectral sensitivity of the photoreceptors in mice suggest little or no response to red light. The peak sensitivity of the photoreceptors that mediate phase shifts in pigmented inbred mouse strains is between 500 nm \[[@B7]\] and 511 nm \[[@B8]\] (blue-green light). The sensitivity of the photoreceptors drop sharply and is vanishingly small at wavelengths above 600 nm (orange light) \[[@B7]-[@B9]\]. In mice lacking rods and cones, the peak sensitivity for phase-shifting is 481 nm, and sensitivity drops to zero at less than 600 nm \[[@B9]\]. In pigmented mice, electroretinographic responses to a flickering monochromatic light and the behavioral responses to a forced-choice discrimination task peak at 510 nm \[[@B10]\]. The light sensitivity in both tests drops sharply as the wavelength approaches 600 nm. Melanopsin, in combination with the classical rod and cone photoreceptors, accounts for the transduction of photic information to the circadian system. We are unaware of studies of the Aschoff effect in rodless and coneless mice, but melanopsin knockout mice have an attenuated Aschoff effect compared to wild-type mice \[[@B11],[@B12]\]. The second question is whether light presented only in response to activity can lengthen period in mice. Pittendrigh and Daan suggested that light pulses during the photosensitive portion of an animal\'s circadian cycle mimic the effect of LL on period \[[@B13]\]. Ferraro and McCormack (1986) confirmed this in rats using feedback lighting (LDfb) \[[@B14]\]. In their LDfb apparatus, the lighting in each rodent\'s cage was controlled by each rodent\'s own locomotor activity. When wheel revolutions reached a certain rate, the cage lighting came on. The lights went out 2 minutes after wheel-running tapered off below the target rate. They compared LL to LDfb at 0.1, 1 and 100 lux of light. They found that circadian period under LDfb obeyed Aschoff\'s rule, and feedback lighting increased circadian period by the same amount as an equivalent irradiance of continuous light. In earlier studies we showed that C57BL/6 and DBA/2 mice differed in their Aschoff affect comparing constant dark to constant full-spectrum LL. At 10 lux, the C57BL/6 mice had an increase in period of 1.20 hours but the period of the DBA/2 mice increased only 0.20 hours \[[@B4]\]. Consequently we predicted that the C57BL/6 mice would have a greater increase in period under the LD feedback regimen than DBA/2 mice. This study tests the hypotheses that a dim red LED provided as feedback to activity elicits an increase in circadian period of locomotor activity and that C57BL/6 and DBA/2 mice have a differential response to the red light stimulus. Methods ======= General housing and care ------------------------ Mice were housed singly in optically clear polycarbonate cages (L × H × W: 11 × 8 × 7 in) with approximately 250 ml of Sani-chip^®^(Harlan Teklad) bedding. They were acclimated under alternating 200 lux light and dark of 12 hours each (LD 12:12) for at least two weeks prior to the start of the study. Food (Teklad 7001 Mouse & Rat Diet 4%) and water were continuously available throughout the study. All animals were maintained in facilities fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care. All research protocols and animal care were approved by the Institutional Animal Care and Use Committee in accordance with the guidelines of the *Guide for the Care and Use of Laboratory Animals*(Institute of Laboratory Animal Resources, Commission on Life Sciences, National Research Council, 1996). Experimental housing and care ----------------------------- For measurement of circadian period, all test mice were kept in a sound attenuating, ventilated room at a constant temperature (23°C) and under continuous darkness (DD). Sound attenuating, opaque dividers were placed between the test cages. Caretakers wore a Pelican Versabrite headlamp fitted with a red safelight beam diffuser. The diffuser/filter transmitted light greater than 600 nm only. Care in the darkroom consisted of ten min per day and each mouse was inspected for less than a minute. Daily visits occurred at random times between 8 am and 5 pm. Locomotor activity assessment ----------------------------- Daily locomotor activity of the mice was monitored with passive infrared detectors (Ademco, Syosset, NY) mounted over each cage. The passive infrared (ir) proximity sensor works by emitting pulses of ir light, and then measuring the distance to objects from the flight time of the reflected signal. Whenever the distance changes, the detector opens or closes a switch. All detectors were tested to ensure response uniformity. Each detector had a red LED that switched on when motion was detected, then switched off 3 to 5 seconds after movement was no longer sensed. The LED could be disabled with a switch mounted on the motion sensor circuit board. Spectral analysis ----------------- The irradiance spectra of the LEDs were determined using an S.I. Photonics fiber optic spectrophotometer (Tucson, AZ). The distance between the fiber optic input probe and the LED was 18 cm (the depth of the mouse cage). A random sample of eight of the 16 motion sensors used in the study was assessed. A hand-held light meter (UDT Instruments, Baltimore, MD) was used to measure illuminance produced by the LED at about 5 cm from the bottom of the cage and 13 cm from the LED. The test was repeated with the LEDs covered with black electrician\'s tape. Assessment of circadian period of locomotor activity ---------------------------------------------------- Activity events were grouped into 5-minute bins by the Stanford Chronobiology Systems\' (Stanford, CA) data processing and automatic storage system integrated into a Dell computer. Clocklab, the biological rhythm analysis software (Actimetrics, Evansville, IL), was used to calculate the period of locomotor activity of each mouse using linear regression through activity onsets of the last 8 to 10 days of each treatment. Experiment 1: Hypothesis -- LEDfb changes circadian period compared to DD ------------------------------------------------------------------------- ### Mouse husbandry C57BL/6 mice were bred in our facility from mice purchased from Jackson Laboratory (Bar Harbor, ME). Two male and six female C57BL/6 mice between the ages of 185 to 295 d were studied. ### Experimental protocol Four mice were put under motion sensors with the LED enabled (LEDfb), and four were put under sensors with the LED disabled (DD). The locomotor activity of the mice was monitored for two weeks (Stage 1). Then treatment was switched, and activity of the mice was monitored for another two weeks (Stage 2). The circadian period for each stage was calculated from the last 8 to 10 days under a given condition. ### Statistical analysis A factorial ANOVA (SAS ver. 9.1) tested for effect of sex, age, lighting condition (LEDfb compared to DD) and sequence of light treatment (LEDfb first compared to DD first). A post-hoc Tukey\'s Studentized Range test compared periods under different lighting protocols. Experiment 2: Hypothesis -- The effect of LEDfb on circadian period can be eliminated by blocking the light source ------------------------------------------------------------------------------------------------------------------ ### Mouse husbandry Twelve male C57BL/6 mice aged 30 d were purchased from Jackson Laboratory and housed singly. ### Experimental protocol Following acclimation to our facility in LD 12:12 for two weeks, they were moved into DD. All mice were put under motion sensors with the LED enabled (LEDfb), but black electrician\'s tape covered the LEDs of six motion sensors (taped LED) for two weeks. For Stage 2, all the LEDs were turned off for two weeks of DD. For Stage 3, the treatment condition of Stage 1 was switched, i.e. all LEDs were enabled but the LEDs previously covered by tape were uncovered and the previously uncovered LEDs were covered. Mouse activity was monitored continuously throughout the experiment. ### Statistical analysis A one-way ANOVA tested for effect of lighting condition (DD compared to LEDfb and taped LED), with a post-hoc Tukey\'s Studentized Range test. Experiment 3: Hypothesis -- C57BL/6 and DBA/2 mice have different response to LEDfb ----------------------------------------------------------------------------------- ### Mouse husbandry Eight male C57BL/6 mice and eight male DBA/2 mice were purchased from Jackson Laboratory, age 4 weeks. ### Experimental protocol Following acclimation to our facility in LD 12:12 for two weeks, mice were moved into DD. Four mice of each strain (DBA/2 and C57BL/6) were put under motion sensors with the LED enabled (LEDfb), and four of each strain were put under sensors with the light disabled (DD). At the end of Stage 1, they were moved to LD 12:12 for one week. When mice returned to the assessment room under DD, the treatment condition was switched. Activities of all mice were monitored for another two weeks. ### Statistical analysis A two-factor ANOVA tested for effect of strain and lighting condition (LEDfb compared to DD) with a post-hoc Tukey\'s Studentized Range test. The change in period from LEDfb compared to DD (Δτ) was calculated for each mouse and compared by a *t*-test. Results ======= Light measurements ------------------ The irradiance spectrum of the LED for a sample of 8 proximity sensors was a narrow band centered on 652 nm. Figure [1](#F1){ref-type="fig"} shows a representative spectrum. The illuminance of the LED was one lux. The illuminance of the LED covered with black electrician\'s tape was zero. ::: {#F1 .fig} Figure 1 ::: {.caption} ###### The irradiance spectrum of a red LED integrated into the passive-infrared motion sensor circuitry. ::: ![](1740-3391-3-8-1) ::: Experiment 1: LEDfb changes circadian period compared to DD ----------------------------------------------------------- Representative actograms of C57BL/6 mice under DD and dim red LEDfb are shown in Figure [2](#F2){ref-type="fig"}. The mean period in DD was 24.05 ± 0.04 h, and under red light LEDfb it was 24.21 ± 0.04 h. A factorial ANOVA testing for effect of sex, age, lighting condition (LEDfb compared to DD) and sequence of light treatment showed no effect of sex, age or sequence but an effect of lighting condition on period \[F~1,5~= 7.72, p = 0.039\]. A post hoc Tukey\'s test showed longer period with LEDfb compared to DD (p = 0.0095). Figure [3](#F3){ref-type="fig"} shows the effects of LEDfb on the circadian period of locomotor activity. ::: {#F2 .fig} Figure 2 ::: {.caption} ###### **Double-plotted actograms of C57BL/6 mice under DD and dim red LEDfb**. Lighting conditions are shown to the right of each actogram. ::: ![](1740-3391-3-8-2) ::: ::: {#F3 .fig} Figure 3 ::: {.caption} ###### **The circadian period of C57BL/6 mice is longer under dim red LEDfb than DD conditions (p = 0.0095)**. Lines show the mean period for each group. ::: ![](1740-3391-3-8-3) ::: Experiment 2: The effect of LEDfb on circadian period can be eliminated by blocking the light source ---------------------------------------------------------------------------------------------------- Representative actograms of C57BL/6 mice under DD, dim red LEDfb, and LEDs coved with black tape are shown in Figure [4](#F4){ref-type="fig"}. The mean period of 12 mice in DD was 23.96 ± 0.03 h; under taped LEDs, it was 23.93 ± 0.03 h; and, under LEDfb, it was 24.07 ± 0.03 h. There was a significant effect of lighting condition by one-way ANOVA \[F~2,33~= 7.02, p = 0.0029\]. A post-hoc Tukey\'s test showed a longer period under the uncovered LED than under the tape-covered LED (p \< 0.01) or DD (p \< 0.025), as summarized in Figure [5](#F5){ref-type="fig"}. Periods did not differ between DD and tape-covered LED. ::: {#F4 .fig} Figure 4 ::: {.caption} ###### **Double-plotted actograms of C57BL/6 mice under DD, dim red LEDfb, and LEDs covered with black tape**. Lighting conditions are shown to the right of each actogram. Arrows show onset of new lighting conditions. ::: ![](1740-3391-3-8-4) ::: ::: {#F5 .fig} Figure 5 ::: {.caption} ###### **Circadian period of C57BL/6 mice is longer under dim red LEDfb than when the light source is covered by black tape (p \< 0.01) or DD (p \< 0.025)**. Lines show the mean period for each group. ::: ![](1740-3391-3-8-5) ::: Experiment 3: C57BL/6 and DBA/2 mice do not have different responses to LEDfb ----------------------------------------------------------------------------- Representative actograms of DBA/2 and C57BL/6 mice under DD and dim red LEDfb are shown in Figure [6](#F6){ref-type="fig"}. For C57BL/6 mice, the mean period under DD was 23.85 ± 0.07 h; the mean period under LEDfb was 24.00 ± 0.07 h. For DBA/2 mice, the mean period under DD was 23.46 ± 0.14 h; the mean period under LEDfb was 23.78 ± 0.08 h. A two-factor ANOVA testing for effect of strain and lighting condition (LEDfb compared to DD) showed a significant effect of both strain \[F~1,14~= 7.73, p = 0.0147\] and lighting condition \[F~1,14~= 10.99, p = 0.0051\] but no interaction. A post-hoc Tukey\'s test showed longer period with LEDfb compared to DD (p \< 0.025). The C57BL/6 mice had different periods from the DBA/2 mice by post-hoc Tukey\'s test (p \< 0.01). Figure [7](#F7){ref-type="fig"} shows the effects of LEDfb on the circadian period of locomotor activity in the two strains of mice. ::: {#F6 .fig} Figure 6 ::: {.caption} ###### **Double plotted actograms of DBA/2 and C57BL/6 mice under DD (top) and dim red LEDfb (bottom)**. The actograms at top and bottom are from one DBA/2 mouse (left) and one C57BL/6 mouse (right). Lighting conditions were separated by two weeks of LD 12:12. ::: ![](1740-3391-3-8-6) ::: ::: {#F7 .fig} Figure 7 ::: {.caption} ###### **The circadian period of both DBA/2 and C57BL/6 mice under DD (filled symbols) and dim red LEDfb (open symbols)**. Lines show the mean period for each group. Overall, mice had longer period under LEDfb than DD (p \< 0.025), and C57BL/6 mice had longer periods than DBA/2 mice (p \< 0.01). ::: ![](1740-3391-3-8-7) ::: The mean increase in period with LEDfb (period under LEDfb minus period under DD, Δτ) for C57BL/6 mice was 0.15 ± 0.05 h. For DBA/2 mice, Δτ was 0.32 ± 0.13 h. The increase in period with LEDfb did not differ between strains by *t*-test (p = 0.26). In summary, circadian period was significantly longer under LEDfb (a small, red LED whose intensity was about 1 lux which came on only when a mouse was active) than that under DD in both C57BL/6 and DBA/2 strains of mice. The LEDs gave off red light in a narrow band centered on 652 nm. Covering the LED with black tape blocked the effect of the dim red light. Furthermore, there was no difference in this effect between the two strains. Discussion ========== This study suggests that the circadian system in mice is responsive to long wavelength red light. The result is surprising because recent studies suggest that melanopsin, in combination with the classical rod and cone photoreceptors, account for the transduction of photic information to the circadian system. There are no studies of spectral sensitivity of the Aschoff effect in mice. However, the spectral sensitivity of photoreceptors mediating circadian phase-shifts in mice is vanishingly small at wavelengths above 600 nm \[[@B7]-[@B9]\]. In most organisms, circadian period under LL is a function of both intrinsic period and photic inputs. The Aschoff effect is understood to result from the cumulative phase-shifting effect of LL on the pacemaker \[[@B5],[@B13],[@B15]\]. Thus an effect of red light on circadian period is unexpected. One possible explanation is that there is another photopigment present in mammals that is sensitive to far-red light and affects period rather than phase. It cannot be excluded that period and phase are affected by different light receptors or light receptive pathways. It seems more likely that low sensitivity to red light via the known circadian photopigments has a cumulative period-lengthening effect on the pacemaker. The timing of light exposure could have amplified this effect. Under LEDfb mice received light between circadian time (CT 12) and CT 24, during their active phase. The period-response curves (τRC) of mice have period-shortening between CT 4 and CT 16 and period-lengthening between CT 16 and CT 4 \[[@B16]\]. LEDfb should cause substantial period-lengthening, with minimal period-shortening. This study has several limitations. Only the Aschoff effect was investigated, and this was not under the usual protocol of constant light. Nevertheless, the increase in circadian period of mice under LEDfb was consistent with a prior activity feedback study in rats, where wheel-running triggered full-spectrum illumination resulting in lengthened circadian period \[[@B14]\]. Covering the LED with black tape blocked the effect of the dim red light. We conclude that ultrasound from the LEDs or the electronics associated with their illumination did not produce the effect. Mice emit ultrasonic cries and this is important in maternal behavior \[[@B17]-[@B19]\]. However, it is unlikely that covering the lights with black tape would block ultrasound. Although this remains a possibility, we are aware of no instances in the literature where ultrasound causes either a phase response or a change in circadian period. The C57BL/6 and DBA/2 mice did not differ in the amount period increased with 1 lux red light LEDfb. This is in contrast to an earlier study where C57BL/6 mice had a greater increase in period with 10 lux full-spectrum LL vs DD than DBA/2 mice \[[@B4]\]. One possible explanation is that DBA/2 mice are more sensitive to period-lengthening effects of red light than of full-spectrum light. Another possibility is that the shape of the τRC differs between the two strains, so that for C57BL/6 mice more of the period-lengthening portion of the τRC lies outside of the time when they received light during LEDfb than for the DBA/2 mice. In this case, constant LL would have more period-lengthening effect than LEDfb. The results of this study suggest that investigators cannot use continuous dim red light to simulate DD, and must be judicious in using red \"safe\" lights for animal care in DD. Further studies are needed to determine whether constant red light of this output spectra and intensity lengthens period more than LEDfb, and whether it can phase-shift the circadian rhythms in mice. Conclusion ========== Mice under a dim, long-wavelength red light that came on intermittently when the animals were active had a circadian period that was long compared to their free-running period under DD. Covering the light with black tape blocked the response. Furthermore, under the conditions described, the magnitude of the mean circadian period increase in DBA/2 and C57BL/6 strains of mice was indistinguishable. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= JRH conceived of the study, participated in its design and statistical analysis and drafted the manuscript. ARH carried out experiment 1, designed and carried out experiment 3, and wrote the Methods section. AMH participated in the statistical analysis and designed and set up experiment 2. ARM carried out experiment 2, participated in the statistical analysis and assisted in drafting the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by a Merit Grant to Dr. Mayeda from the Department of Veterans Affairs.

PubMed Central

2024-06-05T03:55:58.472002

2005-5-31

{ "license": "Creative Commons - Attribution - https://creativecommons.org/licenses/by/4.0/", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1173135/", "journal": "J Circadian Rhythms. 2005 May 31; 3:8", "authors": [ { "first": "John R", "last": "Hofstetter" }, { "first": "Amelia R", "last": "Hofstetter" }, { "first": "Amanda M", "last": "Hughes" }, { "first": "Aimee R", "last": "Mayeda" } ] }